Integrative Zoology 2018 Symposium Plant Animal Interaction

Volume 13 Issue 3 May 2018 The official journal of the International Society of Zoological Sciences
Integrative Zoology
Contents
Editorial 280 Eastern gray squirrels are consistent
shoppers of seed traits: insights from
225 Plant-animal interactions: patterns and discrete choice experiments
mechanisms in terrestrial ecosystems Mekala SUNDARAM, Nathanael I. LICHTI,
Michael A. STEELE, Xianfeng YI and Nicole J. Olynk WIDMAR and Robert K.
Hongmao ZHANG SWIHART
Original Articles 297 Factors shaping life history traits of two

proovigenic parasitoids
Volume 13 Issue 3 May 2018

228 Invasive oaks escape pre-dispersal insect Michal SEGOLI, Shucun SUN, Dori E. NAVA
seed predation and trap enemies in their and Jay A. ROSENHEIM
seeds
Michał BOGDZIEWICZ, Raul BONAL, Josep 307 Distribution and space use of seed-
M. ESPELTA, Ewa M. KALEMBA, Michael A. dispersing rodents in central Pyrenees:
STEELE and Rafał ZWOLAK implications for genetic diversity,
conservation and plant recruitment
238 Seed loss before seed predation: Jon URGOITI, Alberto MUÑOZ, Josep Maria
experimental evidence of the negative ESPELTA and Raúl BONAL
effects of leaf feeding insects on acorn
production 319 Environmental variation shifts the
Tara CANELO, Álvaro GAYTÁN, Guillermo relationship between trees and
GONZÁLEZ-BORNAY and Raul BONAL scatterhoarders along the continuum
from mutualism to antagonism
251 Acorn size and tolerance to seed Gina M. SAWAYA, Adam S. GOLDBERG,
PAGES 224-347
predators: the multiple roles of acorns as Michael A. STEELE and Harmony J.
food for seed predators, fruit for dispersal DALGLEISH
and fuel for growth
Andrew W. BARTLOW, Salvatore J. 331 Effects of rodent abundance and richness
AGOSTA, Rachel CURTIS, Xianfeng YI and on cache pilfering
Michael A. STEELE Jacob W. DITTEL and Stephen B. VANDER
WALL
267 Role of seed size, phenology, oogenesis
and host distribution in the specificity and 339 Phenotypic trait matching predicts
genetic structure of seed weevils (Curculio the topology of an insular plant–bird
spp.) in mixed forests pollination network
Harold ARIAS-LECLAIRE, Raúl BONAL, Matthew BIDDICK and Kevin C. BURNS
Daniel GARCÍA-LÓPEZ and Josep Maria Volume 13 Issue 3 May 2018
ESPELTA
Special issue: Plant–Animal Interactions
CN11-6012/Q
For full journal content,

http://wileyonlinelibrary.com/journal/inz
Editor-in-Chief
Zhibin ZHANG, Institute of Zoology, Chinese Academy of
Sciences (CAS), China
Honorary Editor-in-Chief
John BUCKERIDGE, RMIT University, Australia
Senior Editor
Marcel Holyoak, University of California, Davis, USA
Editorial Advisory Board

Chair: Abraham HAIM, Jean-Marc JALLON, Nils Chr. STENSETH,
Yiyu CHEN, Natural Science University of Haifa, University of Paris XI, University of Oslo,
Foundation of China, China Israel France Norway
Members: Jürgen HEINZE, Rosa-Maria POLYMENI, Yan XIE,Institute
Edwin L. COOPER, University of Regensburg, University of Athens, of Zoology, CAS,
University of California Germany Greece China
Los Angeles, USA
Editorial Board
Nico AVENANT, National Australia University of Rome ‘La Guiming WANG, Mississippi
Museum, South Africa Martin JANSEN, Senckenberg Sapienza’, Italy State University, USA
Harold W. AVERY, Research Institute, Germany David RAUBENHEIMER, Massey Zuoxin WANG, Florida State
Drexel University, USA Patrick A. JANSEN, Smithsonian University, New Zealand University, USA
Thad BARTLETT, University Tropical Research Institute, Alexander RIEDEL, Staatliches Fuwen WEI, Institute of Zoology,
of Texas at San Antonio, Panama Museum für Naturkunde, CAS, China
USA Xiang JI, Nanjing Normal Germany Carsten WOLFF, Humboldt
Luca BORGER, University of University, China Alain ROQUES, Institut National University, Germany
Swansea, UK David Martin LAMBERT, Griffith de la Recherche Agronomique, Xing XU, Institute of Vertebrate
Michael BRUFORD, University, Australia France Paleontology and
Cardiff University, UK Fumin LEI, Institute of Zoology, Ken SASAKI, Kanazawa Institute Paleoanthropology, CAS,
Ahimsa CAMPOS-ARCEIZ, CAS, China of Technology, Japan China
University of Nottingham, UK Philip J. LESTER, Victoria Kazunobu SAWAMOTO, Nagoya Guang YANG, Nanjing
Zanna CLAY, Université de University of Wellington, City University, Japan Normal University, China
Neuchatel, Switzerland New Zealand Jiakun SONG, University of Dexing ZHANG, Institute of
Jiayin DAI, Institute of Zoology, Baoguo LI, Northwest University, Maryland, USA Zoology, CAS, China
CAS, China China James R. SPOTILA, Drexel Jianxu ZHANG, Institute of
Sasha DALL, University of Ming LI, Institute of Zoology, University, USA Zoology, CAS, China
Exeter - Penryn Campus, UK CAS, China Michael STEELE, Wilkes Univer- Yaping ZHANG, Kunming
Yury DGEBUADZE, A. N. Shuqiang LI, Institute of Zoology, sity, USA Institute of Zoology, CAS,
Severtsov Institute, Russian CAS, China Jianghua SUN, Institute of China
Academy of Sciences, Yiming LI, Institute of Zoology, Zoology, CAS, China
Russia CAS, China Jan van der MADE, Museo Executive Editor
Weiguo DU, Institute of Zoology, Burton K. LIM, Royal Ontario Nacional de Ciencas Wenhua XIONG,
CAS, China Museum, Canada Naturales, Spain Institute of Zoology,
Constance DUBUC, University Shujin LUO, Peking University, Maxim V. VINARSKI, CAS, China
of Cambridge, UK China Omsk State Pedadogical
Steffen FOERSTER, Duke Sarita MAREE, University of University, Russia Managing Editors
University, USA Pretoria, South Africa Marvalee H. WAKE, University Wei ZHANG, International
Hongxuan HE, Institute of Jeffrey A. McNEELY, IUCN, of California, Berkeley, USA Society of Zoological
Zoology, CAS, China Switzerland Dehua WANG, Institute of Sciences (ISZS)
Lyn HINDS, CSIRO, Australia Yoshitaka NAGAHAMA, Zoology, CAS, China Jacob WICKHAM,
Ary Anthony HOFFMANN, Ehime University, Japan Ding WANG, Institute of International Society of
University of Melbourne, Maria Rita PALOMBO, Hydrobiology, CAS, China Zoological Sciences (ISZS)
Disclaimer: The Publisher, International Society of Zoological Sciences and Institute of Zoology/Chinese Academy of Sciences and Editors cannot be
held responsible for errors or any consequences arising from the use of information contained in this journal; the views and opinions expressed do not
necessarily reflect those of the Publisher, International Society of Zoological Sciences and Institute of Zoology/Chinese Academy of Sciences and Edi-
tors, neither does the publication of advertisements constitute any endorsement by the Publisher, International Society of Zoological Sciences and Insti-
tute of Zoology/Chinese Academy of Sciences and Editors of the products advertised.
Editorial Correspondence: Editorial Office, Integrative Zoology, Institute of Zoology, No. 1 Beichen West Road, Beijing 100101, China
T/F: 86-010-64807295. Email: inz@ioz.ac.cn
©2018 International Society of Zoological Sciences, Institute of Zoology/Chinese Academy of Sciences and John Wiley & Sons Australia, Ltd
For submission instructions, subscription and all other information visit: wileyonlinelibrary.com/journal/inz
COPYRIGHT AND COPYING

Copyright © 2018 International Society of Zoological Sciences, Institute
PUBLISHER of Zoology/Chinese Academy of Sciences and John Wiley & Sons Australia,
Ltd All rights reserved. No part of this publication may be repro -
Integrative Zoology is published by duced, stored or transmitted in any form or by any means without the
John Wiley & Sons Australia, Ltd prior permission in writing from the copyright holder. Authorization to
42 McDougall Street, Milton, copy items for internal and personal use is granted by the copyright
4064 Australia (PO Box 1226, holder for libraries and other users registered with their local Reproduc-
tion Rights Organisation (RRO), e.g. Copyright Clearance Center (CCC),
Milton, 4064 Australia) 222 Rosewood Drive, Danvers, MA 01923, USA (www.copyright.com),
Tel: +61 (0)7 3859 9755 provided the appropriate fee is paid directly to the RRO. This consent
Production Editor does not extend to other kinds of copying such as copying for general
distribution, for advertising or promotional purposes, for republication, for
Maricris Apostol (Email: INZ@wiley.com) creating new collective works or for resale. Permissions for such reuse can be
obtained using the RightsLink “Request Permissions” link on Wiley Online Library.
Special requests should be addressed to: permissions@wiley.com
Integrative Zoology 2018; 13: 225–227 doi: 10.1111/1749-4877.12320
1 EDITORIAL 1
2 2
3 3
4 4
5 5
6 6
7 Plant-animal interactions: patterns and mechanisms in 7
8 8
9
10
terrestrial ecosystems 9
10
11 11
12 12
13 13
14 14
15 From the early work of Darwin, then Ehrlich and Ra- tions dealt either directly or indirectly with the intrica- 15
16 ven (1964), to the recent novel contribution of mutu- cies of animal-mediated dispersal of oak and/or acorn 16
17 alistic networks developed by Jordano and colleagues predation in oak ecosystems from Asia to North Ameri- 17
18 (Bascompte and Jordano 2007, Nuismer et al. 2013, ca. An eighth paper dealt with the community-wide pro- 18
19 Guimarães et al. 2017), it is clear that plant-animal in- cesses of seed predation and dispersal in another system 19
20 teractions—especially those involving herbivory, seed and the remaining two papers focused on avian polli- 20
21 dispersal, seed predation and pollination—are signifi- nation in New Zealand and the life history strategies of 21
22 cant drivers of ecological processes and the evolution two proovigenic parasitoids of a phytophagous insect in 22
23 and maintenance of biodiversity. The vast majority of North America. 23
24 plant species, for example, are influenced by animal-me- Bartlow et al. (2018) experimentally demonstrat-
24
25 diated seed dispersal and pollination. Hence, under- ed that varying degrees of cotyledon damage in oaks, as
25
26 standing the complexities of these interactions and their often results from partial-predation by a diverse group
26
27 broader geographic patterns is central to better decipher- of insects and vertebrates, have minimal impact on oak
27
28 ing complex ecosystem processes and the potential fate seedling establishment. The authors argue that the acorn
28
29 of biodiversity in the advent of climate change. For this serves as much as a fruit for dispersal and tolerance of
29
30 symposium, we assembled a group of leading authori- partial seed predation as it does a source of energy for
30
31 ties on plant-animal mutualisms and related interactions the developing seedling. Working in a similar system,
31
32 in an effort to highlight new findings, demonstrate sub- Sundaram et al. (2018) introduced a novel approach for
32
33 tle but significant interactions and to further identify and evaluating seed choices by a scatter-hoarding rodent
33
34 focus the direction of this important field of ecological (Eastern gray squirrel; Sciurus carolinensis) when se-
34
35 research. Based on review of approximately 700 papers lecting seeds and nuts for both feeding and caching. Us-
35
36 published in the past few years and our knowledge of ing a multivariate quantitative method, the investigators
36
37 research in this area, we identified invitees from around successfully quantified the simultaneous effects of nu-
37
38 the globe to join us for a colloquium on plant-animal in- merous seed traits (e.g., caloric value, protein content,
38
39 teractions at the 9th International Symposium of Inte- tannin and dormancy period) on behavioral decisions.
39
40 grative Zoology held on August, 27-31, 2017 in Xining, This novel approach to quantifying decision making in
40
41 Qinghai Province, China. these animals has the potential to revolutionize foraging
41
42 42
We selected authors from a broad geographic range studies, historically based exclusively on energetic costs
43 43
who could cover a diversity of topics. The final col- and benefits.
44 44
loquium, represented by contributors from four conti- In one of the first studies of its kind, Dittel and
45 45
nents, included ten papers that are briefly highlighted Vander Wall (2018) investigated the effects of both ro-
46 46
here. Although not our original intent, this volume offers dent abundance and rodent community richness on pil-
47 47
a heavy focus on oak-animal interactions, due in part to ferage rates of artificial scatterhoards of native seed
48 48
the organizers’ research interests but also to the current species and sunflower seeds in seven habitat types dis-
49 49
intensive research on oak (Quercus)–animal interac- tributed across an elevational gradient in western Ne-
50 50
tions across the globe. As such, seven of these contribu- vada, USA. They find that pilferage rates were most
51 51
© 2018 International Society of Zoological Sciences, Institute of Zoology/ 225

Chinese Academy of Sciences and John Wiley & Sons Australia, Ltd
Editorial
1 directly related to the abundance of scatter-hoarding ro- that matures early and specializes on hazelnuts, exhib- 1
2 dents but not the rodent community richness, leading its recent genetic bottlenecks, as evidenced by rare lo- 2
3 the authors to conclude that in these systems “densi- cal haplotypes likely to follow from the discontinuous 3
4 ty-dependent competition” for cached seeds is a primary stands of hazel trees in the region. 4
5 determinant of pilferage rates. We then moved from the seed dispersal to pollination 5
6 In a geographically extensive, 3-year field experi- and to New Zealand, where Biddick and Burns (2018) 6
7 ment in the Central Hardwoods region of North Ameri- explored the structure of a plant-bird pollination net- 7
8 ca, Sawaya et al. (2018) used a published model on con- work. They demonstrated how the network interactions 8
9 ditional mutualism to predict and test the strength of this are largely defined by flower size and bill morpholo- 9
10 relationship between scatterhoarders and four species of gy and that the network is largely nested. They also de- 10
11 hardwood trees. They demonstrated how both seed traits scribed phenotypic trait matching is a reliable predictor 11
12 and environmental conditions interact to influence seed- of pollination networks. Finally, we capped the series 12
13 ling establishment and the strength of these relation- with life history studies of two species of proovigenic 13
14 ships. They found that the strength of this scatterhoard- parasitoid wasps (Anagrus spp.) that complete their en- 14
15 er-tree mutualism is strongest for red oaks (Q. rubra), tire development in the eggs of leafhoppers (Erythro- 15
16 weakest for white oaks (Q. alba) and intermediate for neura spp. and Erasmoneura spp.) and emerge with a 16
17 chestnut oak (Castanea dentata) and hybrid chestnut (C. full lifetime supply of mature eggs (Segoli et al. 2018). 17
18 dentata x Castanea mollissima). Canelo et al. (2018) The authors sought to compare life history strategies of 18
19 employ a careful experimental approach to quantify the the parasitoids in two neighboring habitats: vineyards 19
20 effects of herbivory by a leaf-feeding insect on the Med- and riparian habitats where leafhopper hosts were abun- 20
21 iterranean Holm oak’s (Q. ilex) physiological responses dant and rare, respectively and where parasitoids engage 21
22 in individual shoots that ultimately results in a drop in in different foraging strategies. Results suggest a more 22
23 acorn production. Unlike other correlational studies, this complex mechanism of energy allocation than predicted 23
24 study demonstrated the causal physiological links be- and one that appears independent of host size. 24
25 tween herbivory and reproductive loss. 25
Although only a snapshot of some of the underlying
26 26
Bogdziewicz et al. (2018), working in oak forests of ecological and evolutionary complexities of plant-an-
27 27
Europe, demonstrated how the invasive red oak (Q. ru- imal interactions in a few systems, we believe these
28 28
bra) is nearly free of insect seed predators compared studies provide new perspectives and some novel ap-
29 29
with a common native oak (Q. petrea) in Europe and, as proaches, and reveal complexities in these systems that
30 30
a result, afforded significant dispersal and regeneration have not previously been considered. We hope the read-
31 31
advantages over the native species. Urgoiti et al. (2018) er agrees and is inspired to pursue research that follows
32 32
explored the distribution of small-mammal communi- from these studies.
33 33
ties and genetic isolation of two keystone seed dispers-
34 34
ers in the Pyrenees Mountains of Europe. Their results
35 35
hold important implications for how geographic varia-
36 Michael A. STEELE 36
tion in seed consumers influences native plant dispersal
37 37
and how global warming will likely influence isolation Department of Biology, Wilkes University, USA
38 38
of these keystone seed dispersers.
39 39
40 Working in NE Spain, Arias-Leclaire et al. (2018) ex- Xianfeng YI 40
41 amined how acorn phenology and sexual maturation of 41
female seed weevils of four species (Curculio), as well College of Life Sciences, Jiangxi Normal University,
42 China 42
43 as the trophic breadth of these seed predators and the 43
44 distribution of their host trees (hazel [Corylus avelana] 44
45 and two oak species [Q. ilex and Q. pubescens]), influ- Hongmao ZHANG 45
46 enced the genetic structure of weevil populations. They 46
School of Life Sciences, Central China Normal Uni-
47 find that one weevil species (C. nucum), in particular, 47
versity, China
48 48
49 49
50 50
51 51
226 © 2018 International Society of Zoological Sciences, Institute of Zoology/

Trait matching and network structure
1 REFERENCES Dittel JW, Vander Wall SB (2018). Effects of rodent 1

2 abundance and richness on cache pilfering. Integra- 2
3 Arias-Leclaire H, Bonal R, García-López D, and Espel- tive Zoology 13, 331-8. 3
4 ta JM. (2018). Role of seed size, phenology, oogene- 4
Erhlich PR, Raven PH (1964). Butterflies and plants: a
5 sis and host distribution in the specificity and genet- study in coevolution. Evolution 18, 586-608. 5
6 ic structure of seed weevils (Curculio spp.) in mixed 6
Guimarães Jr. PR, Pires MM, Jordano P, Bascompte J,
7 forests. Integrative Zoology 13, 267-79. Thompson JN (2017). Indirect effects drive coevolu-
7
8 Bartlow AW, Agosta SJ, Curtis R, Yi X, Steele MA 8
tion in mutualistic networks. Nature 550, 511-40.
9 9
(2018). Acorn size and tolerance to seed predators: Nuismer SL, Jordano P, Bascompte (2013). Coevolution
10 10
the multiple roles of acorns as food for seed preda- and the architecture of mutualistic networks. Evolu-
11 11
tors, fruit for dispersal and fuel for growth. Integra- tion 67-2, 338-54.
12 12
13
tive Zoology 13, 251-66. Sawaya GM, Goldberg AS, Steele MA, Dalgleish, HJ 13
14 Bascompte J, Jordano P (2007) Plant-animal mutualistic (2018). Environmental variation shifts the relation- 14
15 networks; the architecture of biodiversity. Annual Re- ship between trees and scatterhoarders along the con- 15
16 view of Ecology and Systematics 38, 567-93. tinuum from mutualism to antagonism. Integrative 16
17 Zoology 13, 319-30. 17
Biddick M, Burns, KC (2018). Phenotypic trait match-
18 ing predicts the topology of an insular plant–bird pol- Segoli, M, Shucun S, Nava, DE, Rosenheim, JA (2018). 18
19
lination network. Integrative Zoology 13, 339-47. Factors shaping life history traits of two proovigenic 19
20 parasitoids. Integrative Zoology 13, 297-306. 20
21 Bogdziewicz M, Bonal R, Espelta JM, Kalemba EM, 21
Sundaram M, Lichti NI, Olynk NJ, Swihart (2018).
22 Steele MA, Zwolak R (2018). Invasive oaks escape 22
Eastern gray squirrels are consistent shoppers of seed
23 pre-dispersal insect predation and trap enemies in traits: insights from discrete choice experiments. In- 23
24 their seeds. Integrative Zoology 13, 228-37. tegrative Zoology 13, 280-96. 24
25 Canelo T, Gaytán, Á, González-Bornay G, Bonal R 25
Urgoiti J, Muñoz A, Espelta JM, Bonal R (2018). Dis-
26 (2018). Seed loss before seed predation: experimen- 26
tribution and space use of seed-dispersing rodents in
27 tal evidence of the negative effects of leaf feeding in- 27
central Pyrenees: implications for genetic diversity,
28 sects on acorn production. Integrative Zoology 13, 28
conservation and plant recruitment. Integrative Zool-
29 29
238-50. ogy 13, 307-18.
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

1 ORIGINAL ARTICLE 1
2 2
3 3
4 4
5 5
6
7
Invasive oaks escape pre-dispersal insect seed predation and trap 6
7
8
9
enemies in their seeds 8
9
10 10
11 11
12 12
13 Michał BOGDZIEWICZ,1,2 Raul BONAL,3,4 Josep M. ESPELTA,2 Ewa M. KALEMBA,5 Michael 13
14 14
15 A. STEELE6 and Rafał ZWOLAK1 15
16 1
Department of Systematic Zoology, Faculty of Biology, Adam Mickiewicz University, Poznań, Poland, 2CREAF, Cerdanyola del 16
17 Valles, Catalonia, Spain, 3Forest Research Group, INDEHESA, University of Extremadura, Plasencia, Spain, 4DITEG Research 17
18 Group, University of Castilla-La Mancha, Toledo, Spain, 5Laboratory of Seed Biochemistry, Institute of Dendrology, Polish 18
19 19
Academy of Sciences, Kórnik, Poland and 6Department of Biology, Wilkes University, Wilkes-Barre, Pennsylvania, USA
20 20
21 21
22 22
23 23
24 Abstract 24
25 25
26 Species introduced to habitats outside their native range often escape control by their natural enemies. Besides 26
27 competing with native species, an alien species might also affect the native herbivores by introducing a new 27
28 source of different quality food. Here, we describe the case of northern red oak (Quercus rubra) invasion in Eu- 28
29 rope. We collected data on insect (moth Cydia spp. and weevil Curculio spp.) seed predation of northern red 29
30 oak in its native (USA, North America) and invasive (Poland, Europe) range, as well as for sessile oaks (Quer- 30
31 cus petrea) in Europe. We also evaluated the quality of acorns as hosts for weevil larvae by collecting infest- 31
32 ed acorns and measuring weevil developmental success, and quantifying acorn traits such as seed mass, tannins, 32
33 lipids and protein concentration. We used DNA barcoding to identify insects to the species level. The predation 33
34 by moths was similar and very low in both species and in both ranges. However, red oaks escape pre-dispersal 34
35 seed predation by weevils in Europe. Weevil infestation rates of northern red oak acorns in their invasive range 35
36 were 10 times lower than that of sessile oaks, and also 10 times lower than that of red oaks in North America. 36
37 Furthermore, even when weevils oviposited into northern red oaks, the larvae failed to develop, suggesting that 37
38 the exotic host created a trap for the insect. This phenomenon might gradually decrease the local abundance of 38
39 the seed predator, and further aid the invasion. 39
40 40
Key words: biotic invasions, ecological trap, enemy release hypothesis, seed predation, weevils
41 41
42 42
43 43
44 44
45 INTRODUCTION 45
46 46
Correspondence: Michał Bogdziewicz, Department of Species introduced to habitats outside their native
47 47
48 Systematic Zoology, Faculty of Biology, Adam Mickiewicz range (i.e. alien species) frequently escape control by 48
49 University, Michał Bogdziewicz, Umultowska 89, 61-064 their natural enemies (Keane & Crawley 2002; Torchin 49
50 Poznań, Poland. et al. 2003; Liu & Stiling 2006). Capitalization on pre- 50
51 Email: michalbogdziewicz@gmail.com dation release often allows the survival rate, recruitment 51

Invasive oaks escape seed predation
1 and growth to be higher in the introduced than in the na- oaks over other local oaks (Major et al. 2013). Nonethe- 1
2 tive range, which gives invaders an advantage over na- less, whether northern red oaks escape seed predation 2
3 tive competitors (Siemann & Rogers 2001; Wolfe 2002; by weevils in their invasive range is unknown (but see 3
4 Colautti et al. 2004; Muller-Scharer et al. 2004). These Myczko et al. 2017). 4
5 changes cause or contribute to the troublesome success Weevil larvae (Coleoptera: Curculionidae) and moth 5
6 of many alien species. caterpillars (Lepidoptera: Tortricidae) represent the most 6
7 Furthermore, in some situations, natural enemies important groups of oak acorn predators, although wee- 7
8 may not only fail to control the population of an invad- vils are responsible for the majority of seed loss (Craw- 8
9 er but may suffer themselves in the interaction (Keeler ley & Long 1995; Branco et al. 2002; Bonal et al. 2007; 9
10 & Chew 2008; Harvey et al. 2010; Yoon & Read 2016). Espelta et al. 2009). Weevil pre-dispersal predation 10
11 For example, lepidopterans that oviposit on and con- shapes recruitment potential of trees (Crawley & Long 11
12 sume host plants do not show strong preference towards 1995; Espelta et al. 2009; Bogdziewicz et al. 2017), and 12
13 native hosts as compared to exotics, despite lower per- severely destroys crops of northern red oaks in their na- 13
14 formance on the latter (Yoon & Read 2016). In con- tive range (Lombardo & McCarthy 2008; Bogdziewicz 14
15 trast, the presence of non-native prey may increase food et al. 2017). While less is known about moth feeding 15
16 availability, which ultimately translates into higher sur- ecology, weevils are feeding specialists, with particular 16
17 vival rates of the predator (Cattau et al. 2010; Brown species best adapted to exploit particular hosts (Bonal et 17
18 et al. 2016; Cattau et al. 2016). Thus, depending on the al. 2011; Govindan et al. 2012; Muñoz et al. 2014; Peg- 18
19 quality of the exotic prey for the native enemy, it can uero et al. 2017). Concerning predation upon red and 19
20 reduce or boost the predator population (Pearse et al. European white oaks, differences may arise because red 20
21 2013; Brown et al. 2016; Yoon & Read 2016). If the ex- oaks acorns have higher tannin concentration, as well 21
22 otic species harms the populations of native predators as thicker shells than European white oaks (Shimada & 22
23 or herbivores, it can further accelerate the invasion of Saitoh 2006; Steele 2008; but see Łuczaj et al. 2014). 23
24 the exotic species, as the natural enemy is gradually re- In fact, differences in the communities of endosymbiont 24
25 moved from the local habitat (Pearse et al. 2013). Yet bacteria associated with weevils are correlated with the 25
26 there are still few studies that, beyond the existence of use of different host plants (Toju & Fukatsu 2011; Mer- 26
27 “predator release effects,” have provided an evaluation ville et al. 2013). Therefore, because European weevils 27
28 of the effects of the invasive plant on the local enemy have a relatively short history of co-occurrence with red 28
29 (but see e.g. Keeler & Chew 2008; Harvey et al. 2010; oaks, we could expect a lower adaptation to this alien 29
30 Brown et al. 2016). host and, thus, reduced acorn predation rates compared 30
31 to native white oaks. Moreover, European oaks belong 31
In this study, we describe a case of a northern red oak
32 to the white oaks (section Quercus), while northern red 32
(Quercus rubra) invasion in Europe, in which the plant
33 oaks belong to the section Lobatae. This has been ob- 33
not only escapes seed predation but also may harm nat-
34 served in some previous studies to be a factor for the 34
ural enemies by trapping them in its acorns. The north-
35 specialization of acorn weevils (Peguero et al. 2017; but 35
ern red oak was introduced to European forests from
36 see Bonal et al. 2016). Host specialization could, how- 36
North America in the 17th century as an ornamental
37 ever, differ between native predators, thus conditioning 37
species, and was first commercially planted at the end
38 their abilities to exploit alien plant species. 38
of the 18th century (Woziwoda et al. 2014b). Nowa-
39 39
days, it is among the most frequent foreign deciduous In this study, we evaluated insect pre-dispersal pre-
40 40
species in Central European forests (Woziwoda et al. dation on northern red oak in North America and Eu-
41 41
2014b). Northern red oaks are considered invasive and rope as well as on sessile oak (Quercus petrea), which
42 42
their abundance correlates negatively with abundance is the dominant native oak in Central European forests.
43 43
and richness of numerous native species (Chmura 2013; We predicted that weevils will destroy a higher propor-
44 44
Woziwoda et al. 2014a). The invasion will likely further tion of northern red oak acorns in North America than in
45 45
accelerate as northern red oaks have great recruitment Europe, and a higher proportion of sessile than northern
46 46
potential in Central Europe: at some sites, their density red oak acorns in Europe. Based on the acorn quality
47 47
of regeneration exceeds all other tree species combined data from past studies (Shimada & Saitoh 2006; Steele
48 48
(Major et al. 2013; Myczko et al. 2014). Dense seedling 2008), in particular on high tannin concentrations and
49 49
banks beneath the canopies suggest that low seed preda- thick shells of red oaks acorns, we predict that the red
50 50
tion may contribute to the regeneration advantage of red oak is a lower quality host for European weevils than is
51 51

M. Bogdziewicz et al.
1 the sessile oak, and this results in lower predation rates is unlikely (Bogdziewicz et al., personal observation; 1
2 on the former species. To evaluate these predictions, we see also Myczko et al. 2017). The number of acorns col- 2
3 measured weevil prevalence in acorns of northern red lected in the samples was used as an index of individual 3
4 oaks (in Europe and North America) and of sessile oaks. tree acorn production. 4
5 Moreover, we evaluated the quality of acorns as hosts In the laboratory, we weighed, measured (width and 5
6 for weevil larvae by collecting infested acorns and mea- length) and assessed insect infestation for individual 6
7 suring weevil developmental success, and quantifying acorns from each treatment (Poland: northern red oaks = 7
8 acorn traits such as seed mass and shell thickness, and 981 acorns, sessile oaks = 1203; USA: northern red oaks 8
9 tannins, lipids and protein concentration. Finally, we as- = 1219). Previous studies provided evidence for chemi- 9
10 sessed whether the ability to exploit alien host seeds dif- cal gradients in acorns that promote consumption of the 10
11 fered among pre-dispersal insect predators, using DNA basal end of the acorns (i.e. avoidance of the apical end) 11
12 barcoding to identify weevil larvae and moth caterpil- by seed predators (Steele et al. 1993; Bogdziewicz et al. 12
13 lars to the species level. 2017). Therefore, we cut each acorn transversely into a 13
14 top (basal) and bottom half (apical) and counted all in- 14
15 15
MATERIALS AND METHODS sect larvae in each half (following earlier protocols; see
16 Steele et al. 1993). 16
17 Study site 17
Another sample of acorns was collected to estimate
18 18
The study was conducted over the course of 3 years the probability of successful weevil emergence, the total
19 19
in Europe (2014–2016) and North America (2013– number of larvae per infested acorn, and the probability
20 20
2015). In Europe, we conducted research in Gorzowska that the larvae predated an acorn embryo. This sample
21 21
Forest (52°47′1.71″N, 15°3′21.87″E), situated in west- consisted only of acorns with oviposition scars but with-
22 22
ern Poland at an altitude of 60–80 m. Gorzowska For- out emergence holes (i.e. acorns that still had all larvae
23 23
est is located in the temperate climate zone, with aver- inside). We collected the acorns from the soil at 20 hap-
24 24
age annual precipitation of 523 mm and average annual hazardly chosen northern red oaks in USA (in 2015),
25 25
temperature of 8 °C (Zwolak et al. 2016). For the study, and 10 red and 10 sessile oaks in Poland (in 2016) at the
26 26
we selected 2 sites in mixed sessile-northern red oak same sites as described above (Poland: northern red oak
27 27
stands. In USA, 2 sites were established in northern red = 237 acorns, sessile oak = 296; USA: northern red oak
28 28
oak-dominated stands in Harvard Forest (42°31′53.28″N, = 352). In the laboratory, we placed the acorns in sepa-
29 29
72°11′23.78″W), Massachusetts, which is located in the rate open plastic vials. Each day we checked the acorns
30 30
temperate climate zone, with average annual precipi- to register emerged larvae. Once all larvae had emerged
31 31
tation of 1100 mm and average annual temperature of (i.e. no emergence was recorded for 3 weeks), we dis-
32 32
10°C. In both regions, the study sites were spaced over sected the acorns to further check whether the acorn em-
33 33
1 km from each other. bryo had been predated and whether there were non-
34 34
emerged larvae inside (following earlier protocols [e.g.
35 Field sampling and laboratory procedures 35
Bonal & Muñoz 2008; Muñoz et al. 2014]).
36 36
37 Each year, we collected acorns under haphazard- DNA barcoding and larval identification 37
38 ly chosen trees (i.e. first trees that we encountered once 38
entering the stand that shed acorns): 13 red and 12 ses- Larval identification is necessary to establish un-
39 39
sile oaks in Poland, and 30 northern red oaks in USA equivocal trophic relationships between insects and their
40 40
(half of that number of the trees per site, red oaks in host plants, and the lack of morphological keys makes
41 41
Europe 6 and 7 trees per site). In Poland, we collected the use of molecular techniques necessary. For DNA-
42 42
samples from the ground beneath each tree in 0.5 × 0.5- based species identification we selected a random sam-
43 43
m squares placed 1 m from each tree trunk in the 4 car- ple of larvae collected from northern red oak acorns
44 44
dinal directions, and in a 2-m-radius circle centered at (North America, N = 24, and Europe, N = 16) and from
45 45
each tree trunk in the USA. Acorns were collected at the sessile oak acorns (Europe, N = 24). We did not pre-de-
46 46
beginning of October, bagged separately according to termine whether larvae were moths or beetles. DNA was
47 47
the tree of origin, transported to the laboratory and re- from tissue and extracted following the salt extraction
48 48
frigerated (4 °C) until further processing. The phenolo- protocol (Aljanabi & Martínez 1997). In the case of
49 49
gy of acorn fall is similar in red and sessile oaks in Po- weevil larvae, we sequenced a fragment (826 bp) of the
50 50
land; therefore, bias due to any phenological difference mitochondrial cytochrome oxidase subunit 1 (cox1) us-
51 51

1 ing primers Pat and Jerry following methods described lengths). This gene tree was built pooling together the 1
2 previously (Hughes & Vogler 2004a). We chose this haplotypes of the individuals sequenced at our study 2
3 fragment of cox1 due to the availability of many refer- sites with reference sequences of the Holartic species of 3
4 ence sequences of European and North American Cur- Curculio spp. available at GenBank from Hughes and 4
5 culio spp. to compare with (Hughes & Vogler 2004a). In Vogler (2004a). The gall-feeding Curculio pyrrhoceras 5
6 the case of Cydia spp. caterpillars we amplified a differ- was used as an outgroup. 6
7 ent fragment (625 bp long) of the same mitochondrial 7
Lipid, tannin and protein concentration in
8 gene (cytochrome oxidase I) using the universal prim- 8
9 ers pair LCOI1490/HCOI2198, common in DNA bar- acorns 9
10 coding (see Folmer et al. 1994 for details on the prim- 10
Previous studies have provided evidence for chemi-
11 er sequences and PCR protocols). Sequencing was 11
cal gradients in acorns that promote consumption of the
12 performed using Big-Dye (Perkin-Elmer) technology 12
basal end of the acorns (i.e. avoidance of the apical end)
13 and an ABI3700 sequencer. Sequence chromatograms 13
by seed predators (Steele et al. 1993). Therefore, we cut
14 were assembled and edited using Sequencher 4.6 (Gene 14
each acorn transversely into a top (basal) and bottom
15 Codes, Ann Arbor, MI, USA). 15
half (apical), and analyzed the halves separately. We
16 Two alignments (one for weevil larvae and another 16
used only non-infested acorns, and measured the con-
17 for caterpillar moths) were created using CLUSTALW 17
centration of each compound based on 10 replicates per
18 supplied through http://align.genome.jp, with default 18
oak species. Each sample consisted of 5–6g of chopped
19 gap open and gap extension penalties. The alignment 19
cotyledons originated from different trees. The detailed
20 sets were collapsed into unique haplotypes and each 20
description of procedures is presented in the Supple-
21 were compared to the reference sequences available at 21
mentary Materials.
22 GenBank (for weevil larvae) or the Barcoding of Life 22
23 Datasystem (for moth caterpillars). We applied the most Data analysis 23
24 conservative limit used in DNA barcoding, which states 24
We analyzed data in R using the generalized linear
25 a maximum genetic divergence of 1% with the reference 25
mixed models (GLMMs) implemented via lme4 pack-
26 sequence for an unambiguous identification to the spe- 26
age (Bates et al. 2015). Each below-described mod-
27 cies level (Ratnasingham & Herbert 2007). In the case 27
el was run separately for European and North American
28 of acorn weevils, we additionally built a neighbor join- 28
data. For the European comparison, we did not include
29 ing gene tree based on Kimura 2 parameter model ge- 29
seed mass as a covariate because acorn mass and spe-
30 netic distances (2KP, Kimura 1980). The objective was 30
cies were strongly confounded: northern red oak acorns
31 to further assess whether the haplotypes corresponding 31
were on average twice heavier than sessile oak acorns
32 to the individuals classified as the same species clus- 32
(Table 1). We estimated the proportion of acorns infest-
33 tered together forming discrete groups separated from 33
ed with weevil and moth larvae, as well as the over-
34 the others by significant genetic discontinuities (branch 34
all infestation rate, each with a separate model. We built
35 35
36 36
37 37
38 38
39 Table 1 Traits of northern red oak (in Europe and North America) and sessile oak acorns 39
40 40
Trait Sessile oak Red oak (Europe) Red oak (North America)
41 41
42 Apical Basal Apical Basal Apical Basal 42
†
43 Tannins 3.59 (1.05) 2.65 (0.85) 3.75 (0.81) 3.72 (0.62) 9.8 NA† 43
44 Lipids 5.14 (0.39) 5.60 (0.73) 14.10 (1.46) 16.49 (1.95) 15.36 (2.75) 18.15 (3.46) 44
45 45
Proteins 0.36 (0.02) 0.28 (0.02) 0.49 (0.03) 0.46 (0.04) NA NA
46 46
47 Mass (g) 1.26 (0.91) 2.85 (1.19) 4.81 (1.48) 47
48 Shell thickness (mm) 0.97 (0.64) 1.71 (0.60) 1.49 (0.54) 48
49 †
Data on tannin concentration in northern red oak acorns in North America following Shimada and Saitoh (2006), measured for the 49
50 whole acorn. Tannins and lipids are expressed as a percentage of acorn dry mass, and proteins in grams per dry mass. Values in pa- 50
51 rentheses represent standard deviations from the mean. Proteins were only measured in the European sample. 51

1 a binomial family, logit link GLMMs (1 per continent) The difference was driven by the decrease in weevil in- 1
2 with tree ID as a random effect. Fixed effects includ- festation in red oak (3–10%) as compared to sessile oak 2
3 ed year, crop size index, oak species and oak species by (31–64%, χ2 = 256.71, df = 1, P < 0.001). Infestation by 3
4 crop size interaction term. Species and interaction terms Cydia sp. did not differ between species (χ2 = 0.03, df = 4
5 were only included in the “European” model. We used 1, P = 0.86), and ranged from 1% to 5% depending on 5
6 the total number of acorns collected under each tree as the year. In the “North American” model, the estimated 6
7 an index of individual crop size. overall proportion of infested northern red oak acorns 7
8 We estimated the probability of successful emergence ranged from 35% to 62% depending on the year (Fig. 8
9 from the infested acorn with binomial family, logit link 1). Weevil infestation ranged from 37% to 48%, while 9
10 GLMM, with tree ID included as random effect. In this moth infestation ranged from 2% to 12%. In all models, 10
11 analysis, the response variable was binomial with failure infestation rates also differed significantly among years 11
12 (0), indicating an oviposition mark on the acorn but no (P < 0.001; Fig. 1), but individual tree acorn production 12
13 larvae emergence, and success (1) indicated acorn with did not have an effect (P < 0.40). 13
14 both an oviposition mark and emerged larvae. We used In Europe, the probability of successful weevil larvae 14
15 oak species as the fixed effect, again, only in the “Euro- emergence from an infested acorn was over 20 times 15
16 pean” model. We also tested for differences in probabil- higher in sessile oaks than in northern red oaks (ap- 16
17 ity of acorn embryo survival in weevil-infested acorns proximately 40% vs 2%, z = 6.13, P < 0.001; Fig. 2). In 17
18 using binomial family, logit link GLMM, with tree ID contrast, the probability of emergence of weevils from 18
19 as random effect. Species was included as a fixed effect northern red oak acorns in USA was similar to the prob- 19
20 in “European” models, while “American” models were ability of emergence of weevils from sessile oak acorns 20
21 fitted with the intercept only. in Europe (Fig. 2). 21
22 22
When the acorns infested by any type of insect (ei-
23 23
24
RESULTS ther weevil or moth caterpillar) were pooled, the prob-
24
ability of embryo survival after infestation also differed
25 In the models fitted to European data, the estimated 25
among oak species. In Europe, this probability was low-
26 overall proportion of infested acorns was higher in ses- 26
er in sessile than in northern red oaks (65% vs 90%, z =
27 sile (33–67%, depending on the year) than in northern 27
−4.40, P < 0.001; Fig. 3), and was the lowest (50%) in
28 red oaks (3–13%, χ2 = 254.53, df = 1, P < 0.001; Fig. 1). northern red oaks in North America.
28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 Figure 1 Estimates of the pre-dispersal predation rates (com- 46
Figure 2 Probability of successful emergence of weevil larvae
47 bined, weevils and moths) of sessile oaks (SO) and northern from infested northern red (in Europe and North America) and 47
48 red oaks (RO, in Europe and North America) across 3 years of sessile oak acorns. Estimates are based on the binomial gener- 48
49 the study. Estimates are based on binomial GLMM and whis- alized linear mixed models (see Methods for details). Whiskers 49
50 kers represent standard errors. Dots represent different years of represent standard errors. 50
51 data collection. 51

1 Using DNA barcoding we successfully identified cies were collected from northern red oak and sessile 1
2 21 out of 24 larvae collected from red oaks in North oak acorns. 2
3 America, 16 out of 16 collected from red oaks in Eu- The concentration of tannins was similar in northern 3
4 rope, and 17 out of 24 from sessile oaks. We identified red and sessile oaks in Europe (Table 1), although red 4
5 2 acorn weevil species in North America: Curculio pro- oak acorns in Europe had almost 2 times lower tannin 5
6 boscideus (12 sequences, 7 haplotypes) and Curculio concentration than that reported for North America pop- 6
7 nasicus (4 sequences, 2 haplotypes), all of them feed- ulations. Lipid concentration was almost 3 times high- 7
8 ing on northern red oak acorns. In Europe we identified er in northern red than in sessile oaks in Europe. Pro- 8
9 only 1; namely, Curculio venosus (8 sequences, 2 hap- tein concentration was also higher in northern red than 9
10 lotypes) and weevil larvae were found both in red and in sessile oaks (Table 1). The mass of northern red oak 10
11 white oaks. The comparison with reference sequenc- acorns in Europe was almost twice lower than in their 11
12 es retrieved a divergence below 1% in all cases. More- native range. Nonetheless, acorns of northern red oaks 12
13 over, the neighbor joining tree showed that the distinct in Europe were almost twice heavier than those of ses- 13
14 haplotypes clustered with their respective reference se- sile oaks. Acorn shell was thicker in northern red oaks 14
15 quences, being separated from the rest of the species in than in sessile oaks (Table 1). 15
16 the phylogeny by longer branches (Fig. 3). Although not 16
17 identified in our study, the other potential species pres- 17
DISCUSSION
18 ent at our site are Curculio glandium, Curculio pellitus 18
19 and Curculio elephas (Burakowski et al. 1995). In the Northern red oaks escape pre-dispersal seed predation 19
20 case of moth caterpillars, 2 species were identified in by weevils in Europe. The predation by moths was sim- 20
21 Europe: Cydia amplana (6 sequences, 1 haplotype) and ilar in both species and in both ranges, but the propor- 21
22 Cydia splendana (4 sequences, 2 haplotypes); both spe- tion of seeds infested by Cydia sp. was low compared 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 Figure 3 Neighbour joining tree based on 39
40 the mitochondrial gene citochrome oxi- 40
dase I showing the positions of the 11 hap-
41 41
lotypes obtained from the weevil larvae at
42 42
the study sites and published reference se-
43 43
quences from Holartic Curculio spp. Se-
44 44
quences correspond to the distinct hap-
45 lotypes. The evolutionary distances were 45
46 computed using the Kimura 2-parameter 46
47 method. The gall-feeding Curculio pyrrho- 47
48 ceras was the outgroup. We identified Cur- 48
49 culio proboscideus and Curculio nasicus in 49
50 North American, and Curculio venosus in 50
51 the European sample. 51

1 to that predated by weevils (see Myczko et al. 2017 for weevils but also seem to effectively trap insects in their 1
2 more detailed analyses of Cydia infestation in northern acorns. In Europe, the probability of successful larvae 2
3 red and sessile oaks in Poland). The lower prevalence of emergence from a northern red oak acorn with an ovi- 3
4 Cydia spp. compared to weevils is common in temper- position scar was lower than 3%, compared to 48% 4
5 ate oak forests (see e.g. Bonal et al. [2007] for a similar in red oaks in North America and over 40% in sessile 5
6 result). The weevil infestation rates of northern red oak oaks. This low success could be the result of larval de- 6
7 acorns in their invasive range were 10 times lower than velopment failure due to the lack of trophic adapta- 7
8 that of sessile oaks, and also 10 times lower than that tion to feed on the alien host tree. Recent studies have 8
9 of red oaks in North America. Plant species with large stressed the importance of endosymbiont bacteria as- 9
10 seeds, like oaks, are particularly likely to be seed-lim- sociated with acorn weevils in the adaptation to differ- 10
11 ited (Moles & Westoby 2002; Clark et al. 2007). Thus, ent trophic sources. The community of these bacteria 11
12 the drastically reduced pre-dispersal seed predation of not only changes among weevil species but also with- 12
13 northern red oaks in their exotic range is likely to fa- in species between populations feeding on different host 13
14 vor their recruitment and accelerate their invasion in plants (Toju & Fukatsu 2011; Merville et al. 2013). The 14
15 European forests. Furthermore, even when weevils, the species feeding on red oak acorns in North America (C. 15
16 main pre-dispersal insect predators of oaks, oviposit proboscideus and C. nasicus) were different from that 16
17 into northern red oaks, the larvae fail to develop. Thus, recorded in Europe (C. venosus). In fact, there is not 17
18 the exotic host apparently creates a “trap” for the insect a single acorn weevil species found in both continents 18
19 (Schlaepfer et al. 2002). (Hughes & Vogler 2004b). Thus, along its evolutionary 19
20 The enemy release hypothesis (ERH) makes 3 pre- history, European species like C. venosus have not faced 20
21 dictions about the effects of natural enemies on alien any red oak species, as red oaks (section Lobatae) are 21
22 plants (Keane & Crawley 2002), and even though our exclusive to America. This might have prevented associ- 22
23 study was not designed to test them, several lines of evi- ations with endosymbiont bacteria that would allow in- 23
24 dence suggest that the escape from seed predation in the sects to feed on seeds with a different chemical compo- 24
25 exotic range may aid the spread of northern red oaks. sition. 25
26 First, ERH assumes that enemies are important regula- One caveat to the argument of the lack of trophic ad- 26
27 tors of plant populations. Although we have not deter- aptation in acorn weevils is that it is not possible to dis- 27
28 mined whether oaks at our sites are seed-limited, seed tinguish whether females laid eggs but the larval de- 28
29 predation commonly limits recruitment in oaks (Craw- velopment stopped in the early stages, or that females 29
30 ley & Long 1995; Haas & Heske 2005; Espelta et al. drilled an oviposition hole but failed to lay eggs. Which 30
31 2009; Bogdziwicz et al. 2017). Second, according to scenario is more likely? On one hand, red oaks have a 31
32 the ERH, enemies should have greater impact on na- shell that is twice the thickness of that of sessile oaks, 32
33 tive than on alien plants. Our results show that the mag- and the weevil rostrum size correlates positively with 33
34 nitude of pre-dispersal seed predation by weevils is al- the size of the acorn host (Hughes & Voegler 2004b). 34
35 most 10 times higher on sessile oaks than on northern Therefore, European weevils may be less successful at 35
36 red oaks, and a similar study by Myczko et al. (2017) drilling oviposition holes in red oak acorns. In such a 36
37 reported even greater differences. Assuming that these scenario, the oviposition scars we observed on red oak 37
38 populations are seed-limited, we can tentatively con- acorns may not always result in egg laying. However, 38
39 clude that the impact of weevil predation on populations recent studies suggest that the rostrum length of weevils 39
40 of sessile oaks is greater than on northern red oaks. Fi- is not critical for exploiting seeds (Bonal et al. 2011). 40
41 nally, according to the ERH, alien species should be Furthermore, C. venosus lays eggs very early in the sea- 41
42 able to capitalize on the enemy release and increase in son, before the acorn shell is fully developed (Venner et 42
43 abundance (Keane & Crawley 2002). Northern red oaks al. 2011; Pélisson et al. 2013). 43
44 have a large recruitment advantage over native oaks in 44
The alternative scenario is that female weevils fail to
45 Central Europe (Major et al. 2013; Myczko et al. 2014) 45
perceive northern red oak acorns as lethal hosts for their
46 but the mechanism is unknown. Our results suggest that 46
larvae. Weevils generally seem to concentrate oviposi-
47 the release from pre-dispersal seed predation might be 47
tioning in seeds of higher quality, although the host-se-
48 an important driver of the difference in recruitment suc- 48
lection process is not clear (Desouhant 1998; Espelta et
49 cess between these species. 49
al. 2009; Bonal et al. 2011; Muñoz et al. 2014). In our
50 50
Northern red oaks not only escape seed predation by population, northern red oaks have similar tannin con-
51 51

1 centration to sessile oaks (although the level of non-tan- MAS recognizes the support of the U. S. National Sci- 1
2 nin phenolics is much higher in red than in sessile oaks ence Foundation (DEB-1556707). 2
3 in Poland: Łuczaj et al. 2014), but much higher lip- 3
4 id and protein concentration. Therefore, a mismatch be- REFERENCES 4
5 tween perceived and real quality of the host seems to 5
6 take place, suggesting an ecological trap (Schlaepfer et Allen CF, Good P, Davis HF, Chisum P, Fowler SD 6
7 al. 2002). If true, northern red oaks will function as sink (1966). Methodology for separation of plant lipids 7
8 habitats for weevils across the landscape (Govindan et and application to spinach leaf and chloroplast lamel- 8
9 al. 2012). The process would slowly result in gradual lae. Journal of the American Oil Chemists Society 43, 9
10 removal of insects from local habitats, which will fur- 223. 10
11 ther decrease predation and accelerate northern red oak Bates D, Maechler M, Bolker B, Walker S (2015). Fit- 11
12 invasion. As a caveat, we have not measured the prefer- ting linear mixed-effects models using lme4. Journal 12
13 ence of weevils to oviposit in red versus sessile oaks in of Statistical Software 67, 1–48. 13
14 Europe. Thus, further studies are required to test wheth- 14
Bogdziewicz M, Crone EE, Steele MA, Zwolak R
15 er red oaks create true ecological traps for weevils (i.e. 15
(2017). Effects of nitrogen deposition on reproduc-
16 are preferred and are lethal hosts; see Schlaepfer et al. 16
17 2002) or are just lethal hosts that are, nevertheless, ex- tion in a masting tree: Benefits of higher seed pro- 17
18 ploited. duction are trumped by negative biotic interactions. 18
19 Journal of Ecology 105, 310–20. 19
To conclude, our study shows that northern red oaks
20 almost completely avoid pre-dispersal seed predation in Bonal R, Munoz A (2008). Seed growth suppression 20
21 their exotic range in contrast to their native range, and constrains the growth of seed parasites: Premature 21
22 in contrast to sessile oaks. Furthermore, European wee- acorn abscission reduces Curculio elephas larval size. 22
23 vils oviposit in the red oaks, but the probability of lar- Ecological Entomology 33, 31–6. 23
24 val emergence from acorns with oviposition scars is Bonal R, Munoz A, Díaz M (2007). Satiation of predis- 24
25 less than 3%. Moreover, even infested northern red oak 25
persal seed predators: The importance of considering
26 acorns have a 90% chance that their embryo survives, 26
both plant and seed levels. Evolutionary Ecology 21,
27 and that these acorns will germinate and produce viable 27
367–80.
28 seedlings (Bonal et al. 2007; Xiao et al. 2007; Muñoz et 28
29 Bonal R, Espelta JM, Vogler AP (2011). Complex selec- 29
al. 2014). We suggest that the mechanisms described in
30 tion on life-history traits and the maintenance of vari- 30
our study might contribute to the invasion of the north-
31 ation in exaggerated rostrum length in acorn weevils. 31
ern red oak across Central European forests. The next
32 Oecologia 167, 1053–61. 32
important step towards testing this hypothesis would be
33 to confirm the demographic impacts of reduced pre-dis- Bonal R, Espelta JM, Muñoz A et al. (2016). Diversi- 33
34 persal seed predation in the northern red oak. ty in insect seed parasite guilds at large geographi- 34
35 cal scale: the roles of host specificity and spatial dis- 35
36 tance. Journal of Biogeography 43, 1620–30. 36
37
ACKNOWLEDGMENTS 37
Branco M, Branco C, Merouani H, Almeida MH (2002).
38 We thank Agnieszka Amborska-Bogdziewicz, Paula 38
Germination success, survival and seedling vigour of
39 Bednarz, Jarret Colvin, Sylwia Dziemian-Zwolak, Phil- 39
Quercus suber acorns in relation to insect damage.
40 lip Esempio, Austin Ford, Urszula Eichert, Shealyn Ma- 40
Forest Ecology and Management 166, 159–64.
41 rino, Ewelina Ratajczak, Danuta Ratajczak, Kinga Stęp- 41
42 niak, Aleksandra Wróbel and Milena Zduniak for their Brown LM, Breed GA, Severns PM, Crone EE (2016). 42
43 help in field and laboratory work, and Samuel Venner Losing a battle but winning the war: Moving past 43
44 and an anonymous reviewer for constructive reviews. preference–performance to understand native herbi- 44
45 This study was supported by the Polish National Science vore–novel host plant interactions. Oecologia 183, 45
46 Foundation grant Preludium no. 2015/17/N/NZ8/01565, 441–53. 46
47 while MB was supported by Polish Foundation for Sci- Burakowski B, Mroczkowski M, Stefańska J (1995). 47
48 ence scholarship ‘Start’, and Etiuda NSF grant no. Chrząszcze-Coleoptera. Ryjkowce-Curculionidae, 48
49 2015/16/T/NZ8/00018. DNA sequencing was supported część 2. Katalog Fauny Polski tom: XXIII, Warsza- 49
50 by PLAGANADO AGL2014-54739-R awarded to RB. wa. 50
51 51

1 Colautti RI, Ricciardi A, Grigorovich IA, MacIsaac JH Keane RM, Crawley MJ (2002). Exotic plant invasions 1
2 (2004). Is invasion success explained by the enemy and the enemy release hypothesis. Trends in Ecology 2
3 release hypothesis? Ecology Letters 7, 721–33. & Evolution 17, 164–70. 3
4 Cattau CE, Martin J, Kitchens WM (2010). Effects of an Keeler MS, Chew FS (2008). Escaping an evolutionary 4
5 exotic prey species on a native specialist: Example of trap: preference and performance of a native insect 5
6 the snail kite. Biological Conservation 143, 513–20. on an exotic invasive host. Oecologia 156, 559–68. 6
7 7
Cattau CE, Fletcher RJ, Reichert BE, Kitchens WM Kimura M (1980). A simple method for estimating evo-
8 8
(2016). Counteracting effects of a non-native prey lutionary rates of base substitutions through compar-
9 9
on the demography of a native predator culminate in ative studies of nucleotide sequences. Journal of Mo-
10 10
positive population growth. Ecological Applications lecular Evolution 16, 111–20.
11 26, 1952–68. 11
12 Kumar S, Stecher G, Tamura K (2016). MEGA7: Mo- 12
Chmura D (2013). Impact of alien tree species Quer- lecular Evolutionary Genetics Analysis Version 7.0
13 13
cus rubra l. on understorey environment and flora: A for Bigger Datasets. Molecular Biology and Evolu-
14 14
study of the Silesian Upland (Southern Poland). Pol- tion 33, 1870–4.
15 15
ish Journal of Ecology 61, 431–41. Liu H, Stiling P (2006). Testing the enemy release hy-
16 16
17 Clark CJ, Poulsen JR, Levey DJ, Osenberg CW (2007). pothesis: A review and meta-analysis. Biological In- 17
18 Are plant populations seed limited? A critique and vasions 8, 1535–45. 18
19 meta-analysis of seed addition experiments. Ameri- Lombardo JA, McCarthy BC (2008). Silvicultural treat- 19
20 can Naturalist 170, 128–42. ment effects on oak seed production and predation by 20
21 Crawley M, Long C (1995). Alternate bearing, predator acorn weevils in southeastern Ohio. Forest Ecology 21
22 satiation and seedling recruitment in Quercus robur L. and Management 255, 2566–76. 22
23 Journal of Ecology 83, 683–96. Łuczaj L, Adamczak A, Duda M (2014). Tannin content 23
24 Desouhant E (1998). Selection of fruits for oviposition in acorns (Quercus spp.) from Poland. Dendrobiolo- 24
25 by the chestnut weevil, Curculio elephas. Entomolo- gy 72, 103–11. 25
26 gia Experimentalis Et Applicata 86, 71–8. 26
Major KC, Nosko P, Kuehne C, Campbell D, Bauhus J
27 Espelta JM, Cortés P, Molowny-Horas R, Retana J 27
(2013). Regeneration dynamics of non-native north-
28 (2009). Acorn crop size and pre-dispersal predation 28
ern red oak (Quercus rubra L.) populations as influ-
29 determine inter-specific differences in the recruitment 29
enced by environmental factors: A case study in man-
30 of co-occurring oaks. Oecologia 161, 559–68. 30
aged hardwood forests of southwestern Germany.
31 Govindan BN, Kery M, Swihart RK (2012). Host selec- 31
Forest Ecology and Management 291, 144–53.
32 tion and responses to forest fragmentation in acorn 32
33
Merville A, Venner S, Henri H et al. (2013). Endosym- 33
weevils: inferences from dynamic occupancy models. biont diversity among sibling weevil species compet-
34 Oikos 121, 623–33. 34
35
ing for the same resource. BMC Evolutionary Biolo- 35
Haas JP, Heske EJ (2005). Experimental study of the ef- gy 13, 28.
36 36
fects of mammalian acorn predators on red oak acorn Moles AT, Westoby M (2002). Seed addition experi-
37 37
survival and germination. Journal of Mammalogy 86, ments are more likely to increase recruitment in larg-
38 38
1015–21. er-seeded species. Oikos 99, 241–8.
39 39
40
Harvey JA, Biere A, Fortuna T et al. (2010). Ecological Muller-Scharer H, Schaffner U, Steinger T (2004). Evo- 40
41
fits, mis-fits and lotteries involving insect herbivores lution in invasive plants: Implications for biological 41
42
on the invasive plant, Bunias orientalis. Biological control. Trends in Ecology & Evolution 19, 417–22. 42
Invasions 12, 3045–59.
43 Muñoz A, Bonal R, Espelta JM (2014). Acorn–weevil 43
44 Hughes J, Vogler AP (2004a). The phylogeny of acorn interactions in a mixed-oak forest: Outcomes for lar- 44
45 weevils (genus Curculio) from mitochondrial and nu- val growth and plant recruitment. Forest Ecology and 45
46 clear DNA sequences: the problem of incomplete 46
Management 322, 98–105.
47 data. Molecular Phylogenetics and Evolution 32, 47
601–15. Myczko Ł, Dylewski Ł, Zduniak P, Sparks TH, Try-
48 janowski P (2014). Predation and dispersal of acorns 48
49 Hughes J, Vogler AP (2004b). Ecomorphological adap- 49
by European Jay (Garrulus glandarius) differs be-
50 tation of acorn weevils to their oviposition site. Evo- 50
tween a native (Pedunculate Oak Quercus robur) and
51 lution 58, 1971–83. 51

1 an introduced oak species (Northern Red Oak Quer- can Midland Naturalist 130, 229–38. 1
2 cus rubra) in Europe. Forest Ecology and Manage- 2
Toju H, Fukatsu T (2011). Diversity and infection prev-
3 ment 331, 35–9. 3
alence of endosymbionts in natural populations of the
4 Myczko Ł, Dylewski Ł, Chrzanowski A, Sparks TH 4
chestnut weevil: Relevance of local climate and host
5 (2017). Acorns of invasive Northern Red Oak (Quer- plants. Molecular Ecology 20, 853–68. 5
6 cus rubra) in Europe are larval hosts for moths and 6
Torchin ME, Lafferty KD, Dobson AP, McKenzie VJ,
7 beetles. Biological Invasions 19, 2419–25. 7
Kuris AM (2003). Introduced species and their miss-
8 8
Peguero G, Bonal R, Sol D, Muñoz A, Sork VL, Espe- ing parasites. Nature 421, 628–30.
9 9
lta JM (2017). Tropical insect diversity: Evidence of Venner S, Pélisson PF, Bel-Venner MC, Débias F, Ra-
10 greater host specialization in seed-feeding weevils. 10
11 jon E, Menu F (2011). Coexistence of insect species
11
Ecology competing for a pulsed resource: Toward a unified
12 12
Pearse IS, Harris DJ, Karban R, Sih A (2013). Predict- theory of biodiversity in fluctuating environments.
13 13
ing novel herbivore–plant interactions. Oikos 122, PLoS ONE 6, e18039.
14 14
1554–64. Wolfe LM (2002). Why alien invaders succeed: Sup-
15 15
16 Pélisson PF, Bel-Venner MC, Giron D, Menu F, Venner port for the escape-from-enemy hypothesis. Ameri-
16
17 S (2013). From income to capital breeding: when di- can Naturalist 160, 705–11. 17
18 versified strategies sustain species coexistence. PloS Woziwoda B, Kopec D, Witkowski J (2014a). The neg-
18
19 One 8, e76086. ative impact of intentionally introduced Quercus ru-
19
20 Ratnasingham S, Hebert PDN (2007). BARCODING, bra L. on a forest community. Acta Societatis Botani-
20
21 BOLD: The Barcode of Life Data System (www.bar- corum Poloniae 83, 39–49. 21
22 codinglife.org). Molecular Ecology Notes 7, 355–64. 22
Woziwoda B, Potocki M, Sagan J, Michal Z, Tomusiak
23 Schlaepfer MA, Runge MC, Sherman PW (2002). Eco- 23
R, Wilczynski S (2014b). Commercial forestry as a
24 logical and evolutionary traps. Trends in Ecology & 24
vector of alien tree species – The case of Quercus ru-
25 Evolution 17, 474–80. 25
bra L. introduction in Poland. Baltic Forestry 20,
26 Shimada T, Saitoh T (2006). Re-evaluation of the rela- 131–41. 26
27 tionship between rodent populations and acorn mast- 27
Xiao Z, Harris MK, Zhang Z (2007). Acorn defenses
28 ing: A review from the aspect of nutrients and de- 28
to herbivory from insects: Implications for the joint
29 fensive chemicals in acorns. Population Ecology 48, 29
evolution of resistance, tolerance and escape. Forest
30 341–52. 30
Ecology and Management 238, 302–8.
31 Siemann E, Rogers WE (2001). Genetic differences in 31
32
Yoon S, Read Q (2016). Consequences of exotic host
32
growth of an invasive tree species. Ecology Letters 4, use: Impacts on Lepidoptera and a test of the ecologi-
33 514–8. 33
34
cal trap hypothesis. Oecologia 181, 985–96.
34
Singleton VL, Orthofer R, Lamuela-Raventos RM Zwolak R, Bogdziewicz M, Wróbel A, Crone EE (2016).
35 35
(1999). Analysis of total phenols and other oxidation Advantages of masting in European beech: timing of
36 36
substrates and antioxidants by means of Folin–Cio- granivore satiation and benefits of seed caching sup-
37 37
calteu reagent. Oxidants and Antioxidants, Pt A 299, port the predator dispersal hypothesis. Oecologia
38 38
152–78. 180, 749–58.
39 39
40
Steele MA (2008). Evolutionary interactions between 40
tree squirrels and trees: A review and synthesis. Cur-
41 SUPPLEMENTARY MATERIALS 41
rent Science 95, 871–6.
42 42
43 Steele MA, Knowles T, Bridle K, Simms EL (1993). Additional supporting information may be found in 43
44 Tannins and partial consumption of acorns: implica- the online version of this article. 44
45 tions for dispersal of oaks by seed predators. Ameri- S1 Lipid tannin, and protein concentration in acorns 45
46 46
47 47
48 Cite this article as: 48
49 49
Bogdziewicz M, Bonal R, Espelta JM (2018). Invasive oaks escape pre-dispersal insect seed predation and trap
50 50
enemies in their seeds. Integrative Zoology 13, 228–37.
51 51

2 2
3 3
4 4
5 5
6
7
Seed loss before seed predation: experimental evidence of the 6
7
8
9
negative effects of leaf feeding insects on acorn production 8
9
10 10
11 11
12 12
13 Tara CANELO,1 Álvaro GAYTÁN,1 Guillermo GONZÁLEZ-BORNAY1 and Raul BONAL1,2,3 13
14 14
1
15 Forest Research Group, INDEHESA, University of Extremadura, Plasencia, Spain, 2DITEG Research Group, University of 15
16 Castilla-La Mancha, Toledo, Spain and 3CREAF, Cerdanyola del Vallès, Catalonia, Spain 16
17 17
18 18
19 19
20 20
21 Abstract 21
22 Insect herbivory decreases plant fitness by constraining plant growth, survival and reproductive output. Most 22
23 studies on the effects of herbivory in trees rely on correlational inter-individual comparisons and could, thus, 23
24 be affected by confounding factors linked to both herbivory and plant performance. Using the Mediterranean 24
25 Holm oak (Quercus ilex) as a study model, we followed an experimental approach in which leaf-feeding insects 25
26 (mainly Lepidoptera caterpillars) were excluded from some shoots in all study trees. Shoots subjected to herbi- 26
27 vore exclusion exhibited lower defoliation rates and produced more acorns than control shoots. Defoliation con- 27
28 strained shoot growth throughout the study period, but had no effect on the number of female flowers produced 28
29 per shoot. Acorn production was, however, lower in control shoots due to their higher abortion rates, and also to 29
30 their greater mortality risk during summer drought, as shoots with fewer leaves were less likely to survive. Plant 30
31 reaction to herbivory inhibits certain physiological pathways involved in plant growth, which, together with the 31
32 effects of physical damage, reduces the amount and efficiency of the photosynthetic tissue. This increases their 32
33 vulnerability to environmental stresses, such as water deficit, which limit resource assimilation. Defoliation is 33
34 likely a key factor affecting oak regeneration, as it may be a significant source of seed loss prior to pre-dispersal 34
35 acorn predation. Further experimental studies could help to elucidate its effects in contrasting environments. In 35
36 Mediterranean regions, the harsher droughts predicted by climate change models could worsen the effects of in- 36
37 sect herbivory on oak reproductive output. 37
38 38
39 Key words: drought stress, insect defoliation, oak, seed abortion, shoot mortality 39
40 40
41 41
42 42
43 43
INTRODUCTION
44 44
45 Herbivory constitutes one of the main antagonistic 45
46 interactions between plants and insects, being the trig- 46
Correspondence: Raul Bonal, Forest Research Group,
47 ger of numerous coevolutionary processes (see review 47
48 INDEHESA, Centro Universitario de Plasencia, Avda. Virgen in Strauss & Zangerl 2002). The loss of photosynthet- 48
49 del Puerto 2, University of Extremadura, 10600 Plasencia, ic tissue reduces plant productivity (Nabity et al. 2009), 49
50 Spain. constraining their growth and threatening their surviv- 50
51 Email: raulbonal@unex.es al (Rausher & Feeny 1980; Davidson et al. 1999; Piper 51

Seed loss before seed predation
1 & Fajardo 2014; Camarero et al. 2015). Herbivory may sound seeds available for recruitment. In some oak spe- 1
2 also reduce plant fecundity directly or indirectly. Di- cies the proportion of acorns infested by insects averag- 2
3 rect effects include the consumption of flowers by her- es 50% (Bonal et al. 2007), but in certain trees and years 3
4 bivorous insects (review in Crawley 1989), whereas in- the losses may be much higher (Bonal & Muñoz 2007; 4
5 direct effects result from the lower number of flowers Espelta et al. 2008; Xia et al. 2016). Post-dispersal 5
6 produced due to the reduction of shoot growth (Samson acorn consumption may also be noteworthy and reduce 6
7 & Werk 1986). In addition, seed set and growth is cost- the number of sound seeds available for dispersal (Bon- 7
8 ly, and a decrease of available resources increases seed al & Muñoz 2007; Zhang et al. 2008; González-Rodrí- 8
9 abortion rates (Rausher & Feeny 1980; Marquis 1992; guez & Villar 2012). Nonetheless, and in spite of the 9
10 Niesenbaum 1996; Espelta et al. 2008; Sánchez-Hu- relevance of all these factors and bottlenecks along the 10
11 manes & Espelta 2011). The negative effects of defoli- regeneration cycle, constraints on acorn production may 11
12 ation on plant reproductive output may be extended in start before pre-dispersal predation. 12
13 time and persist for many years (Morrow & LaMarche Some factors related with early acorn losses are abi- 13
14 1978; McConnel 1988). otic, especially those related with pollination fail- 14
15 Most studies on the effects of insect herbivory on ure (Pearse et al. 2016). In addition, in areas with se- 15
16 plant fitness have followed a correlational approach, es- vere droughts, like the evergreen Mediterranean forests, 16
17 pecially in the case of woody species (Landsber & Oh- seed abortion during the summer may provoke sharp de- 17
18 mart 1989; Simmons et al. 2014; Camarero et al. 2015). creases in the number of acorns produced (Espelta et al. 18
19 Those including manipulation (i.e. removal or reduc- 2008). At this early stage, insect defoliation can also re- 19
20 tion of herbivory) have been mostly carried out on her- duce the size of the acorn crop significantly. Different 20
21 baceous plants (but see Crawley 1985) due to the ease studies have shown strong decreases after massive de- 21
22 of performing such experimental treatments (Hawkes foliation events (Davidson et al. 1999; Nakajima 2015). 22
23 & Sullivan 2001; see review in Massad 2013). How- Nonetheless, reports on the effects of the average defo- 23
24 ever, correlational studies may sometimes mislead our liation rates are less common and mostly correlation- 24
25 conclusions when other factors related to herbivory and al. Moreover, the few experimental studies that have ap- 25
26 acorn production are involved. For example, recent re- proached this subject on individual trees have quantified 26
27 ports have shown that, within the same population, ge- the effects once the acorns have matured (e.g. Crawley 27
28 netic diversity and relatedness between trees affect cat- 1985). They have analyzed neither the stage of flower or 28
29 erpillar herbivory rates in oaks (see Fernandez-Conradi seed development at which defoliation has a stronger ef- 29
30 et al. 2017) and these factors might, also, be related to fect, nor the consequences of the interactions between 30
31 plant reproductive output. defoliation and the environmental conditions on acorn 31
32 production. 32
Oaks Quercus spp. are among the most prevalent tree
33 33
species in the temperate forests of the northern hemi- We carried out an experimental study to determine
34 34
sphere and may suffer severe defoliations (Schultz & the effects of insect defoliation on acorn production us-
35 35
Baldwin 1982; Davidson et al. 1999), especially with ing the Holm oak (Quercus ilex) as a model. The Holm
36 36
the periodic outbreaks of the gypsy moth (Lymantria oak is the most widespread tree in the Mediterranean
37 37
dispar Linnaeus, 1758) among the most dramatic ever forests of the Iberian Peninsula, where it inhabits ar-
38 38
recorded in forest ecosystems (Gottschalk et al. 1998; eas with strong summer droughts and suffers defoliation
39 39
Liebhold et al. 2000; Asaro & Chamberlin 2015). Defo- mainly by moth caterpillars (Lepidoptera) (Ruiz-Car-
40 40
liation, however, has not usually been included among bayo et al. 2016). Contrary to previous studies with oth-
41 41
the main factors impacting the decrease of natural re- er oak species, in which defoliators were excluded from
42 42
generation, which is one of the main problems of oak certain trees and compared with control ones (Crawley
43 43
forests worldwide. Rather, studies on recruitment fail- 1985), we randomly excluded caterpillars from a num-
44 44
ure have focused their attention on other issues, such as ber of branches of all trees and compared them with
45 45
pre-dispersal and post-dispersal acorn predation, seed control branches in the same oaks, thus controlling for
46 46
dispersal or the lack of favorable microhabitats for seed- potential individual effects. Moreover, we followed the
47 47
ling establishment (e.g. Callaway 1992; Gómez 2004; effects of defoliation along the whole season: from early
48 48
Bonal et al. 2007; Zhang et al. 2008; Bonal & Muñoz shoot elongation to the final stage of acorn growth after
49 49
2009; Pulido et al. 2010; Xia et al. 2016). summer drought. In addition, analyzing whether the ef-
50 50
Pre-dispersal seed predation reduces the number of fects of defoliation differed through the season, we ap-
51 51

T. Canelo et al.
1 plied 2 different exclusion treatments: in some branches insect species, mainly moth caterpillars (Lepidoptera) 1
2 caterpillars were excluded from early new shoot elonga- (Ruiz-Carbayo et al. 2016). According to a previous 2
3 tion and in others, exclusion experiments were carried sampling at our study site the more abundant species 3
4 out 1 month after the caterpillars had started feeding. feeding on Q. ilex are: Catocala nymphagoga (Esper, 4
5 Our concrete objectives were: (i) to analyze wheth- 1787), Dryobota labecula (Esper, 1788), Dryoboto- 5
6 er the exclusion of caterpillars increased the final acorn des eremita (Fabricius, 1775), Dryobotodes monochro- 6
7 production; (ii) to assess whether defoliation reduces ma (Esper, 1790), Phycita torrenti (Agenjo, 1972), Tor- 7
8 oak growth, shoot length and/or the number of leaves tricodes alternella (Denis & Schiffermüller, 1775) and 8
9 per shoot; and (iii) to investigate the mechanisms by Tortrix viridana (Linnaeus, 1785) (unpublished results). 9
10 which defoliation may reduce acorn production, analyz- Caterpillars start feeding from early budburst in ear- 10
11 ing the effect of shoot length and the number of leaves ly April and continue throughout the spring until early 11
12 on female flower production and acorn set success. June, when the sclerified leaves become too tough to be 12
13 eaten (Elkinton et al. 1996). This phenology is, howev- 13
14 er, subjected to variability due to the stochasticity of the 14
MATERIALS AND METHODS
15 Mediterranean climate (Gilioli et al. 2013). The species 15
16
Study area composition of caterpillar guilds also changes during 16
17 the spring, with the family Tortricidae the first to feed 17
18 The study was performed at a Holm oak Q. ilex L. sa- on the incipient new shots and the Geometridae species 18
19 vannah located in Malpartida de Plasencia, in the prov- the last (Soria 1987). Their reproductive strategies also 19
20 ince of Cáceres, western Spain (39°56′N, 5°58′O) be- differ and go from short-cycle species, which ovipos- 20
21 tween April and September 2016. Oak savannahs (so it before summer, to long-cycle ones that overwinter as 21
22 called dehesas in Spain) are agroecosystems resulting pupae. Caterpillar damage in the Holm oak also exhib- 22
23 from the human use of Mediterranean forests over cen- its interannual variability and severe defoliations, espe- 23
24 turies. Tree clearing has produced landscapes with oaks cially by Tortrix viridana L. (Lepidoptera: Tortricidae) 24
25 interspersed within a grassland matrix, in which live- (Schroeder & Degen 2008). 25
26 stock raising is the main use of land (Moreno & Pulido 26
27 2009). Our study was carried out in a dehesa of 50 ha Experimental design and sampling methodology 27
28 with a mean density of 14 trees/ha. Fifteen adult Holm oaks of similar size were selected 28
29 for the experiment, excluding those with evident symp- 29
Study species
30 toms of decay. At each tree, 12 branches were assigned 30
31 The broadleaved evergreen Holm oak (Q. ilex L.) randomly to 4 different categories: Control (C, no treat- 31
32 spreads over the western Mediterranean basin, being ment), Treatment 1 (T1, caterpillars excluded from the 32
33 the most widely distributed tree species in the Iberian onset of new shoot elongation onwards), Treatment 2 33
34 Peninsula (Blanco et al. 1997). It inhabits a wide range (T2, excluded 1 month after caterpillars started feeding) 34
35 of habitats but predominates in dry Mediterranean re- and Negative Control (NC, treated with water). At each 35
36 gions with strong summer droughts, which will presum- tree, 1 branch was assigned randomly to 1 treatment 36
37 ably suffer the consequences of global warming more and, starting from it, the rest were allocated to the other 37
38 severely (Petit et al. 2005). In these areas, July and Au- treatments following a clockwise design (see Fig. 1a). 38
39 gust are the warmest months, with temperatures that can Ten shoots per branch were arbitrarily selected on 1 39
40 reach 40°C and almost no precipitation except for some April (a total of 1800 shoots initially marked includ- 40
41 storms later in the season with an average rainfall of just ing all trees: 15 trees, ×12 branches/tree, ×10 shoots per 41
42 6 mm in July and 7 mm in August (AEMET, Spanish branch). All shoots were individually identified by sur- 42
43 Meteorological Agency). rounding their base with a soft wire with a numbered 43
44 Holm oak acorns mature in 1 year and budburst takes label. The soft wire was adjusted just below the axil- 44
45 place in early April. Flowering occurs in April–May and lary bud, and was wide enough not to damage the shoot. 45
46 acorns grow throughout the summer until early autumn. Caterpillars were excluded by spraying insecticide dis- 46
47 Summer drought, with its extreme temperatures and wa- solved in water (cypermethrin 0.005% vol.) over the 47
48 ter deficit, constitutes the main cause of premature acorn treated branches 2 times per week. When it was raining, 48
49 abortion (Espelta et al. 2008). the spraying was repeated the day after the rain stopped. 49
50 Shoots subjected to Treatment 1 (T1) were sprayed from 50
Fresh new oak shoots are defoliated by a number of
51 51

1 the onset of shoot elongation (1 April) to 1 June. In the deed, enough to reduce defoliation rates. 1
2 case of Treatment 2 (T2), spraying started 1 month later Control, T1 and T2 shoots were monitored 3 times: 2
3 (1 May) and ended also on 1 June. In the Control group on 1 May, 1 June and 5 September. The Negative-Con- 3
4 (C) shoots were not treated at all and suffered natural trol shoots were measured at the end of the experiment 4
5 defoliation levels throughout the study period. We also on 5 September, to assess whether the application of the 5
6 established a Negative-Control (NC) in which branches experimental treatment (water spraying) might have had 6
7 were sprayed from the beginning just with regular water any independent effect on the variables of interest, and 7
8 containing no insecticide at all (see Fig. 1b). so rule out any potential confounding effects (Fig. 1). At 8
9 Cypermethrin is a pyrethroid insecticide (Χ22H19Cl2NO3) each revision we recorded whether the shoot was still 9
10 with an environmental persistence lower than 90 days. present or it had been prematurely abscised, registered 10
11 The European Commission and the United States En- shoot length to the nearest millimeter and the number of 11
12 vironmental Protection Agency (EPA) allow its use on leaves. The number of female flowers was recorded in 12
13 plants (Pesticide Properties DataBase: http://sitem.herts. June to compare it with the number of grown acorns in 13
14 ac.uk/aeru/footprint/index2.htm). Its bioaccumulation September, just after the summer drought. In Septem- 14
15 has been detected on aquatic animals such as freshwa- ber, we harvested all shoots and took them to the labora- 15
16 ter fish when used in high concentrations (Corcellas et tory to record their length, the number of leaves and the 16
17 al. 2015) but there were no rivers in our study area. This number of acorns produced. We also calculated the final 17
18 insecticide can be extensively used in croplands (Al- defoliation rates by dividing at each shoot the number 18
19 Sarar et al. 2006; Varikou et al. 2015) and is also rec- of leaves with signs of herbivory by the total number of 19
20 ommended for gardening and horticulture, thus having leaves. 20
21 no adverse effect on plant growth or productivity (what 21
22 Statistical analyses 22
excludes a potential bias of insecticide application on
23 our results). We used the dose provided by the com- We used the Tukey test of additivity to assess wheth- 23
24 mercial product Cythrin spray (0.005% vol.) approved er the experimental design adjusted to a complete ran- 24
25 by the Spanish Ministry of Agriculture (license number domized block, and so the effects of the experimental 25
26 25581). This product is aimed at phytophagous insects treatments on defoliation rates and acorn production per 26
27 and our results showed that that concentration was, in- shoot in September did not differ among trees. The dif- 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 Figure 1 Experimental design. (a) Scheme 40
depicting the distribution of all the experi-
41 41
mental treatments and controls at the can-
42 42
opy of each tree (T1, T2, C and CN) cor-
43 43
responding to Treatment 1, Treatment 2,
44 44
Control and Negative Control, respective-
45 ly. (b) Chronology of the experimental 45
46 treatments: filled black dots show the tim- 46
47 ing of insecticide spraying in T1 and T2. 47
48 Empty triangles show the duration of the 48
49 water spraying (no insecticide) on the Neg- 49
50 ative Controls. Control shoots received no 50
51 treatment throughout the study period. 51

T. Canelo et al.
1 ferences between treatments (C, T1, T2 and NC) were RESULTS 1

2 analyzed using an ANOVA to assess the differences in 2
3 defoliation rates and in the number of acorns produced. The Tukey test of additivity showed that the effect 3
4 Post-hoc analyses for pairwise treatment comparisons of the experimental treatments (C, T1, T2 and CN) did 4
5 were performed with Tukey tests. not differ between trees (block factor) either in the de- 5
6 foliation rates (F = 0.029; df = 1461; P = 0.87), nor in 6
We analyzed whether the temporal changes in shoot
7 length differed between treatments using a linear mixed the number of acorns produced (F = 0.122; df = 1461; 7
8 model (LMM) in which shoot identity was included as a P = 0.72). We could, thus, perform an ANOVA to assess 8
9 random factor. For the analysis on the number of leaves the effect of the experimental treatment on both defolia- 9
10 per shoot, a generalized linear mixed model (GLMM) tion rates and the number of leaves per shoot. Caterpil- 10
11 with a Poisson distribution of errors was conducted, in- lar exclusion (i.e. insecticide spraying) reduced the per- 11
12 cluding shoot length as a continuous covariate and shoot centage of leaves defoliated (ANOVA F3,1462 = 47.595; 12
13 identity as a random factor. Pairwise post-hoc Tukey P < 0.001; Fig 2a). In both exclusion treatments (T1 13
14 tests were performed in both cases (shoot length and and T2) the final defoliation rates recorded in Septem- 14
15 number of leaves) to assess differences between treat- ber were significantly lower than in C shoots (T1 esti- 15
16 ments at different times of the study period. mate: −16.098; t = −7.697; P < 0.001 and T2 estimate: 16
17 −10.879; t = −5.178; P = 0.001) (Fig. 2a). There were 17
The differences between treatments (C, T1 and T2)
18 18
in the number of female flowers in June were evaluat-
19 19
ed using an ANOVA. To test whether acorn set success
20 20
differed between treatments (C, T1 and T2) we used
21 21
2 GLMMs with a binomial distribution of errors and
22 22
shoot identity as a random factor. In the first test, we
23 23
compared the number of female flowers in June versus
24 24
the number of grown acorns in September in all shoots
25 25
that had flowers in June. In those that were not present
26 26
in September (shoots prematurely abscised during the
27 27
summer) acorn set success was zero. In the second test,
28 28
we did the same but only including those shoots that
29 29
had flowers in June and were alive in September. Doing
30 30
this it is possible to estimate the relative impact of shoot
31 31
mortality during the summer drought on acorn set suc-
32 32
cess.
33 33
34 Shoot survival throughout the study period was an- 34
35 alyzed using a Kaplan–Meier analysis. In addition, we 35
36 investigated the factors affecting shoot mortality during 36
37 summer drought with a generalized linear model (GLM). 37
38 The dependent variable was shoot mortality risk (bino- 38
39 mial distribution: variable takes the value 1 when the 39
40 shoot died and zero when it did not); independent vari- 40
41 ables included the shoot length, the number of leaves 41
42 and the number of flowers. 42
43 All data analyses were performed with the statisti- 43
Figure 2 Bar plot showing (a) the defoliation rates (number of
44 cal analysis program R Core Team (2016) and using 44
leaves with signs of herbivory/total number of leaves; mean ±
45 the following libraries: rJava (Urbanek 2016), xlsxjars 45
SE) and (b) the number of acorns per shoot (mean ± SE) mea-
46 (Dragulescu 2014a), xlsx (Dragulescu 2014b), asbio 46
sured at oak shoots subjected to all the different treatments
47 (Aho 2016), multcomp (Hothorn et al. 2008), nlme (Pin- and controls (T1, T2, C and NC) at the end of the study peri- 47
48 heiro et al. 2017), lme4 (Bates et al. 2015), Matrix (Bates od (September). Different letters on the bars indicate statisti- 48
49 & Maechler 2016), survival (Therneau 2015) and surv- cally significant differences between groups (P < 0.05). The as- 49
50 Misc (Dardis 2016). terisks show the groups with the same letters among which the 50
51 differences are marginally significant (0.10 > P > 0.05). 51

1 not significant differences between C and NC, whereas T1 compared to T2 shoots (in the latter caterpillars were 1
2 defoliation rates were lower (marginally significant) in excluded 1 month later) (estimate: 5.219; t = 2.575; P = 2
3 0.05) (Fig. 2a). Defoliation decreased acorn production, 3
4 as there were differences between treatments (ANO- 4
5 VA F3,1462 = 5.106; P = 0.002) (Fig. 2b). The number of 5
6 acorns in September was higher in shoots subjected to 6
7 T1 than either in the C (estimate = 0.065; t = 2.689; P = 7
8 0.036) or the NC shoots (estimate = 0.072; t = 3.156; P 8
9 = 0.009). The rest of the pairwise comparisons were not 9
10 significant (P > 0.05 in all cases) (Fig. 2b). 10
11 11
Caterpillar defoliation constrained shoot develop-
12 12
13
ment. The LMM showed a significant interaction be- 13
14 tween time and treatments in shoot length (F4,2333= 9.467; 14
15 P < 0.001; Fig. 3a) as T1 shoots grew more than the 15
16 others throughout the study period. In June only a mar- 16
17 ginal significant result (comparison between T1 and C) 17
18 was found (estimate = 5.726; Z = 2.804; P = 0.092), but 18
19 in September, T1 shoots were significantly longer than 19
20 C shoots (estimate = 7.325; Z = 3.426; P = 0.014; Fig. 20
21 3a). Differences in the number of leaves per shoot were 21
22 found between treatments as well. Longer shoots had 22
23 more leaves (estimate = 0.0086; Z = 32.64; P < 0.001) 23
24 but, irrespective of their length, the number of leaves 24
25 was higher in T1 compared with the C shoots in June 25
26 (estimate = 0.1625; Z = 4.919; P < 0.001; Fig. 3b). The 26
27 number of leaves was also higher in T2 compared to C 27
28 shoots, but the result was only marginally significant 28
Figure 3 Temporal changes in shoot length (mean ± SE) and
29 (estimate = 0.10; Z = 2.979; P = 0.067). In September 29
number of leaves per shoot (mean ± SE) registered along the
30 the gap between treatments decreased, as even the dif- 30
study period in shoots subjected to Treatment 1, Treatment 2
31 and Control (T1, T2 and C, respectively). Different letters on ferences between C and T1 did not reach significance 31
32 the bars indicate statistically significant differences between (estimate = 0.097; Z = 3.008; P = 0.062; Fig. 3b). 32
33 groups (P < 0.05). The asterisks show the groups with the same 33
Caterpillar feeding had no effect on the number of
34 letters among which the differences are marginally significant 34
female flowers in June, which did not differ between
35 (0.10 > P > 0.05). 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
Figure 4 Graph of the results of the Ka-
45 plan–Meier analysis showing the temporal 45
46 changes in mean survival probability of the 46
47 oak shoots subjected to the 2 treatments, 47
48 T1 (orange line) and T2 (purple line) and 48
49 the Control (green line) during the study 49
50 period. Dashed lines represent the standard 50
51 errors of the mean of each group. 51

T. Canelo et al.
1 Table 1 Acorn set rates (number of female flowers in June divided by the number of acorns in September) in C, T1 and T2: (a) set 1
2 rates calculated including all shoots bearing flowers in June and setting to zero the number of acorns in those shoots that were pre- 2
3 maturely abscised (i.e. died) during the summer and (b) set rates including only those shoots that were alive both in June and Sep- 3
4 tember and, thus, reflecting only acorn losses due to abortion, not to shoot mortality. 4
5 a 5
6 Treatment Number of shoots Flowers Acorns Set rate 6
7 Control 34 57 13 0.23 7
8 Treatment 1 49 88 39 0.44 8
9 Treatment 2 54 92 35 0.38 9
10 b 10
11 Control 25 40 13 0.33 11
12 Treatment 1 39 76 39 0.51 12
13 Treatment 2 46 81 35 0.43 13
14 14
15 15
16 16
17 17
treatments (C, T1 and T2) (ANOVA F2,1273 = 2.107; P =
18 18
0.122). The number of acorns in September, however,
19 19
did vary between them (Fig. 2b). The first GLMM in-
20 20
cluding all the shoots bearing female flowers in June (n
21 21
22
= 137) and setting to zero the number of acorns in those 22
23
shoots that died and were abscised during the summer 23
24 drought (i.e. not present in September), showed that 24
25 acorn set success was significantly higher in T1 than in 25
26 C (estimate = 1.380; Z = 2.478; P = 0.013; Table 1a). 26
27 Acorn set rates were also higher in T2 compared to C, 27
28 although the result was marginally significant (estimate 28
29 = 1.055; Z = 1.945; P = 0.052). To figure out the effect 29
30 on acorn abortion, a second GLMM test was analyzed 30
31 only with those shoots that had flowers in June and had 31
32 survived the dry season (n = 112). In this case, the dif- 32
33 ferences between treatments decreased (Table 1a); only 33
the comparison between T1 and C was significant (esti- Figure 5 Relationship between the number of leaves per shoot
34 34
and shoot mortality risk during the summer drought. Shoot
35 mate = 1.095; Z = 1.990; P = 0.045; Table 1b). 35
mortality risk is coded as 1 when the shoot died from June to
36 Shoot survival probability changed throughout the 36
September and as zero when it did not. The function shows that
37 study period (Fig. 4). The Kaplan–Meier analyses 37
shoots with more leaves were less likely to die.
38 showed that summer survival (from June to September) 38
39 decreased less in those shoots sprayed with insecticide 39
40 (T1 and T2) compared to C shoots subjected to natural 40
41
defoliation rates (χ2 = 26.1; df = 2; P < 0.001; Fig. 4). A DISCUSSION 41
42 42
further general lineal model of the causes underpinning Defoliation significantly reduces acorn production
43 43
this higher mortality risk showed that those shoots with in the Holm oak Q. ilex; that is, experimental shoots
44 44
a lower number of leaves in June had a higher chance sprayed with insecticide exhibited lower defoliation
45 45
of dying (i.e. being prematurely abscised) during the rates and produced more acorns. Defoliation decreased
46 46
summer drought period (Fig. 5) (estimate = −0.06; Z = shoot growth throughout the study period, but the num-
47 47
−2.875; P = 0.004), whereas neither shoot length (esti- ber of female flowers at the end of the spring did not
48 48
49
mate = 0.004; Z = 1.259; P = 0.208) nor the number of differ between treatments. However, acorn abortion 49
50 flowers (estimate = −0.015; Z = − 0.128; P = 0.898) had rates and premature shoot abscission during the summer 50
51 any significant effect. drought were higher in the smaller and heavily defoliat- 51

1 ed control shoots than in the experimental shoots, which may also reduce vessel size (Uggla et al. 1996). Our re- 1
2 provoked a sharp reduction of the acorn crop. sults suggest, however, that the physiological pathways 2
3 With respect to the few previous experimental ap- that constrain plant growth may be reverted if defoli- 3
4 proaches that have used trees as study models to assess ation disappears or decreases its intensity. The shoots 4
5 the effects of defoliation on seed production (e.g. Craw- subjected to Treatment 2 suffered defoliation at the be- 5
6 ley 1985), ours comprises a number of novelties. We in- ginning of the study period, as insecticide spraying start- 6
7 cluded a negative control group in the analyses, which ed 1 month later; however, these shoots could partially 7
8 ruled out any effect of the experimental procedure itself recover their normal growth rates and finally produced 8
9 (i.e. liquid spraying) on the variables of interest, thus more acorns than control shoots (which suffered defoli- 9
10 confirming the negative relationship of caterpillar de- ation throughout the spring). 10
11 foliation with shoot length, number of leaves and acorn In the case of our study model, Q. ilex, experimen- 11
12 production. In addition, we monitored the effects of de- tal increase of jasmonates on leaves has been shown 12
13 foliation throughout the study period, which evidenced to decrease photosynthetic rates and stomata conduc- 13
14 the progressive negative effects on shoot growth. More tance, thus reducing CO2 assimilation rates (Filella et al. 14
15 importantly, this temporal monitoring showed the im- 2006). These changes that alter photosynthetic efficien- 15
16 pact of summer drought on shoot premature abscission cy and water balance make plants especially vulnerable 16
17 (i.e. mortality). To the best of our knowledge, this neg- to environmental stresses, such as water deficit (Quentin 17
18 ative effect had gone previously unnoticed in oaks, be- et al. 2011; Sánchez-Humanes & Espelta 2011). During 18
19 cause measurements in previous studies of acorn pro- drought periods, such as occurs in Mediterranean sum- 19
20 duction were made just once in autumn comparing mers, plants face a trade-off between carbon gain and 20
21 control and treated trees (e.g. Crawley 1985). That sort water loss (Cowan & Farquhar 1977) so that an optimal 21
22 of approach attributes acorn losses from defoliation just stomata functioning is essential to overcome severe wa- 22
23 to seed abortion, while this work has shown that they ter-stress conditions (Xu & Baldocchi 2003). The Holm 23
24 are the consequence of both seed abortion and summer oak Q. ilex has evolved physical and physiological ad- 24
25 shoot abscission (i.e. mortality). aptations to the Mediterranean climate, such as scle- 25
26 rotic leaves and the capability of closing its stomata to 26
There are a number of mechanisms involved in plant
27 avoid excessive evapotranspiration. Stomatal closure 27
responses to defoliation that may be underpinning the
28 limits the number of hours in which CO2 may be as- 28
decreased shoot growth observed in our study. Plants are
29 similated during summer drought (Chaves et al. 2002), 29
able to recognize herbivory damage thanks to fine-tuned
30 which may explain the high mortality rates observed in 30
sensors of the elicitors of the saliva (Kessler & Baldwin
31 the more heavily defoliated shoots. To the natural en- 31
2002; Darwish et al. 2008). The reaction may start even
32 vironmental limitation for CO2 assimilation during the 32
before the damage occurs, as oviposition can also be de-
33 summer, defoliated shoots would add a reduced photo- 33
tected (Balbyshev & Lorenzen 1997; Doss et al. 2000;
34 synthetic tissue (i.e. fewer leaves), a diminished photo- 34
Meiners & Hilker 2000; Meiners et al. 2000). When
35 synthetic efficiency and a reduced stomata conductance. 35
non-lethal herbivory stress occurs, compounds like re-
36 Moreover, the direct mechanical damage would increase 36
active oxygen species (ROS) trigger plant responses by
37 their vulnerability to drought, as dehydration risk is 37
modifying the transport of phytohormones and/or plant
38 higher in damaged leaves (Aldea et al. 2005). This high- 38
metabolites (Potters et al. 2007).
39 er risk could also explain the lower number of leaves in 39
40 The physiological response to insect damage may re- 40
duce plant growth. Insect defoliation activates the syn- the control shoots (not only explained by their shorter
41 length) as dehydration could lead to premature leaf ab- 41
42 thesis of jasmonates (Filella et al. 2006; Nabity et al. 42
2009; Erb et al. 2012), which interact with the pathways scission.
43 43
of other compounds such as abscisic acid, cytokinins, The negative effects of defoliation on resource as-
44 44
brassinosteroids, gibberellins and auxins (Erb 2012). similation may also underlie the lower final reproduc-
45 45
Auxins (indole-3-acetic acid, IAA) deserve special at- tive output in control shoots. The initial investment in
46 46
tention, due to their importance for plant growth and de- reproduction (number of flowers in June) did not dif-
47 47
velopment (Geldner et al. 2003; Teale et al. 2006). Jas- fer between treatments, suggesting that, contrary to oth-
48 48
monates reduce auxin activity, which, in turn, decreases er cases (Crawley 1989), caterpillars did not decrease
49 49
photosynthetic PSII efficiency and CO2 fixation rates oak reproductive output by eating flowers. However,
50 50
(Agtuca et al. 2014); in angiosperms auxin inhibition the final production of acorns per shoot did differ. Seed
51 51

T. Canelo et al.
1 abortion was higher in control shoots (not sprayed with 2012). Considering that defoliated shoots are more vul- 1
2 insecticide), probably because the investment in repro- nerable to die during the summer drought, such a cli- 2
3 duction was programmed before spring and they could matic scenario will seriously hamper oak natural regen- 3
4 not adapt their inversion to the level of damage (Mar- eration. The economic losses would also increase in that 4
5 quis 1992). Acorn growth is based on the photosyn- context, especially in the Spanish oak savannahs (de- 5
6 thates produced during the seeding season (Hoch et al. hesas), where acorns constitute a key food source for 6
7 2013; Fernández-Martínez et al. 2015) and the loss of livestock in autumn (Moreno & Pulido 2009). Howev- 7
8 leaf tissue would reduce the capability of storing re- er, the natural value of such agroecosystems, includ- 8
9 sources available for acorn production. The lower re- ed in the Directive Habitats of the European Union, ad- 9
10 source availability may also account for the higher sum- vise against any massive insecticide spraying to control 10
11 mer mortality (premature abscission) recorded in control caterpillar populations. Rather, management measures 11
12 shoots, which reduced acorn production significantly. should promote their natural enemies, such as insectivo- 12
13 Even though adult Holm oaks are well adapted to the rous birds (Sanz 2001) or parasitoids. 13
14 dry Mediterranean climate and rarely die (Lopez-Iglesias In summary, these results put forward that seed loss 14
15 et al. 2014), shoot mortality during the summer may be may start before pre-dispersal acorn predation. Defoli- 15
16 very significant. ation should be included in future studies on oak natu- 16
17 In accordance with previous studies, our results show ral regeneration, preferentially in contrasting ecological 17
18 that, due to the modular architecture of trees (Takena- contexts, as its negative effects could vary depending on 18
19 ka 1994), oaks did not mobilize resources from health- the environmental conditions. At least in Mediterranean 19
20 ier shoots to more damaged shoots (Sánchez-Humanes oak forests, subjected to strong summer droughts, de- 20
21 et al. 2011; Hoch et al. 2013). Acorn abortion or shoot foliation poses an important threat to oak reproductive 21
22 mortality could, thus, be barely counterbalanced by the output. 22
23 shoots nearby. Treated and control shoots were in the 23
24 same trees, but their mortality and reproductive output 24
25
ACKNOWLEDGMENTS 25
differed greatly. In addition, irrespective of the treat-
26 ment, those shoots with a higher number of leaves in This work was financed by the project AGL2014- 26
27 June had a higher probability of surviving summer 54739-R from the Spanish Ministry of Economy and 27
28 drought. In September there were no differences in the Competitiveness and the European Regional Develop- 28
29 number of leaves between control and experimental ment Fund (FEDER, European Union). RB was funded 29
30 shoots, as the control shoots that were more heavily de- by a contract of the Atracción de Talento Investigador 30
31 foliated were prematurely abscised during the summer. Programme (Gobierno de Extremadura TA13032). TC 31
32 was funded by a FPI fellowship (Ministry of Economy 32
From a conservation perspective, these results stress
33 and Competitiveness BES-2015-071382). Comments of 33
the importance of defoliation as a limitation of natu-
34 the senior editor (Marcel Holyoak) and 3 anonymous re- 34
ral oak recruitment. Defoliation may reduce the num-
35 viewers improved a previous version of the manuscript. 35
ber of acorns available for regeneration even more than
36 36
pre-dispersal acorn predators. In the Holm oak, infes-
37 37
tation rates by acorn-boring insects average 50% of REFERENCES
38 38
the crop, but a percentage of the infested acorns may
39 Agtuca B, Rieger E, Hilger K et al. (2014). Carbon-11 39
still germinate and produce viable seedlings (Bon-
40 reveals opposing roles of auxin and salicylic acid in 40
al et al. 2007). In comparison, acorn production in con-
41 regulating leaf physiology, leaf metabolism, and re- 41
trol shoots was almost half of that of the experimental
42 source allocation patterns that impact root growth in 42
shoots, and the difference could be wider considering
43 Zea mays. Journal of Plant Growth Regulation 33, 43
that more severe defoliations rates have been reported in
44 328–39. 44
oaks (Schroeder & Degen 2008).
45 Aho K (2016). asbio: A collection of statistical tools for 45
46 Within the global change context, the negative effects 46
biologists. R package version 1.3–4.
47 of defoliation on oak recruitment may worsen. Climate 47
models predict an increase of air temperatures and water Aldea M, Hamilton JG, Resti JP, Zangerl AR, Beren-
48 baum MR, DeLucia EH (2005). Indirect effects of 48
49 deficit in the Mediterranean areas, which will presum- 49
ably favor leaf feeding insects (Hódar & Zamora 2004; insect herbivory on leaf gas exchange in soybean.
50 Plant, Cell & Environment 28, 402–11. 50
51 Dymond et al. 2010; Carnicer et al. 2011; Jacquet et al. 51

1 Al-Sarar A, Hall FR, Downer RA (2006). Impact of ceedings of the National Academy of Sciences 108, 1
2 spray application methodology on the development 1474–8. 2
3 of resistance to cypermethrin and spinosad by fall ar- Chaves MM, Pereira JS, Maroco J et al. (2002). How 3
4 myworm Spodoptera frugiperda (JE Smith). Pest plants cope with water stress in the field. Photosyn- 4
5 Management Science 62, 1023–31. thesis and growth. Annals of Botany 89, 907–16. 5
6 Asaro C, Chamberlin LA (2015). Outbreak histo- 6
Corcellas C, Eljarrat E, Barceló D (2015). First report of
7 ry (1953–2014) of spring defoliators impacting 7
pyrethroid bioaccumulation in wild river fish: A case
8 oak-dominated forests in Virginia, with emphasis on 8
study in Iberian river basins (Spain). Environment In-
9 gypsy moth (Lymantria dispar L.) and fall canker- 9
ternational 75, 110–16.
10 worm (Alsophila pometaria Harris). American Ento- 10
11 Cowan IR, Farquhar GD (1977). Stomatal function in 11
mologist 61, 174–85. relation to leaf metabolism and environment. In: Jen-
12 12
Balbyshev NF, Lorenzen JH (1997). Hypersensitivi- nings DH, ed. Integration of Activity in the Higher
13 13
ty and egg drop: A novel mechanism of host-plant Plant, SEB Symposium XXXI. Cambridge University
14 14
resistance to Colorado potato beetle (Coleoptera: Press, Cambridge, pp. 471–505.
15 15
Chrysomelidae). Journal of Economic Entomology Crawley MJ (1985). Reduction of oak fecundity by
16 16
90, 652–7. low-density herbivore populations. Nature 314, 163–
17 17
18 Bates D, Maechler M, Bolker B, Walker S (2015). Fit- 64. 18
19 ting linear mixed-effects models using lme4. Journal Crawley MJ (1989). Insect herbivores and plant popu- 19
20 of Statistical Software 67, 1–48. lation dynamics. Annual Review of Entomology 34, 20
21 Bates D, Maechler M (2016). Matrix: Sparse and dense 531–62. 21
22 matrix classes and methods. R package version 1.2– Dardis C (2016). survMisc: Miscellaneous functions for 22
23 7.1. survival data. R package version 0.5.4. 23
24 Blanco E, Casado MA, Costa M et al. (1997). Los Darwish SA, Pan L, Ide C, Bede JC (2008). Caterpil- 24
25 Bosques Ibéricos. Una Interpretación Geobotánica. lar-specific gene expression in the legume, Medicago 25
26 Planeta, Spain, Madrid. truncatula. Plant Molecular Biology Reporter 26, 12– 26
27 Bonal R, Muñoz A (2007). Multi-trophic effects of un- 27
31.
28 gulate intraguild predation on acorn weevils. Oecolo- 28
29
Davidson CB, Gottschalk KW, Johnson JE (1999). Tree 29
gia 152, 533–40. mortality following defoliation by the European gyp-
30 30
Bonal R, Muñoz A (2009). Seed weevils living on the sy moth (Lymantria dispar L.) in the United States: A
31 31
edge: Pressures and conflicts over body size in the review. Forest Science 45, 74–84.
32 32
endoparasitic Curculio larvae. Ecological Entomolo- Dragulescu AA (2014a). xlsx: Read, write, format Excel
33 33
gy 34, 304–9. 2007 and Excel 97/2000/XP/2003 files. R package
34 34
35
Bonal R, Muñoz A, Díaz M (2007). Satiation of predis- version 0.5.7. 35
36
persal seed predators: The importance of considering Dragulescu AA (2014b). xlsxjars: Package required POI 36
37
both plant and seed levels. Evolutionary Ecology 21, jars for the xlsx package. R package version 0.6.1. 37
38
367–80. 38
Doss RP, Oliver JE, Proebsting WM et al. (2000).
39 Callaway RM (1992). Effect of shrubs on recruitment of Bruchins: Insect-derived plant regulators that stimu- 39
40 Quercus douglasii and Quercus lobata in California. late neoplasm formation. Proceedings of the National 40
41 Ecology 73, 2118–28. Academy of Sciences of the United States of America 41
42 Camarero JJ, Sangüesa-Barreda G, Vergarechea M 97, 6218–23. 42
43 (2015). Prior height, growth, and wood anatomy dif- Dymond CC, Neilson ET, Stinson G et al. (2010). Fu- 43
44 ferently predispose to drought-induced dieback in ture spruce budworm outbreak may create a carbon 44
45 two Mediterranean oak speciesk. Annals of Forest source in Eastern Canadian forests. Ecosystems 13, 45
46 Science 73, 341–51. 917–31. 46
47 Carnicer J, Coll M, Ninyerola M, Pons X, Sanchez G, 47
Elkinton JS, Healy WM, Buonaccorsi JP, Boettner GH,
48 Penuelas J (2011). Widespread crown condition de- 48
Hazzard M, Smith HR (1996). Interactions among
49 cline, food web disruption, and amplified tree mortal- 49
gypsy moths, white-footed mice, and acorns. Ecolog-
50 ity with increased climate change-type drought. Pro- 50
ical Society of America 77, 2332–42.
51 51

T. Canelo et al.
1 Erb M, Meldau S, Howe GA (2012). Role of phytohor- tacks a relict, boreal pine species. Biodiversity and 1
2 mones in insect-specific plant reactions. Trends in Conservation 13, 493–500. 2
3 Plant Science 17, 250–59. Hothorn T, Bretz F, Westfall P (2008). Simultaneous in- 3
4 Espelta JM, Cortés P, Molowny-Horas R, Sánchez-Hu- ference in general parametric models. Biometrical 4
5 manes B, Retana J (2008). Masting mediated by sum- Journal 50, 346–63. 5
6 mer drought reduces acorn predation in mediterra- 6
Jacquet JS, Orazio C, Jactel H (2012). Defoliation by
7 nean oak forests. Ecology 89, 805–17. 7
processionary moth significantly reduces tree growth:
8 8
Fernandez-Conradi P, Jactel H, Hampe A, Leiva MJ, A quantitative review. Annals of Forest Science 69,
9 9
Castagneyrol B (2017). The effect of tree genetic di- 857–66.
10 10
versity on insect herbivory varies with insect abun- Kessler A, Baldwin IT (2002). Plant responses to insect
11 dance. Ecosphere 8, e01637. 11
12 herbivory: The emerging molecular analysis. Annual 12
Fernández-Martínez M, Garbulsky M, Peñuelas J, Peg- Review of Plant Biology 53, 299–328.
13 13
uero G, Espelta JM (2015). Temporal trends in En- Landsber J, Ohmart C (1989). Levels of insect defolia-
14 14
hanced Vegetation Index and spring weather predict tion in forests: Patterns and concepts. Trends in Ecol-
15 15
seed production in Mediterranean oaks. Plant Ecolo- ogy and Evolution 4, 96–100.
16 16
gy 216. 1061–72.
17 Liebhold A, Elkinton J, Williams D, Muzika RM (2000). 17
18 Filella I, Peñuelas J, Llusià J (2006). Application of jas- What causes outbreaks of the gypsy moth in North 18
19 monic acid enhances the emissions of monoterpenes America? Population Ecology 42, 257–66. 19
20 and methyl salicylate and decreases the uptake of 20
Lopez-Iglesias B, Villar R, Poorter L (2014). Functional
21 formaldehyde by Quercus ilex. New Phytologist 169, 21
traits predict drought performance and distribution of
22 135–44. 22
mediterranean woody species. Acta Oecologica, 56,
23 Geldner N, Anders N, Wolters H et al. (2003). The Ara- 10–18. 23
24 bidopsis GNOM ARF-GEF mediates endosomal re- 24
cycling, auxin transport, and auxin-dependent plant Marquis RJ (1992). A bite is a bite is a bite? Constraints
25 25
growth. Cell 112, 219–30. on response to folivory in Piper arieianum (Piperace-
26 26
ae). Ecological Society of America. Ecology 73, 143–
27 Gilioli G, Bodini A, Baumgärtner J (2013). Metapopula- 27
52.
28 tion modelling and area-wide pest management strat- 28
29 egies evaluation. An application to the Pine proces- Massad TJ (2013). Ontogenetic differences of herbiv- 29
30 sionary moth. Ecological Modelling 260, 1–10. ory on woody and herbaceous plants: A meta-analy- 30
31
sis demonstrating unique effects of herbivory on the 31
Gómez JM (2004). Importance of microhabitat and
32
young and the old, the slow and the fast. Oecologia 32
acorn burial on Quercus ilex early recruitment:
33
172, 1–10. 33
Non-additive effects on multiple demographic pro-
34 cesses. Plant Ecology 172, 287–97. McConnel SP (1988). Effects of gypsy moth defoliation 34
35 on acorn production and viability, litterfall, and lit- 35
González-Rodríguez V, Villar R (2012). Post-disper-
36 ter layer depth and biomass in North-Central Virginia 36
sal seed removal in four Mediterranean oaks: Species
37 and Western Maryland, thesis, Faculty of the Virginia 37
and microhabitat selection differ depending on large
38 Polytechnic Institute and State University. 38
herbivore activity. Ecological Research 27, 587–94.
39 Meiners T, Hilker M (2000). Induction of plant syno- 39
Gottschalk KW, Coeber JJ, Feicht DL (1998). Tree mor-
40 mones by oviposition of a phytophagous insect. Jour- 40
tality risk of oak due to gypsy moth. European Jour-
41 nal of Chemical Ecology 26, 221–32. 41
nal of Pathology 28, 121–32.
42 Meiners T, Westerhaus C, Hilker M (2000). Specificity 42
Hawkes CV, Sullivan JJ (2001). The impact of herbiv-
43 of chemical cues by a specialist egg parasitoid during 43
ory on plants in different resource conditions: A me-
44 host location. Entomologia Experimentalis et Appli- 44
ta-analysis. Ecology 82, 2045–58.
45 cata 95, 151–9. 45
46 Hoch G, Siegwolf RTW, Keel SG, Körner C, Han Q 46
(2013). Fruit production in three masting tree species Moreno G, Pulido FJ (2009). The functioning, manage-
47 ment and persistence of dehesas. In: Rigueiro-Ro- 47
48 does not rely on stored carbon reserves. Oecologia 48
171, 653–62. dríguez A et al. ed., Agroforestry in Europe: Cur-
49 rent Status and Future Prospects. Springer Science + 49
50 Hódar JA, Zamora R (2004). Herbivory and climatic 50
Business Media B.V., pp. 127–60.
51 warming: A Mediterranean outbreaking caterpillar at- 51

1 Morrow PA, LaMarche CV (1978). Tree ring evidence Ruiz-Carbayo H, Bonal R, Espelta JM, Hernández M, 1
2 for chronic insect suppression of productivity in Sub- Pino J (2016). Community assembly in time and 2
3 alpine eucalyptus. Science 201, 1244–6. space: The case of Lepidoptera in a Quercus ilex L. 3
4 Nabity PD, Zavala JA, DeLucia EH (2009). Indirect savannah-like landscape. Insect Conservation and 4
5 suppression of photosynthesis on individual leaves Diversity 10, 21–31. 5
6 by arthropod herbivory. Annals of Botany 103, 655– Samson DA, Werk KS (1986). Size-dependent effects 6
7 63. in the analysis of reproductive effort in plants. The 7
8 American Naturalist 127, 667–80. 8
Nakajima H (2015). Defoliation by gypsy moths nega-
9 9
tively affects the production of acorns by two Japa- Sánchez-Humanes B, Espelta JM (2011). Increased
10 10
nese oak species. Trees – Structure and Function 29, drought reduces acorn production in Quercus ilex
11 11
1559–66. coppices: Thining mitigates this effect but only in the
12 12
Niesenbaum, R. A. (1996). Linking herbivory and polli- sort term. Forestry 84, 73–82.
13 13
14 nation: Defoliation and selective fruit abortion in Lin- Sánchez-Humanes B, Sork VL, Espelta JM (2011). 14
15 dera Benzoin. Ecology 77, 2324–31. Trade-offs between vegetative growth and acorn pro- 15
16 Pearse IS, Koenig WD, Kelly D. (2016). Mechanisms of duction in Quercus lobata during a mast year: The 16
17 mast seeding: Resources, weather, cues and selection. relevance of crop size and hierarchical level within 17
18 New Phytologist 212, 546–62. the canopy. Oecologia 166, 101–10. 18
19 Petit RJ, Hampe A, Cheddadi R (2005). Climate chang- Sanz JJ (2001). Experimentally increased insectivorous 19
20 es and tree phylogeography in the Mediterranean. bird density results in a reduction of caterpillar den- 20
21 TAXON 54, 877–85. sity and leaf damage to Pyrenean oak. Ecological Re- 21
22 Pinheiro J, Bates D, DebRoy S, Sarkar D (2016). nlme: search 16, 387–94. 22
23 Linear and Nonlinear Mixed Effects Models_. R Schroeder H, Degen B (2008). Spatial genetic structure 23
24 package version 3.1–128 in populations of the green oak leaf roller, Tortrix vir- 24
25
Piper FI, Fajardo A (2014). Foliar habit, tolerance to de- idana L. (Lepidoptera, Tortricidae). European Jour- 25
26
foliation and their link to carbon and nitrogen stor- nal of Forest Research 127, 447–53. 26
27 Schultz JC, Baldwin IT (1982). Oak leaf quality de- 27
age. Journal of Ecology 102, 1101–11.
28 clines in response to defoliation by gypsy moth lar- 28
29
Potters G, Pasternak TP, Guisez Y, Palme KJ, Jansen 29
MAK (2007). Stress-induced morphogenic respons- vae. Science 217, 149–51.
30 30
es: Growing out of trouble? Trends in Plant Science Simmons MJ, Lee TD, Ducey MJ, Elkinton JS, Boettner
31 31
12, 98–105. GH, Dodds KJ (2014). Effects of invasive win-
32 32
Pulido F, García E, Obrador JJ, Moreno G (2010). Mul- ter moth defoliation on tree radial growth in Eastern
33 33
tiple pathways for tree regeneration in anthropogen- Massachusetts, USA. Insects 5, 301–318.
34 34
35 ic savannas: Incorporating biotic and abiotic drivers Soria S (1987). Lepidópteros defoliadores de Quercus 35
36 into management schemes. Journal of Applied Ecolo- pyrenaica Willdenow, 1805. Ministerio de Agricultu- 36
37 gy 47, 1272–81. ra y Medio Ambiente, Madrid. 37
38 Quentin AG, O’Grady AP, Beadle CL, Worledge D, Strauss SY, Zangerl AR (2002) Plant insect interactions 38
39 Pinkard EA (2011). Responses of transpiration and in terrestrial ecosystems. In: Herrera CM, Pellmyr 39
40 canopy conductance to partial defoliation of Eucalyp- O, eds. Plant–Animal Interactions: An Evolutionary 40
41 tus globulus trees. Agricultural and Forest Meteorol- Ecology Approach. Blackwell Publishing, Malden, 41
42 ogy 151, 356–64. MA, pp. 77–106. 42
43 Rausher M D, Feeny P (1980). Herbivory, plant density, Takenaka A (1994). A simulation model of tree architec- 43
44 and plant reproductive success: The effect of Battus ture development based on growth response to local 44
45 philenor on Aristolochia reticulata. Ecology 61, 905– light environment. Journal of Plant Research 107, 45
46 917. 321–30. 46
47 R Core Team (2016). R: A language and environment Teale W D, Paponov I A, Palme K (2006). Auxin in ac- 47
48 for statistical computing. R Foundation for Statisti- tion: Signalling, transport and the control of plant 48
49 cal Computing, Vienna, Austria. Available from URL growth and development. Nature Reviews Molecular 49
50 https://www.R-project.org/. Cell Biology 7, 847–59. 50
51 51

T. Canelo et al.
1 Therneau T (2015). A package for survival analysis in S. Xia K, Harrower WL, Turkington R, Tan HY, Zhou ZK 1
2 version 2.38. (2016). Pre-dispersal strategies by Quercus schottk- 2
3 Uggla C, Moritz T, Sandberg G, Sundberg B (1996). yana to mitigate the effects of weevil infestation of 3
4 Auxin as a positionnal signal in pattern formation in acorns. Scientific Reports 6, 37520. 4
5 plants. Proceedings of the National Academy of Sci- Xu L, Baldocchi DD (2003). Seasonal trends in pho- 5
6 ences of the United States of America 93, 9282–6. tosynthetic parameters and stomatal conductance of 6
7 blue oak (Quercus douglasii) under prolonged sum- 7
Urbanek S (2016). rJava: Low-level R to Java Interface.
8 mer drought and high temperature. Tree Physiology 8
R package version 0.9–8.
9 23, 865–77. 9
10 Varikou K, Garantonakis N, Birouraki A, Ioannou A, 10
Kapogia E (2015). Improvement of bait sprays for Zhang SB, Hu H, Li ZR (2008). Variation of photosyn-
11 11
the control of Bactrocera oleae (Diptera: Tephritidae). thetic capacity with leaf age in an alpine orchid, Cyp-
12 12
Crop Protection 81, 1–8. ripedium flavum. Acta Physiologiae Plantarum 30,
13 13
381–8.
14 14
15 15
17 17
18 Canelo T, Gaytán A, González-Bornay, Bonal R (2018). Seed loss before seed predation: experimental evidence 18
19 of the negative effects of leaf feeding insects on acorn production. Integrative Zoology 13, 238–50. 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Acorn size and tolerance to seed predators: the multiple roles of 6
7
8
9
acorns as food for seed predators, fruit for dispersal and fuel for 8
9
10
11 growth 10
11
12 12
13 13
14 Andrew W. BARTLOW,1,2 Salvatore J. AGOSTA,1,3 Rachel CURTIS,1,4 Xianfeng YI1,5 and Michael 14
15 15
16 A. STEELE1 16
17 1
Department of Biology and The WIESS Institute for Environmental Science and Sustainability, Wilkes University, Wilkes Barre, 17
18 Pennsylvania, USA 18
19 19
20 20
21 21
22 Abstract 22
23 23
Fitness of parents and offspring is affected by offspring size. In oaks (Quercus spp.), acorns vary considerably
24 24
in size across, and within, species. Seed size influences dispersal and establishment of oaks, but it is not known
25 25
whether size imparts tolerance to seed predators. Here, we examine the relative extent to which cotyledon size
26 26
serves as both a means for sustaining partial consumption and energy reserves for developing seedlings during
27 27
early stages of establishment. Acorns of 6 oak species were damaged to simulate acorn predation by vertebrate
28 28
and invertebrate seed predators. Seedling germination/emergence and growth rates were used to assess seedling
29 29
performance. We predicted that if cotyledons are important for dispersal, acorns should show tolerance to par-
30 30
tial seed consumption. Alternatively, if the cotyledon functions primarily as an energy reserve, damage should
31 31
significantly influence seedling performance. Acorns of each species germinated and produced seedlings even
32 32
after removing >50% of the cotyledon. Seed mass explained only some of the variation in performance. With-
33 33
in species, larger acorns performed better than smaller acorns when damaged. Undamaged acorns performed as
34 34
well or better than damaged acorns. There was no pattern among individual species with increasing amounts of
35 35
damage. In some species, simulated invertebrate damage resulted in the poorest performance, suggesting alter-
36 36
native strategies of oaks to sustain damage. Large cotyledons in acorns may be important for attracting seed dis-
37 37
persers and sustaining partial damage, while also providing energy to young seedlings. Success of oak establish-
38 38
ment may follow from the resilience of acorns to sustain damage at an early stage.
39 39
40 Key words: acorns, damage tolerance, dispersal, Quercus, seed size 40
41 41
42 42
43 43
44 Correspondence: Andrew W. Bartlow, 257 South 1400 East, Rm 201, Salt Lake City, Utah 84112, USA. 44
45 45
Email: bartlow.andrew@gmail.com
46 2
46
Present address: Department of Biology, University of Utah, Salt Lake City, Utah, USA.
47 3
47
48 Present address: Center for Environmental Studies and Department of Biology, Virginia Commonwealth University, Richmond, 48
49 Virginia, USA. 49
4
50 Present address: Department of Veterinary Integrative Biosciences, Texas A&M University, College Station, Texas, USA. 50
51 5
Present address: College of Life Sciences, Jiangxi Normal University, Nanchang, Jiangxi, China. 51

A. W. Bartlow et al.
1 INTRODUCTION Robertson et al. 1990; Yi & Zhang 2008; Sage et al. 1

2 2011), but, nevertheless, permit seed growth after nonle- 2
3 Offspring size is a widely-studied life history trait thal damage by a predator. In general, larger seeds may 3
4 with implications for dispersal, reproductive success, better tolerate damage because of larger initial ener- 4
5 population dynamics and community structure (Bernar- gy reserves (Bonfil 1998; Mack 1998; Pizo et al. 2006; 5
6 do 1996; Eriksson & Jakobsson 1998; Leishman 2001). Xiao et al. 2007; Mendoza & Dirzo 2009; Yi & Yang 6
7 A trade-off in resource allocation exists between the 2010; Yang & Yi 2012). Hence, tolerance of such dam- 7
8 benefits of producing few relatively large offspring ver- age, often influenced by seed size, is a potentially signif- 8
9 sus the production of many relatively small offspring icant selective advantage for plant species such as oaks. 9
10 (Smith & Fretwell 1974). This tradeoff concerns both 10
With respect to resource allocation, the production
11 the fitness of the parents and the offspring, which, un- 11
of large seeds is expensive, which translates to few-
12 der many circumstances, may conflict. From the off- 12
er seeds. In the context of significant predation pres-
13 spring’s perspective, large size may be especially ad- 13
sure, producing smaller seeds may be a better solution
14 vantageous for early growth and survival. However, for 14
than producing fewer larger seeds. Moreover, in some
15 parents, producing “many small offspring” rather than 15
circumstances, small seeds may be better dispersers
16 “few large offspring” may be an advantage, depending 16
(e.g. some wind dispersed species: Howe & Smallwood
17 on conditions (Bernardo 1996; Alcántara & Ray 2003; 17
1982) or may have higher survivorship due to the pref-
18 Gómez 2004; Agosta 2008). 18
erences of seed predators (e.g. Bekker et al. 1998; Espe-
19 In plants, larger seeds generally produce larger seed- lta et al. 2009; Bartlow et al. 2011). 19
20 lings with higher relative growth rates, presumably be- 20
In species that produce seeds with large cotyledons,
21 cause of the greater metabolic energy available for es- 21
such as oaks (Quercus spp.) and chestnuts (Castanea
22 tablishment, growth and development (Tripathi & Khan 22
spp.), larger seeds with high energy content are afford-
23 1990; Leishman & Westoby 1994; Seiwa 2000; Green 23
ed a distinct dispersal advantage, especially when fac-
24 & Juniper 2004; Khan 2004; Lehtila & Ehrlen 2005; 24
ing rodent predation (Xiao et al. 2004, 2005; Moore et
25 Du & Huang 2008; Tilki 2010; Elwell et al. 2011). Yet, 25
al. 2007; Chang et al. 2009; Lichti et al. 2017). Such
26 there is increasing evidence for conflicting advantag- 26
selective pressures, coupled with the potential influ-
27 es of large and small seeds during the processes of seed 27
ence of seed size on the ability to tolerate partial dam-
28 dispersal and seed predation. Larger seeds may have a 28
age (Steele et al. 1993) and enhance seedling perfor-
29 dispersal advantage (see below) or higher tolerance of 29
mance (Gómez 2004), may account for larger acorn size
30 post-germination damage (Harms & Dalling 1997; Dal- 30
than is needed for growth in many oak species. Numer-
31 ling & Harms 1999), whereas small seeds may have 31
ous vertebrates (e.g. eastern gray squirrels [Sciurus car-
32 higher survival in the face of seed predation if seed 32
olinensis Gmelin, 1788], blue jays [Cyanocitta cristata
33 predators prefer larger seeds (Alcántara & Ray 2003; 33
Linnaeus, 1758] and grackles [Quiscalus quiscula Lin-
34 Gómez 2004; however, see Moles et al. 2003). 34
naeus, 1758], for example, selectively consume only the
35 Seed size may also play a role in tolerance to seed 35
basal half of acorns (Steele et al. 1993), likely because
36 damage by seed predators. Although it is assumed that 36
of chemical gradients in the acorns that make the bas-
37 seed predators kill seeds in the process of seed con- 37
al half more attractive (Steele et al. 1993; Bogdziewicz
38 sumption, some predators inflict nonlethal damage, al- 38
et al. 2017, 2018; Steele et al. unpubl. data). Predators
39 lowing partially consumed nuts, fruits and seeds to sur- 39
such as weevils, which do not directly affect seed dis-
40 vive and establish (Steele et al. 1993; Branco et al. 40
persal, may also place selective pressure on seed size,
41 2002; Perea et al. 2011; Yang & Yi 2012). Previous 41
energy content and chemical gradients in seeds. Weevils
42 studies suggest that, in some species, seed damage may 42
lay eggs in acorns and the developing larvae consume
43 have minimal effects on seedling performance (Arm- 43
the acorn from the inside. Weevils damage seeds and af-
44 strong & Westoby 1993; Dalling et al. 1997; Branco et 44
fect germination (Yi & Yang 2010). Weevils can also af-
45 al. 2002; Mendoza & Dirzo 2009; Hou et al. 2010; Yi & 45
fect the interactions between vertebrate seed dispers-
46 Yang 2010) and even increase germination rates (Bran- 46
ers and acorns, such as when weevil infestation affects
47 co et al. 2002; Giertych & Suszka 2011). In other spe- 47
the choice to cache or consume the acorn (Steele et al.
48 cies, seed damage may decrease overall seedling height 48
1996). Here, we sought to understand how intraspecif-
49 (Robertson et al. 1990; Mack 1998; Yi & Zhang 2008; 49
ic and interspecific variation in acorn size influence both
50 Giertych & Suszka 2011) and survival (Janzen 1976; 50
seedling performance and tolerance to seed damage in
51 51

Seed size and damage in oaks
1 oaks. minating (Smallwood et al. 2001). The other 3 species 1

2 We specifically hypothesized that in addition to their in our study (white, bur and chestnut oak) are all white 2
3 role in seed germination and early seedling establish- oak species (section Quercus). Acorns of this group typ- 3
4 ment, acorns are highly robust propagules adapted to ically have lower levels of lipids and tannins, and ger- 4
5 sustain significant damage following partial consump- minate in autumn, during or soon after seed fall (Small- 5
6 tion by seed predators. To test this hypothesis, we sim- wood et al. 2001). 6
7 ulated specific damage by seed predators and examined The acorns of bur oak, black oak and chestnut oak 7
8 the functional relationship between seed size and ger- used in this study were purchased from Sheffield’s Seed 8
9 mination/seedling performance during the critical stage Company (Locke, NY, USA) in late autumn 2010, and 9
10 of seedling establishment in 6 species of oaks. We pre- were reported to originate from the ground beneath mul- 10
11 dicted that if the cotyledon functions primarily as an en- tiple trees of each species. We collected acorns of red 11
12 ergy source for germination and growth of young seed- oak, white oak and pin oak from the ground beneath 12
13 lings, we should observe: (i) a strong positive functional trees (n ≥ 8 of each species) in north-eastern Pennsyl- 13
14 relationship between seed size and seedling perfor- vania during the autumn of 2010 (41°14′N, 75°52′W). 14
15 mance; (ii) a strong positive functional relationship be- All acorns were stored in humid conditions at 4°C until 15
16 tween damage and the degree to which performance is the time of experiments. Multiple maternal sources for 16
17 reduced relative to undamaged (control) acorns; and (iii) acorns of each oak species ensured variation in individ- 17
18 an advantage for large seeds (and large-seeded species) ual acorn size and, thus, allowed us to test the effects of 18
19 in response to damage. In contrast, we predicted that if both inter-specific and intra-specific variation in acorn 19
20 the cotyledon serves also as a fruit with excess energy size on performance and tolerance to damage. 20
21 able to be lost to seed predators/dispersers, we should 21
Prior to experiments, composite samples of acorns of
22 observe: (i) significant residual variation in the rela- 22
each species were created, from all source trees of the
23 tionships between seed size and young seedling perfor- 23
same species. Individual acorns of each species used in
24 mance; (ii) a general ability to suffer significant cotyle- 24
the experiment were then randomly selected from these
25 don loss and still survive; and (iii) similar responses to 25
composite samples. Acorns with any signs of damage
26 damage in small-seeded acorn species compared with 26
(e.g. oviposition scars or exit holes of weevils [Curculio
27 large-seeded species. Although not a set of mutually ex- 27
spp.], desiccation of the cotyledon or fungal growth) af-
28 clusive hypotheses, this comparison of the functional 28
ter close inspection were rejected. We chose not to rely
29 roles of the acorn cotyledon as an “energy-reserve” and 29
on flotation of seeds to determine soundness because
30 a “fruit” provides a logical framework for understand- 30
previous experience with this technique proved inade-
31 ing how acorn size relates to competing selective pres- 31
quate for detecting light weevil damage (M.A. Steele,
32 sures that ultimately influence acorn germination and 32
personal observation). Instead, we relied on individu-
33 oak seedling survival and establishment. 33
al manipulation of the acorn by rolling it and applying
34 34
pressure to the outer pericarp (shell), coupled with close
35 35
36
MATERIALS AND METHODS visual inspection of the pericarp.
36
37 Experimental design 37
Study species and acorn collection
38 38
Based on the types of partial damage that acorns are
39 We studied the effect of acorn damage in 6 North 39
reported to experience (primarily rodent, bird and insect
40 American oak species: northern red oak (Quercus ru- 40
damage; Steele et al. 1993, 1998; Perea et al. 2011), we
41 bra), pin oak (Quercus palustris), black oak (Quercus 41
chose 4 experimental treatments representing varying
42 velutina), chestnut oak (Quercus prinus = Quercus mon- 42
degrees and types of cotyledon damage: (i) 25% coty-
43 tana), bur oak (Quercus macrocarpa) and white oak (Q. 43
ledon removal from the basal end of the acorn; (ii) 50%
44 alba). All 6 species are sympatric across much of east- 44
cotyledon removal from the basal half of the acorn; (iii)
45 ern USA, with the exception of bur oak, which is found 45
1 drilled hole (3 mm in diameter and across the width of
46 primarily in midwestern USA. These 6 species belong to 46
the acorn); and (iv) 4 drilled holes (each of which was
47 1 of 2 sections in the genus Quercus. Red, pin and black 47
3 mm in diameter and across the width of the acorn).
48 oak are all red oak species (section Lobatae). Acorns of 48
The 25 and 50% treatments simulated rodent and bird
49 this group have high levels of lipids (attractive to seed 49
damage as reported in the literature (Steele et al. 1993,
50 predators) and tannins (defense against seed predators), 50
1998; Perea et al. 2011), while the drilled holes simulat-
51 exhibit a delayed dormancy and overwinter before ger- 51

1 ed weevil damage. Experimental acorns in which rodent that each tray consisted of 8 blocks of varying treatment 1
2 and bird damage was simulated were cut transversely types. Throughout the experiment, each tray was rotated 2
3 with a rodent guillotine (Harvard Apparatus, item num- every 2 weeks to account for variation in temperature, 3
4 ber: 550020) to remove approximately 25% or 50% of light and any other position effect in the growing room. 4
5 the cotyledon from the basal half of the acorn. Acorns Growing conditions included natural light and constant 5
6 in which we sought to simulate weevil damage were temperature (19–23°C). Each plant was watered with 6
7 drilled with a Dremel (Model 300 with 3-mm drill bit) approximately the same amount of water at the same 7
8 at approximately the mid-point between apical and bas- time every 2–3 days. 8
9 al ends. This allowed us to standardize “weevil” dam- 9
Data collection
10 age among acorns. Based on a long-term dataset mon- 10
11 itoring acorn weevils, the 4-hole treatment represented Germination of acorns and the survival of seedlings 11
12 a high weevil infestation commonly seen in oak pop- throughout the growing period were recorded. Because 12
13 ulations (M.A. Steele, unpublished data). Each drilled we were interested primarily in the effects of damage 13
14 hole resulted in only a single opening in the acorn shell. on germination, emergence and performance of young 14
15 Whole, intact acorns served as controls. The number of seedlings, the seedlings were grown for 2 months and 15
16 acorns in each treatment ranged from 40 to 46. survival was assessed throughout the experiment. Previ- 16
17 Prior to treatment, all acorns were weighed (±0.001g). ously we found that partially damaged acorns often sur- 17
18 To reduce the risk of experimenter-induced microbial in- vive a full growing season if they make it past the first 18
19 fection affecting germination, acorns were wiped clean 2 months of growth (Steele et al., unpublished data). 19
20 with 70% isopropanol before being subjected to treat- At the end of the experiment, seedlings were harvest- 20
21 ment. The blade of the cutting instrument was cleaned ed over a 3-day period (19 April 2011 to 21 April 2011). 21
22 with an alcohol swab between each acorn, and the drill Seedlings were removed from the soil, with all debris 22
23 bit was washed in 70% ethanol prior to treatment of carefully removed from the roots, and were weighed to 23
24 each acorn. These methods do not limit pathogens in- the nearest ±0.001 g. Seedlings were dried for 4 days at 24
25 fecting acorns after processing that may arise through- 40°C and were again weighed to the nearest ±0.001g. 25
26 out the growing period. Any experimental acorns show- This final mass was then divided by the number of days 26
27 ing signs of rot or damage that were revealed during grown to calculate the growth rate, hereafter referred to 27
28 treatment preparation were retained for the experiment as the average daily increase in biomass (ADB). 28
29 because comparable damage in control acorns would 29
30 Data analysis 30
have remained undetected. This prevented the introduc-
31 tion of bias between experimental and control acorns. All statistical analyses were performed using the pro- 31
32 All acorns were weighed after treatment (±0.001g), and gram R (R Core Development Team 2012) and JMP 32
33 this final post-treatment mass was used to quantify the Pro 10 (SAS Institute, Cary, NC, USA). Mean seed size 33
34 percent cotyledon loss due to each treatment. (mass) and standard deviation were calculated for each 34
35 species and compared across species. Kolmogorov– 35
To control for soil, temperature and light differenc-
36 Smirnoff tests were used to determine if the masses of 36
es that may occur under field conditions, acorns were
37 each species were normally distributed. Several species 37
planted in the laboratory and were all subject to the
38 failed to meet assumptions of normality, so the acorn 38
same growing conditions. Acorns were planted imme-
39 masses of all species were log10 transformed. An ANO- 39
diately after damage treatments. All acorns were pro-
40 VA, followed by Tukey’s post-hoc test, was used to as- 40
cessed and planted from 9 to 15 February 2011 and were
41 sess interspecific differences in seed size. 41
planted in individual cells within a seed tray consist-
42 42
ing of 72 cells (5 × 5 cm) arranged in 2 × 4 blocks, with To determine the relationship between seed size and
43 43
each block consisting of 9 cells (3 × 3 cells). Acorns of early seedling performance, separate linear regressions
44 44
the same treatment, but different species, were planted of seed mass versus ADB were performed with the con-
45 45
in a pseudo-random manner within the same tray. The trol acorns of each species. Initial seed mass was log10
46 46
acorns were planted in a medium consisting of a 1:1 ra- transformed, and ADB was square root transformed to
47 47
tio of Pro-Mix (Premier Tech) to commercial topsoil to better approximate normal distributions (a linear mod-
48 48
control for any differences related to soil chemistry that el including all species with interaction terms showed
49 49
may influence acorn growth. After all acorns were plant- an interactive effect of species and initial seed mass on
50 50
ed, the blocks were detached and further randomized so ADB). In cases where relationships appeared curvilin-
51 51

1 ear, a polynomial function was fit to the data using AIC tionally more cotyledon loss because they have high- 1
2 scores to determine the best fit. In addition, separate lo- er initial energy reserves above this threshold. If so, we 2
3 gistic regressions were performed to determine relation- predicted a significant interaction between initial seed 3
4 ships between seed size and survival (germination and mass and damage level in the ANCOVAs, and an in- 4
5 emergence). Again, this was done separately for each creasing advantage (steeper slope) of seed size with a 5
6 species and for control acorns only. As above, initial greater proportional loss of cotyledon (i.e. the strongest 6
7 seed mass was log10 transformed prior to analysis. effect of seed size on performance should have occurred 7
8 Before testing for damage effects, we confirmed that in the 50% damage treatment). Second, to determine 8
9 treatments represented discrete and independent levels if larger-seeded species performed better in each dam- 9
10 of damage by comparing the cotyledon lost in all treat- age treatment, acorn masses (mean ± SE) of the 6 spe- 10
11 ments using ANOVA, followed by Tukey’s post-hoc cies were plotted against the ADB (mean ± SE) and per- 11
12 test. These data were first square root transformed to ap- cent survival. Linear regression was used to determine 12
13 proximate normal distributions. how well the initial seed mass of a species predicted the 13
14
To test the hypothesis that seed size affects robust- ADB and survival across damage treatments. Finally, to 14
15
ness to damage, we analyzed the data in 3 ways, each of determine if small-seeded species suffered a greater rel- 15
16
which provided a complementary test of our predictions. ative reduction in performance, the relationship between 16
17
First, linear and logistic analysis of covariance (AN- damage and the degree to which performance was re- 17
18
COVAs) was used to determine how initial seed mass duced relative to undamaged control acorns was exam- 18
19
and damage treatment predicted ADB and survival. In ined. In each damage treatment, acorn masses (mean 19
20
these models, initial seed mass was first log10 trans- ± SE) of the 6 species were plotted against the corre- 20
21
formed. For the linear models, ADB was square root sponding mean ADB of each treatment divided by the 21
22
transformed. In these linear and logistic regression mod- mean for each control ADB. In addition, the percent 22
23
els, there were significant interactions between species survival of each treatment was divided by the percent 23
24
and treatment. Therefore, ANCOVAs were run separate- survival of the control and plotted against mean acorn 24
25
ly for each species using treatment as the factor and ini- masses. 25
26 26
tial seed mass as the covariate. Because some treatments
27 27
were planted on different days, the number of days each RESULTS
28 28
seedling was grown was included as a second covariate.
29 Variation in acorn mass within and among species 29
If the relationship between performance and damage is
30 is summarized in Table 1. Differences in acorn mass 30
a simple function of the amount of energy reserves, then
31 among species were significant (ANOVA, F5,1198 = 1033, 31
we expected larger seeds to perform better at each lev-
32 P < 0.0001; Table 1). All species significantly differed, 32
el of damage. However, the relationship between per-
33 except for pin oak and white oak acorns (Tukey post- 33
formance and damage may be a more complex function
34 hoc, P = 0.43) In our study, there was a 4-fold differ- 34
of seed size, if, for example, there is a threshold energy
35 ence between the largest chestnut oak acorns (mean ± 35
level needed for successful germination and growth. In
36 SD; 5.90 ± 1.43 g) and the smallest white oak acorns 36
this case, larger seeds may be able to withstand propor-
37 (1.50 ± 0.74 g). 37
38 38
39 39
40 40
41 Table 1 The mean masses and standard deviations of acorns from the 6 oak species. All species were significantly different from 41
42 one another, except Quercus palustris and Quercus alba (Tukey’s post-hoc, P = 0.43). 42
43 Species Sample size Range Mean (g) Standard deviation 43
44 44
45 Chestnut oak (Quercus montana) 198 0.84–10.4 5.90 1.43 45
46 Bur oak (Quercus macrocarpa) 199 2.78–7.41 5.07 0.96 46
47 Red oak (Quercus rubra) 202 2.34–8.97 4.66 1.28 47
48 Black oak (Quercus velutina) 200 1.40–3.45 2.25 0.49 48
49 49
Pin oak (Quercus palustris) 203 1.06–3.14 1.68 0.35
50 50
51 White oak (Quercus alba) 202 0.54–4.31 1.50 0.74 51

Figure 2
1 For control acorns, seed mass was a significant pos- 70 1

2 itive predictor of ADB in each species except black 2
3 oak (Fig. 1). Across species, R2 ranged from reasonably 60 3
Percent cotyledon loss

4 high in black oak (0.74) and white oak (0.73), to mod- 4
5 erate in pin oak (0.39), to 0.11 in bur oak. With respect 50 5
6 to survival (germination and emergence) of the control 6
40
7 acorns, logistic regression revealed a significant positive 7
8 effect of initial seed mass in only pin oak. Black oak 30 8
9 showed a significant negative relationship between seed 9
10 mass and survival. Interestingly, there was no significant 20 10
11 relationship between seed mass and survival in the other 11
12 4 species (Table 2). 10 12
13 Four distinct damage classes were created during the 13
0
14 damage process based on the percentage of cotyledon 14
15 removed (mean ± SE % of cotyledon loss; 1 hole, 2.08%
Control 1 hole 4 holes 25% 50% 15
16 ± 0.093%; 4 holes, 5.06% ± 0.25%; “25%,” 17.54% ± Damage treatment 16
17 0.28%; “50%,” 43.60% ± 0.48%; Fig. 2). These 4 class- 17
Figure 2 The 4 experimental categories showing the amount of
18 es of damage approximated the varying levels of dam- 18
cotyledon removed after acorns were subject to damage. The
19 age we sought and were significantly different from one 19
amount removed was significantly different for each treatment.
20 another (ANOVA, F4,1198 = 6879, P < 0.001; Fig. 2). 20
21
22
Figure 1 21
22
23 23
24 24
25 a b 25
5 2
R = 0.359, P = 0.0005
5
26 26
4 4
27 27
ADB (mg/day)
28 3 3
28
29 2 2 29
30 1 1 2
R = 0.387, P = 0.0003
30
31 31
0 0
32 0.5 0.6 0.7 0.8 0.9 0.0 0.1 0.2 0.3 0.4
32
33 33
34 c d 34
5 5
35 2
35
4 R = 0.737, P < 0.0001 4
36 36
ADB (mg/day)
37 3 3 37
38 38
2
R = 0.327, P = 0.0054
2 2
39 1 1 39
40 40
0 0
41 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
41
42 42
43
e f 43
Figure 1 The linear relationships between 5 5
44 44
initial seed mass and average daily increase 4 4
2
R = 0.728, P = 0.0001
45 45
ADB (mg/day)
in biomass (ADB) of the control acorns of

46 the 6 oak species: (a) red oak, (b) pin oak,
3 3
46
47 (c) black oak, (d) chestnut oak, (e) bur oak 2 2 47
48 and (f) white oak. The ADB was square 1 2
R = 0.11, P = 0.042 1 48
49 root transformed and the acorn mass was 49
0 0
50 log transformed. The ADB increased as 0.5 0.6 0.7 0.8 0.9 0.2 0.0 0.2 0.4 0.6
50
51 seed mass increased in 5 of the species. Acorn mass (g) Acorn mass (g) 51

1 1
2
Control
1 hole
4 holes
Figure 3 2
3 25% cut
50% cut
3
4 4
5 5
6 100
40
6
Percentage of seeds surviving damage

38
7 36 7
8 34 8
9 80 30
32
9
10 10
11 26
26
11
24 24
12 60 25 23 22 12
13 20
13
14 17 14
16
15 40 15
13 15
16
15 14 15
16 16
13
12 12
17 Figure 3 The survival of each species in 9

10 9 17
18 response to the 4 experimental treatments 20 18
and the control acorns, not subject to dam-
6
5
19 19
age. The species are ordered smallest to
20 20
largest by mass (see Table 1). Values indi-
21 0 21
cate numbers of individuals surviving. White oak Pin oak Black oak Red oak Bur oak Chestnut oak
22 22
23 Figure 4 23
24 24
25 Table 2 Summary of generalized linear model with binomi- Red oak 25
26 al distribution of seed size versus survival (germination and 4 Pin oak
26
ADB (mg/day) LS means ± SE
Black oak
27 emergence) for control acorns of each species using the initial Chestnut oak
Bur oak 27
28 seed masses as predictors White oak 28
29 Species Coefficient Degrees of freedom P-value
3 29
30 30
31 Chestnut oak 9.592 1 0.058 31
2
32 Bur oak 1.865e-08 1 1 32
33 Red oak 7.939 1 0.077 33
34 34
Black oak −8.401 1 0.032* 1
35 35
36 Pin oak 11.028 1 0.036* 36
37 White oak 2.3926 1 0.15 Control 1 hole 4 holes 25% 50% 37
38 Treatment 38
* indicates significance at P = 0.05.
39 Figure 4 Least squares (LS) means of the average daily in- 39
40 crease in biomass (ADB; mg/day) ± SE of the 6 species of oaks 40
41 in each damage treatment. ADB was square root transformed. 41
42 Acorns in all 4 of the damage classes germinated and The higher the LS mean for a given species within a treatment, 42
43 survived until harvest (Fig. 3). Bur oak had the highest the faster the average daily increase in biomass. Performance 43
44 survival in every damage class, maintaining >60% sur- generally decreased as damage increased, except for the 25% 44
45 vival even at 50% damage. Four species (red, pin, bur damage treatment. 45
46 and white oak) showed higher survival when 25% of the 46
47 cotyledon was removed than in either of the treatments 47
48 simulating weevil damage (1 and 4 holes), even though separate ANCOVAs for each species. There was no sig- 48
49 far less cotyledon was removed in the weevil treatments. nificant interaction between treatment and initial acorn 49
50 To test for the effects of damage, initial acorn mass, mass for any species, indicating that within treatments, 50
51 and their potential interaction on ADBs, we conducted ADBs responded similarly to seed size (Table 3). There- 51

1 Table 3 Analysis of covariance results using average daily increase in biomass (ADB) as the response variable 1
2 2
Species Source factor Degrees of freedom Type III sum of squares F-statistic P-value
3 3
Chestnut oak Treatment 4 5.83 1.66 0.17
4 4
Initial mass 1 7.34 8.34 0.005*
5 5
Initial mass × treatment 4 5.25 1.49 0.21
6 Days grown 1 5.61 6.38 0.014* 6
7 7
Bur oak Treatment 4 16.13 6.59 <0.001*
8 8
Initial mass 1 5.45 8.91 0.003*
9 9
Initial mass × treatment 4 1.37 0.56 0.69
10 Days grown 1 1.14 1.86 0.17 10
11 11
Red oak Treatment 4 19.27 11.47 <0.001*
12 12
Initial mass 1 6.89 16.41 <0.001*
13 Initial mass × treatment 4 0.64 0.38 0.83 13
14 Days grown 1 0.01 0.03 0.87 14
15 15
Black oak Treatment 4 7.96 5.87 <0.001*
16 Initial mass 1 0.10 0.29 0.59 16
18 Days grown 1 0.42 1.23 0.27 18
19 Pin oak Treatment 4 12.07 8.87 <0.001* 19
20 Initial mass 1 0.68 2.00 0.16 20
22 Days grown 1 0.05 0.15 0.70 22
23 White oak Treatment 4 1.86 2.41 0.062 23
24 Initial mass 1 11.86 6146 <0.001* 24
26 Days grown 1 0.05 0.26 0.62 26
27 Each species was analyzed separately. Shown are the degrees of freedom, type III sum of squares, F-statistics, and P-values of the 27
28 main effects of treatment, initial mass, and days grown with an interaction between seed mass and treatment. * indicates signifi- 28
29 cance at P = 0.05. 29
30 30
31 31
32 32
33 33
34
fore, there was no evidence for an increasing advantage holes drilled). In addition, these 2 species (chestnut oak 34
35
for large seeds when subject to increasing amounts of and black oak) performed better after 50% cotyledon re- 35
36
damage. For 2 species, pin oak and black oak, the main moval than after 25% removal (Fig. 4). 36
37
effect of initial acorn mass was not significant (Table 3). For survival, there was no significant interaction be- 37
38
For the 4 other species, there was a significant main ef- tween treatment and initial acorn mass for any species. 38
39
fect of initial acorn mass on ADB: ADBs increased with This suggests that the relationship between mass and 39
40
increasing seed mass across treatments (Table 3). Fi- survival is similar across damage treatments. In all spe- 40
41
nally, treatment had a significant effect on ADBs for 4 cies but 1 (white oak), there were significant negative 41
42
of the 6 species (Table 3, Fig. 4). In general, the control effects of treatment on acorn survival (Table 4). The ini- 42
43
group (intact acorns) of each species performed as well tial seed mass successfully predicted the survival of 4 43
44
or better than damage treatments. For the 4 species with of the species; the seed masses of red oak and bur oak 44
45
a significant effect of treatment, there was no pattern re- were not significant predictors of their survival. 45
46
garding seedling performance as damage increased. All Average seed mass was a significant predictor of 46
47
6 species showed a decrease in performance when sub- mean ADB across species for each treatment and the 47
48
ject to weevil damage (Fig. 4). However, all species ex- control (Fig. 5). The 3 largest oak species produced 48
49
cept black oak and chestnut oak appeared to respond acorns that had the highest ADB in all 4 damage class- 49
50
better when 25% of the cotyledon was removed com- es and the control treatment. However, acorn mass was 50
51
pared to treatments simulating weevil damage (1 and 4 a poor predictor of acorn survival in each damage treat- 51

Figure 5
1 a b 8
1
2 R2 = 0.99 P < 0.0001 2
Mean ADB ± SE (mg/day)

10 R2 = 0.93 P = 0.0019 7
3 8
6 3
4 5 4
5 6 4
5
3
6 4 6
2
7 2 1
7
8 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 8
9 9
10 c 12 d 10
5
11 11

10
12 8
4
12
13 6
3 13
14 2 2
R = 0.97 P = 0.0003 14
4 R2 = 0.83 P = 0.012
15 1
15
2
16 16
17 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 17
18 Figure 5 The mean seed size (± SE) and e 16
Mean acorn mass ± SE (g)
18
the mean absolute average daily increase
19 Red oak 19
in biomass (ADBs; ± SE) for the 6 species 14

20 Pin oak 20
of oaks in the 4 damage treatments and the 12
21 Black oak 21
control treatment. In each treatment, the 10
22 22
mean seed mass was a significant predictor 8 Chestnut oak
23 of ADB among species; the largest species
R2 = 0.83 P = 0.011 23
6 Bur oak
24 produced seedlings with the highest ADB. 24
White oak
25 (a) 1 hole, (b) 4 holes, (c) 25% damage, (d)
4
0 1 2 3 4 5 6 7
25
26 50% damage, (e) control. Mean acorn mass ± SE (g) 26
27 27
28 Table 4 Summary of generalized linear models with binomial distributions using survival (germination and emergence) as the re- 28
29 sponse variable with 2 predictor variables and the interaction 29
30 30
31 Species Source factor Degrees of freedom χ2 P-value 31
32 Chestnut oak Treatment 4 12.59 0.014* 32
33 Initial mass 1 11.61 0.001* 33
34 Initial mass × treatment 4 2.60 0.63 34
35 Bur oak Treatment 4 23.45 <0.001* 35
36 Initial mass 1 < 0.001 0.99 36
38 Red oak Treatment 4 33.62 <0.001* 38
39 Initial mass 1 1.83 0.18 39
41 Black oak Treatment 4 20.10 <0.001* 41
42 Initial mass 1 24.03 <0.001* 42
44 Pin oak Treatment 4 40.41 <0.001* 44
45 Initial mass 1 9.80 0.0017* 45
47 White oak Treatment 4 4.86 0.30 47
48 Initial mass 1 10.64 0.001* 48
50 Each species was analyzed separately. Reported are the degrees of freedom, χ -statistics, and the P-values for the main effects of 50
2
51 initial mass, treatment, and their interaction. Asterisks indicate significance (P < 0.05). 51

Figure 6
1 1
a b
2 1.0 1.0 2
3 3
Proportion surviving
0.8 R2 = 0.62 P = 0.062 0.8
R2 = 0.44 P = 0.15
4 0.6 0.6 4
5 0.4 0.4 5
6 0.2 0.2
6
7 7
0.0 0.0
8 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 8
9 9
10 c d 10
1.0 1.0
11 R2 = 0.42 P = 0.17 R2 = 0.034 P = 0.73 11
0.8 0.8
12 12
0.6 0.6
13 13
0.4 0.4
14 14
15 0.2 0.2
15
16 0.0 0.0 16
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
17 Mean acorn mass ± SE (g)
17
18 18
e
19 1.0
Red oak 19
20 Figure 6 The mean seed size of the 6 spe- 0.8

Pin oak 20
21 cies (± SE) of oaks and the proportion sur- 0.6
Black oak 21
22 viving in each damage treatment and the 0.4
Chestnut oak 22
23 control treatment: (a) 1 hole, (b) 4 holes, (c) 0.2
R2 = 0.19 P = 0.39 Bur oak 23
24 25% damage, (d) 50% damage and (e) con- White oak 24
Figure 7
0.0
25 trol. Overall, seed size did a poor job pre- 0 1 2 3 4 5 6 7 25
26 dicting survival. Mean acorn mass ± SE (g)
26
27 27
28 28
a b
Relative ADB (treatment/control)
29 29
1.0 1.0
30 30
0.8 0.8 R2 = 0.67 P = 0.047
31 31
0.6 0.6
32 32
0.4 0.4
33 R2 = 0.081 P = 0.58 33
0.2 0.2
34 34
35 35
0.0 0.0
1 2 3 4 5 6 7 1 2 3 4 5 6 7
36 36
37 c d 37
Relative ADB (treatment/control)
38 1.0 1.0
38
39 0.8 0.8 R2 = 0.20 P = 0.37
39
40 0.6 0.6
40
41 0.4 R2 = 0.27 P = 0.288 0.4
41
Figure 7 The mean seed size and the rel-
42 0.2 0.2 42
ative average daily increase in biomass
43 0.0 0.0 43
(ADB) defined as the mean treatment
44 44
1 2 3 4 5 6 7 1 2 3 4 5 6 7
ADB divided by the mean control ADB Mean acorn mass ± SE (g) Mean acorn mass ± SE (g)
45 of each species in each treatment: (a) 1 45
46 hole, (b) 4 holes, (c) 25% damage and (d)
Red oak 46
47 50% damage. When scaled relative to con-
Pin oak 47
48 trol acorns, there was no significant rela- Black oak 48
49 tionship between mean seed size and mean Chestnut oak 49
50 ADB in any damage treatment, except the Bur oak 50
51 4 hole treatment. White oak 51

!
Figure 8 Seed size and damage in oaks
1 1
a b
Relative survival (treatment/control)

2 1.2
1.0
2
3 1.0
0.8
3
4 0.8
4
0.6
5 0.6
5
0.4
6 0.4
R2 = 0.58 P = 0.077 6
0.2 R2 = 0.31 P = 0.25
7 0.2
7
8 0.0 0.0
8
1 2 3 4 5 6 7 1 2 3 4 5 6 7
9 9
10 c d 10
Relative survival (treatment/control)

1.2
11 1.0
1.0 11
R2 = 0.18 P = 0.407
12 0.8
0.8 12
13 0.6
0.6 13
14 0.4 0.4 14
15 0.2
R2 = 0.10 P = 0.54
0.2 15
16 0.0 0.0 16
17 Figure 8 The mean seed size and the rela- 1 2 3 4 5 6 7 1 2 3 4 5 6 7 17
18 tive survival (proportion of treatment sur- Mean acorn mass ± SE (g) Mean acorn mass ± SE (g)
18
viving divided by the proportion of control
19 19
surviving) of each species in each treat-
20 Red oak 20
ment(a) 1 hole, (b) 4 holes, (c) 25% dam-
21 Pin oak 21
age and (d) 50% damage. Survival relative
22 Black oak 22
to the control acorns was not dependent on
23 mean seed size. The relative reduction in Chestnut oak 23
24 survival was similar across all species re- Bur oak 24
25 gardless of size. White oak 25
26 26
27 27
28 28
29 ment (Fig. 6); none of the relationships were significant. further supported by studies suggesting that a suite of 29
30 When damaged acorns were analyzed relative to con- acorn characteristics may divert damage by avian and 30
31 trol acorns, only 1 treatment (4 holes) showed a signifi- rodent acorn predators and dispersers, and even insect 31
32 cant positive relationship between mean species-specif- seed predators, away from the apical (embryo-contain- 32
33 ic acorn mass and mean ADB (Fig. 7). Likewise, there ing) end of the acorn (Steele et al. 1993, 1998). When 33
34 were no significant relationships between mean initial acorns are abundant, these predators often inflict par- 34
35 mass and survival relative to controls (Fig. 8). There- tial damage to only the basal end of acorns (Steele et al. 35
36 fore, small-seeded species did not suffer a greater rela- 1998; Perea et al. 2011), allowing these partially dam- 36
37 tive reduction in performance compared to large-seeded aged seeds to still germinate and potentially establish 37
38 species in response to damage, except in the treatment (Steele et al. 1993; Steele et al. unpubl. data). Long- 38
39 simulating heavy weevil infestation (4-holes). term studies, for example, indicate that weevil larvae are 39
40 found significantly more often in the basal half of the 40
41 acorn compared with the apical half closer to the seed 41
DISCUSSION
42 (Steele et al. unpublished). Acorn characteristics consid- 42
43 The acorn cotyledon, although certainly critical for ered central to influencing this partial acorn consump- 43
44 seedling survival, establishment and growth, also serves tion are chemical gradients, including higher levels of 44
45 a secondary role of manipulating the dispersal process phenolic defense compounds (tannins) in the apical end 45
46 (Steele & Smallwood 2002; Moore et al. 2007; Steele et of the acorn, near the embryo; and higher levels of nu- 46
47 al. 2014). However, as we demonstrate here, the acorn trients (some minerals and lipids; Steele et al. unpub- 47
48 also serves yet another function: to tolerate partial dam- lished) at the basal end. We suggest that these chemical 48
49 age by potential seed predators. gradients direct partial seed damage away from the em- 49
50 Our argument that the cotyledon in the acorn func- bryo, thereby allowing seed survival (germination and 50
51 tions in part to promote tolerance of seed predators is emergence) and young seedling establishment (Steele et 51

1 al. 1993; Steele et al. unpubl. data). As we show here, ing prior to treatment. Sanitizing prior to treatment lim- 1
2 at least 6 oak species can germinate and establish even its experimenter-induced pathogens, but does not limit 2
3 when 50% of the cotyledon is removed from the basal pathogen infestation that may occur throughout the en- 3
4 half of the acorn. Although such damage may somewhat tire growing period. We suggest that this difference may 4
5 reduce seedling performance, we postulate that this may result from the greater exposure of the cotyledon to air, 5
6 represent a critical strategy for tolerating seed predation. which likely results in oxidation of tannins to more tox- 6
7 In the life of an oak (>150 years in some species), such ic quinones (Appel 1993). Such a reaction may explain 7
8 a strategy only requires occasional establishment suc- why “half-eaten acorns” are not necessarily more sus- 8
9 cess to significantly influence lifetime reproductive suc- ceptible to pathogens and should be tested in the future. 9
10 cess. Thus, our results, especially in the context of pre- In addition, treatment had an effect on performance, 10
11 vious studies, support the hypothesis that the cotyledon with lower ADBs and survival in damage treatments 11
12 acts as both a fruit and an energy reserve. (Fig. 4). However, there was no evidence of an increas- 12
13 We predicted that if acorns, indeed, serve as a fruit ing advantage for larger seeds with increased damage 13
14 for dispersal and also to tolerate damage then we should (i.e. there were no significant interactions between ini- 14
15 observe: (i) significant residual variation in the relation- tial acorn mass and treatment) and therefore, no evi- 15
16 ships between seed size and seedling performance; (ii) a dence for a threshold level of energy needed for growth 16
17 general ability to suffer significant damage and still sur- and survival. Although it seems likely that some mini- 17
18 vive; and (iii) similar performance in response to dam- mal amount of cotyledon is ultimately needed for suc- 18
19 age between small and large-seeded species. We ob- cessful germination, we did not detect it even after re- 19
20 served that numerous individuals of all species were moving approximately 50% of the cotyledon from 20
21 able to germinate and grow into young seedlings with acorns over a 4-fold size range. 21
22 nearly half of the initial cotyledon removed. Second, 22
At the interspecific level, large-seeded species (red
23 among surviving acorns, anywhere from 26% to 89% of 23
oak, bur oak and chestnut oak) had higher absolute
24 the variation in average daily increase in biomass (ABD) 24
ADBs than small-seeded species (white oak, pin oak
25 was unexplained by initial acorn mass. Thus, the expec- 25
and black oak) in all treatments (damage and control). In
26 tation that larger acorns correspond to larger seedlings 26
contrast, there were no significant interspecific relation-
27 was only partially supported by these data. Moreover, in 27
ships between seed size and survival in any treatment or
28 black oak, intermediate-sized acorns produced the larg- 28
the control. When scaled relative to control acorns, all
29 est seedlings and initial acorn mass was negatively re- 29
species responded similarly to damage (growth: Fig. 7;
30 lated to acorn survival, a result that is unexpected. In all 30
survival: Fig. 8). Small-seeded species did not suffer a
31 other species, “bigger” was “better,” but the explanatory 31
greater proportional reduction in performance compared
32 power of initial acorn size ranged from high in pin oak 32
to large-seeded species, except in the 4-hole treatment
33 and white oak to very low in bur oak. Seed mass was 33
for relative ADB (Fig. 7).
34 not a significant factor that explained survival in 4 of 34
35 Collectively, these results demonstrate that a signifi- 35
the 6 species (Table 2). cant portion of the energy packaged in acorns is not es-
36 36
Results of the ANCOVAs indicated that larger seeds sential for early seedling development and survival.
37 37
performed better in each damage treatment, with a sig- Based on these results, we suggest that acorns act in part
38 38
nificant main effect of initial mass on ADBs for 4 of the as a fruit to attract potential dispersers and to sustain
39 39
6 species and a significant effect on survival in 4 of the partial seed damage by seed predators. Up to 50% (and
40 40
6 species. In chestnut oak and white oak, treatment was possibly more) of the cotyledon reserve is not essential
41 41
not significant in explaining ADB. Some seedlings perfor acorn germination and early seedling survival, even
42 42
formed better when 25% of the cotyledon was removed though performance may be reduced when this amount
43 43
compared to when 4 holes were drilled (Fig. 3). Cut- of cotyledon is lost.
44 44
ting acorns therefore may have functionally different ef- Several oak species found worldwide are now known
45 45
fects than drilling into acorns. Observations of our dam- to tolerate insect and rodent damage and still germi-
46 46
aged acorns suggested that those in which we simulated nate and produce seedlings (Steele et al. 1993; Bon-
47 47
partial damage by birds and rodents (removal of 25% fil 1998; Xiao et al. 2007; Hou et al. 2010; Yi & Yang
48 48
or 50% of the basal portion) seemed to show far less in- 2010; Giertych & Suszka 2011; Perea et al. 2011; this
49 49
festation of pathogens (fungi and mold) than those in study). Furthermore, at least 2 species of white oaks in
50 50
which we simulated weevil damage, despite sanitiz- North America (Quercus alba and Q. montana) can tol-
51 51

1 erate the complete removal of acorns from developing dispersed greater distances and to presumably better 1
2 taproots, suggesting that the acorn is not essential for cache sites than large acorns that must be dispersed in- 2
3 growth and survival once the taproot reaches a minimal dividually (Bartlow et al. 2011). Such gape-limited cor- 3
4 threshold length (Yi et al. 2012, 2013). These observa- vids and smaller rodents may, therefore, exert disruptive 4
5 tions, coupled with those reported herein, suggest that selection on seed size, with the other factors discussed 5
6 acorns are tremendously robust and well adapted for tol- above favoring larger seed size. In the present study, all 6
7 erance of seed and seedling damage. 6 oak species are consumed and dispersed by rodents, 7
8 In the current experiments, the larger seeded species while the 3 small-seeded species (pin oak, black oak 8
9 had higher ADBs in each damage treatment. Therefore, and white oak) are selectively consumed and dispersed 9
10 a larger-seeded oak species would have an advantage by jays as well (Darley-Hill & Johnson 1981; Moore & 10
11 over a smaller-seeded species when subject to similar Swihart 2006). 11
12 damage, given the same growing conditions. Howev- Partial acorn consumption and subsequent tolerance 12
13 er, this does not appear to be the case in terms of surviv- to damage by seed consumers may be more important 13
14 al; large and small-seeded species showed a similar sur- than previously thought and may play a major role in 14
15 vival response to damage. The apparent advantage of the dispersal, establishment and regeneration of oaks. 15
16 large-seeded species, in terms of ADB, does not seem Why the 6 oak species in the current study had different 16
17 to follow from the phylogeny of these 6 species. Of the responses to damage is unclear. Future studies that rep- 17
18 3 large-seeded oaks, 2 are in the white oak section, and licate field conditions, especially stressful conditions, 18
19 the other is a red oak. Although the white oak acorns in should be considered to further elucidate the importance 19
20 our studies were small, this species often produces sub- of cotyledon reserves to developing seedlings. In addi- 20
21 stantially larger acorns in more southerly portions of its tion, future work should consider the functional roles of 21
22 geographic range. acorns as fruit and/or energy reserves to understand how 22
23 seed size influences pre-dispersal and post-dispersal tol- 23
Whereas variation in acorn size may in part result
24 erance to damage, seed dispersal, and seedling estab- 24
from the need for acorns to tolerate damage and still
25 lishment for other oak species worldwide. 25
survive, acorn size is also known to directly influence
26 26
dispersal by rodents and corvids. Rodents, for exam-
27 27
28
ple, are known to selectively cache larger seeds (includ- ACKNOWLEDGMENTS 28
ing acorns) and disperse them significantly greater dis-
29 This study was supported by funding from the How- 29
tances than smaller seeds of the same species (Jansen
30 ard Hughes Medical Institute to M.A.S., the Natu- 30
et al. 2004; Xiao et al. 2005; Moore et al. 2007; Wang
31 ral Science Foundation of China (No. 31172101) to 31
& Chen 2009; Lichti et al. 2017). Recently it has also
32 X.F.Y., the National Basic Research Program of China 32
been shown that larger acorns are selectively dispersed
33 (No. 2007CB109100) to X.F.Y., and the H. Fenner Re- 33
by Eastern gray squirrels (Sciurus carolinensis) to more
34 search Fund of Wilkes University (M.A.S.). M.A.S also 34
secure locations in open sites where the probability of
35 recognizes support of a Bullard Fellowship from Har- 35
cache pilferage is reduced (Steele et al. 2014), and the
36 vard Forest, Harvard University and the National Sci- 36
probability of seedling establishment, if not recovered
37 ence Foundation (DEB 15556707). Support during final 37
by the cache owner, is higher than smaller acorns of
38 preparation of the manuscript was provided by the Na- 38
both the same and different species (Steele et al. 2014).
39 tional Science Foundation Graduate Research Fellow- 39
40 If acorn size is favored for dispersal, growth and par- 40
ship Program to A.W.B. and R.C. We also thank S. Ma-
41 tial damage by seed predators, then why are small- 41
rino for assistance with experiments and data collection.
42 er-seeded oak species still common? The answer to 42
The authors have no industry links or affiliations nor
43 this may in part follow from the classic trade-off be- 43
conflicts of interest.
44 tween offspring size and number. However, it may also 44
45 in part follow from the size preferences of other disper- 45
sal agents, such as jays, which are gape-limited and pre- REFERENCES
46 46
47 fer smaller-seeded acorns for dispersal (Moore & Swi- Agosta SJ (2008). Fitness consequences of host use in 47
48 hart 2006; Bartlow et al. 2011). Such gape limitation the field: Temporal variation in performance and a 48
49 also favors multiple-prey loading by jays in which small life history tradeoff in the moth Rothschildia lebeau 49
50 acorns are swallowed and placed in the crop. These mul- (Saturniidae). Oecologia 157, 69–82. 50
51 tiple loads (up to 8 acorns for 1 dispersal event) are then 51

1 Alcantara JM, Rey PJ (2003). Conflicting selection pres- Dalling JW, Harms KE, Aizprua R (1997). Seed dam- 1
2 sures on seed size: Evolutionary ecology of fruit size age tolerance and seedling respouting ability of Prio- 2
3 in a bird-dispersed tree, Olea europaea. Journal Evo- ria copaifera in Panamá. Journal of Tropical Ecology 3
4 lutionary Biology 16, 1168–76. 13, 481–90. 4
5 Appel HM (1993). Phenolics in ecological interactions: Darley-Hill S, Johnson WC (1981). Acorn dispersal 5
6 The importance of oxidation. Journal of Chemical by the blue jay (Cyanocitta cristata). Oecologia 50, 6
7 Ecology 19, 1521–52. 231–2. 7
8 8
Armstrong DP, Westoby M (1993). Seedlings from large Du Y, Huang Z (2008). Effects of seed mass and emer-
9 9
seeds tolerate defoliation better: A test using phylo- gence time on seedling performance in Castanop-
10 10
genetically independent contrasts. Ecology 74, 1092– sis chinensis. Forest Ecology and Management 255,
11 2495–501. 11
100.
12 12
Bartlow AW, Kachmar M, Lichti N, Swihart RK, Strat- Elwell AL, Gronwall DS, Miller ND, Spalding EP,
13 13
ford JA, Steele MA (2011). Does multiple seed load- Durham Brooks TL (2011). Separating parental en-
14 14
ing in blue jays result in selective dispersal of smaller vironment from seed size effects on next generation
15 15
acorns? Integrative Zoology 6, 235–43. growth and development in Arabidopsis. Plant, Cell
16 16
Bernardo J (1996). The particular maternal effect of & Environment 34, 291–301.
17 17
18 propagule size, especially egg size: Patterns, models, Eriksson OVE, Jakobsson A (1998). Abundance, distri- 18
19 quality of evidence and interpretations. American Zo- bution and life histories of grassland plants: a com- 19
20 ologist 36, 216–36. parative study of 81 species. Journal of Ecology 86, 20
21 Bekker RM, Bakker JP, Grandin U et al. (1998). Seed 922–33. 21
22 size, shape and vertical distribution in the soil: In- Espelta JM, Bonal R, Sánchez-Humanes B (2009). 22
23 dicators of seed longevity. Functional Ecology 12, Pre-dispersal acorn predation in mixed oak forests: 23
24 834–42. interspecific differences are driven by the interplay 24
25 Bogdziewicz M, Crone EE, Steele MA, Zwolak R among seed phenology, seed size and predator size. 25
26 (2017). Effects of nitrogen deposition on reproduc- Journal of Ecology 97, 1416–23. 26
27 tion in a masting tree: Benefits of higher seed pro- Giertych MJ, Suszka J (2011). Consequences of cutting 27
28 duction are trumped by negative biotic interactions. off distal ends of cotyledons of Quercus robur acorns 28
29 Journal of Ecology 105, 310–20. before sowing. Annals of Forest Science 68, 433–42. 29
30 Gómez JM (2004). Bigger is not always better: Conflict- 30
Bogdziewicz M, Bonal R, Espelta JM, Kalemba EM,
31 ing selective pressures on seed size in Quercus ilex. 31
Steele MA, Zwolak R (2018). Invasive oaks escape
32 Evolution 58, 71–80. 32
pre-dispersal insect seed predation and trap enemies
33 Green PT, Juniper PA (2004). Seed-seedling allometry 33
in their seeds. Integrative Zoology 13, 228-37.
34 in tropical rain forest trees: Seed mass-related pat- 34
35
Bonfil C (1998). The effects of seed size, cotyledon 35
reserves, and herbivory on seedling survival and terns of resource allo9cation and the ‘reserve effect’.
36 Journal of Ecology 92, 397–408. 36
37
growth in Quercus rugosa and Q. laurina (Fagaceae). 37
American Journal of Botany 85, 79–87. Harms KE, Dalling JW (1997). Damage and herbivo-
38 38
Branco M, Branco C, Merouani H, Almeida MH (2002). ry tolerance through resprouting as an advantage of
39 39
Germination success, survival and seedling vigour of large seed size in tropical trees and lianas. Journal of
40 40
Quercus suber acorns in relation to insect damage. Tropical Ecology 13, 617–21.
41 41
42 Forest Ecology and Management 166, 159–64. Hou XG, Yi X, Yang Y, Liu W (2010). Acorn germina- 42
Chang G, Xiao Z, Zhang Z (2009). Hoarding decisions tion and seedling survival of Q. variabilis: Effects of
43 43
by Edward’s long-tailed rats (Leopoldamys edwardsi) cotyledon excision. Annals of Forest Science 67, 711.
44 44
45 and South China field mice (Apodemus draco): The Howe HF, Smallwood J (1982). Ecology of seed disper- 45
46 responses to seed size and germination schedule in sal. Annual Review of Ecology and Systematics 13, 46
47 acorns. Behavioural Processes 82, 7. 201–28. 47
48 Dalling JW, Harms KE (1999). Damage, tolerance and Jansen PA, Bongers F, Hemerik L (2004). Seed mass 48
49 cotyledonary resource use in the tropical tree Gus- and mast seeding enhance dispersal by a neotropical 49
50 tavia superba. Oikos 85, 257–64. scatter-hoarding rodent. Ecological Monographs 74, 50
51 569–89. 51

1 Janzen DH (1976). Reduction of Mucuna andreana (Le- R Development Core Team (2012). R: A language and 1
2 guminosae) seedling fitness by artificial seed damage. environment for statistical computing. R Founda- 2
3 Ecology 57, 826–8. tion for Statistical Computing, Vienna, Austria. ISBN 3
4 Khan ML (2004). Effects of seed mass on seedling suc- 3-900051-07-0, URL http://www.R-project.org/. 4
5 cess in Artocarpus heterophyllus L., a tropical tree Robertson AI, Giddins R, Smith TJ (1990). Seed preda- 5
6 species on north-east India. Acta Oecologia 25, 103– tion by insects in tropical mangrove forests: Extent 6
7 10. and effects on seed viability and the growth of seed- 7
8 lings. Oecologia 83, 213–9. 8
Lehtila K, Ehrlen J (2005). Seed size as an indicator of
9 9
seed quality: A case study of Primula veris. Acta Oe- Sage RD, Koenig WD, McLaughlin BC (2011). Fitness
10 10
cologia 28, 207–12. consequences of seed size in the valley oak Quercus
11 11
Leishman MR (2001). Does the seed size/number trade- lobata Neé (Fagaceae). Annals of Forest Science 68,
12 12
off model determine plant community structure? An 477–84.
13 13
14 assessment of the model mechanisms and their gen- Seiwa K (2000). Effects of seed size and emergence 14
15 erality. Oikos 93, 294–302. time on tree seedling establishment: Importance of 15
16 Leishman MR, Westoby M (1994). The role of seed size developmental constraints. Oecologia 123, 208–15. 16
17 in seedling establishment in dry soil conditions – Ex- Smallwood PD, Steele MA, Faeth SH (2001). The ul- 17
18 perimental evidence from semi-arid species. Journal timate basis of the caching preferences of rodents, 18
19 of Ecology 82, 249–58. and the oak-dispersal syndrome: tannins, insects, and 19
20 seed germination. American Zoologist 41, 840–51. 20
Lichti NI, Steele MA, Swihart RK (2017). Seed fate and
21 decision-making processes in scatter-hoarding ro- Smith CC, Fretwell SD (1974). The optimal balance be- 21
22 dents. Biological Reviews 92, 474–504. tween size and number of offspring. The American 22
23 Naturalist 108, 499–506. 23
Mack AL (1998). An advantage of large seed size: Tol-
24 Steele MA, Smallwood PD (2002). Acorn dispersal by 24
erating rather than succumbing to seed predators.
25 birds and mammals. In: McShea WJ, Healy WM, 25
Biotropica 30, 604–8.
26 eds. Oak Forest Ecosystems: Ecology and Manage- 26
27 Mendoza E, Dirzo R (2009). Seed tolerance to preda- 27
ment for Wildlife. Johns Hopkins University Press,
28 tion: Evidence from the toxic seeds of the buckeye 28
Baltimore, pp 182–95.
29 tree (Aesculus californica: Sapindaceae). American 29
Journal of Botany 96, 1255–61. Steele MA, Knowles T, Bridle K, Simms EL (1993).
30 30
Tannins and partial consumption of acorns: Impli-
31 Moles AT, Warton DI, Westoby M (2003). Do cations for dispersal of oaks by seed predators. The
31
32 small-seeded species have higher survival through American Midland Naturalist 130, 229–38.
32
33 seed predation than large-seeded species? Ecology 33
Steele MA, Hadj-Chikh LZ, Hazeltine J (1996). Cach-
34 84, 3148–61. 34
35
ing and feeding decisions by Sciurus carolinen- 35
Moore JE, Swihart RK (2006). Nut selection by captive sis: Responses to weevil-infested acorns. Journal of
36 blue jays: Importance of availability and implications 36
37
Mammalogy 77, 305–14. 37
for seed dispersal. Condor 108, 377–88.
38 Steele MA, Gavel K, Bachman W (1998). Dispersal of 38
Moore JE, McEuen AB, Swihart RK, Contreras TA, half-eaten acorns by gray squirrels: Effects of physi-
39 39
Steele MA (2007). Determinants of seed removal dis- cal and chemical seed characteristics. In: Steele MA,
40 40
tance by scatter-hoarding rodents in deciduous for- Merritt JF, Zegers DA, eds. Ecology and Evolution-
41 41
ests. Ecology 88, 2529–40. ary Biology of Tree Squirrels. Virginia Museum of
42 42
43 Perea R, San Miguel A, Gil L (2011). Leftovers in seed Natural History, Martinsville, pp. 223–32. 43
44 dispersal: Ecological implications of partial seed con- Steele MA, Bugdal M, Yuan A et al. (2011). Cache 44
45 sumption for oak regeneration. Journal of Ecology placement, pilfering, and a recovery advantage in a 45
46 99, 194–201. seed-dispersing rodent: Could predation of scatter 46
47 Pizo MA, Von Allmen C, Morellato PC (2006). Seed hoarders contribute to seedling establishment? Acta 47
48 size variation in the palm Euterpe edulis and the ef- Oecologia 37, 554–60. 48
49 fects of seed predators on germination and seedling Steele MA, Contreras TA, Hadj-Chikh LZ, Agosta SJ, 49
50 survival. Acta Oecologia 29, 311–5. Smallwood PD, Tomlinson CN (2014). Do scatter 50
51 51

1 hoarders trade off increased predation risks for low- Xiao Z, Harris MK, Zhang Z (2007). Acorn defenses 1
2 er rates of cache pilferage? Behavioral Ecology 25, to herbivory from insects: Implications for the joint 2
3 206–15. evolution of resistance, tolerance and escape. Forest 3
4 Tilki F (2010) Influence of acorn size and storage dura- Ecology and Management 238, 302–8. 4
5 tion on moisture content, germination and survival of Yang Y, Yi X (2012). Partial acorn consumption by 5
6 Quercus petraea (Mattuschka). Journal of Environ- small rodents: Implication for regeneration of white 6
7 mental Biology 31, 325–8. oak, Quercus mongolica. Plant Ecology 213, 197– 7
8 205. 8
Tripathi RS, Khan ML (1990). Effects of seed weight
9 9
and microsite characteristics on germination and Yi XF, Yang YQ (2010). Large acorns benefit seedling
10 10
seedling fitness in two species of Quercus in a sub- recruitment by satiating weevil larvae in Quercus ali-
11 11
tropical wet Hill forest. Oikos 57, 289–96. ena. Plant Ecology 209, 291–300.
12 12
13 Wang B, Chen J (2009). Seed size, more than nutrient Yi XF, Zhang ZB (2008). Influence of insect-infested 13
14 or tannin content, affects seed caching behavior of cotyledons on early seedling growth of Mongolian 14
15 a common genus of Old World rodents. Ecology 90, oak, Quercus mongolica. Photosynthetica 46, 139– 15
16 3023–32. 42. 16
17 Xiao Z, Zhang Z, Wang Y (2004). Dispersal and ger- Yi X, Yang Y, Curtis R, Bartlow AW, Agosta SJ, Steele 17
18 mination of big and small nuts of Quercus serrata in MA (2012). Alternative strategies of seed predator 18
19 a subtropical broad-leaved evergreen forest. Forest escape by early-germinating oaks in Asia and North 19
20 Ecology and Management 195, 141–50. America. Ecology and Evolution 2, 487–92. 20
21 Xiao Z, Zhang Z, Wang Y (2005). Effects of seed size Yi X, Curtis R, Bartlow AW, Agosta SJ, Steele MA 21
22 on dispersal distance in five rodent-dispersed faga- (2013). Ability of chestnut oak to tolerate acorn prun- 22
23 ceous species. Acta Oecologia 28, 221–9. ing by rodents: The role of the cotyledonary petiole. 23
24 Naturwissenschaften 100, 81–90. 24
25 25
26 26
27 27
29 Bartlow AW, Agosta SJ, Curtis R, Yi X, Steele MA (2018). Acorn size and tolerance to seed predators: the mul- 29
30 tiple roles of acorns as food for seed predators, fruit for dispersal and fuel for growth. Integrative Zoology 13, 30
31 251–66. 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Role of seed size, phenology, oogenesis and host distribution in the 6
7
8
9 specificity and genetic structure of seed weevils (Curculio spp.) in 8
9
10 10
11 mixed forests 11
12 12
13 13
14 1,2 3,4
Harold ARIAS-LECLAIRE Raúl BONAL, Daniel GARCÍA-LÓPEZ and Josep Maria 5 14
15 15
16 ESPELTA1 16
17 1 2
CREAF, Centre for Ecological Research and Forestry Applications, Cerdanyola del Vallès, 08193, Spain, Exact and Natural 17
18 Sciences School, State Distance University of Costa Rica, UNED, Mercedes de Montes de Oca, San José, Costa Rica, 3Forest 18
19 4 19
Research Group, INDEHESA, University of Extremadura, Plasencia, Spain, DITEG Research Group, University of Castilla-La
20 5
20
Mancha, Toledo, Spain and Zoology Department, Autonomous University of Barcelona, Cerdanyola del Vallès, Spain
21 21
22 22
23 23
24 Abstract 24
25 25
26 Synchrony between seed growth and oogenesis is suggested to largely shape trophic breadth of seed-feeding in- 26
27 sects and ultimately to contribute to their co-existence by means of resource partitioning or in the time when in- 27
28 festation occurs. Here we investigated: (i) the role of seed phenology and sexual maturation of females in the 28
29 host specificity of seed-feeding weevils (Curculio spp.) predating in hazel and oak mixed forests; and (ii) the 29
30 consequences that trophic breadth and host distribution have in the genetic structure of the weevil populations. 30
31 DNA analyses were used to establish unequivocally host specificity and to determine the population genetic 31
32 structure. We identified 4 species with different specificity, namely Curculio nucum females matured earlier and 32
33 infested a unique host (hazelnuts, Corylus avellana) while 3 species (Curculio venosus, Curculio glandium and 33
34 Curculio elephas) predated upon the acorns of the 2 oaks (Quercus ilex and Quercus pubescens). The high spec- 34
35 ificity of C. nucum coupled with a more discontinuous distribution of hazel trees resulted in a significant genet- 35
36 ic structure among sites. In addition, the presence of an excess of local rare haplotypes indicated that C. nucum 36
37 populations went through genetic expansion after recent bottlenecks. Conversely, these effects were not ob- 37
38 served in the more generalist Curculio glandium predating upon oaks. Ultimately, co-existence of weevil spe- 38
39 cies in this multi-host-parasite system is influenced by both resource and time partitioning. To what extent the 39
40 restriction in gene flow among C. nucum populations may have negative consequences for their persistence in a 40
41 time of increasing disturbances (e.g. drought in Mediterranean areas) deserves further research. 41
42 42
Key words: Corylus avellana, Curculio spp., genetic structure, Quercus spp., trophic breadth
43 43
44 44
45 45
46 46
47 INTRODUCTION 47
48 48
49 Correspondence: Josep Maria Espelta, CREAF, Cerdanyola del Seed predation by insects may play a crucial role in 49
50 Vallès, Spain. plant population dynamics, by reducing the reproductive 50
51 Email: josep.espelta@uab.cat output (Bonal et al. 2007; Espelta et al. 2008) and con- 51
© 2017 The Authors. Integrative Zoology published by International Society of Zoological Sciences, 267
Institute of Zoology/Chinese Academy of Sciences and John Wiley & Sons Australia, Ltd
This is an open access article under the terms of the Creative Commons Attribution License, which
permits use, distribution and reproduction in any medium, provided the original work is properly cited.
H. Arias-LeClaire et al.
1 straining the regeneration process (Espelta et al. 2009b). seeds of different hosts in multi-specific systems has 1
2 Trophic breadth and specificity of seed-feeding insects been seldom explored. 2
3 is often explained by differences among plant hosts in The breadth of the trophic niche of seed-feeding in- 3
4 chemical or morphological traits (Bernays & Graham sects (specialist vs generalist) may influence the num- 4
5 1988; Forister et al. 2015). Differences in phytochemis- ber of species that predate upon different seeds and it 5
6 try (mainly nitrogen-based defensive compounds) have has consequences for the dynamics of the communi- 6
7 been observed to be highly relevant for the diversifi- ty of hosts (Espelta et al. 2009b). However, beyond the 7
8 cation of phytophagous insects and their diet breadth effects on plant fitness, differences in the trophic niche 8
9 (Kergoat et al. 2005). Concerning other seed features, may also influence the population dynamic of the seed 9
10 size is a trait claimed to influence ecomorphological di- consumers (Ylioja et al. 1999) depending on life-his- 10
11 versification in many endophytic insects (e.g. body size tory traits such as dispersal ability and landscape attri- 11
12 and rostrum shape), promoting differences in their tro- butes (i.e. abundance and spatial distribution of hosts). 12
13 phic niche, ecological adaptations and species radiation Spatial connectivity among plant-hosts has been shown 13
14 (Hughes & Voegler 2004a; Bonal et al. 2011; Pegue- to be especially relevant for insects with low disper- 14
15 ro et al. 2017). In addition to chemical and morpholog- sal ability (Thomas et al. 2001; Kruess 2003), resulting 15
16 ical differences, seeding phenology and stochasticity in a stronger genetic structure and reduced gene flow in 16
17 in the availability of this resource have also been sug- the insect populations located on more isolated hosts. In 17
18 gested as key factors influencing the guild of insect spe- the long run, host isolation may even result in coloni- 18
19 cies predating upon a particular plant host (Espelta et zation credits for some insect species, especially those 19
20 al. 2008, 2009b; Coyle et al. 2012; see also Pélisson et with a narrower diet breadth (Ruiz-Carbayo et al. 2016) 20
21 al. 2013a). As insects are short-living organisms, syn- and poor dispersal ability (Pélisson et al. 2013b; Hei- 21
22 chronization of their life-cycle with the resources upon neger et al. 2014). Conversely, generalist species may 22
23 which they depend is critical (Bale et al. 2002, 2007; show a more continuous distribution in the landscape, 23
24 Hood & Ott 2010). Therefore, processes such as adult benefiting from the spatial overlap of the different host 24
25 emergence (Espelta et al. 2017) and oogenesis (Trudel plants upon which they feed (Newman & Pilson 1997), 25
26 et al. 2002; Son & Lewis 2005) have to be tightly con- and show no genetic structure differences among popu- 26
27 nected with the presence of seeds for oviposition (Bonal lations owing to gene flow. Interestingly, for seed-feed- 27
28 et al. 2010). In particular, oogenesis (i.e. egg maturation ing insects a comparison of the genetic structure of their 28
29 in females) has been predicted to differ depending on populations and the spatial structure of their potential 29
30 the stochasticity of seeds availability. Thus, proovigen- hosts could provide strong evidence about differences 30
31 esis (i.e. females have already mature eggs at the onset of the trophic niche breadths. Moreover, the use of mo- 31
32 of their adult life) would be favored in species predating lecular techniques (DNA barcoding) may help to detect 32
33 upon hosts that regularly produce seeds while synovi- cryptic speciation and trophic niche segregation among 33
34 genesis (i.e. females start their adult life with imma- morphologically similar species (Peguero et al. 2017), 34
35 ture eggs) would be advantageous for species exposed and also to establish species specificity in an unequiv- 35
36 to more random fluctuations of seed production (Jervis ocal way in comparison to classifications based on the 36
37 et al. 2008; Richard & Casas 2009), as they can better presence or absence of a species on a particular plant, 37
38 adjust the amount of energy invested in reproduction to especially when the lack of morphological differences 38
39 the amount of seeds (but see Pélisson 2013b). Ultimate- at certain stages (e.g. larvae) make species identification 39
40 ly, the co-existence of the different seed consumers in impossible otherwise (Govindan et al. 2012). Unfortu- 40
41 a multi-host community could be mediated by resource nately, this combination of landscape ecology (i.e. host 41
42 partitioning (e.g. insects predate preferentially upon dif- connectivity) and population genetics when studying the 42
43 ferent species according to different seed traits; see Es- breadth of the trophic niche and dispersal ability of phy- 43
44 pelta et al. 2009a), time-partitioning (e.g. insects exhibit tophagous insects remains largely unexplored. 44
45 differences in life span and the timing of seed predation; 45
The main aims of this study have been to investigate
46 see Pélisson et al. 2012) or the trade-off among disper- in a multi-host and multi-seed-predator system the role
46
47 sal versus dormancy ability to cope with resource scar- 47
of seed size, seed phenology and ooegenesis in the host
48 city (Pélisson et al. 2012). Yet, the importance of the specificity of seed-parasite weevils (Curculio spp.) and
48
49 interplay among seed size, seeding phenology and oo- 49
to analyze the consequences that potential differences
50 genesis in driving the guild of insects predating upon 50
in trophic specialization and host distribution may have
51 51
268 © 2017 The Authors. Integrative Zoology published by International Society of Zoological Sciences,
Causes of host specificity in seed weevils
1 in the genetic structure of weevil populations. Curculio Study area and species 1
2 spp. (Coleoptera: Curculionidae) are seed parasites that 2
3 differ in their dispersal ability (Venner et al. 2011), dia- The study was carried out in mixed forests with the 3
4 pause duration (Pélisson et al. 2013a,b), oogenesis (with presence of oaks (Q. ilex and Q. pubescens and com- 4
5 both proovigenic and synovigenic species; Pélisson et mon hazel trees (Co. avellana) in Catalonia (north-east 5
6 al. 2013a) and the breadth of their trophic niche (Muñoz Spain, Fig. 1). The evergreen Q. ilex and the winter-de- 6
7 et al. 2014; Bonal et al. 2015; Peguero et al. 2017). We ciduous Q. pubescens are extensively distributed in pure 7
8 conducted this study in Catalonia (northeast Spain) in and mixed forests in all the western rim of the Mediter- 8
9 mixed forests including oaks (Quercus ilex, Quercus pu- ranean basin (Espelta et al. 2008), while the common 9
10 bescens) and common hazel trees (Corylus avellana) hazel (Co. avellana) often appears in scattered groups 10
11 with 4 different weevil species present (Curculio nucum in mixed deciduous forests or cultivated in monospe- 11
12 Linnaeus, Curculio glandium Marsham, Curculio veno- cific stands (AliNiazee 1998). Acorns in Quercus spp. 12
13 sus Gravenhorst and Curculio elephas Gyllenhaal). In- and hazelnuts in Co. avellana mature in 1 year and both 13
14 terestingly, in this region oaks show a much more con- are subjected to intense pre-dispersal seed predation 14
15 tinuous distribution and later seeding, while hazels often by weevils (Curculio spp.), a group of granivorous in- 15
16 appear in more discontinuous patches and have an earli- sects extensively distributed in the northern hemisphere 16
17 er production of fruits (Gracia et al. 2004). Concerning (Hughes & Voegler 2004a). In Catalonia, the most com- 17
18 weevils, the 4 species overwinter underground, but they mon weevil species predating upon acorns are C. glan- 18
19 differ in the duration of their diapause, the phenology of dium and C. elephas (Espelta et al. 2009b), the lat- 19
20 emergence, oogenesis and dispersal ability. Adults of C. ter also depredating upon chestnuts (Castanea spp.), 20
21 glandium, C. venosus and C. nucum emerge in spring 2 while in hazelnuts the unique species described up to 21
22 years after larvae buried into the soil, while C. elephas now has been C. nucum, a hypothesized highly specif- 22
23 exhibits variable diapause and adults emerge in early au- ic seed parasite (Guidone et al. 2007; Bel-Venner et al. 23
24 tumn (Bonal et al. 2010; Espelta et al. 2017) for up to 3 24
25 years (Pelisson et al. 2013b). Concerning oogenesis, in 25
26 C. glandium, C. venosus and C. nucum females are re- 26
27 productively immature (synovigenic) and ovarian devel- 27
28 opment is accomplished after 1 or 2 months of the feed- 28
29 ing period (Bel-Venner et al. 2009), while C. elephas 29
30 females are proovigenic and food intake is not required 30
31 for ovarian development (Pélisson et al. 2012). Regard- 31
32 ing host selection, previous studies have suggested that 32
33 C. nucum is highly specialized in hazelnuts (Bel-Ven- 33
34 ner et al. 2009), while the other weevils depredate upon 34
35 several oak species (Muñoz et al. 2014). However, these 35
36 results have not been confirmed by means of DNA anal- 36
37 yses as no study has been conducted in mixed hazel– 37
38 oak forests. Considering the traits of the species in- 38
39 volved in this multi-host and multi-predator system and 39
40 the spatial distribution of hosts, we hypothesize that: (i) 40
41 seed size and the synchronization of seeding phenology 41
42 and oogenesis will be responsible for the guild of wee- 42
43 vils predating upon the different plants; and (ii) the nar- 43
44 rower trophic breadth of C. nucum and the more patchy 44
45 distribution of hazels in comparison to the more gener- 45
46 alist habit of the other weevils and the continuous dis- Figure 1 (a) Location of study sites in Catalonia (north-east 46
47 tribution of oaks will result in differences in the genetic Spain). (b) Distribution of Corylus avellana, Quercus ilex and 47
48 structure of weevil populations of these species. Quercus pubescens according to the presence of this species in 48
49 plots inventoried in the Catalan Forest Inventory (Gracia et al. 49
50 2004). RI, Ripoll (5 plots); OL, Olot (4 plots); MO, Montseny (4 50
MATERIALS AND METHODS
51 plots); MA, Maresme (5 plots); PR, Prades (5 plots). 51
1 2009). However, it must be highlighted that except for scars. We calculated the volume of both sound and in- 1
2 the weevil species predating upon oaks, ascription of fested seeds by measuring the length and width to the 2
3 weevil species to a plant host is based on the observa- nearest 0.01 mm with a digital caliper (see Espelta et 3
4 tion of adults in the foliage of that particular species, but al. 2009a). Infested seeds were placed individually in 4
5 no study has addressed this issue comprehensively (e.g. plastic trays for individual monitoring. Each seed was 5
6 identifying by means of molecular techniques the spe- checked daily to register the emergence of larvae, which 6
7 cies of the larvae inside chestnuts or hazelnuts). were immediately transferred to 2 mL Eppendorf with 7
8 96% alcohol. Once larvae stop emerging (approximate- 8
During early summer on hazelnuts (AliNiazee 1998)
9 ly 3 weeks after seeds were collected) seeds were dis- 9
and early autumn on acorns (see Bonal & Muñoz 2009)
10 sected to check for the presence of non-emerged larvae. 10
female weevils perforate the seed cover with their snout
11 From hazelnuts only Curculio larvae emerged, while for 11
and oviposit commonly a single egg so the larvae devel-
12 acorns the 6% of larvae corresponded to the Cydia spp. 12
op feeding on the seed kernel. At the middle of summer
13 moth. 13
in C. nucum (Bel-Venner et al. 2009) or late autumn in C.
14 14
elephas and C. glandium (Espelta et al. 2009a), larvae In parallel, during the abovementioned field cam-
15 15
16
exit the seed and bury into the ground to overcome the paigns adult weevil were captured by shaking the can- 16
17
diapause period and undergo full metamorphosis. opy and collecting the fallen individuals in an inverted 17
umbrella held beneath the foliage for species identifi-
18 Sampling design 18
19 cation at the laboratory. To establish whether females 19
In 2013 we established a total of 23 sampling plots were sexually mature, they were dissected under a mi-
20 20
grouped into 5 geographical clusters (Sites) in a north croscope to observe abdominal segments and ovary ma-
21 21
to south latitudinal gradient (see Fig. 1). This sampling turity. We considered the presence of eggs as a sign of
22 22
procedure was selected to account for the possible ef- female ready for oviposition and the absence of eggs as
23 23
fects of latitude on the duration of the vegetative sea- females that were still immature or had already ovipos-
24 24
son and, thus, on the seeding phenology of oaks and ha- ited (Pélisson et al. 2013a).
25 25
zelnuts, their overlap and the overlap among these host
26 DNA barcoding and larval species identification 26
species and the weevils predating upon their seeds. Pre-
27 27
sumably a tighter vegetative season in northern and A total of 1657 Curculio larvae emerged from ha-
28 28
colder sites would lead to more similar patterns of seed zelnuts and acorns. In order to establish unequivo-
29 29
30
production while these could be more relaxed and lon- cal trophic relationships between insects and their host 30
31
ger in southern and warmer places. Ultimately, this plants we used molecular techniques (DNA barcoding) 31
32 could lead to differences in the guild of weevils predat- as larvae cannot be determined according to morpho- 32
33 ing upon these plants. Plots were selected by searching logical characters. Therefore, from 342 larvae select- 33
34 for the presence of trees of Co. avellana and Q. ilex or Q. ed randomly among the ones emerged in the labora- 34
35 pubescens based on the Catalan Forest Inventory (Gracia tory from the 3 hosts we extracted DNA from a small 35
36 et al. 2004) and field observations of their reproductive piece of larval tissue (approximately 2-mm long) us- 36
37 status (i.e. presence of seeds). From late July (end of ha- ing the NucleoSpin-Tissue kit according to the manu- 37
38 zel seeding season) to early October (end of the acorn facturer’s instructions (MACHEREY-NAGEL GmbH, 38
39 crop) we carried out 3 sampling campaigns: (i) late July Düren, Germany; www.mn-net.com). We amplified a 39
40 to early August; (ii) late August to early September; and fragment (826 bp) of the mitochondrial cytochrome oxi- 40
41 (iii) late September to early October to account for pos- dase subunit 1 (cox1) using primers Pat and Jerry (please 41
42 sible differences in the phenology of seed infestation by see Hughes & Vogler [2004b] for details on primer se- 42
43 the different weevil species present. In every plot and quences and PCR protocols). We chose this fragment of 43
44 in each sampling period we randomly collected a mini- cox1 due to the availability of many reference sequenc- 44
45 mum of 100 seeds from each species (Co. avellana and es from correctly determined adults of European Cur- 45
46 Quercus spp.) under the canopies of several randomly culio spp. For comparison (Hughes & Vogler 2004b). 46
47 selected trees. Seeds were taken to the laboratory and Sequencing was performed using Big-Dye (Perkin-El- 47
48 classified as sound or infested to assess infestation rates mer) technology and an ABI3700 sequencer. Sequence 48
49 per species and sampling period. Infested seeds are eas- chromatograms were assembled and edited using Se- 49
50 ily recognizable by the presence of female oviposition quencher 4.6 (Gene Codes, Ann Arbor, MI, USA). For 50
51 51
1 species identification we discarded those sequences that (Dupanloup et al. 2002). This method identifies the op- 1
2 after edition were shorter than 500 base pairs. Edited se- timal grouping option (K) that maximizes the among- 2
3 quences were aligned using CLUSTALW supplied via group component (FCT) of the overall genetic variance. 3
4 http://align.genome.jp, with default gap open and gap We defined the number of populations (K) and ran 100 4
5 extension penalties. The alignment sets were collapsed simulated annealing processes. We simulated different 5
6 into unique haplotypes and each of this compared to numbers of populations, ranging from K = 2 to K = 4, to 6
7 the Curculio spp. reference sequences available at Gen- determine the best population clustering option. 7
8 Bank. We applied the most conservative limit used in 8
9 DNA barcoding, which states a maximum genetic diver- 9
RESULTS
10 gence (number of different nucleotides by the total num- 10
11 ber of compared nucleotides) of 1% with respect to the Molecular analyses allowed the identification of the 11
12 reference sequence for an unambiguous identification to larvae emerged from the seeds of the 3 host species (Co. 12
13 the species level (Ratnasingham & Herbert 2007). avellana, Q. ilex and Q. pubescens) as all sequences 13
14 showed a divergence below 1% with respect to Curcu- 14
15 Data analysis lio spp. reference sequences from GenBank. This diver- 15
16 To evaluate the occurrence of the different wee- gence was much lower than inter-specific differences, 16
17 vil species in the 3 potential hosts Co. avellana, Q. ilex which in all cases exceeded 8%. All larvae correspond- 17
18 and Q. pubescens) across the 5 study sites, we conducted to 4 species; namely, C. elephas, C. glandium, C. nu- 18
19 ed a Pearson’s χ2-test. Similarly, we used χ2-tests for the cum and C. venosus. As shown in Figure 2, weevil spe- 19
20 comparison of the presence of male and female wee- cies were not randomly distributed among hosts; that 20
21 vil proportion, as well as that of immature and mature is, C. nucum was exclusively present in hazelnuts while 21
22 females among sampling periods. A generalized linear the other 3 weevils emerged uniquely from acorns (χ26 22
23 mixed model (GLMM), following a binomial distribu- = 263.9, P < 0.001). C. glandium and C. elephas were 23
24 tion, was used to test for the effects of the study site (RI, more abundant in Q. ilex (respectively, χ 214 = 91.8, P < 24
25 Ripoll; OL, Olot; MO, Montseny; MA, Maresme; PR, 0.001, and χ 28 = 23.3, P < 0.001) while there were not 25
26 Prades), sampling period (1, 2, 3) and host species (Co. significant differences in the presence of Curculio ve- 26
27 avellana, Q. ilex and Q. pubescens) on the seed preda- nosus between the 2 oak species (χ 25 = 7.47, P > 0.05). 27
28 tion rate by weevils. The factor “plot” was included as The different presence of larvae of the 4 weevil species 28
29 a random effect in the GLMM analyses to account for 29
30 the repeated nature of the measurements and other un- 30
31 explained variation. Analyses of deviance Type II Wald 31
32 χ2-tests were performed to establish the significance of 32
33 each different independent variable in the model. A gen- 33
34 eral linear mixed model was applied to test for the ef- 34
35 fects of host species, sampling period and seed condi- 35
36 tion (sound or infested) on seed size (volume in mm3) 36
37 with the factor “plot” included as a random effect. 37
38 For population genetic analyses we chose those spe- 38
39 cies in which there were a minimum of 10 individu- 39
40 als per population with sequences longer than 750 bp; 40
41 namely, Curculio glandium and Curculio nucum. We 41
42 used Arlequin software (Excoffier et al. 2005) to calcu- 42
43 late standard molecular diversity indices (gene diversi- 43
44 ty and nucleotide diversity) and to perform analyses of 44
45 the molecular variance (AMOVAs). Signatures of pop- 45
46 ulation demographic changes (bottlenecks or expan- 46
47 47
sions) were examined by Tajima’s D (Tajima 1989) and
48 Figure 2 Proportion of the different weevil species infesting 48
Fu’s F (Fu 1997) as implemented in Arlequin software.
49 the seeds of the 3 host species (Corylus avellana, Quercus ilex 49
We also tested whether there was any geographic pat-
50 and Quercus pubescens) according to the DNA analyses of the 50
tern in the population genetic structure using Samova 1.0
51 larvae emerging from the seeds. 51
1 in the 3 hosts, especially among hazelnut and the 2 oaks, 1

2 was not due to the lack of a particular species in a given 2
3 site as we captured adult specimens of all weevil species 3
4 along the entire gradient. Moreover, as hazelnuts were 4
5 bigger than acorns during almost the entire seeding sea- 5
6 son (Table 1 and Fig. 3), the abovementioned differenc- 6
7 es in weevil specificity among these 2 groups of hosts 7
8 does not seem to be due to the exclusion of certain wee- 8
9 vil species from hazelnuts by a too small seed size. 9
10 Male and female weevils occurred in nearly the same 10
11 frequency with no significant variation along the sam- 11
12 pling periods (χ 22 = 2.28, P > 0.05). Yet the propor- 12
13 tion of females with presence of eggs and without eggs 13
14 showed significant differences through the season (χ 22 = 14
15 33.7, P < 0.001) and for the different weevil species. As 15
16 shown in Figure 4, through the season the presence of 16
17 females with eggs was earlier in C. nucum, followed by 17
18 C. glandium and C. elephas. In the 2 synovigenic spe- 18
19 cies, C. nucum had a decreasing pattern (χ 22 = 10.5, P 19
20 < 0.01) in the presence of females with eggs, while C. 20
21 glandium exhibited an increasing pattern (X22 = 35.3, P 21
22 < 0.001). In the proovigenic C. elephas we did not find 22
23 females on the very first sampling period but as soon 23
24 as they appeared during the second and third sampling 24
25 dates they were already sexually mature (χ 22 = 16.5, P 25
26 < 0.001). Consistently with the seasonal patterns of the 26
27 presence of females ready to oviposit, we found that in- 27
28 festation rates showed significant variation among study 28
29 sites (χ 24 = 16.5, P < 0.001), sampling periods (χ 21 = 5.5, 29
30 P < 0.05) and host species (χ 22 = 6.4, P < 0.05). Over- 30
31 all, infestation was higher in northern localities and it 31
32 increased as the seeding season progressed (see coeffi- 32
33 cients for the different effects in Table 2). Concerning 33
34 host species, infestation rates showed contrasting tem- 34
35 poral patterns in hazelnut versus oaks (Table 2, Fig. 5), 35
36 in agreement with host seeding phenology and oogene- 36
37 37
38 38
39 39
40 40
Table 1 Estimates for the significant effects of tree host, sam-
41 41
pling period and seed condition (sound or infested) over seed
42 42
size (volume in mm3) according to the linear mixed model
43 43
44 Effects Estimate Standard error t-value 44
45 Intercept 3.199 0.0278 153.912*** 45
Figure 3 Proportion of females with eggs (black column) and
46 without eggs (white columns) for the 3 weevil species (Corylus Date 0.01695 0.004249 3.990*** 46
47 avellana, Quercus ilex and Quercus pubescens) captured in the 47
Host, Q. pubescens −0.1876 0.008793 −21.336***
48 3 sampling periods during the seeding season: Jul–Aug = from 48
Host, Q. ilex −0.4808 0.009984 −48.159***
49 the end of July to the beginning of August; Aug–Sep = from 49
50 the end of August to the beginning of September; Sep–Oct = Seed condition, sound −0.03143 0.006962 −4.515*** 50
51 from the end of September to the beginning of October. *P < 0.05, **P < 0.01 and ***P < 0.001. 51
1 a Table 2 Estimates for the significant effects of study site, tree 1

2 host and sampling period on weevil infestation rates according 2
3 to the generalized linear mixed model 3
4 Effects Estimate Standard error z-value 4
5 Intercept −2.8280 0.5462 −5.177*** 5
6 6
Site Olot 1.0830 0.4776 2.268*
7 7
8 Site Ripoll 1.5611 0.4788 3.260** 8
9 Host, Q. pubescens 0.9615 0.3810 2.523* 9
10 Date 0.4634 0.1973 2.349* 10
11 11
*P < 0.05, **P < 0.01 and ***P < 0.001.
12 12
13 13
14 14
15 b 15
16 16
17 17
18 18
19 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 c 29
30 30
31 31
32 32
33 33
Figure 5 Mean ± SE seed infestation rate of the 3 host species
34 34
(Corylus avellana, Quercus ilex and Quercus pubescens) along
35 35
the sampling dates. Jul–Aug = from the end of July to the be-
36 36
ginning of August; Aug–Sep= from the end of August to the
37 beginning of September; Sep–Oct = from the end of September 37
38 to the beginning of October. 38
39 39
40 40
41 41
42 sis in females; that is, in hazelnuts infestation occurred 42
43 earlier and slightly decreased through the season, while 43
44 it was absent during the first sampling date in the 2 oak 44
Figure 4 Mean ± SE volume (mm3) of sound (open dots) and
45 species, and progressively increased towards the end of 45
infested (black dots) seeds of the 3 host species (Corylus avel-
46 the season (Fig. 5). 46
lana, Quercus ilex and Quercus pubescens) along the sampling
47 dates. Jul–Aug = from the end of July to the beginning of Au- The population genetic analyses showed marked dif- 47
48 gust; Aug–Sep = from the end of August to the beginning of ferences between C. nucum and C. glandium. Mean ge- 48
49 September; Sep–Oct = from the end of September to the begin- netic diversity was higher in C. nucum (Table 3), main- 49
50 ning of October. Notice the difference in the scale of the y-axis ly due to the higher number of distinct haplotypes; that 50
51 for the 3 host species. is, an ANOVA in which the population was included as 51
1 Table 3 Values of gene diversity, nucleotide diversity, Tajima’s of our study, the first grouping the nearby populations 1
2 D and Fu’s F recorded at each population for Curculio nucum of Montseny (MO) and Maresme (MA) and another one 2
3 (a) and Curculio glandium (b) including the rest (see Fig. 1). No significant geograph- 3
4 (a) Curculio nucum ical pattern of molecular variance was found in C. glan- 4
5
Gene Nucleotide dium. 5
6 diversity diversity Tajima’s D Fu’s F 6
7 7
Ripoll 0.87 0.0015 −1.63* −4.54*** DISCUSSION
8 8
9 Olot 0.75 0.0024 −1.96** −8.34*** Seed infestation by weevils did not occur randomly 9
10 Montseny 0.89 0.0032 −1.40* −6.87*** but with 2 opposite breadths of host specificity; name- 10
11 Maresme 0.88 0.0034 −0.41 −1.20 ly, the highly specialized C. nucum infested a unique 11
12 host (hazelnuts), while up to 3 species (C. glandium, C. 12
Prades 0.59 0.0012 −1.69** −5.27***
13 elephas and C. venosus) predated almost indistinctive- 13
(b) Curculio glandium ly upon 2 oaks (Q. ilex and Q. pubescens). These dif-
14 14
15 Ripoll 0.71 0.0012 −1.10 −2.61* ferences in trophic specificity coupled with differenc- 15
16 Olot 0.57 0.0009 −0.46 −0.84 es in the geographical distribution of the hosts resulted 16
17 Montseny 0.69 0.0013 −1.22 −2.61* in 2 distinct patterns concerning the genetic character- 17
18 istics of weevils’ populations; that is, we only found a 18
Maresme 0.69 0.0013 −0.75 −1.95
19 significant genetic structure among the populations in 19
20 Prades 0.63 0.0034 −1.79 1.46 the highly specialist C. nucum. Ultimately, the results of 20
21 *P < 0.05, **P < 0.01 and ***P < 0.001. these genetic analyses confirmed the specialist or gener- 21
22 alist trophic breadth of the different weevil species ac- 22
23 cording to the identification of the larvae found in the 23
24 seeds and they stress how molecular techniques may 24
25 a random factor showed that mean gene diversity was help to establish unequivocal trophic relations for seed 25
26 significantly higher in C. nucum (F1,4 = 9.40; P = 0.03). feeding insects. 26
27 A total of 31 haplotypes were retrieved from the 118 se- Previous studies have suggested that seed size has 27
28 quences of C. nucum included in the analyses versus been a relevant trait promoting ecomorphological adap- 28
29 just 13 from 96 sequences in C. glandium (see Tables tations in the genus Curculio and driving species diver- 29
30 4 and 5). In C. nucum, 48% of the individuals had the sification (Hughes & Vogler 2004a; see also Peguero et 30
31 most common haplotype but there were many rare hap- al. 2017). In the end, a tight relationship between seed 31
32 lotypes sometimes found in just one individual and/or at and weevils’ body size would result in differences in 32
33 a single population (Table S1). In the case of C. glandi- the ability of larger and smaller weevils to infest seeds 33
34 um, there were much fewer rare haplotypes and the two of different size (differences in trophic breadth); that 34
35 most frequent ones were found in 80% of the individu- is, small species would be able to infest both small and 35
36 als (Table S2). The high proportion of rare haplotypes large seeds while species with a larger body size would 36
37 in C. nucum suggests population expansion after recent be restricted to larger seeds to obtain enough resourc- 37
38 bottlenecks and, accordingly, both Tajima’s D and Fu’s es to complete larvae development (see Bonal & Muñoz 38
39 F had negative and significant values in all populations 2008; Espelta et al. 2009a; Bonal & Muñoz 2011; Peg- 39
40 except one. In the case of C. glandium only for Fu’s uero et al. 2017). Yet, this does not seem to be the case 40
41 test were the values significant in two populations, thus in our study system where hazels, the species infested 41
42 showing that most populations were in equilibrium (Ta- by a single species (C. nucum), showed the largest seeds 42
43 jima 1989; Fu 1997). The AMOVA revealed a more re- during most of the season (Fig. 3) and experienced the 43
44 stricted gene flow between populations in the case of C. lower infestation rate (see Table S3). Instead of an in- 44
45 nucum, in which differentiation among populations ex- fluence of seed size, our results suggest that the exclu- 45
46 plained 5.02% of the total molecular variance (degrees sive infestation of hazelnuts by C. nucum could be more 46
47 of freedom = 4; P < 0.01), whereas in C. glandium in- related to a different pattern of sexual maturation of fe- 47
48 ter-population differences were not significant. The re- males among the 2 weevil species emerging from the 48
49 sults of the SAMOVA were marginally significant for C. soil in spring, specifically an earlier maturation in C. nu- 49
50 nucum (FCT = 0.08; degrees of freedom = 1; P = 0.08) cum in comparison to C. glandium (Fig. 4). These dif- 50
51 and defined two clusters within the geographical range 51
1 ferences could be due to differences between the 2 wee- able diapause) to cope with seed scarcity. This seems to 1
2 vils in the requirements of resource acquisition as it has be the case for C. glandium and C. elephas; that is high 2
3 been demonstrated that sexual maturation in females of dispersal ability (up to 11 km) in the former species and 3
4 these synovigenic species critically requires some feed- an extended diapause (up to 3 years) in the later (see 4
5 ing at adulthood before reproductive development takes Venner et al. 2011; Pélisson et al. 2012). Yet, other fac- 5
6 place (Bel-Venner et al. 2009; Pélisson et al. 2012). The tors not covered in this study, such as the risk of parasit- 6
7 early maturation in C. nucum would be advantageous ism or survival of larvae during diapause, may also help 7
8 to oviposit in hazelnuts before the hardening of the nut- equalize their success to infest (Bonal et al. 2011). Sim- 8
9 shell, as this is a fast process occurring during seed ilarly, future studies with more intense and appropriate 9
10 growth and the main mechanism in hazels to avoid in- sampling schemes should address the relationship be- 10
11 festation (Guidone et al. 2007). Moreover, oviposition tween the number of adults of the different species and 11
12 of C. nucum would be expected to occur soon after mat- the number of larvae to disentangle the different preda- 12
13 ing as weevils do not adjust laying eggs to the moment tion rates upon each species and the influence of other 13
14 of highest seed availability, but they oviposit as soon as environmental factors. 14
15 females have mature eggs (Bonal et al. 2010). This be- Ultimately, differences in the trophic breadth leave a 15
16 havior is probably linked to the temporal unpredictabil- contrasting genetic signature in the populations of the 16
17 ity of seed crop size (Bonal & Muñoz 2008; Espelta et 2 species of weevils. A much higher number of local 17
18 al. 2008) and other constraints they have to cope with, rare haplotypes were found in the monophagus C. nu- 18
19 such as the need of rainfall episodes to soften the soil cum, along with a marginally significant genetic struc- 19
20 and allow the emergence of adults (Bonal et al. 2010, ture among populations, contrary to the more general- 20
21 2015; Espelta et al. 2017). Only during the 2 earlier ist C. glandium (see, for a similar example in aphids, 21
22 samplings, females of C. nucum seemed to preferential- Gaete-Eastman et al. 2004). Inter-specific differences 22
23 ly choose bigger seeds, a behavior related with the need in genetic characteristics of phytophagous insects could 23
24 to select a minimum seed size to ensure larvae develop- arise from differences in their dispersal ability or in the 24
25 ment and also owing to the availability of more seeds spatial distribution (isolated vs continuous) of the host 25
26 for oviposition (see Espelta et al. 2009a). (Peterson & Denno 1998; Kubish et al. 2014). Unfortu- 26
27 In comparison to the extreme host–parasite specificity nately, in comparison to the precise information about 27
28 of C. nucum, the other 3 weevils (C. venosus, C. glan- the dispersal ability of C. glandium (approximately 10 28
29 dium and C. elephas) predated indistinctively upon the km in Pélisson et al. 2012), we lack detailed knowledge 29
30 2 oaks with no evidence of a strategy in the partitioning about the dispersal range of C. nucum, except some ev- 30
31 of this resource according to the identity of the host spe- idence of weevils moving away from local sites to feed 31
32 cies or to seed size. The avoidance of competitive exclu- during adulthood and prior to mating (Bel-Venner et 32
33 sion among these species could be explained by several al. 2009). Yet the fact that C. nucum and C. glandium 33
34 mechanisms contributing to stabilize their coexistence; are sister species (Hughes & Voegler 2004a) and they 34
35 that is, time partitioning (Pélisson et al. 2012; see also share many similar life-history traits, such as ecomor- 35
36 Espelta et al. 2009a) and/or diversification of disper- phological adaptations and body size, adult emergence 36
37 sal-dormancy strategies (Pélisson et al. 2012). On the in spring, synovigenic females and a fixed diapause of 37
38 one hand, time partitioning in breeding activity can exist 2 years (see Hughes & Vogler 2004a; Bel-Venner et al. 38
39 when 2 competing species differ in the speed of energy 2009; Pélisson et al. 2012a,b), make us consider that 39
40 acquisition to be allocated to reproduction by females they may have a similar dispersal ability. Therefore, the 40
41 and the duration of their lifespan; that is, one species ac- differences we observed in their genetic characteristics 41
42 quires resources faster and it is able to oviposit earlier would be probably due to their different diet breadth and 42
43 on seeds, but it is exposed to a higher risk of seed abor- the more patchy and discontinuous distribution of hazels 43
44 tion, while the other oviposits later but has a longer life in comparison to the more abundant and constant pres- 44
45 span allowing it to lay eggs during a larger time frame ence of oaks (Gracia et al. 2004; see also Fig. 1), along 45
46 (see Pelisson et al. 2012 for C. pellitus and C. glandi- the geographical gradient sampled (approximately 225 46
47 um). On the other hand, stabilization can be reached by km from Ripoll to Prades). 47
48 means of different dispersal versus dormancy strategies 48
Connectivity may be critical for population survival
49 with some species relying on a high dispersal ability 49
(Fahrig & Merrian 1985; Fahrig & Paloheimo 1988) and
50 and others depending on dormancy strategies (e.g. vari- 50
metapopulation dynamics (Levins 1970), especially in
51 51
1 front of a disturbance: for example, the negative impact patchily distributed hosts (Co. avellana). To what extent 1
2 of severe drought episodes for the emergence of adult this restriction in gene flow (dispersal) may have nega- 2
3 weevils (Bonal et al. 2015; Espelta et al. 2017). In that tive consequences for the persistence of the populations 3
4 sense, our molecular data show that such disturbances of these highly specialized seed-feeding pests in a time 4
5 may have occurred and left their signature in C. nucum of increasing disturbances (e.g. drought events in Med- 5
6 population genetics. The significant negative values re- iterranean areas) is a fascinating question that deserves 6
7 trieved in the neutrality tests (Tajima’s D and Fu’s F) in- further research. 7
8 dicate that most of the C. nucum populations sampled 8
9 went through population expansion after recent bottle- ACKNOWLEDGMENTS 9
10 necks. Almost half of the individuals had the same hap- 10
11 lotype and there was an excess of rare haplotypes that This research was supported by the projects FOR- 11
12 differed little from the most common one. The lower ASSEMBLY (CGL2015-70558-P) and PLAGANADO 12
13 gene flow between populations (marginally significant (AGL2014-54739-R) of the Spanish Ministry of Econ- 13
14 genetic structure among populations) in C. nucum would omy, and the projects BEEMED (SGR913) (Generalitat 14
15 favor such bottlenecks as the patchy distribution of ha- de Catalunya) and PII1C09-0256-9052 (Regional Gov- 15
16 zel trees would complicate the arrival of immigrants. ernment of Castilla-La Mancha and the European So- 16
17 None of this happened in the case of C. glandium feed- cial Fund). R.B. was funded by a contract of the Pro- 17
18 ing on the widespread oak trees. Yet the interpretation of gram Atracción de Talento Investigador (Gobierno de 18
19 the results for C. nucum must be cautious as the shallow Extremadura). 19
20 genetic structure observed suggests that a fair amount of 20
21 gene flow still occurs, enough to overcome drift. More- REFERENCES 21
22 over, in addition to the current distribution of hazelnuts, 22
23 other abiotic environmental conditions could also be in- AliNiazee MT (1998). Ecology and management of ha- 23
24 volved in the genetic structure observed in C. nucum zelnuts pests. Annual Review of Entomology 43, 395– 24
25 (e.g. geological barriers or altitude for Trichobaris soror 41. 25
26 in De la Mora et al. 2015) Bale JS, Masters GJ, Hodkinson ID, Awmack et al. 26
27 (2002). Herbivory in global climate change research: 27
28 Direct effects of rising temperature on insect herbi- 28
CONCLUSION
29 vores. Global Change Biology 8, 1–16. 29
30 The use of molecular analyses allowed us to precise- Barat M, Tarayre M, Atlan A (2007). Plant phenology 30
31 ly identify the weevil species depredating upon the vari- and seed predation: Interactions between gorses and 31
32 ous potential hosts in these mixed deciduous forests and weevils in Brittany (France). Entomologia Experi- 32
33 to unequivocally confirm the high specificity of the ha- mentalis et Applicata 124, 167–76. 33
34 zelnut C. nucum and the more flexible and wider trophic Bel-Venner MC, Mondy N, Arthaud F et al. (2009). 34
35 breadth of the rest of the weevils (C. venosus, C. glan- Ecophysiological attributes of adult overwintering in 35
36 dium and C. elephas) depredating upon acorns. In this insects: Insights from a field study of the nut weevil, 36
37 multi-host and multi-parasite system, co-existence of the Curculio nucum. Physiological Entomology 34, 61– 37
38 various weevil species seems to be mediated by a com- 70. 38
39 bination of extreme resource partitioning (i.e. among 39
C. nucum and the rest of species) and a combination of Bernays EA, Graham M (1988). On the evolution of
40 40
time partitioning and differences in dispersal-dormancy host specificity in phytophagous arthropods. Ecology
41 41
strategies among the 3 species depredating upon oaks. 69, 886–92.
42 42
43 Interestingly, although sometimes suggested, differences Bonal R, Muñoz A (2008). Seed growth suppression 43
44 in seed size did not have any effect in driving host spec- constrains the growth of seed parasites: Premature 44
45 ificity or the trophic breadth of the weevil species pres- acorn abscission reduces Curculio elephas larval size. 45
46 ent. Moreover, our results highlight that differences in Ecological Entomology 33, 31–6. 46
47 specificity in trophic breadth and in the spatial distribu- Bonal R, Muñoz A (2009). Seed weevils living on the 47
48 tion of hosts at a large geographical scale may result in edge: Pressures and conflicts over body size in the 48
49 the presence of genetic structure among the populations endoparasitic Curculio larvae. Ecological Entomolo- 49
50 of highly specific parasites (C. nucum) depredating upon gy 34, 304–9. 50
51 51
1 Bonal R, Muñoz A, Díaz M (2007). Satiation of predis- genetics data analysis. Evolutionary Bioinformatics 1
2 persal seed predators: The importance of considering Online 1, 47–50. 2
3 both plant and seed levels. Evolutionary Ecology 21, Fahrig L, Merriam G (1985). Habitat patch connectivity 3
4 367–80. and population survival. Ecology 66, 1762–8. 4
5 Bonal R, Muñoz A, Espelta JM (2010). Mismatch be- 5
Fahrig L, Paloheimo J (1988). Effect of spatial arrange-
6 tween the timing of oviposition and the seasonal op- 6
ment of habitat patches on local population size.
7 timum. The stochastic phenology of Mediterranean 7
Ecology 69, 468–75.
8 acorn weevils. Ecological Entomology 35, 270–78. 8
9 Forister ML, Novotny V, Panorska AK et al. (2015). 9
Bonal R, Espelta JM, Vogler AP (2011). Complex selec- The global distribution of diet breadth in insect herbi-
10 10
tion on life-history traits and the maintenance of vari- vores. PNAS 112, 442–7.
11 11
ation in exaggerated rostrum length in acorn weevils.
12 Fu YX (1997). Statistical tests of neutrality of mutations 12
Oecologia 167, 1053–61.
13 against population growth, hitchhiking and back- 13
14 Bonal R, Hernández M, Ortego J, Muñoz A, Espelta, JM ground selection. Genetics 147, 915–25. 14
15 (2012). Positive cascade effects of forest fragmenta- 15
Gaete-Eastman C, Figueroa CC, Olivares-Donoso R,
16 tion on acorn weevils mediated by seed size enlarge- 16
Niemeyer HM, Ramírez CC (2004). Diet breadth and
17 ment. Insect Conservation and Diversity 5, 381–8. 17
its relationship with genetic diversity and differentia-
18 Bonal R, Hernández M, Espelta JM, Muñoz A, Apari- 18
tion: The case of southern beech aphids (Hemiptera:
19 cio JM (2015). Unexpected consequences of a drier 19
Aphididae). Bulletin of Entomological Research 94,
20 world: Evidence that delay in late summer rains bias- 20
219–27.
21 es the population sex ratio of an insect. Royal Society 21
Open Science 2, 150198. Govindan BN, Kéry M, Swihart RK (2012). Host selec-
22 22
tion and responses to forest fragmentation in acorn
23 Coyle DR, Mattson WJ Jr, Jordan MS, Raffa KF (2012). 23
weevils: inferences from dynamic occupancy models.
24 Variable host phenology does not pose a barrier to in- 24
Oikos 121, 623–33.
25 vasive weevils in a northern hardwood forest. Agri- 25
26 cultural and Forest Entomology 14, 276–85. Gracia C, Ibàñez JJ, Burriel JA, Mata T, Vayreda J 26
27 (2004). Inventari Ecològic i Forestal de Catalunya. 27
Dupanloup S, Schneider I, Excoffier L (2002). A simu-
28 CREAF, Bellaterra. 28
lated annealing approach to define the genetic struc-
29 ture of populations. Molecular Ecology 11, 2571–81. Guidone L, Valentini N, Rolle L, Me G, Tavella L (2007). 29
30
Espelta JM, Cortés P, Molowny-Horas R, Sánchez-Hu- Early nut development as a resistance factor to the at- 30
31
manes B, Retana J (2008). Masting mediated by sum- tacks of Curculio nucum (Coleoptera: Curculioni- 31
32
mer drought reduces acorn predation in Mediterra- dae). Annals of Applied Biology 150, 323–9. 32
33 Jervis MA, Ellers J, Harvey JA (2008). Resource acqui- 33
nean oak forests. Ecology 89, 805–17.
34 sition, allocation, and utilization in parasitoid repro- 34
35
Espelta JM, Bonal R, Sánchez-Humanes B (2009a). 35
Pre-dispersal acorn predation in mixed oak forests: ductive strategies. Annual Review of Entomology 53,
36 361–85. 36
37
interspecific differences are driven by the interplay 37
among seed phenology, seed size and predator size. Heiniger C, Barot S, Ponge JF et al. (2014). Effect of
38 38
Journal of Ecology 97, 1416–23. habitat spatiotemporal structure on collembolan di-
39 39
Espelta JM, Cortés P, Molowny-Horas R, Retana J versity. Pedobiologia 57, 103–17.
40 40
41 (2009b). Acorn crop size and pre-dispersal predation Hood GR, Ott JR (2010). Developmental plasticity and 41
42 determine inter-specific differences in the recruitment reduced susceptibility to natural enemies follow- 42
43 of co-occurring oaks. Oecologia 161, 559–68. ing host plant defoliation in a specialized herbivore. 43
44 Espelta JM, Arias-LeClaire H, Fernández-Martínez M, Oecologia 162, 673–83. 44
45 Doblas-Miranda E, Muñoz A, Bonal R (2017). Be- Hughes J, Vogler AP (2004a). Ecomorphological adap- 45
46 yond predator satiation: Masting but also the effects tation of acorn weevils to their oviposition site. Evo- 46
47 of rainfall stochasticity on weevils drive acorn preda- lution 58, 1971–83. 47
48 tion. Ecosphere 8, e01836. Hughes J, Vogler AP (2004b). The phylogeny of acorn 48
49 Excoffier L, Laval G, Schneider S (2005). Arlequin ver. weevils (genus Curculio) from mitochondrial and nu- 49
50 3.0: An integrated software package for population clear DNA sequences: The problem of incomplete 50
51 51
1 data. Molecular Phylogenetics and Evolution 32, Peterson MA, Denno RF (1998). The influence of dis- 1
2 601–15. persal and diet breadth on patterns of genetic isola- 2
3 Kergoat GJ, Delobel A, Fédière G, Le Rü B, Silvain JF tion by distance in phytophagous insects. The Ameri- 3
4
(2005). Both host–plant phylogeny and chemistry can Naturalist 152, 428–46. 4
5 Ratnasingham S, Hebert PDN (2007). Bold: The Bar- 5
have shaped the African seed-beetle radiation. Molec-
6 code of Life Data System (www.barcodinglife.org). 6
ular Phylogenetics and Evolution 35, 602–11.
7 Molecular Ecology Notes 7, 355–64. 7
8 Kruess A (2003). Effects of landscape structure and hab- 8
itat type on a plant–herbivore–parasitoid community. Richard R, Casas J (2009). Stochasticity and controlla-
9 9
Ecography 26, 283–90. bility of nutrient sources in foraging: Host–feeding
10 10
and egg resorption in parasitoids. Ecological Mono-
11 Levins R (1970). Extinctions. Some mathematical ques- graphs 79, 465–83.
11
12 tions in biology. In: Gerstenhaber R, ed. Lectures on 12
13 Ruiz–Carbayo H, Bonal R, Espelta JM, Hernández M, 13
Mathematics in Life Sciences 2. The American Math-
14 Pino J (2017). Community assembly in time and 14
ematical Society, Providence, RI, pp. 77–107.
15 space: The case of Lepidoptera in a Quercus ilex L. 15
Kubisch A, Holt RD, Poethke HJ, Fronhofer EA (2014). savannah–like landscape. Insect Conservation and
16 16
Where am I and why? Synthesizing range biology Diversity 10, 21–31.
17 17
and the eco–evolutionary dynamics of dispersal. Oi- Solar A, Stampar F (2011). Characterisation of select-
18 18
19
kos 123, 5–22. ed hazelnut cultivars: Phenology, growing and yield- 19
20 Merville A, Venner S, Henri H et al. (2013). Endosym- ing capacity, market quality and nutraceutical value. 20
21 biont diversity among sibling weevil species compet- Journal of the Science of Food and Agriculture 91, 21
22 ing for the same resource. BMC Evolutionary Biolo- 1205–12. 22
23 gy, 13, 28. Son Y, Lewis EE (2005). Effects of temperature on the 23
24 Muñoz A, Bonal R, Espelta JM (2014). Acorn–weevil reproductive life history of the black vine weevil, 24
25 interactions in a mixed-oak forest: Outcomes for lar- Otiorhynchus sulcatus. Entomologia experimentalis 25
26 val growth and plant recruitment. Forest Ecology and et applicate 114, 15–24. 26
27 Management 322, 98–105. Tajima F (1989). Statistical method for testing the neu- 27
28 tral mutation hypothesis by DNA polymorphism. Ge- 28
29
Newman D, Pilson D (1997). Increased probability of 29
netics 123, 585–95.
30 extinction due to decreased genetic effective popu- 30
lation size: experimental populations of Clarkia pul- Thomas CD, Bodsworth EJ, Wilson RJ et al. (2001).
31 Ecological and evolutionary processes at expanding 31
32 chella. Evolution 51, 354–62. 32
range margins. Nature 411, 577–81.
33 Peguero G, Bonal R, Sol D, Muñoz A, Sork VL, Espe- 33
Toju H, Fukatsu T (2011). Diversity and infection prev-
34 lta JM (2017). Tropical insect diversity: Evidence of 34
alence of endosymbionts in natural populations of the
35 greater host specialization in seed-feeding weevils. 35
chestnut weevil: Relevance of local climate and host
36 Ecology 98, 2180–90. 36
plants. Molecular Ecology 20, 853–68.
37 Pélisson PF, Bel–Venner M, Rey B et al. (2012). Con- 37
38 Trudel R, Lavallee R, Guertin C (2002). The effect of 38
trasted breeding strategies in four sympatric sibling cold temperature exposure and long–day photoperi-
39 39
insect species: When a proovigenic and capital breed- od on the termination of the reproductive diapause of
40 40
er copes with a stochastic environment. Functional newly emerged female Pissodes strobi (Coleoptera:
41 41
Ecology 26, 198–206. Curculionidae). Agricultural and Forest Entomology
42 42
43 Pélisson PF, Bel-Venner MC, Giron D, Menu F, Venner 4, 301–8. 43
44 S (2013a). From income to capital breeding: When Venner S, Pélisson PF, Bel-Venner MC, Débias F, Ra- 44
45 diversified strategies sustain species coexistence. jon E, Menu F (2011). Coexistence of insect species 45
46 PloS One 8, e76086. competing for a pulsed resource: toward a unified 46
47 Pélisson PF, Bernstein C, Francois D, Menu F, Venner theory of biodiversity in fluctuating environments. 47
48 S (2013b). Dispersal and dormancy strategies among PloS One 6, e18039. 48
49 insect species competing for a pulsed resource. Eco- Ylioja T, Roininen H, Ayres MP, Rousi M, Price PW 49
50 logical Entomology 38, 470–77. (1999). Host-driven population dynamics in an her- 50
51 bivorous insect. PNAS, 10735–40. 51
1 SUPPLEMENTARY MATERIALS bearing each haplotype in the 5 study populations 1

2 Additional supporting information may be found in Table S3 Mean ± SE density of host plants and the per- 2
3 the online version of this article. centage of sound and infested seeds per location and 3
4 host plant Density of host plants was calculated as 4
Table S1 Number of Curculio nucum individuals bear-
5 the mean of the nearest inventoried plots included in 5
ing each haplotype in the 5 study populations
6 the Catalan Forest Inventory (Gracia et al. 2004). 6
7 Table S2 Number of Curculio glandium individuals 7
8 8
9 9
Cite this article as:
10 10
11 Arias-LeClaire H, Bonal R, García-López D, Espelta JM (2018). Role of seed size, phenology, oogenesis and host 11
12 distribution in the specificity and genetic structure of seed weevils (Curculio spp.) in mixed forests. Integrative 12
13 Zoology 13, 267–79. 13
14 14
15 15
16 16
17 17
18 18
19 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51
2 2
3 3
4 4
5 5
6
7
Eastern gray squirrels are consistent shoppers of seed traits: 6
7
8
9 insights from discrete choice experiments 8
9
10 10
11 11
12 12
13 13
Mekala SUNDARAM,1,* Nathanael I. LICHTI,1 Nicole J. Olynk WIDMAR2 and Robert K.
14 14
15 SWIHART1 15
16 1
Department of Forestry and Natural Resources, Purdue University, West Lafayette, Indiana, USA and 2Department of Agricultural 16
17 17
Economics, Purdue University, West Lafayette, Indiana, USA
18 18
19 19
20 20
21 21
22 Abstract 22
23 23
24 Seeds of many hardwood trees are dispersed by scatter-hoarding rodents, and this process is often mediat- 24
25 ed by the traits of seeds. Although numerous studies have linked seed traits to seed preference by rodents, lit- 25
26 tle is known about how rodents forage for seeds when multiple desirable and undesirable seed traits are avail- 26
27 able simultaneously. Here, we adopt a novel method of designing choice experiments to study how eastern gray 27
28 squirrels (Sciurus carolinensis) select for 6 traits (caloric value, protein content, tannin concentration, kernel 28
29 mass, dormancy period and toughness of shell) among seeds. From n = 426 seed-pair presentations, we found 29
30 that squirrels preferentially consumed seeds with short dormancy or tougher shells, and preferentially cached 30
31 seeds with larger kernel mass, tougher shells and higher tannin concentrations. By incorporating random ef- 31
32 fects, we found that squirrels exhibited consistent preferences for seed traits, which is likely due to the fitness 32
33 consequences associated with maintaining cached resources. Furthermore, we found that squirrels were will- 33
34 ing to trade between multiple traits when caching seeds, which likely results in more seed species being cached 34
35 in the fall. Ultimately, our approach allowed us to compute the relative values of different seed traits to squir- 35
36 rels, despite covariance among studied traits across seed species. In addition, by investigating how squirrels 36
37 trade among different seed traits, important insights can be gleaned into behavioral mechanisms underlying seed 37
38 caching (and, thus, seed survival) dynamics as well as evolutionary strategies adopted by plants to attract seed 38
39 dispersers. We describe how discrete choice experiments can be used to study resource selection in other eco- 39
40 logical systems. 40
41 41
Key words: cached resources, discrete choice experiment, resource selection, trade-offs, trait-dependent
42 42
43 43
44 44
45 45
46 46
Correspondence: Mekala Sundaram, 191, Biomedical Center, INTRODUCTION
47 47
48 171 Meeting St., Providence, RI 02912, USA. Scatter-hoarding rodents influence tree regenera- 48
49 Email: mekala_sundaram@brown.edu tion processes in 2 ways that have opposing effects on 49
50 * Current address: Room 191, Biomedical Center, 171 Meeting a tree’s fitness (Vander Wall 1990). First, by consuming 50
51 Street, Providence RI 02912, USA. large proportions of seeds, rodents can reduce numbers 51

Trait-based seed selection by squirrels
1 of seeds available for recruitment (Chambers & MacMa- et al. 2010). Traits are linked together under a unitless 1
2 hon 1994). Alternatively, scatter-hoarding rodents may currency of utility. Thus, it is possible to examine how 2
3 disperse seeds horizontally away from the parent plant variations in different traits alter the utility of a prod- 3
4 and bury seeds in caches, thereby enhancing chances of uct, how variations in 1 trait can offset utility changes 4
5 germination for unrecovered seeds (Chambers & Mac- resulting from variations in a second trait, and, hence, 5
6 Mahon 1994). Both predation and dispersal behaviors of how a decision-maker trades among traits (Louviere et 6
7 scatter-hoarding rodents are influenced by traits of seeds al. 2010; Browning & Zupan 2015). A second benefit is 7
8 (Wang & Chen 2008; Wang et al. 2012; Sundaram et al. the ability to maximize the amount of information cap- 8
9 2015). Therefore, understanding how seed traits affect tured from choice data through the use of DCEs (Lou- 9
10 the perceived value of seeds to granivores is important viere et al. 2010). DCEs involve construction of choice 10
11 because of potential ramifications for seed mortality and scenarios with the goal of quantitatively eliciting pref- 11
12 dispersal (Wang & Smith 2002; Wang et al. 2012). erences for attractive and unattractive traits (Louviere 12
13 Multiple seed traits influence rodent foraging deci- et al. 2010). By designing trait levels of alternatives 13
14 sions (Xiao & Zhang 2006; Wang et al. 2012). On en- ahead of time and then presenting designed alternatives 14
15 countering a seed, scatter-hoarding rodents likely eval- to consumers, DCEs can maximize the information cap- 15
16 uate quality and condition (Preston & Jacobs 2009; tured in choice data (Louviere et al. 2010). More spe- 16
17 Delgado et al. 2014). Seeds germinating immediate- cifically, this is achieved by ensuring orthogonality and 17
18 ly or seeds with a lower mass are of a low storage value equal representation of trait levels in the design (Kuhfeld 18
19 and tend to be consumed immediately and carried short- 2010). Both orthogonality and equal representation of 19
20 er distances (Smallwood et al. 2001; Sundaram et al. trait levels allow for choices made by individuals to be 20
21 2015). In contrast, seeds with thick shells or seeds with tied to specific traits (Louviere et al. 2010). DCMs with- 21
22 large kernel masses are of higher storage value and tend out DCEs have been used previously by ecologists to 22
23 to be transported and cached at greater distances (Sund- study prey selection and habitat use as well as tradeoffs 23
24 aram et al. 2015). Given that multiple seed traits influ- made with respect to habitat and prey traits (McCrack- 24
25 ence a rodent’s decision to cache or consume seed and en et al. 1998; Cooper & Millspaugh 1999; McDonald 25
26 that different clades of hardwood trees produce predict- et al. 2006; Cooper et al. 2007). While DCEs have been 26
27 able combinations of seed traits (Sundaram et al. 2015), extensively employed in consumer economics to con- 27
28 selection of a seed type may not result in maximization struct trait-based experiments (Louviere et al. 2010), to 28
29 of all desired traits and minimization of less desirable our knowledge they have not yet been used in ecology. 29
30 traits. Furthermore, changes in rodent preference due In this paper, we describe and combine a DCE and 30
31 to 1 seed trait may be offset by changes in another trait. DCM approach to evaluate how preferences for seeds 31
32 Thus, the following question arises: How do rodents in eastern gray squirrels (Sciurus carolinensis Gmelin, 32
33 trade between multiple traits? 1788) are related to seed traits. We utilize a DCE to in- 33
34 crease the number of unique combinations of traits 34
Consumer economic models routinely quantify “at-
35 presented to squirrels and then model the resulting 35
tractiveness” of product traits, and the tradeoffs made by
36 choice data with a DCM to quantify squirrels’ willing- 36
customers selecting for multiple product traits (Brown-
37 ness-to-trade 1 seed trait against another. We then dis- 37
ing & Zupan 2015). Whenever selection occurs among
38 cuss the implications of willingness-to-trade estimates 38
discrete items (e.g. discrete prey items), economists typ-
39 from the perspective of the squirrel and the perspective 39
ically use a class of methods called discrete choice mod-
40 of the tree. 40
els and experiments (DCMs and DCEs) to study trait-
41 41
based selections (Louviere et al. 2010). These methods
42 42
43
offer 2 major benefits. First, with DCMs traits are mod- MATERIALS AND METHODS 43
eled together using a common currency of “utility,” de-
44 44
fined as the attractiveness of a product (or satisfaction Overview of discrete choice experiments
45 45
derived from consuming a product) to the decision-mak-
46 Discrete choice experiments offer a means to maxi- 46
er (Louviere et al. 2010). The utility is typically inferred
47 mize the statistical power associated with detection of 47
from choices made by the consumer given all of the al-
48 trait-based selections whenever a consumer selects be- 48
ternatives that are available (Louviere et al. 2010). Util-
49 tween 2 or more items. Subjects in DCEs select between 49
ity models such as DCMs describe how product attrac-
50 alternatives in a set of pre-determined, designed choice 50
tiveness varies as a function of product traits (Louviere
51 scenarios (Kuhfeld et al. 1994; Louviere et al. 2010). 51

M. Sundaram et al.
1 1
2 2
3 3
4 4
5 5
6 6
7 7
8 8
9 9
10 10
11 11
12 12
13 13
14 14
15 15
16 16
17 17
18 18
19 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 Figure 1 Four steps in designing and analysing data from discrete choice experiments (DCE). Steps outlined for: (a) a hypothetical 38
39 discrete choice experiment with 2 traits and (b) the discrete choice experiment in this study with 4 traits. The 4 steps include: (1) se- 39
40 lect traits, (2) categorize traits, (3) determine all possible alternatives or combinations of traits, (4) pair trait combinations in choice 40
scenarios, and (5) data collection and analysis. In a(1), traits X and Y are selected, which represent price and nutrient quantity in
41 41
milk brands. In a(2), traits X and Y are categorized into low and high values. In a(3), all combinations of low and high for traits X
42 42
and Y are listed, such that items are represented across rows and traits across columns. In a(4), pairings of trait combinations are
43 43
considered in a candidate design. Multiple choice scenarios are designed so that each scenario presents 2 alternatives, and the group
44 44
of scenarios allows for all possible contrasts among trait levels to be examined. In a(5), choice scenarios are presented to subjects
45 and selections are analysed. In b(1), 4 seed traits are selected based on patterns of inheritance across phylogenetic clades (more de- 45
46 tails on phylogenetic analysis in Sundaram et al. 2015). In b(2), seed traits are categorized into low and high or low, medium and 46
47 high. In b(3), examples of all possible combinations of low, medium or high for the 4 traits are listed. Some combinations are repre- 47
48 sented by naturally occurring seeds, including Fagus grandifolia, Corylus americana, Carya glabra and Castanea mollissima. Oth- 48
49 ers do not occur in natural seeds and are eliminated from the list, shown as “No seed.” In b(4), pairs of seeds are arranged in a de- 49
50 sign that attempts to capture all possible contrasts with the minimal number of scenarios. In b(5), choice scenarios are presented to 50
51 squirrels and selections are analyzed. 51

1 Individually, each scenario represents a choice between paired and tested (Kuhfeld 2010; Louviere et al. 2010). 1
2 items, each of which contains a specific combination of 2
Selection of seed traits and design of choice
3 traits. Provided that researchers are willing to discretize 3
4 traits into levels, careful selection of the items present- experiments 4
5 ed in each scenario will allow them to construct a set of 5
Four seed traits were used to design choice experi-
6 scenarios in which all possible combinations of traits are 6
ments and were selected from a suite of 11 seed traits
7 presented while maintaining a balanced and orthogo- 7
for which we had estimates. Summaries of all 11 seed
8 nal statistical design (Kuhfeld et al. 1994). For example, 8
traits (including measurements and correlations) across
9 consider a hypothetical experiment involving 2 traits 9
23 hardwood tree species are provided in Sundaram et
10 (Fig. 1a, step 1). The 2 traits may represent the price and 10
al. (2015). Briefly, species seed trait estimates are high-
11 quantity of nutrients in several different brands of milk, 11
ly correlated to one another and show distinct patterns
12 where consumer preference for price relative to nutri- 12
of inheritance across phylogenetic clades. When a phy-
13 ents is unknown. After discretizing the 2 traits into high 13
logenetic PCA was employed on all traits, 3 axes with
14 and low levels (akin to treatment levels in analysis of 14
varying degrees of phylogenetic autocorrelation were
15 variance; Fig. 1a, step 2), all possible combinations of 15
found to be important determinants of eastern gray
16 the trait levels are represented by 4 different alternatives 16
squirrel behavior (Sundaram et al. 2015; Fig. 1b, step 1).
17 (Fig. 1a, step 3). These combinations of traits, which 17
The first phylogenetically correlated axis differentiated
18 are found in the different brands of milk, are then paired 18
seeds with high lipid concentration and thick shells (such
19 (Fig. 1a, step 4), presented to consumers, choices are re- 19
as Juglans and Carya) from species with high carbohy-
20 corded, and the selections are interpreted as a function 20
drate concentration and thin shells (such as Quercus).
21 of the milk-brand traits (price and nutrient content) us- 21
The second phylogenetically correlated axis differen-
22 ing discrete choice models or DCMs (Fig. 1a, step 5; 22
tiated high protein seeds (such as Juglans and Casta-
23 Louviere et al. 2010). 23
nea seeds) from low protein and high tannin seeds (such
24 Importantly, discrete choice pairings (Fig. 1a, step 24
as Quercus section Quercus). The third axis showed
25 4) represent a balanced and orthogonal design. The de- 25
no correlation to phylogeny and differentiated seeds of
26 sign is balanced because the number of “lows” in the 26
higher kernel mass from seeds of lower kernel mass.
27 design equals number of “highs” represented for each 27
All 3 axes significantly predicted squirrel foraging be-
28 trait. The design is also orthogonal because trait X and 28
haviors such as handling time and transport distance to
29 Y are not correlated to one another in the final design, 29
cache site (Sundaram et al. 2015); thus, we selected and
30 which is important to decouple the effects of the traits 30
combined seed traits from these axes to obtain 4 seed
31 on the choices made by subjects. In more complex ex- 31
traits for designing choice experiments (Fig. 1b, step 1):
32 periments that involve multiple traits or trait levels (e.g. 32
(1) total caloric value of a seed (calories, product of ker-
33 Fig. 1b), an orthogonal and balanced set of choice sce- 33
nel mass and calories/gm); (2) shell toughness (Nm or
34 narios can be selected using design algorithms (Kuhfeld 34
Newton-meter, product of shell thickness and shell hard-
35 et al. 1994). 35
ness); (3) protein content (product of kernel mass and
36 36
In contrast to standard, observational study designs, protein concentration); and (4) Tannin content (prod-
37 37
DCEs can more easily parse out the effects of correlated uct of kernel mass and tannin concentrations, measured
38 38
traits on preference by pairing items so that correlations as % tannic acid equivalent or %TAE in radial diffusion
39 39
among trait levels in the final design are minimized, ide- assays).
40 40
ally to zero (Hensher et al. 2005; Kuhfeld 2010). In ad- To design the DCE scenarios, we discretized trait
41 41
dition, DCEs can take advantage of fractional factori- values into categories of low and high (2 levels), or
42 42
al designs that maximize the proportion of information low, medium and high (3 levels) by identifying natural
43 43
captured relative to the full factorial design while min- breaks in seed trait values across the candidate seed set
44 44
imizing the number of unique scenarios presented. The (Fig. 1b, step 2; see Table S2 and Suppl. Fig. 1 in Sup-
45 45
proportion of information captured in a candidate design porting Information 1 for break points). Although traits
46 46
is typically measured using an optimality criterion such may be divided into more than 3 categories, increasing
47 47
as D-efficiency (Hensher et al. 2005; Kuhfeld 2010; the number of discretized levels significantly increas-
48 48
Louviere et al. 2010). Consequently, the designed ap- es the number of combinations to be tested (Hensher et
49 49
proach avoids logistical difficulties of using full factorial. 2005). Therefore, we limited the number of levels in-
50 50
al designs where every trait level combination must be vestigated to preserve the tractability of the design. Note
51 51

M. Sundaram et al.
1 that the discretization used here applied only to the se- trait levels generated in the list do not exist in real seeds. 1
2 lection of items for the experimental design; analysis is Therefore, we eliminated 14 unrealistic combinations 2
3 based on the actual, continuous trait values of the items from the design (represented as “No seed” in Fig. 1b, 3
4 presented. step 3). 4
5 Next, a list of all unique combinations of low or me- Finally, we used PROC PLAN and OPTEX in SAS 5
6 dium or high attribute levels of the 4 seed traits was 9.2 to identify an optimal set of choice scenarios us- 6
7 compiled (Fig. 1b, step 3). Because 3 traits had 2 levels ing the remaining 10 seed types (Fig. 1b, step 4). PROC 7
8 and 1 trait had 3 levels (Table S2 in Supplementary Ma- PLAN generates a full factorial design with all possi- 8
9 terials 1), this step produced 24 (23 × 31) unique combi- ble choice scenarios (Kuhfeld et al. 1994). In our study, 9
10 nations of seed trait levels, each representing a particu- this consisted of 45 (10 choose 2) choice scenarios. 10
11 lar type of seed. Many of these trait-level combinations PROC OPTEX uses the modified Federov algorithm 11
12 are represented by naturally occurring seed species (e.g. to drop and add candidate choice scenarios and identi- 12
13 Fagus grandifolia seeds have low kernel mass, low pro- fy those that maximize an optimality criterion (Kuhfeld 13
14 tein content, low tannin content and low toughness lev- et al. 1994). For our optimality metric, we used D-effi- 14
15 els (Fig. 1b, step 3). However, many combinations of ciency, a measure of the proportion of information cap- 15
16 16
17 17
18 18
19 19
Table 1 Paired seeds (or choice scenarios) designed based on seed traits and presented to squirrels
20 20
21 Species 1 Species 2 21
22 Seed Kernel Protein Tannin Tough- Seed type Kernel Protein Tannin Tough- Seed type 22
23 pairs mass mass mass ness mass mass mass ness 23
24 1 Low Low Low Medium Corylus americana High High High Low Quercus rubra† or Quercus 24
25 prinus 25
26 2 Low Low Low Low Fagus grandifolia,† Low High Low High Carya tomentosa 26
Quercus velutina,
27 27
Quercus
28 28
muehlenbergii,
29 Quercus palustris, 29
30 chestnut hybrid, Q. 30
31 macrocarpa, Quercus 31
32 alba or Quercus 32
33 coccinea 33
34 3 Low Low High Medium Carya ovata or Carya Low Low Low Medium Corylus americana 34
35 cordiformis† 35
36 4 Low High Low Low Castanea dentata Low Low High Medium Carya ovata or Carya 36
37 cordiformis† 37
38 5 Low High Low High Carya tomentosa Low High Low Low Castanea dentata 38
39 6 Low High Low High Carya tomentosa High High Low High Juglans nigra,† Juglans 39
cinerea or Juglans regia
40 40
7 High High Low High J. nigra,† J. cinerea, Low Low Low Low F. grandifolia,† Q. velutina,
41 41
or J. regia Q. muehlenbergii, Q.
42 palustris, BC3 chestnut 42
43 hybrid, Q. macrocarpa, Q. 43
44 alba or Q. coccinea 44
45 8 High High Low High †
J. nigra, J. cinerea, High High High Medium Quercus michauxii 45
46 or J. regia 46
47 9 High High High Medium Q. michauxii Low High Low High Carya tomentosa 47
48 10 High High High Low Q. rubra† or Q. prinus High High Low Low Castanea mollissima 48
49 †
These seeds were used in the first phase of data collection. Species 1 and 2 were presented to eastern gray squirrels (Sciurus caro- 49
50 linensis) by balancing seeds on golf tees. In the second phase of data collection, whenever multiple species could represent attribute 50
51 combinations, one species was randomly selected from the list. 51

1 tured relative to a fully orthogonal and balanced design, the same time. For each presentation, an observer bal- 1
2 where 100% signifies a completely orthogonal and balanced a pair of seeds (Table 1) on golf tees spaced ap- 2
3 anced design (Kuhfeld et al. 1994). Our final design had proximately 3cm apart at the base of a tree and waited 3
4 a D-efficiency of 92% and consisted of only 10 choice for a squirrel to approach the seed pair. Observers noted 4
5 scenarios, a 78% reduction in logistical requirements the seed species selected first and its fate (cached or eat- 5
6 relative to the full factorial design (Table 1; Fig. 1b, step en). The next seed pair was then presented on the golf 6
7 4). tees, until the squirrel stopped making selections or un- 7
8 Designing the choice experiments on the basis of til data for all 10 choice scenarios was collected from 8
9 these 4 seed traits allowed us to maximize the amount the squirrel. The order in which the 10 paired trait levels 9
10 of information captured, particularly with respect to the were presented was randomized for each location. 10
11 4 traits. This step is especially important in this study The second phase of data collection occurred be- 11
12 given that traits of seeds are correlated to one another tween 4 November 2012 and 5 January 2013. During 12
13 and redundant combinations of seed traits exist within a this phase, we replaced seed species for those seed types 13
14 phylogenetic clade (Sundaram et al. 2015). Of course, in the design that could be represented by more than 1 14
15 other traits or variables can be considered post hoc in species (Table 1). When more than 2 species could rep- 15
16 analysis. The results of these post hoc assessments then resent a particular seed type, we randomized the spe- 16
17 reflect an approach more like other observational DCMs cies assignments across presentations. The final design 17
18 (McCracken et al. 1998; Cooper & Millspaugh 1999; included the same 10 choice scenarios as phase 1, but 18
19 McDonald et al. 2006; Cooper et al. 2007). We used av- a total of 20 seed species were used to represent the 10 19
20 erage trait values for each species in our analysis, which seed types in the design (Table 1). Phase 2 thus enabled 20
21 could produce misleading results if intraspecific varia- us to verify that choices truly were dependent on seed 21
22 tion in traits was large relative to interspecific variation. traits rather than on seed species alone. Presentations 22
23 Fortunately, this was not the case; interspecific variation followed the same procedure as phase 1. It was not pos- 23
24 in seed trait estimates was significantly higher than in- sible to record data blind because our study involved fo- 24
25 traspecific variation in seed trait estimates, based on 3 cal animals in the field. 25
26 different tests (details in Supplementary Materials 1). 26
27 Data analysis 27
28 Data collection and field methods 28
Seed selection by squirrels was analyzed using 2 dif-
29 To implement a DCE in consumer economic studies, ferent types of DCM. We initially fit simple multinomial 29
30 subjects are individually presented with choice scenar- logit models to the data. Then we fit a random parame- 30
31 ios and their selections are recorded. Each scenario is ters logit model to test for heterogeneity in preferenc- 31
32 replicated multiple times. Ideally, scenarios are replicates, a post hoc indicator of non-independence among 32
33 ed across multiple independent subjects, and each sub- the choice trials. Finally, we used the results from these 33
34 ject is presented with all scenarios in a randomized se- models to calculate willingness-to-trade estimates. 34
35 quence. When scenarios cannot be presented to subjects These estimates quantify the degree to which the squir- 35
36 in a completely controlled manner (a common situation rels in the study viewed different seed traits as tradable. 36
37 in field studies), a random effects model may be used to The next 3 sections describe these models and metrics 37
38 search for evidence of non-independence post hoc (see in greater detail. 38
39 section on “Random parameters logit model”). 39
Multinomial logit model
40 40
We collected choice data for the 10 choice scenari-
41 Choice experiments can be analyzed using DCMs 41
os in our study during 2 phases. The first phase occurred
42 based on random utility theory. In economics, utility is 42
between 1 September 2012 and 5 January 2013. Nine
43 interpreted as satisfaction derived from consuming or 43
seed species were used to represent the 10 seed types
44 using a product. Although utility cannot be measured di- 44
(Table 1; Fig. 1b step 4). For each choice, a pair of seeds
45 rectly, it can be inferred from choices made by individ- 45
was presented to a free-ranging eastern gray squirrel on
46 uals that select among a set of options. Random utili- 46
the Purdue University campus. Presentation locations
47 ty models assume a decision-maker (a squirrel in this 47
were separated by at least 150 m to minimize the chance
48 study) always maximizes utility by selecting the alterna- 48
that data were collected from the same small group of
49 tive with the highest utility from all available options (i.e. 49
squirrels. Exceptions to this proximity rule were made
50 1 of the seed species in a presentation) (Manski 1977). 50
only if we could present seeds to different squirrels at
51 Mathematically, utility for the ith selection event (i.e. 51

M. Sundaram et al.
1 seed pair presentation) and jth alternative (i.e. jth seed 1

2 species) is a random variable (Uij) expressed as: (6) 2
3 3
4 4
(1) Therefore, the model may be fit by setting the utility of
5 5
6 the reference alternative r to 0 and solving j − 1 simulta- 6
7 Here, Vij is the systematic portion of utility and εij is an neous linear equations for the log odds of selecting j − 1 7
8 independently and identically distributed error term with alternatives over r in a choice scenario (Louviere et al. 8
9 a Gumbel distribution. In our current use, each alterna- 2000). Derivations and further mathematical details can 9
10 tive represents a collection of traits, and selection of a be found elsewhere (McCracken et al. 1998; Louviere et 10
11 particular seed species implies selection of the trait val- al. 2000). 11
12 ues associated with that species. To determine the pref- Initially, we separated choice events into 2 datasets, 12
13 erences associated with specific traits, we defined the those for which caching was observed and those for 13
14 systematic portion of utility (Vij) as which consumption was observed. We then fit models 14
15 to each dataset using as covariates the 4 seed trait vari- 15
16 , (2) ables on which the design was based. In addition, we 16
17 created a third alternative of opting to ignore a seed 17
18 pair in order to accurately model all possible alterna- 18
where xij1, xij2, …, xijn are the n trait values associated tives available to a squirrel. This third alternative had
19 19
with the jth alternative in the ith choice scenario, β0j is a value of 0 for all seed traits and took a value of 1 for
20 20
an intercept explaining utility of the jth alternative as a an indicator variable called “ignore” if the squirrel in-
21 21
result of factors other than the n traits included in the vestigated the presented seeds but left the area without
22 22
model, and β1, β2, …, βn are preference coefficients as- taking either seed (if a seed was selected, “ignore” =
23 23
24
sociated with each of the n traits. 0). We interpret the utility of “ignore” as the value that 24
25 The DCM assumes that decision-makers select alter- the squirrel perceives in leaving to pursue other activ- 25
26 native k if Uik > Uij, k ≠ j. Therefore, the probability of ities; in other words, the utility of ignore represents 26
27 a decision-maker selecting the kth alternative is given by the missed opportunity cost associated with selecting 27
28 (Louviere et al. 2000): a seed. We separated our 4 combined variables back 28
29 into individual seed traits (e.g. calories in kernel was 29
30 (3) separated back into kernel mass and calories per gram 30
31 in models; Fig. 1a) and iteratively added each trait to 31
32 models (see Supplementary Materials 2 for a list of all 32
With a Gumbel error distribution, equation (3) reduc- models examined). At each stage, we used likelihood
33 33
es to the multinomial logit model (Boxall & Adamowicz ratio tests to examine if the addition of a seed trait im-
34 34
2002): proved fit either as a fixed effect or as a random effect
35 35
36 (Supplemental Materials 2). 36
37 (4) Both caching and seed consumption choice events 37
38 were influenced by similar traits; therefore, for our fi- 38
39 The model may be fit using the method of maximum nal models, we combined the caching and consumption 39
40 likelihood, provided that one of the potential alterna- datasets and used an indicator variable for caching (1 = 40
41 tives in the overall experiment (i.e. one of the seed cached, 0 = consumed) and interaction terms to estimate 41
42 types) is defined as a reference case and assigned a fixed different coefficients for traits when seeds were cached 42
43 utility, typically 0. The odds of selecting alternative k versus when seeds were consumed. The final model in- 43
44 over a reference alternative r are (Louviere et al. 2000): cluded kernel mass (g), toughness of shell (Nm), tannin 44
45 concentration of kernel (g per 100 g of kernel), dorman- 45
46 cy period (measured by average number of cold strat- 46
47 (5) ification days required to germinate a given seed type, 47
48 Sundaram et al. 2015) and the corresponding interaction 48
49 and the logarithm of the odds of choosing k over r is terms with caching events. Therefore, the model can be 49
50 (Louviere et al. 2000): expressed as: 50
51 51

1 Vij=β1 Kernelj+β2 Tanninj+β3 Toughj+β4 Dormancyj+β5 corporated into the model by including a random utility 1
2 Kernelj*cachej+β6 Tanninj * cachej+β7 Toughj * term for each trait in a choice set. Therefore, Equation (1) 2
3 cachej+β8 Dormancyj * cachej+β9 Ignorej. (7) can be rewritten as (Tonsor et al. 2005): 3
4 4
5 (9) 5
Furthermore, we evaluated whether “unfamiliar” seeds
6 6
influenced seed selections. We added an indicator vari-
7 7
able to the model in Equation (7) to quantify how seed where ujt is the random utility estimate for the tth indi-
8 8
familiarity influenced utility of a seed (1 = unfamil- vidual on the jth alternative distributed identically and
9 9
iar seeds which included chestnut species, 0 = famil- independently across individuals and alternatives. The
10 10
iar seeds). A seed species was considered familiar if the systematic portion of utility is still specified as a func-
11 11
tree species occurred within the geographic extent of the tion of n traits as in Equation (2). However, the addition
12 12
study area. This additional parameter resulted in the fol- of the random utility term across individuals allows for
13 13
lowing model structure: the estimation of variation around β coefficients due to
14 14
15 individual differences in preferences for each trait (Lou- 15
16 Vij=β1 Kernelj+β2 Tanninj+β3 Toughj+β4 Dormancyj+β5 viere et al. 2000) (equivalently, ujt = gtXj, so that Vij are 16
17 Kernelj * cachej+β6 Tanninj * cachej+β7 Toughj * cachej+β8 fixed effects and gt ~ N(0, Σ) are random, individual ef- 17
18 Dormancyj * cachej+β9 Novelj+β10 Ignorej. (8) fects on preference). If the estimated standard deviation 18
19 of ujt is close to 0, then variability across individuals is 19
20 considered to be negligible and suggests homogeneity 20
Finally, we examined if seed selections by squirrels
21 of preference for the trait. A standard deviation signifi- 21
were different in the first and second phases of data col-
22 cantly different from 0 indicates heterogeneity of prefer- 22
lection. Given that seed types used to represent the de-
23 ences for a specific trait (Hensher et al. 2005). 23
sign were different in the 2 phases of data collection,
24 this question examines whether combinations of seed We fit a random parameters logit model for the traits 24
25 traits explain seed selection by squirrels rather than spe- in the best-fit model of the previous section, under the 25
26 cies identity. First, we compared mean β ± 1 SE in- assumption that all selection events made at the same 26
27 tervals of coefficients for traits from multinomial log- location were made by the same squirrel. Although we 27
28 it models fit separately to the first and second phases did not mark individual squirrels in this study, observers 28
29 of data collection to verify that choices were driven by were able to visually track individuals between presen- 29
30 combinations of seed traits rather than seed types. If in- tations at a given location. While we cannot be certain 30
31 tervals from the 2 models overlap for each trait, no sig- that the same individuals did not visit multiple loca- 31
32 nificant differences between models can be concluded at tions, we believe this to be unlikely given our attempts 32
33 α = 0.05 (Payton et al. 2003). Second, we added an in- to segregate locations geographically. Therefore, we es- 33
34 dicator variable to quantify how the phase of data col- timated heterogeneity in squirrel preferences for traits 34
35 lection influenced seed selection (phase = 1 if the choice across locations as a post hoc check for variability in in- 35
36 event belonged to the second phase of data collection, dividual preferences. The model was fit using programs 36
37 phase = 0 if the choice event occurred in the first phase NLOGIT 6 and LIMDEP 11 (Greene 2011). 37
38 of data collection). Likelihood ratio tests were then used Willingness-to-trade 38
39 to examine if including phase and interactions terms as 39
An advantage of discrete choice models is the abili-
40 covariates improved fit. 40
ty to estimate an animal’s willingness to exchange value
41 41
Random parameters logit model derived from different traits using willingness-to-trade,
42 42
If individuals in a population are heterogeneous in also referred to as the marginal rate of substitution (Coo-
43 43
their preferences for modeled traits, then inferences per & Millspaugh 1999). The marginal utility of a con-
44 44
made from the multinomial logit model can be incorrect tinuous trait g is defined as the partial derivative of utili-
45 45
(Olynk et al. 2010). Conversely, if no heterogeneity ex- ty with respect to g, . The exchange rate between g
46 46
ists among individual choice events, then either a single and another trait, h, is obtained by scaling the marginal
47 47
squirrel was responsible for all choices (impossible, giv- utility of h so that it offsets a unit change in g and main-
48 48
en our sampling design) or squirrels behave consistently tains the same constant utility value. That is, we need to
49 49
50 with one another and individuality can, therefore, be ig- know the value Wg,h, where and is inter- 50
51 nored from a statistical perspective. Heterogeneity is in- preted as the amount of trait h that must be increased or 51

M. Sundaram et al.
1 decreased to offset a unit increase in trait g in order to WTT estimates for such traits because ratios involving 1
2 maintain a constant utility. In a multinomial logit model 0 would either yield 0 or approach infinity and generate 2
3 that lacks interaction terms, , so, uninformative estimates. Using the Krinsky and Robb 3
4 (1986) method, we also calculated the 95% CIs around 4
5
(10) WTT estimates (Supplementary Materials 3). If 95% CI 5
6 around a WTT estimate incorporates 0, then the WTT 6
7 estimate is deemed not significant. 7
8 For the multinomial logit models specified in Equations 8
9 (7) and (8), the WTT estimate of kernel for tannins can 9
be given as
RESULTS
10 10
11 We presented a total of 426 seed pairs to eastern 11
12 (11) gray squirrels across 27 locations on the Purdue cam- 12
13 pus, of which 307 (71.8%) trials resulted in seeds being 13
14 cached, 118 (27.7%) resulted in seeds being consumed, 14
where “cache” is the indicator variable used to repre- and 1 seed pair was ignored. The coefficient estimates
15 15
sent seed fate. Therefore, for a consumed seed the above were not significantly different between first and second
16 16
equation becomes . Note that these calculations phases of data collection (mean β ± 1 SE intervals over-
17 17
are valid only if no heterogeneity exists in preferenc- lapped for all traits). Furthermore, including phase as
18 18
es. We estimated the 95% confidence intervals around an indicator variable did not improve models of squir-
19 19
mean marginal utilities and willingness-to-trade esti- rel selection of seed traits (P = 1, Table 2), providing
20 20
mates for all seed traits from our best-fit multinomi- verification that selection of seed traits by squirrels did
21 21
al logit model using the method of Krinsky and Robb not change when seed types represented in the design
22 22
(1986). The Krinsky and Robb method relies on random changed. Therefore, we pooled data from the 2 phases
23 23
24
draws of β estimates using the mean β estimate for each of data collection for purposes of fitting models. 24
25
seed trait and the estimated variance accompanying 25
Grams of protein, protein concentration, lipid con-
26
each β. For cached seeds, random draws for the inter- 26
centration and energy concentration failed to explain
27
action terms with caching were added to random draws 27
seed selection by squirrels for either consumed or
28
for the beta estimates of consumed seeds to obtain mar- 28
cached seeds. From likelihood ratio tests, grams of pro-
29
ginal utilities. If the 95% CI around the marginal utility 29
tein did not influence seed selection (P = 0.88, P = 0.51
30
of a trait overlapped 0, then the trait was considered to 30
for caching and consumption events; Supplementary
31
have a utility that is not significant. We did not compute 31
Materials 2). In addition, when protein concentration re-
32 32
33 33
34 Table 2 Comparisons of models fit to seed selection events 34
35 35
36 Model AIC Δ AIC Pseudo R2 Log- Number of P-value 36
37 likelihood parameters 37
38 Ignore + Tannin + Kernel + Tough + Dormancy + Tannin (*cache) 531.4 3.9 0.094 −256.72 9 — 38
39 + Kernel (*cache) + Tough (*cache) + Dormancy (*cache) 39
40 Ignore + Phase*Tannin + Phase*Kernel + Phase*Tough + 537.9 10.4 0.120 −251.97 17 1 40
41 Phase*Dormancy + Phase*Tannin (*cache) + Phase*Kernel 41
42 (*cache) + Phase*Tough (*cache) + Phase*Dormancy (*cache) 42
43 Ignore + Tannin + Kernel + Tough + Dormancy + Tannin (*cache) 531.4 3.9 0.094 −256.72 9 — 43
44 + Kernel (*cache) + Tough (*cache) + Dormancy (*cache) 44
45 Ignore + Unfamiliar + Tannin + Kernel + Tough + Dormancy 527.5 0 0.104 −253.74 10 0.014 45
46 + Tannin (*cache) + Kernel (*cache) + Tough (*cache) + 46
47 Dormancy (*cache) 47
48 The bold-faced model (Equation 8 in text) was selected as the best-fit model. For each model, AIC value, ΔAIC value, pseudo R2 48
49 value, log-likelihood value and number of parameters is provided. The P-value associated with the likelihood ratio test statistic is 49
50 provided for each pair of models to examine how model fit changes with added predictors ( — in the P-value column represents the 50
51 null model in a likelihood ratio test). 51

1 placed shell toughness in the model for cached seeds, fit 1

2 of the model worsened (ΔAIC = 27; Supplementary Ma- 2
3 terials 2) and protein concentration was not a significant 3
4 predictor of seed selection. Energy concentration (cal- 4
5 ories per g) did not improve fit for cached or consumed 5
6 seeds (P = 0.54, P = 0.21 for caching and consumption 6
7 events; Supplementary Materials 2). Percentage lipid 7
8 concentration also did not improve fit for cached seeds (P 8
9 = 0.33; Supplementary Materials 2). Kernel mass, tan- 9
10 nin concentration and shell toughness consistently in- 10
11 fluenced seed selection and, therefore, all caching and 11
12 consumption events were considered together in mod- 12
13 els (Table 2). The model incorporating unfamiliarity of 13
14 seeds (Equation 8) was the “best” model as determined 14
15 with likelihood ratio tests (Table 2). 15
16 Coefficients for interactions of seed traits with cach- 16
17 ing were significant. In particular, the utility of seeds 17
18 chosen for consumption was positively associated with Figure 2 Utility contour lines plotted for varying seed tough- 18
19 toughness of seed and negatively with species’ dorman- ness and dormancy period. Contours represent lines of indif- 19
20 ference at utilities of 1, 0.02 and −1, assuming a familiar seed 20
cy periods (Table 3). However, the utility of seeds being
21 with 0.5% TAE tannin concentration and 1-g kernel mass that 21
cached was positively related to tannin concentration,
22 is consumed instead of cached. The lines of indifference rep- 22
kernel size and dormancy period. Finally, a significant
23 resent combinations of toughness and dormancy viewed as 23
negative coefficient was found for the “ignore” vari- tradable by eastern gray squirrels (Sciurus carolinensis) at the
24 able and a significant positive coefficient was associated 24
25 specified utility. Dormancy and toughness plotted for Quercus 25
with seed unfamiliarity (Table 3). The random parame- macrocarpa, Quercus prinus, Juglans nigra, Juglans cinerea
26 ters logit model showed no significant variation in pref- 26
and Castanea dentata. The 2 Quercus species and 2 Juglans
27 erences among locations with respect to any trait (P > 0.5 27
species occur close to line of indifference at 0.02, indicating
28 for all trait standard deviations, suggesting that the es- 28
that these seeds are approximately equally likely to be con-
29 timated standard deviations of random trait coefficients 29
sumed in pairwise trials. In contrast, C. dentata occurs below
30 are not significantly different from 0). When cach- the −1 line of indifference, indicating that this species is less 30
31 ing and consumption events were examined separate- likely to be consumed when presented with one of the Quercus 31
32 ly, models incorporating random effects did not improve or Juglans species. 32
33 33
34 34
35 35
36 Table 3 Best-fit model (Equation 8) parameterized for choices made by eastern gray squirrels (Sciurus carolinensis) 36
37 Coefficient Standard error t-ratio P-value Lower CI Upper CI 37
38 38
Tannin (%) −0.071 0.058 −1.220 0.222 −0.184 0.043
39 39
40 Tannin*Caching 0.212** 0.067** 3.164** 0.002** 0.081** 0.344** 40
41 Kernel (g) −0.104 0.139 −0.748 0.455 −0.377 0.169 41
42 Kernel*Caching 0.514** 0.169** 3.051** 0.002** 0.184** 0.845** 42
43 43
Toughess (Nm) 0.477** 0.160** 2.983** 0.003** 0.163** 0.790**
44 44
45 Toughess*Caching −0.148 0.183 −0.809 0.419 −0.507 0.211 45
46 Ignore −5.616** 1.172** −4.791** 0.000** −7.913** −3.319** 46
47 Dormancy (days) −0.021** 0.007** −3.133** 0.002** −0.035** −0.008** 47
48 48
Dormancy*Caching 0.023** 0.008** 2.848** 0.004** 0.007** 0.040**
49 49
50 Unfamiliar 0.660** 0.272** 2.423** 0.015** 0.126** 1.194** 50
51 **Variables significant at α of 0.05. CI, confidence interval. 51

M. Sundaram et al.
1 Table 4 Marginal utilities and willingness-to-trade estimates for all seed traits 1
2 Marginal utilities of consumed seeds Marginal utilities of cached seeds 2
3 3
Variable Lower CI Mean Upper CI Lower CI Mean Upper CI
4 4
5 Kernel mass (g) −0.366 −0.100 0.165 0.212 0.412 0.607 5
6 Toughness (Nm) 0.152 0.472 0.804 0.106 0.328 0.548 6
7 Dormancy (days) −0.035 −0.021 −0.009 −0.007 0.002 0.011 7
8 8
Tannin −0.183 −0.072 0.041 0.034 0.142 0.249
9 9
concentration (%)
10 10
Willingness-to-trade column attribute (x) for row Willingness-to-trade column attribute (x) for row
11 attribute (y) when seeds are consumed attribute (y) when seeds are cached 11
12 Kernel Toughness Dormancy Tannin Kernel Toughness Dormancy Tannin 12
13 mass (g) (Nm) (days) concentration (%) mass (g) (Nm) (days) concentration (%) 13
14 Kernel mass (g) NA — — — NA −1.254 — −2.899 14
15 15
Toughness (Nm) — NA 22.365 — −0.798 NA — −2.313
16 16
Dormancy (days) — 0.045 NA — — — NA —
17 17
18 18
19 19
20 20
Tannin — — — NA −0.345 −0.432 — NA
21 concentration (%) 21
22 22
Mean marginal utilities and 95% confidence intervals for seed traits under conditions of caching and consumption. Mean willing-
23 ness-to-trade estimates for all paired seed traits (— provided when WTT could not be computed because the marginal utility for a 23
24 seed trait overlapped 0). 24
25 25
26 26
27 27
28 28
model fit (P = 1.00, P = 0.76 for caching and consump- content could compensate for reduced toughness.
29 29
tion datasets; Supplementary Materials 2) To aid in interpretation of WTT estimates, we plot-
30 30
31 Willingness-to-trade (WTT) estimates were calculat- ted lines of indifference for consumed seeds using coef- 31
32 ed separately for consumed and cached seeds, and var- ficient estimates from Equation (8) (Fig. 2). Lines of in- 32
33 ied for different combinations of traits (Table 4). Mean difference represent combinations of 2 traits for which 33
34 WTT estimates were not computed for 8 of 12 seed trait utility remains constant. The slope of the line is equal to 34
35 combinations because at least 1 of the marginal utilities the WTT estimate for the 2 traits. For consumed seeds 35
36 in the combination was not significantly different from 0 we created 3 lines of indifference to represent tradeoffs 36
37 (Table 4, confidence intervals for WTT estimates over- between toughness and dormancy period at utilities of 1, 37
38 lap 0 for 8 of 12 seed trait combinations in Supplemen- 0.02 and −1. For all 3 lines, we set tannin concentration 38
39 tary Materials 3). For consumed seeds, WTT estimates at 0.5% TAE, kernel mass at 1 g and assumed seed type 39
40 between toughness and dormancy period were positive, was familiar. Overlaying the seed species with the high- 40
41 indicating that for perceived value of a seed to remain est estimated utility when consumed in Fig. 2, Juglans 41
42 constant, any change in dormancy period would need cinerea, Juglans nigra, Quercus macrocarpa and Quer- 42
43 to be accompanied by a corresponding, parallel change cus prinus fall close to the line of indifference at utili- 43
44 in toughness, or vice versa. For cached seeds, negative ty of 0.02, indicating that these seeds are approximately 44
45 WTT estimates were computed between kernel mass equivalent in terms of their desirability for consump- 45
46 and toughness, kernel mass and tannins, and toughness tion. All other seeds were of lower utility when select- 46
47 and tannins. Therefore, for a seed to maintain a con- ed for consumption and, if plotted, would occur below 47
48 stant desirability for caching relative to other species, the −1 line of indifference (see e.g. Castanea dentata in 48
49 a decrease in kernel mass could be offset by increased Fig. 2). 49
50 toughness or increased tannin content, or a mixture of For cached seeds, more complex trading scenarios 50
51 both. Similarly, an increase in kernel mass or tannin occurred. In this case, finite WTT estimates for 3 traits 51

1 DISCUSSION 1
2 2
3 Seed selection by eastern gray squirrels 3
4 4
5
The methods and models outlined here allow some 5
6
general inferences to be made about seed selection by 6
7
eastern gray squirrels and also provide novel insights 7
8
regarding the effects of seed traits on squirrel forag- 8
9
ing and caching behaviors. Species identity, reflected 9
10
by the different phases of data collection in this study, 10
11
did not alter coefficient estimates predicting seed selec- 11
12 tion by squirrels. This result strongly suggests that seed 12
13 trait combinations, as opposed to species identities, in- 13
14 fluenced selections made by the eastern gray squirrels 14
15 in our experiment. Understanding the selection of seed 15
16 Figure 3 Utility contour surface plotted for varying seed traits by squirrels gives insight into the evolution of seed 16
17 toughness, kernel mass and tannin concentration. Surface of in- construction strategies in hardwood trees and may also 17
18 difference (black dashed line, equation: toughness = 3.03 − 1.24 allow us to anticipate the animals’ functional responses 18
19
× kernel − 4.24 × tannin) plotted at a utility of 1 for a cached, to changes in neighborhood or community composition. 19
20
familiar seed representing combinations of toughness, kernel Furthermore, it emphasizes the argument that seed se- 20
mass and tannins viewed as tradable by eastern gray squirrels lection by rodents is context-dependent, where the com-
21 21
(Sciurus carolinensis). Toughness, kernel size and tannin con- bination of traits of available seeds influences seed se-
22 22
centration plotted for Quercus macrocarpa, Quercus velutina, lection as opposed to species identity.
23 Carya ovata and Juglans species. Q. macrocarpa, Q. velutina 23
24 The preferences for seed traits observed in this study 24
and C. ovata occur on the same surface of indifference, indicat-
25 corroborate previous findings. We found that the per- 25
ing that these seeds are tradable with one another. In contrast,
26 the 3 Juglans species occur well above the plotted surface of
centage of tannins in seed kernels significantly increased 26
27 indifference, indicating that they are more likely to be cached utility for caching and decreased utility for consump- 27
28 when presented with Q. macrocarpa, Q. velutina and C. ovata. tion. Steele et al. (1993) show that eastern gray squirrels 28
29 are sensitive to tannins and reject portions of kernels 29
30 that contain relatively high concentrations of tannins 30
31 when consuming seeds, but also preferentially cache 31
32 high-tannin seeds. The mass of a seed may be increased 32
33 yield a 3-D surface that represents combinations of ker- either by increased shell mass (and, thus, toughness) or 33
34 nel mass, toughness and tannin concentrations with by increased kernel mass, and our results show that both 34
35 equivalent utility (Fig. 3; the dashed line grid shows the factors positively influence utility for caching. Seed 35
36 surface at utility = 1). Coefficients of the surface equa- coat toughness was a significant and positive predictor 36
37 tion equal the WTT estimates and are interpreted for 2 of seed selection for caching by eastern gray squirrels 37
38 tradable traits while maintaining the third trait a con- (Table 3), which supports previous assertions that se- 38
39 stant. For example, if tannin concentrations remain un- lection by squirrels in the fall follows a time-minimiza- 39
40 changed, then a constant utility is predicted if a 1 g in- tion strategy rather than an energy-maximization strat- 40
41 crease in kernel mass is accompanied by a 1.25 Nm egy (Thompson & Thompson 1980). Specifically, seeds 41
42 decline in toughness (Table 4). We plotted observed val- with tougher shells have a higher consumption time and 42
43 ues of toughness, tannins and kernel mass for Quercus are cached more often to reduce time spent handling 43
44 velutina, Q. macrocarpa, Carya ovata, and all of the Jug- and, hence, increase time available to cache additional 44
45 lans species used in our study. Three species, Q. veluti- seeds in the fall (Thompson & Thompson 1980; Jacobs 45
46 na, Q. macrocarpa and C. ovata, occur close to the plot- 1992; Sundaram et al. 2015). It should be noted, howev- 46
47 ted surface of indifference, suggesting that these species er, that in our seed set toughness is highly correlated to 47
48 have equivalent desirability for caching. In contrast, all lipid concentration and, thus, to energy content; seeds of 48
49 Juglans species occur well above the surface and are, Betulaceae and Juglandaceae (including walnuts and ha- 49
50 therefore, much more desirable than the other species as zelnuts) have evolved both thicker shells and higher lip- 50
51 cached resources. id concentrations when compared to seeds of Fagaceae 51

M. Sundaram et al.
1 (including oaks and chestnuts) (Sundaram et al. 2015). typically studied in consumer economics. When it oc- 1
2 Thus, the selection by squirrels of seeds with tough curs, heterogeneity in responses of free-ranging animals 2
3 shells in conjunction with long dormancy periods results is sometimes attributed to differences in experience or 3
4 in the selection of the highly energetically valuable wal- age and differences in effort expended by individuals 4
5 nuts (J. nigra and J. cinerea) for consumption (Table 3 (Mauck et al. 2012). Homogenous responses such as 5
6 and Fig. 2). Utility for caching was also influenced pos- those we observed are likely a result of natural selection. 6
7 itively by kernel mass (Table 3). All else being equal, For free-ranging squirrels, creation and maintenance of 7
8 large kernels reflect high total caloric content, thus mak- caches during fall is critical to ensure survival during 8
9 ing seeds more attractive to rodents and increasing the winter when food is scarce (Nixon et al. 1975). In a sep- 9
10 probability of seeds being cached (Grubb & Burslem arate study, we determined that significant proportions 10
11 1998; Yang et al. 2012). of foraging behaviors of eastern gray squirrels were ex- 11
12 Apart from corroborating existing findings, we tested plained by hardwood tree phylogeny, and that the seed 12
13 and found additional seed trait trends that merit further traits explored in this study are likely constrained by 13
14 attention. Total protein content of a seed was not a sig- phylogeny to different degrees (Sundaram et al. 2015). 14
15 nificant predictor of utility, in contrast to systems where The consistency in selection of seed traits that we ob- 15
16 rodents were found to maximize protein intake (Hen- served also indicates that squirrels may exert selective 16
17 derson 1990). Although several studies have found that pressures on hardwood trees by consistently caching 17
18 seeds with higher protein content are selected more fre- seeds with large kernel sizes, tougher shells and higher 18
19 quently (reviewed by Lichti et al. 2017), these studies tannin concentrations so long as the seed attributes are 19
20 generally have relied on observational, post hoc correla- heritable. Some nut morphological characters, such as 20
21 tions and could not unambiguously attribute selection weight and thickness, show very high broad-sense heri- 21
22 to protein content. Finally, we found a significant pos- tability, indicating that the environment influences these 22
23 itive effect of unfamiliar seeds on utility, implying that traits to a minimal extent (Lan-Ying et al. 2009). To our 23
24 squirrels were more likely to choose unfamiliar chest- knowledge, no studies have explored heritability of tan- 24
25 nut seeds regardless of seed fate. While the addition of nin concentrations of hardwood tree seeds, although a 25
26 this variable significantly improved the fit of the model genetic basis has been established for seed coat tannins 26
27 (Table 2), we cannot determine from our results wheth- in other plant species (Caldas & Blair 2009). 27
28 er squirrels are more likely to select chestnuts due to an Our analysis of squirrels’ marginal utilities and will- 28
29 unobserved trait or if squirrels are more likely to select ingness-to-trade among different seed traits provides 29
30 chestnuts because they are unfamiliar. Studies directly new insights into the interactions between trees and 30
31 comparing familiar and unfamiliar seeds are required to scatter-hoarding rodents. Previous theoretical and em- 31
32 address this question. pirical work has shown that seed dispersal effectiveness 32
33 in hoarded trees depends in part on the composition of 33
Squirrels in our study were strikingly consistent in
34 seed crops available to rodents and the rodents’ percep- 34
their preferences for seed traits. Such consistency in se-
35 tion of different species values for consumption or cach- 35
lection of seed traits across individuals (see section on
36 ing (Lichti et al. 2014; Sundaram et al. 2017). Trees that 36
“random parameters logit model” and Supplementary
37 increase the consumption utility of their seeds relative 37
Materials 2), and extreme preferences for a trait (WTT
38 to their neighbors while decreasing their relative cach- 38
estimates approach 0 for 8 of 12 seed trait combinations;
39 ing utility could face significant negative selective pres- 39
Supplementary Materials 3), is not well documented in
40 sure. In contrast, any tree that can decrease consumption 40
the ecological literature. Consistency in seed preference
41 utility and increase caching utility stands to benefit from 41
of rodents has been noted in other studies (Lobo et al.
42 hoarder behavior. For example, either a decrease in the 42
2009) as well as reviews that have compared preferenc-
43 length of overwinter dormancy or an increase in tough- 43
es among rodent consumers (Vander Wall 2010; Lichti
44 ness (and energy content) will make seeds more attrac- 44
et al. 2017). In contrast to rodents, consumer economic
45 tive for consumption (Table 3). However, assuming that 45
discrete choice models evaluating the utility of a prod-
46 neighbors remain static, simultaneous increases in both 46
uct with respect to product traits typically find hetero-
47 toughness and dormancy as seen in the shift from white 47
geneity in responses of human consumers (Ortega et al.
48 oaks (Quercus section Quercus) to walnuts (Juglans) 48
2011). Human heterogeneity is attributed to differences
49 could maintain a tree’s desirability for consumption at a 49
in taste of consumers (Olynk & Ortega 2013) and prob-
50 constant level and avoid placing the tree at a disadvan- 50
ably reflects the low fitness consequences of decisions
51 51

1 tage to its neighbors. ences for hardwood seeds. Developmental and evolu- 1
2 In contrast to consumption, WTT estimates for tionary constraints have resulted in a high degree of cor- 2
3 cached seeds were negative (Table 4). As a result, in- relation for some seed traits (Sundaram et al. 2015); 3
4 creased kernel mass (e.g. Q. macrocarpa), increased consequently, some combinations of traits are not found 4
5 toughness (e.g., C. ovata), increased tannin concentra- in real seeds of any species. This issue of biological- 5
6 tions (e.g. Q. velutina) or increases in any combina- ly missing trait combinations is likely to occur in many 6
7 tion of these traits would increase a seed’s desirability systems. Missing traits may also arise if traits are inade- 7
8 for caching (Fig. 3). Given such conditions, squirrel be- quately sampled, although this is not likely for the seeds 8
9 havior might be expected to drive diversification of seed used here (Sundaram et al. 2015). If only a few trait 9
10 types as various species develop along alternative evo- combinations are missing, a choice experiment may still 10
11 lutionary pathways. For instance, nuts from the family be designed with existing traits. D-efficiency of the re- 11
12 Juglandaceae have evolved tough shells, whereas acorns sulting design (or the amount of information captured 12
13 from Quercus section Lobatae have evolved high tannin by the resulting design relative to a fully orthogonal and 13
14 concentrations (Sundaram et al. 2015). Yet, our results balanced design), however, will be lower when com- 14
15 indicate that these seeds are functionally similar (Fig. 2). pared to a design incorporating missing trait combina- 15
16 From a squirrel’s perspective, the flexibility that results tions. 16
17 from trading off a larger number of traits may allow The interpretation of coefficients and intercepts must 17
18 hoarders to reduce the time spent examining seeds in the be carefully considered when planning discrete choice 18
19 fall when competition to cache is high and when there is experiments based on traits. Our experiment used a de- 19
20 a need to cache high numbers of seeds to ensure avail- sign that was optimized over combinations of trait val- 20
21 ability of food during winter (Thompson & Thompson ues, not combinations of seed species. During data 21
22 1980). collection, each combination of trait values could be 22
23 represented by any of a number of actual seed species. 23
To further test the hypothesis that seed traits such
24 Consequently, structuring our models so that the inter- 24
as kernel size and tannin concentration are truly inter-
25 cept terms corresponded to one of the alternatives in a 25
changeable from the squirrel’s point of view at the esti-
26 presentation (i.e. for alternative 1 or alternative 2) would 26
mated WTT values, an artificial seed experiment could
27 not have yielded interpretable results because species 27
be devised wherein these traits are manipulated. For ex-
28 representing the alternative changed across scenarios. 28
ample, an artificial seed with 2.9% TAE tannin concen-
29 Instead, we estimated an intercept for the third alterna- 29
tration could be paired with a second seed having a 0%
30 tive available to squirrels (i.e. ignoring a seed pair). We 30
TAE but a larger kernel mass (larger by 1 g). If the 2
31 interpret this coefficient as the perceived value of leav- 31
constructed seeds have the same total perceived value,
32 ing to forage elsewhere or pursue other activities. It 32
as is suggested by our model, then a squirrel should be
33 therefore provides an estimate of the missed opportuni- 33
equally likely to cache either of the constructed seeds,
34 ty cost associated with selecting seeds. Allowing for the 34
all else being equal. Furthermore, a discrete choice ex-
35 option to ignore alternatives and not make a choice be- 35
perimental framework may be useful to evaluate selec-
36 tween available alternatives is important in certain deci- 36
tion of traits in an artificial seed study to shed light on
37 sion-making frameworks, but is often overlooked in tra- 37
how squirrels perceive artificial seeds and assess wheth-
38 ditional ecological resource selection models. 38
er results from artificial seed experiments can be extrap-
39 39
olated to hardwood tree seeds. We designed our experiment with algorithms that
40 40
maximize D-efficiency and used a multinomial log-
41 Applying discrete choice experiments to an 41
it modeling approach, but alternatives may be appropri-
42 ecological system 42
ate in some instances. D-efficiency is computed from
43 43
the geometric mean of the determinants of candidate de-
44 We believe our study is the first to present discrete 44
signs. Although D-efficiency is considered to be a stan-
45 choice experiments to free-ranging animals based on 45
dard metric representing the total information captured
46 a design informed by pre-determined traits. In theory, 46
by a design, other metrics have been proposed, includ-
47 a main advantage of DCEs is the ability to create bal- 47
ing A-efficiency and G-efficiency (Kuhfeld et al. 1994).
48 anced and orthogonal fractional designs with correlated 48
A-efficiency is computed from the arithmetic mean of
49 traits. In practice, while we successfully parsed out the 49
determinants of candidate designs, while G-efficiency
50 effects of moderately correlated traits, we faced some 50
is derived from the maximum standard error for predic-
51 constraints in applying the approach to squirrel prefer- 51

M. Sundaram et al.
1 tion from a candidate design (Kuhfeld et al. 1994). Sim- nin concentration, kernel mass and toughness of a seed. 1
2 ilarly, alternatives exist to the multinomial logit frame- Moreover, we found that selection for traits is remark- 2
3 work used here, including Bayesian choice models that ably consistent across squirrels. Our approach allowed 3
4 offer additional flexibility when compared to tradition- us to compute tradeoffs between seed traits and show 4
5 al choice models. For example, Bayesian methods may that squirrels may trade among multiple traits when 5
6 include non-normal distributions for coefficients and al- caching seeds and between fewer traits when consum- 6
7 low for situations where the full choice set is unknown ing seeds. Our results suggest that tradeoffs by squir- 7
8 (Train 2003). rels during fall may have substantial implications for the 8
9 Overall, DCEs and DCMs offer a powerful tool for evolution and diversification of nut-bearing trees. 9
10 modeling and designing experiments to investigate re- 10
11 11
source selection by consumers. Traditional foraging ACKNOWLEDGMENTS
12 studies generally model consumer preference for 1 cur- 12
13 rency at a time (e.g. caloric value of food item, Stephens The Fred M. Van Eck Forest Foundation for Purdue 13
14 et al. 2007) or examine preferences across artificial food University and the McIntire-Stennis program provid- 14
15 items (e.g. clay “seeds” with manipulated attributes; ed funding. We thank P. A. Zollner and M. A. Steele for 15
16 Wang & Chen 2008). As previous studies have already valuable comments. 16
17 described, DCMs are broadly applicable in studies of 17
18 resource selection and allow consumer preference to be REFERENCES 18
19 examined for multiple attributes and, thus, multiple cur- 19
20 Baasch DM, Tyre AJ, Millspaugh JJ, Hygnstrom SE, 20
rencies simultaneously (Buskirk & Millspaugh 2006).
21 Vercauteren KC (2010). An evaluation of three statis- 21
However, most DCM applications in ecology rely on
22 tical methods used to model resource selection. Eco- 22
observational data and lack clearly defined choice sets
23 logical Modelling 221, 565–74. 23
or the set of available resources that animals select from
24 is generally unknown (Baasch et al. 2011). Designed Boxall PC, Adamowicz W (2002). Understanding heter- 24
25 choice experiments help to address both of these issues. ogenous preferences in random utility models: A la- 25
26 By using multiple different resource pairs to construct tent class approach. Environmental & Resource Eco- 26
27 the same orthogonal contrasts among levels of focal nomics 23, 421–46. 27
28 traits, the DCE controls the choice set used in model- Browning EK, Zupan MA (2015). The theory of con- 28
29 ing and averages over any other functional traits that re- sumer choice. In: Microeconomics: Theory and Ap- 29
30 sources may possess as well as the effects of environ- plications. John Wiley & Sons, Hoboken, pp. 42–79. 30
31 mental context on decisions. As a result, DCEs reduce Buskirk SW, Millspaugh JJ (2006). Metrics for studies 31
32 the risk that spurious results might be generated by lurk- of resource selection. Journal of Wildlife Manage- 32
33 ing variables and, therefore, support stronger inferences. ment 70, 358–66. 33
34 Although our specific approach is most directly appli- 34
Caldas GV, Blair MW (2009). Inheritance of seed con-
35 cable to diet and some mate selection systems, it may 35
densed tannins and their relationship with seed-coat
36 be possible to construct similar approaches for other as- 36
color and pattern genes in common bean (Phaseolus
37 pects of resource selection by conducting a post hoc 37
vulgaris L.). Theoretical and Applied Genetics 119,
38 classification of observed selection events, followed by 38
131–42.
39 a random subsampling of the classified events for analy- 39
40 Chambers JC, MacMahon JA (1994). A day in the life 40
sis.
41 of a seed – Movements and fates of seeds and their 41
42 implications for natural and managed systems. Annu- 42
43
CONCLUSIONS al Review of Ecology and Systematics 25, 263–92. 43
44 Our study illustrates a designed approach to studying Cooper AB, Millspaugh JJ (1999). The application of 44
45 seed trait selection by eastern gray squirrels. We suggest discrete choice models to wildlife resource selection 45
46 that the designed approach is useful for studying effects studies. Ecology 80, 566–75. 46
47 of traits on resource selection, especially when testing Cooper AB, Pettorelli N, Durant SM (2007). Large car- 47
48 in a full factorial design is not feasible. We showed that nivore menus: Factors affecting hunting decisions by 48
49 eastern gray squirrels choosing between 2 seeds select cheetahs in the Serengeti. Animal Behaviour 73, 651– 49
50 for specific seed traits, including dormancy period, tan- 9. 50
51 51

1 Delgado MM, Nicholas M, Petrie DJ, Jacobs LF (2014). Choice Methods. Cambridge University Press, Cam- 1
2 Fox squirrels match food assessment and cache effort bridge, UK. 2
3 to value and scarcity. PloS One 9, e92892. Manski CF (1977). The structure of random utility mod- 3
4 Greene WH (2011). LIMDEP Student Reference Guide. els. Theory and Decision 8, 229–54. 4
5 Econometric Software, 1–244. 5
Mauck RA, Huntington CE, Doherty PF (2012). Experi-
6 6
Grubb PJ, Burslem DFRP (1998). Mineral nutrient con- ence versus effort: What explains dynamic heteroge-
7 7
centrations as a function of seed size within seed neity with respect to age? Oikos 121, 1379–90.
8 8
crops: implications for competition among seedlings McCracken ML, Manly BFJ, Heyden MV (1998). The
9 9
and defence against herbivory. Journal of Tropical use of discrete-choice models for evaluating resource
10 10
Ecology 14, 177–85. selection. Journal of Agricultural, Biological, and
11 11
12 Henderson CB (1990). The influence of seed apparen- Environmental Statistics 3, 268–79. 12
13 cy, nutrient content and chemical defenses on dietary McDonald TL, Manly BFJ, Nielson RM, Diller LV 13
14 preference in Dipodomys ordii. Oecologia 82, 333– (2006). Discrete-choice modeling in wildlife studies 14
15 41. exemplified by northern spotted owl nighttime hab- 15
16 Hensher DA, Rose JM, Greene WH (2005). Applied itat selection. Journal of Wildlife Management 70, 16
17 Choice Analysis: A Primer. Cambridge University 375–83. 17
18 Press, New York, NY. Nixon CM, McClain MW, Donohoe RW (1975). Effects 18
19 Jacobs LF (1992). The effect of handling time on the de- of hunting and mast crops on a squirrel population. 19
20 cision to cache by gray squirrels. Animal Behaviour The Journal of Wildlife Management 39, 1–25. 20
21 43, 522–4. Olynk NJ, Ortega DL (2013). Consumer preferences 21
22 Krinsky I, Robb AL (1986). On approximating the sta- for verified dairy cattle management practices in pro- 22
23 tistical properties of elasticities. The Review of Eco- cessed dairy products. Food Control 30, 298–305. 23
24 nomics and Statistics 68, 715–19. 24
Olynk NJ, Tonsor GT, Wolf CA (2010). Consumer will-
25 25
Kuhfeld WF (2010). Marketing research methods in ingness to pay for livestock credence attribute claim
26 26
SAS. SAS Institute, Cary, NC. Available from URL: verification. Journal of Agricultural and Resource
27 27
http://citeseerx.ist.psu.edu/viewdoc/download?- Economics 35, 261–80.
28 28
doi=10.1.1.163.8176&rep=rep1&type=pdf. Ortega DL, Wang HH, Wu LP, Olynk NJ (2011). Mod-
29 29
30
Kuhfeld WF, Tobias RD, Garratt M (1994). Efficient ex- eling heterogeneity in consumer preferences for se- 30
31
perimental-design with marketing-research applica- lect food safety attributes in China. Food Policy 36, 31
32
tions. Journal of Marketing Research 31, 545–57. 318–24. 32
33 Zhou LY, Xiao Q, Zhang L, Zhang SJ, Wang PB (2009). Payton ME, Greenstone MH, Schenker N (2003). Over- 33
34 Analysis of combining ability and heritability about lapping confidence intervals or standard error inter- 34
35 nut quality of walnut. Botany Research Journal 2, vals: what do they mean in terms of statistical signifi- 35
36 24–7. cance? Journal of Insect Science 3, 34–40. 36
37 Lichti NI, Steele MA, Zhang H, Swihart RK (2014). Preston SD, Jacobs LF (2009). Mechanisms of cache 37
38 Mast species composition alters seed fate in North decision making in Fox squirrels (Sciurus Niger). 38
39 American rodent-dispersed hardwoods. Ecology 95, Journal of Mammalogy 90, 787–95. 39
40 1746–58. Smallwood PD, Steele MA, Faeth SH (2001). The ulti- 40
41 Lichti NI, Steele MA, Swihart RK (2017). Seed fate and mate basis of the caching preferences of rodents, and 41
42 decision-making processes in scatter-hoarding ro- the oak-dispersal syndrome: Tannins, insects, and 42
43 dents. Biological Reviews 92, 474–504. seed germination. American Zoologist 41, 840–51. 43
44 Lobo N, Duong M, Millar JS (2009). Conifer-seed pref- Steele MA, Knowles T, Bridle K, Simms EL (1993). 44
45 erences of small mammals. Canadian Journal of Zo- Tannins and partial consumption of acorns: implica- 45
46 ology 87, 773–80. tions for dispersal of oaks by seed predators. Ameri- 46
47 can Midland Naturalist 130, 229–38. 47
Louviere JJ, Flynn TN, Carson RT (2010). Discrete
48 48
choice experiments are not conjoint analysis. Journal Stephens DW, Brown JS, Ydenberg RC (2007). Forag-
49 49
of Choice Modelling 3, 57–72. ing: Behavior and Ecology. University of Chicago
50 50
Louviere JJ, Hensher DA, Swait JD (2000). Stated Press, Chicago, USA.
51 51

M. Sundaram et al.
1 Sundaram M, Willoughby JR, Lichti NI, Steele MA, Wang B, Ye C, Cannon CH, Chen J (2012). Dissecting 1
2 Swihart RK (2015). Segregating the effects of seed the decision making process of scatter-hoarding ro- 2
3 traits and common ancestry of hardwood trees on dents. Oikos 122, 1027–34. 3
4 eastern gray squirrel foraging decisions. Plos One 10, Wang BC, Smith TB (2002). Closing the seed dispersal
4
5 e0130942. 5
loop. Trends in Ecology & Evolution 17, 379–85.
6 Sundaram M, Lichti NI, Steele MA, Dalgleish HJ, Swi- 6
7 Xiao ZS, Zhang ZB (2006). Nut predation and disper- 7
hart RK (2017). Frequency-dependent hoarding by
8 Sciurus carolinensis occurs with seeds of similar persal of Harland Tanoak Lithocarpus harlandii by scat- 8
9 ceived value. Journal of Mammalogy 98, 124–34. ter-hoarding rodents. Acta Oecologica 29, 205–13. 9
10 Thompson DC, Thompson PS (1980). Food habits and Yang YQ, Yi XF, Niu KK (2012). The effects of kernel 10
11 caching behavior of urban grey squirrels. Canadian mass and nutrition reward on seed dispersal of 3 tree 11
12 Journal of Zoology 58, 701–10. species by small rodents. Acta Ethologica 15, 1–8. 12
13 13
Tonsor GT, Schroeder TC, Fox JA, Biere A (2005). Eu-
14 14
15
ropean preferences for beef steak attributes. Journal SUPPLEMENTARY MATERIALS 15
of Agricultural and Resource Economics 30, 367–80. Additional supporting information may be found in
16 16
17 Train KE (2003). Discrete Choice Methods with Simula- the online version of this article. 17
18 tion. Cambridge University Press, New York, NY. Supplementary Material 1 Tests confirming inter- 18
19 Vander Wall SB (1990). Food Hoarding in Animals. The specific variation in seed traits is higher than intraspecif- 19
20 University of Chicago Press, Chicago, IL. ic variation in seed traits. 20
21 Vander Wall SB (2010). How plants manipulate the Supplementary Material 2 Model selection and 21
22 scatter-hoarding behaviour of seed-dispersing ani- model fit summary for cached and consumed seed selec- 22
23 mals? Philosophical Transactions of the Royal Soci- 23
tion events.
24 ety B-Biological Sciences 365, 989–97. 24
25 Supplementary Material 3 Willingness-to-trade 25
Wang B, Chen J (2008). Tannin concentration enhances
26 seed caching by scatter-hoarding rodents: An exper- (WTT) 95% confidence intervals for all seed trait com- 26
27 iment using artificial “seeds”. Acta Oecologica 34, binations. 27
28 379–85. 28
29 29
30 30
32 32
Sundaram M, Lichti NI, Widmar NJO, Swihart RK (2018). Eastern gray squirrels are consistent shoppers of seed
33 33
traits: insights from discrete choice experiments. Integrative Zoology 13, 280–96.
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Factors shaping life history traits of two proovigenic parasitoids 6
7
8 8
9 9
10 10
11 Michal SEGOLI,1 Shucun SUN,2 Dori E. NAVA3 and Jay A. ROSENHEIM4 11
12 1
12
13 Mitrani Department of Desert Ecology, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, 13
14 Midreshet Ben-Gurion, Israel, 2Department of Ecology, School of Life Science, Nanjing University, Nanjing, China, 3Embrapa 14
15 Clima Temperado (Embrapa Temperate Agriculture), Pelotas, Rio Grande do Sul, Brazil and 4Department of Entomology and 15
16 Nematology, University of California Davis, Davis, California, USA 16
17 17
18 18
19 19
20 Abstract 20
21 What shapes the relative investment in reproduction versus survival of organisms is among the key questions in 21
22 life history. Proovigenic insects mature all their eggs prior to emergence and are short lived, providing a unique 22
23 opportunity to quantify their lifetime investments in the different functions. We investigated the initial eggloads 23
24 and longevity of 2 proovigenic parasitoid wasps: Anagrus erythroneurae and Anagrus daanei, (Hymenoptera: 24
25 Mymaridae) that develop within leafhopper eggs in both agricultural vineyards and natural riparian habitats in 25
26 Northern California. We collected Vitis spp. leaves containing developing parasitoids from 3 natural sites (Knight 26
27 Landing, American River and Putah Creek) and 3 agricultural vineyards (Solano Farm, Davis Campus and Vil- 27
28 lage Homes). We recorded eggloads at parasitoid emergence and female parasitoid longevity with or without 28
29 honey-feeding. Theory predicts that parasitoids from vineyards (where hosts are abundant) would have higher 29
30 initial eggloads and lower longevity compared with parasitoids from riparian habitats (where hosts are scarce). 30
31 Although host density and parasitoid eggloads were, indeed, higher in vineyards than in riparian habitats, para- 31
32 sitoid longevity did not follow the predicted pattern. Longevity without feeding differed among field sites, but 32
33 it was not affected by habitat type (natural vs agricultural), whereas longevity with feeding was not significantly 33
34 affected by any of the examined factors. Moreover, longevity was positively, rather than negatively, correlated 34
35 with eggloads at the individual level, even after correcting for parasitoid body size. The combined results sug- 35
36 gest a more complex allocation mechanism than initially predicted, and the possibility of variation in host quali- 36
37 ty that is independent of size. 37
38 38
39 Key words: Anagrus parasitoids, eggload, leafhoppers, life-history tradeoffs, longevity 39
40 40
41 41
42 42
43 43
INTRODUCTION
44 44
45 Life history theory seeks to explain optimal alloca- 45
46 tion of resources to different developmental, physiolog- 46
Correspondence: Michal Segoli, Mitrani Department of Desert
47 ical and reproductive functions of the organism. While 47
48 Ecology, The Jacob Blaustein Institutes for Desert Research, the approach of “one solution per species” is tempting, 48
49 Ben-Gurion University of the Negev, Midreshet Ben-Gurion, it is widely acknowledged that environmental condi- 49
50 8499000, Israel. tions are likely to alter optimal life-history traits, either 50
51 Email: msegoli@bgu.ac.il at the phenotypic (i.e. plastic responses) or genotypic 51

M. Segoli et al.
1 (i.e. evolutionary responses) level (Stearns 1992; Roff oids. Nevertheless, very few studies have investigated 1
2 2002). life history responses in proovigenic parasitoids in re- 2
3 Parasitoids are insects (mostly wasps) that lay their lation to environmental conditions. Part of the reason 3
4 eggs in or on the body of another organism (mostly an- could be difficulties in handling and dissecting proovig- 4
5 other insect) and whose offspring develop while feed- enic parasitoids due to their small size. 5
6 ing on the host, eventually killing it. Parasitoids have We studied egg load and longevity of 2 strictly 6
7 long been used for life history studies due to their var- proovigenic parasitoids of the genus Anagrus (A. eryth- 7
8 ied strategies, relative ease of rearing and manipulat- roneurae S. Trjapitzin & Chiappini, 1994 and A. daanei 8
9 ing in the laboratory, and high responsiveness to envi- Triapitsyn, 1998). These parasitoids attack the eggs of 9
10 ronmental conditions (Godfray 1994). The reproductive leafhoppers that are considered major pests in vineyards 10
11 success of parasitoids is considered to be limited mostly in Northern California. Both the parasitoids and the 11
12 by the availability of mature eggs, or by the availability leafhoppers occur also on wild grapes in riparian habi- 12
13 of suitable hosts in the environment (Rosenheim 1999). tats in California. This provides an excellent opportunity 13
14 Hence, what shapes the relative investment in egg pro- to study the life history traits of parasitoids originating 14
15 duction versus longevity is a key question in parasitoid from contrasting habitats. Previous work demonstrat- 15
16 life history. ed higher initial eggloads for parasitoids from vine- 16
17 yards (approximately 25 eggs) compared to their conge- 17
Despite early opinions that female parasitoids rare-
18 ners from riparian habitats (fewer than 20 eggs) (Segoli 18
ly deplete their eggs in nature (van Alphen et al. 1992;
19 & Rosenheim 2013). This is consistent with theory, as 19
Visser et al. 1992; Sevenster et al. 1998), recent empir-
20 leafhoppers are more abundant and may reach very high 20
ical data combined with theoretical developments sug-
21 densities in agricultural vineyards. In this study we ex- 21
gest that egg limitation in parasitoids, as well as in other
22 tended this work to compare parasitoid longevity among 22
insects, is likely to occur under field conditions (Heim-
23 these different habitats. Based on theory, we predicted 23
pel & Rosenheim 1998; Casas et al. 2000; Rosenheim
24 higher longevity for parasitoids from riparian habitats 24
et al. 2008; Segoli & Rosenheim 2013; Phillips & Kean
25 (where hosts are scarce) compared to those from vine- 25
2017). Moreover, even in cases where egg limitation is
26 yards (where hosts are abundant). Longevity was quan- 26
rare, it may impose strong selection on female repro-
27 tified either with or without feeding the parasitoids with 27
ductive traits, due the disproportionately high repro-
28 a sugar source (honey). Sugar supplementation was pre- 28
ductive contribution of egg-limited females (i.e. those
29 viously shown to extend parasitoid lifespan from less 29
that lay all of their eggs) to the next generation (Rosen-
30 than a day to few days (Segoli & Rosenheim 2013). We 30
heim 2011). Hence, the risk of egg limitation may lead
31 predicted that differences in longevity would be more 31
to higher investment in egg production in environments
32 pronounced under starvation conditions, where the par- 32
with many oviposition opportunities. In contrast, in en-
33 asitoid needs to rely on teneral reserves (Jervis et al. 33
vironments with limited host availability, higher invest-
34 2008), than when provided with honey. 34
ment in longevity, at the expense of egg production, is
35 35
expected, as parasitoids need an extended amount of
36 36
37
time to search for scarce hosts (Rosenheim 1996; Ellers MATERIALS AND METHODS 37
et al. 2000; Gandon et al. 2009).
38 38
The majority of parasitoids are synovigenic; that is, Study species
39 39
40 they continue to mature eggs throughout their adult life Anagrus spp. parasitoid wasps are among the most 40
41 (Godfray 1994; Jervis et al. 2001). Hence, decisions re- important natural enemies of Erythroneura and Eras- 41
42 garding the relative investment in survival versus egg moneura spp. leafhoppers in California (Doutt & Naka- 42
43 production may change during the lifetime of the para- ta 1973; Daane & Costello 2000; Bentley 2009). They 43
44 sitoid, making it difficult to quantify. In contrast, strictly complete their entire development (egg to adult) inside 44
45 proovigenic parasitoids emerge with their full egg com- the leafhopper egg, consuming the egg as they devel- 45
46 plement and do not mature additional eggs following op. These parasitoids are solitary (develop singly inside 46
47 emergence. In this case, the relative investment in egg the host), pro-ovigenic (emerge with their full lifetime 47
48 production is a “one-time decision” that is made during complement of eggs already matured) and do not re- 48
49 parasitoid development inside the host, prior to its emer- sorb eggs (Jepsen et al. 2007). Anagrus are short lived, 49
50 gence. This provides a unique opportunity to study in- even under the most benign laboratory conditions (En- 50
51 traspecific variation in lifetime investment of parasit- glish-Loeb et al. 2003). 51

Life history traits of proovigenic parasitoids
1 Erythroneura and Erasmoneura spp. leafhoppers de- °C, 12:12 h light : dark cycle) and checked daily un- 1
2 posit eggs singly or in clusters under the epidermal tis- til all died. Longevity was calculated as the number of 2
3 sue or along the veins of Vitis spp. leaves. Each egg is days from parasitoid emergence until its death. Parasit- 3
4 approximately 0.8 mm long. The freshly deposited egg oids that were found stuck in honey were excluded from 4
5 is colorless and transparent. When eggs are parasit- the sample as this might have expedited their death. Fol- 5
6 ized, they become brown or red. The leafhoppers com- lowing death, each parasitoid was sexed, females were 6
7 plete approximately 3 generations per growing season identified to species (Triapitsyn et al. 2010), and the 7
8 (June to September) and may reach very high densities length of a hind tibia was measured. 8
9 in vineyards towards the end of the season (Daane & 9
Eggload estimates
10 Costello 2000). 10
11 Females were then dissected to determine their egg- 11
12
Collection of parasitoids 12
loads. Dissection was done under a dissecting micro-
13 Grapevine leaves with apparent leafhopper damage scope using Minuten insect pins (Fine Science Tools, 13
14 were collected from 3 riparian habitats and 3 vineyards USA, Inc.). The number of eggs upon death was as- 14
15 located in Yolo, Solano and Sacramento counties in Cal- sumed to represent the number of eggs upon emergence. 15
16 ifornia, during August 2016. Some of the leaves were This is because these parasitoids do not resorb eggs, did 16
17 put immediately in emergence cages (approximately 10 not have oviposition opportunities in the vials and were 17
18 cages per agricultural site and 20 per natural site; ap- never observed to lay eggs outside a host. Males were 18
19 proximately 40 leaves per cage) and kept at room tem- not used in this study. 19
20 perature for one week. The cages were empty paper car- 20
21 Sample sizes 21
ton containers with a transparent funnel and a vial on
22 top. Parasitoids emerging from leafhopper eggs on the Sample sizes (Table 1) were highly unequal across 22
23 leaves were attracted to the light and collected from the sites because of strong inter-site variation in leafhopper 23
24 vials. The remaining leaves were kept in the refrigerator and parasitoid abundance, and despite the higher inten- 24
25 (4 °C) and were used to replenish the emergence cages sity of sampling (larger number of emergence cages) in 25
26 throughout the week. Emergence cages were monitored low-density sites. 26
27 hourly between 07:00 and 20:00 hours. 27
28 Host density estimates 28
29 Longevity without feeding 29
To test the assumption of higher host density in ag-
30 Newly emerged parasitoids were put in individual ricultural fields we examined 20 of the leaves collect- 30
31 glass vials, kept in an incubator (29 °C, 12:12 h light : ed from each site. The percentage of leaf area that was 31
32 dark cycle) and checked hourly during 12 h. Experimen- damaged by leafhopper feeding (visible as white stipling 32
33 tal temperature reflected temperatures in the field during on the leaves) was estimated. Host density estimates 33
34 this season. Following death, parasitoid sex was deter- differed significantly among field sites and were high- 34
35 mined and females were used for egg load and body size er in agricultural sites (Fig. 1, GLM, habitat type F1,119 = 35
36 estimates (see below). Longevity was calculated as the 64.91, P < 0.001; field site F4,119 = 16.34, P < 0.001), al- 36
37 number of hours that passed from parasitoid emergence though based on the Tukey post-hoc test, one of the ag- 37
38 until its death. Only parasitoids that emerged between ricultural sites (Solano Farm) did not differ significantly 38
39 0700 and 1000 hours were used for this estimate. This from the natural sites. 39
40 is because it was impossible to determine the timing of 40
41 emergence for parasitoids that were already emerged at Statistical analysis 41
42 0700 hours, and it was sometimes impossible to record Host density and parasitoid body size were compared 42
43 the time of death for parasitoids emerging after 1000 using general linear models with habitat type (agricul- 43
44 hours (they sometimes survived beyond the observation tural vs. natural), field site (nested within habitat type) 44
45 period). and parasitoid species (only for body size) as indepen- 45
46 dent factors. Eggloads were compared using general lin- 46
47 Longevity with sugar feeding 47
ear models with habitat type, field site (nested within
48 Newly emerged parasitoids were put into individual habitat type), feeding treatment (feeding vs. no feeding), 48
49 vials with a small piece of filter paper soaked with hon- parasitoid species and tibia length as independent fac- 49
50 ey solution. Parasitoids were kept in an incubator (29 tors. Because longevity with or without feeding was es- 50
51 51

M. Segoli et al.
1 Table 1 Sample sizes of females of the different field sites 1

2 2
Habitat type Field site Eggloads Longevity (hr) without feeding Longevity (d) with feeding
3 Anagrus daanei Anagrus daanei Anagrus daanei 3
4 Anagrus erythroneurae Anagrus erythroneurae Anagrus erythroneurae 4
5 5
Agricultural Village Homes 4 4 0
6 29 21 8 6
7 7
Davis Campus 4 4 0
8 8
43 28 15
9 9
10 Solano Farm 24 12 12 10
2 1 1
11 11
12 Natural Knights Landing 11 5 6 12
13 3 0 3 13
14 Putah Creek 21 9 12 14
15 3 2 1 15
16 American River 37 27 9 16
17 7 2 5 17
18 18
19 19
20 20
21 21
22 22
RESULTS
23 23
24 Body size 24
25 25
26 Hind tibia length of female parasitoids did not dif- 26
27 fer among habitat types, field sites or parasitoid spe- 27
28 cies (GLM, habitat type F1,187 = 0.02, P = 0.88; field site 28
29 F4,187 = 0.88, P = 0.47; Species F1,187 = 0.34, P = 0.56, N 29
30 = 188 females). 30
31 31
Eggloads
32 32
33 Mean eggloads of females varied among sites and 33
34 were higher in agricultural vineyards (Fig. 2, GLM, 34
35 field site F4,187 = 2.39, P = 0.05; habitat type F1,187 = 35
36 20.46, P < 0.001). In addition, eggload variation among 36
37 Figure 1 Means ± SE leafhopper density estimated as % leaf- females was positively correlated with tibia length (F1,187 37
38 hopper damage on the leaf at the different field sites (n = 20 = 41.36, P < 0.001), although it explained a small por- 38
39 leaves per site). tion of the variation (R2 = 0.29 for A. erythroneurae and 39
40 R2 = 0.21 for A. daanei). Eggloads were not affected 40
41 by either the parasitoid species (F1,187 = 0.20, P = 0.66) 41
42 or by the feeding treatment (F1,187 = 3.28, P = 0.07, N = 42
43 timated using different units (hour vs days), it was an- 188). Residual eggloads based on this model were used 43
44 alyzed separately using the same model as for eggload for the regression between longevity and eggload (see 44
45 (excluding the feeding-treatment factor). To test the re- below). 45
46 lationship between longevity and eggloads at the indi- 46
vidual level, we regressed longevity (with or without Effect of feeding
47 47
48 feeding) against the residuals of eggloads, based on the As expected, feeding increased female lifespan from 48
49 model above. Data that were not distributed normally less than a day (mean = 5.44 ± 2.36 hours, n = 115) to 49
50 were log or square root transformed to meet model as- several days (mean = 4.08 ± 0.33 days, n = 72). 50
51 sumptions. Means are reported ± SE throughout. 51

1 Longevity without feeding by tibia length or by parasitoid species (Fig. 3, GLM, 1

2 habitat type F1,114 = 1.07, P = 0.30; Field site- F4,114 = 2
3 Longevity without feeding differed significantly 3.41, P = 0.01; tibia length- F1,114 = 0.01, P = 0.93; Spe- 3
4 among field sites but was not affected by habitat type, cies- F1,114 = 0.44, P = 0.51). The main variation was 4
5 5
6 6
7 7
8 8
9 9
10 10
11 11
12 12
13 13
14 14
15 15
16 16
17 17
18 18
19 Figure 2 Means ± SE eggloads of female 19
20 parasitoids of the different field sites. 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
Figure 3 Means ± SE longevity (h) of un-
34 34
fed female parasitoids of the different field
35 35
sites.
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 Figure 4 Means ± SE longevity (days) of 49
50 fed female parasitoids of the different field 50
51 sites. 51

M. Segoli et al.
1 among the different natural sites, while the agricultural ronment. This is consistent with previous results (Segoli 1
2 sites had similar values. & Rosenheim 2013), further confirming the robustness 2
3 of this observation. To our knowledge, this is the only 3
Longevity with feeding
4 example of inter-population variation in initial eggloads 4
5 Longevity with feeding was not significantly affected in a strictly proovigenic parasitoid. Given that eggloads 5
6 by any of the independent factors (Fig. 4, GLM, habitat are determined prior to parasitoid emergence, variation 6
7 type F1,71 = 2.15, P = 0.09; field site F4,71 = 2.75, P = 0.10; in eggloads cannot be explained as an immediate plas- 7
8 tibia length F1,71 = 0.75, P = 0.39; species F1,71 = 0.27, P tic response to host availability experienced by adult fe- 8
9 = 0.6). This may partly be due to the small sample size males in the environment. Alternative explanations in- 9
10 for some of the species/site combinations. clude variability in host quality among the different 10
11 habitats, transgenerational phenotypic plasticity (ma- 11
12 Relationship between eggload and longevity ternal effects) or genetic differences among parasitoids 12
13 We found a weak positive, rather than a negative, re- of the different populations. We will next discuss these 13
14 lationship between longevity without feeding and resid- possibilities. 14
15 ual eggload (controlling for tibia length, habitat type, Host size is one of the most basic aspects of host 15
16 field site and species [see above]; linear regression, P = quality and in other host-parasitoid systems has of- 16
17 0.01, R2 = 0.06; Fig. 5). Longevity with feeding was not ten been shown to affect both the number and quality 17
18 related to residual eggload (linear regression, P = 0.37, of emerging parasitoids (Godfray 1994). For example, 18
19 R2 = 0.01). in the parasitoid Uscana lariophaga, which parasitiz- 19
20 es eggs of bruchid beetles, females reared in small host 20
21 eggs developed slower, were smaller and produced few- 21
22
DISCUSSION 22
er eggs compared to those reared in large host eggs
23 (Spitzen & van Huis 2005). Previous work in the Ana- 23
Variation in eggloads
24 grus-leafhopper-grapes system found no consistent dif- 24
25 We studied life history traits of 2 species of proovi- ferences in leafhopper egg size on leaves collected from 25
26 genic parasitoid wasps originating from either natural vineyards or from wild grapes, suggesting the lack of 26
27 riparian habitats or agricultural vineyards. Consistent differential host quality between these habitats (Sego- 27
28 with our predictions, parasitoids from agricultural habi- li & Rosenheim 2013). This conclusion is strengthened 28
29 tats emerged with more eggs, and, hence, were prepared by the lack of difference in mean parasitoid body size 29
30 for the plentiful oviposition opportunities in their envi- between these habitats, as demonstrated in the current 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
Figure 5 Relationship between longevity
49 49
of unfed female parasitoids and their egg-
50 load. 50
51 51

1 study. Nevertheless, other aspects of host quality may ferences in the timing of allocation than in total invest- 1
2 also be of importance. For example, fertilization and ir- ment (Ellers & van Alphen 1997). 2
3 rigation in vineyards may affect the nutritional compo- Here we examined the potential tradeoff between 3
4 sition of leafhopper eggs, independent of their size, sub- eggload and longevity for proovigenic species, at both 4
5 sequently affecting parasitoid traits. This hypothesis is the population and individual level. Because longevity 5
6 yet to be examined. of parasitoids that do not feed is dependent exclusively 6
7 Transgenerational plasticity of reproductive traits in on their teneral reserves (Jervis et al. 2008), we expect- 7
8 relation to environmental conditions experienced by ovi- ed that variation in longevity will be more pronounced 8
9 positing females has been demonstrated in many insects in food deprived females. Indeed, the longevity of fe- 9
10 (Mousseau & Dingle 1991; Mousseau & Fox 1998), in- males fed with honey was substantially greater (up to 10
11 cluding parasitoids (Morag et al. 2011). The possibili- 10-fold) and was not dependent on the specific site or 11
12 ty of transgenerational response to host availability in habitat of origin. The longevity of starved females, how- 12
13 Anagrus parasitoids was previously tested by expos- ever, differed significantly among field sites, suggesting 13
14 ing A. erythroneurae females to either high or low host inter-population variation in parasitoid energy reserves. 14
15 densities and examining the eggloads of their daughters Nevertheless, in contrast to our predictions, longevi- 15
16 (Andreazza & Rosenheim 2015). However, no differ- ty was not consistently higher for parasitoids originat- 16
17 ences across host density treatments were found, argu- ing from natural habitats. Moreover, longevity was pos- 17
18 ing against the operation of transgenerational phenotyp- itively, rather than negatively, correlated with eggload 18
19 ic plasticity. at the individual level, even after correcting for parasit- 19
20 oid body size and site of origin. This contradicted previ- 20
Initial fecundity has been found to be heritable in
21 ous results that instead suggested a negative relationship 21
many insects, including Anagrus parasitoids (Cronin &
22 between female eggload and longevity at the individual 22
Strong 1996; Andreazza & Rosenheim 2015). Although
23 level (Segoli & Rosenheim 2013). 23
the observed inter-population variation in Anagrus is
24 24
likely to be at least partially genetic, the relative contri- There are several general explanations for the ab-
25 25
bution of genetic variation is currently unknown. Future sence of an expected tradeoff in life history traits (Stea-
26 26
work will aim to conduct common garden experiments rns 1992; Agrawal et al. 2010). One possible expla-
27 27
(in which parasitoids of different origins are reared on nation is that a tradeoff may not be expressed under
28 28
the same host cohort), transplant experiments (in which certain environmental conditions. For example, parasit-
29 29
parasitoids originating of one habitat are reared on hosts oids from natural habitats could perhaps have the ability
30 30
of another habitat and vice versa) and evolutionary ex- to survive longer in the field, but this ability might not
31 31
periments (in which parasitoids are reared at high or low be expressed under the artificial conditions experienced
32 32
host densities over multiple generations) to address this in the lab. In particular, confining parasitoids in a vial
33 33
question. could substantially alter their movement behavior (van
34 34
den Assem 1996), thereby affecting their energy expen-
35 Tradeoff between eggload and longevity 35
diture. However, documenting the realized longevity of
36 36
Higher investment in egg production is likely to such small insects in the field, or quantifying their en-
37 37
come at the expense of other functions. Indeed, stud- ergy reserves at emergence (Ellers 1996), may be ex-
38 38
ies have demonstrated both phenotypic and genetic tremely challenging.
39 39
tradeoffs between reproduction and longevity in many Another potential reason is that allocation patterns
40 40
insects (Miyatake 1997; Tatar 2001; Roff 2002; Ed- and selection pressures may be more complex or simply
41 41
ward & Chapman 2011). In parasitoids, such tradeoffs different than initially assumed. For example, higher in-
42 42
were demonstrated at the species level (i.e. long-lived vestment in egg production could come at the expense
43 43
species emerge with relatively fewer eggs; Jervis et al. of energy reserves, but could also potentially trade off
44 44
2001, 2003) and the individual level (i.e. individuals in- with egg quality, female mobility, immune function or
45 45
vesting more in egg production die sooner; Ellers et al. stress tolerance. Hence, certain functions may not al-
46 46
2000; Zhang et al. 2011; Segoli & Rosenheim 2013). ways directly compete for resources (Agrawal et al.
47 47
Tradeoffs at the population level were also suggested, 2010). In particular, the tradeoff between egg number
48 48
but these were either confounded with parasitoid repro- and size was suggested to be of high importance (Ber-
49 49
ductive mode (sexual vs asexual; Pelosse et al. 2007; rigan 1991; Rosenheim 1996). Indeed, previous results
50 50
Ameri et al. 2015), or were more likely to represent dif- suggest that Anagrus eggs in agricultural sites in Cali-
51 51

M. Segoli et al.
1 fornia are smaller (Segoli & Rosenheim 2013). Hence, and wild grapes. This is publication number 952 of the 1
2 the tradeoff between egg number and size may be more Mitrani Department of Desert Ecology. 2
3 pronounced than that of eggload and longevity in this 3
4 system. Moreover, additional environmental factors, REFERENCES 4
5 such as food availability, habitat complexity, predation 5
6 and competition, may impose stronger or even oppos- Agrawal AA, Conner JK, Rasmann S (2010). Tradeoffs 6
7 ing selection on longevity, leading to inter-site variation and negative correlations in evolutionary ecology. In: 7
8 that is independent of host density. For example, high- Bell MA, Eanes WF, Futuyma DJ, Levinton JS, eds. 8
9 er competition among females at higher densities may Evolution After Darwin: the First 150 Years. Sinauer 9
10 compromise their oviposition rates, as was shown for Associates, Sunderland, MA, pp. 243–68. 10
11 another Anagrus species (Cronin & Strong 1993). Ameri M, Rasekh A, Mohammadi Z (2015). A compari- 11
12 Finally, the expectation for a tradeoff is based on the son of life history traits of sexual and asexual strains 12
13 assumption of a resource pool whose size is constant of the parasitoid wasp, Lysiphlebus fabarum (Braco- 13
14 across individuals. If instead there is individual variation nidae: Aphidiinae). Ecological Entomology 40, 50– 14
15 in total resource availability, individuals with larger to- 61. 15
16 tal resource pools may have both greater fecundity and Andreazza F, Rosenheim JA (2015). Absence of trans- 16
17 longevity, thus masking the occurrence of a tradeoff. generational phenotypic plasticity in fecundity in the 17
18 This does not mean that these functions do not trade off; parasitoid Anagrus erythroneurae (Hymenoptera: 18
19 rather, that in order to detect the tradeoff there is a need Mymaridae). Journal of Insect Science 15, 138; DOI: 19
20 to control for individual quality, either statistically or 10.1093/jisesa/iev122. 20
21 experimentally (Stearns 1992). Although we statistical- 21
Bentley WJ (2009). The integrated control concept and
22 ly controlled for female body size, as mentioned above, 22
its relevance to current integrated pest management
23 this may not be a reliable measure of the quality of re- 23
in California fresh market grapes. Pest Management
24 source available for parasitoid development. In particu- 24
Science 65, 1298–304.
25 lar in egg parasitoids, body size may be constrained by 25
26 Berrigan D (1991). The allometry of egg size and num- 26
the size of the host egg, which could be relatively con-
27 ber in insects. Oikos 60, 313–21. 27
stant. Hence, parasitoid traits may be dependent on the
28 composition of nutrients inside the host, which, in turn, Casas J, Nisbet RM, Swarbrick S, Murdoch WW (2000). 28
29 may depend on the nutritional value of the plant (Turl- Eggload dynamics and oviposition rate in a wild pop- 29
30 ings & Benrey 1998; Sarfraz et al. 2009). Such effects ulation of a parasitic wasp. Journal of Animal Ecolo- 30
31 may lead to positive correlations between different life gy 69, 185–93. 31
32 history components as observed in this study. This may Cronin JT, Strong DR (1993). Superparasitism and mu- 32
33 also explain the previously documented tradeoff (Segoli tual interference in the egg parasitoid Anagrus deli- 33
34 & Rosenheim 2013) that was observed in a small sam- catus (Hymenoptera, Mymaridae). Ecological Ento- 34
35 ple of parasitoids of a uniform origin (parasitoids of the mology 18, 293–302. 35
36 same site, emerging on the same day). Cronin JT, Strong DR (1996). Genetics of oviposition 36
37 In conclusion, parasitoid life history traits varied success of a thelytokous fairyfly parasitoid, Anagrus 37
38 among field sites, and although we confirmed the greater delicatus. Heredity 76, 43–54. 38
39 fecundity of parasitoids originating in agricultural sites, 39
Daane KM, Costello MJ (2000). Variegated and west-
40 we found no evidence of a tradeoff with parasitoid lon- 40
ern grape leafhoppers. In: Christensen PL, ed. Rai-
41 gevity at either the population or individual level. The 41
sin Production Manual. University of California, Ag-
42 lack of apparent tradeoff between eggload and longevi- 42
ricultural and Natural Resources, Oakland, CA, pp.
43 ty may be due to inappropriate experimental conditions, 43
173–81.
44 complex allocation patterns or variation in host quality 44
45
Doutt RL, Nakata J (1973). The Rubus leafhopper and 45
that is independent of host size.
46
its egg parasitoid: An endemic biotic system useful in 46
47
grapepest management. Environmental Entomology 2, 47
ACKNOWLEDGMENTS 381–6.
48 48
49 We thank Beth Flory, Bryan Pellissier, Village Homes Edward DA, Chapman T (2011). Mechanisms underly- 49
50 and the County of Sacramento Department of Region- ing reproductive trade-offs: costs of reproduction. In: 50
51 al Parks for permission to sample and access vineyards Flatt T, Heyland A, eds. Mechanisms of life history 51

1 evolution: the genetics and physiology of life histo- Miyatake T (1997). Genetic trade-off between early fe- 1
2 ry traits and trade-offs. Oxford University Press, Ox- cundity and longevity in Bactrocera cucurbitae (Dip- 2
3 ford, UK, pp. 137–52. tera: Tephritidae). Heredity 78, 93–100. 3
4 Ellers J (1996). Fat and eggs: An alternative method to Morag N, Keasar T, Harari A, Bouskila A (2011). 4
5 measure the trade-off between survival and reproduc- Trans-generational effects of maternal rearing density 5
6 tion in insect parasitoids. Netherlands Journal of Zo- on offspring development time in a parasitoid wasp. 6
7 ology 46, 227–35. Physiological Entomology 36, 294–8. 7
8 8
Ellers J, Driessen G, Sevenster JG (2000). The shape of Mousseau TA, Dingle H (1991). Maternal Effects in In-
9 9
the trade-off function between egg production and sects: Examples, constraints, and geographic varia-
10 life span in the parasitoid Asobara tabida. Nether- 10
tion. In: Dudley EC, ed. The Unity of Evolutionary
11 lands Journal of Zoology 50, 29–36. 11
Biology. Dioscorides Press, pp. 745–61.
12 12
Ellers J, Sevenster JG, Driessen G (2000). Egg load Mousseau TA, Fox CW (1998). Maternal effects as ad-
13 13
evolution in parasitoids. American Naturalist 156, aptations. Oxford University Press, New York.
14 14
650–65. Pelosse P, Bernstein C, Desouhant E (2007). Differential
15 15
16 Ellers J, van Alphen JJM (1997). Life history evolution energy allocation as an adaptation to different habi- 16
17 in Asobara tabida: Plasticity in allocation of fat re- tats in the parasitic wasp Venturia canescens. Evolu- 17
18 serves to survival and reproduction. Journal of Evo- tionary Ecology 21, 669–85. 18
19 lutionary Biology 10, 771–85. Phillips CB, Kean JM (2017). Response of parasitoid 19
20 English-Loeb G, Rhainds M, Martinson T, Ugine T egg load to host dynamics and implications for egg 20
21 (2003). Influence of flowering cover crops on Ana- load evolution. Journal of Evolutionary Biology 30, 21
22 grus parasitoids (Hymenoptera : Mymaridae) and 1313–24. 22
23 Erythroneura leafhoppers (Homoptera : Cicadelli- Roff DA (2002). Life history evolution. Sinauer Associ- 23
24 dae) in New York vineyards. Agric. Forest Entomol. 5, ates, Sunderland, MA. 24
25 173–81. 25
Rosenheim JA (1996). An evolutionary argument for
26 Gandon S, Varaldi J, Fleury F, Rivero A (2009). Evolu- 26
egg limitation. Evolution 50, 2089–94.
27 tion and manipulation of parasitoid egg load. Evolu- 27
28 tion 63, 2974–84. Rosenheim JA (1999). The relative contributions of time 28
29
and eggs to the cost of reproduction. Evolution 53, 29
Godfray HCJ (1994). Parasitoids: Behavioral and evo-
30
376–85. 30
lutionary ecology. Princeton University Press, Princ-
31 eton, NJ. Rosenheim JA (2011). Stochasticity in reproductive op- 31
32 portunity and the evolution of egg limitation in in- 32
Heimpel GE, Rosenheim JA (1998). Egg limitation
33 sects. Evolution 65, 2300–312. 33
in parasitoids: A review of the evidence and a case
34 study. Biological Control 11, 160–8. Rosenheim JA, Jepsen SJ, Matthews CE, Smith DS, Ro- 34
35 senheim MR (2008). Time limitation, egg limitation, 35
Jepsen SJ, Rosenheim JA, Matthews CE (2007). The
36 the cost of oviposition, and lifetime reproduction by 36
impact of sulfur on the reproductive success of Ana-
37 an insect in nature. American Naturalist 172, 486–96. 37
grus spp. parasitoids in the field. Biocontrol 52, 599–
38 Sarfraz M, Dosdall LM, Keddie BA (2009). Host plant 38
612.
39 nutritional quality affects the performance of the par- 39
40
Jervis MA, Heimpel GE, Ferns PN, Harvey JA, Kidd 40
asitoid Diadegma insulare. Biological Control 51,
41
NAC (2001). Life-history strategies in parasitoid 41
34–41.
wasps: a comparative analysis of “ovigeny”. Journal
42 Segoli M, Rosenheim JA (2013). Limits to the repro- 42
of Animal Ecology 70, 442–58.
43 ductive success of two insect parasitoid species in the 43
44 Jervis MA, Ferns PN, Heimpel GE (2003). Body size 44
and the timing of egg production in parasitoid wasps: field. Ecology 94, 2498–504.
45 45
a comparative analysis. Functional Ecology 17, 375– Segoli M, Rosenheim JA (2013). The link between host
46 46
83. density and egg production in a parasitoid insect:
47 47
Jervis MA, Ellers J, Harvey JA (2008). Resource acqui- comparison between agricultural and natural habitats.
48 48
sition, allocation, and utilization in parasitoid repro- Functional Ecology 27, 1224–32.
49 49
50 ductive strategies. Annual Review of Entomology 53, Segoli M, Rosenheim JA (2013). Spatial and temporal 50
51 361–85. variation in sugar availability for insect parasitoids in 51

M. Segoli et al.
1 agricultural fields and consequences for reproductive Turlings TCJ, Benrey B (1998). Effects of plant metab- 1
2 success. Biological Control 67, 163–9. olites on the behavior and development of parasitic 2
3 Sevenster JG, Ellers J, Driessen G (1998). An evolutionary wasps. Ecoscience 5, 321–33. 3
4 argument for time limitation. Evolution 52, 1241–4. van Alphen JJM, Visser ME, Nell HW (1992). Adaptive 4
5 5
Spitzen J, van Huis A (2005). Effect of host quality of superparasitism and patch time allocation in solitary
6 6
Callosobruchus maculatus (Coleoptera: Bruchidae) parasitoids – Searching in groups vs sequential patch
7 7
on performance of the egg parasitoid Uscana lario- visits. Functional Ecology 6, 528–35.
8 8
phaga (Hymenoptera: Trichogrammatidae). Bulletin van den Assem J (1996). Mating behavior. In: Jervis
9 9
of Entomological Research 95, 341–7. M, Kidd N, eds. Insect Natural Enemies: Practical
10 10
11 Stearns SC (1992). The evolution of life histories. Ox- Approaches to their Study and Evaluation [1996]. 11
12 ford University Press, Oxford; New York. Chapman & Hall, London. pp. 163–221. 12
13 Tatar M (2001). Senescence. In: Fox CW, Roff DA, Visser ME, van Alphen JJM, Nell HW (1992). Adaptive 13
14 Fairbain DJ, eds. Evolutionary Ecology: Concepts superparasitism and patch time allocation in solitary 14
15 and Case Studies. Oxford University Press, New parasitoids – The influence of prepatch experience. 15
16 York, pp. 128–41. Behavioral Ecology and Sociobiology 31, 163–71. 16
17 Triapitsyn SV, Rugman-Jones PF, Jeong G, Morse JG, Zhang YB, Liu WX, Wang W, Wan FH, Li Q (2011). 17
18 Lifetime gains and patterns of accumulation and mo- 18
Stouthamer R (2010). Morphological and molecular
19 19
differentiation of the Anagrus epos species complex bilization of nutrients in females of the synovigen-
20 20
(Hymenoptera: Mymaridae), egg parasitoids of leaf- ic parasitoid, Diglyphus isaea Walker (Hymenoptera:
21 21
hoppers (Hemiptera: Cicadellidae) in North America. Eulophidae), as a function of diet. Journal of Insect
22 22
Zootaxa 2428, 1–21. Physiology 57, 1045–52.
23 23
24 24
25 25
26 26
27 Segoli M, Sun S, Nava DE, Rosenheim JA (2018). Factors shaping life history traits of two proovigenic parasit- 27
28 oids. Integrative Zoology 13, 297–306. 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Distribution and space use of seed-dispersing rodents in central 6
7
8
9 Pyrenees: implications for genetic diversity, conservation and 8
9
plant recruitment
10 10
11 11
12 12
13 13
14 Jon URGOITI,1 Alberto MUÑOZ,2 Josep Maria ESPELTA1 and Raúl BONAL3,4,5 14
15 15
1
16 CREAF, Cerdanyola del Vallés, Catalonia, Spain, 2Department of Didactics of Experimental Sciences, Faculty of Education, 16
17 University Complutense of Madrid, Madrid, Spain, 3Forest Research Group, INDEHESA, University of Extremadura, Plasencia, 17
18 Spain, 4DITEG Research Group, University of Castilla-La Mancha, Toledo, Spain and 5Research group of Genetic and Cultural 18
19 Biodiversity, Research centre for Cinegetic Resources (CSIC-UCLM-JCCM), Ciudad Real, Spain 19
20 20
21 21
22 22
23 Abstract 23
24 The function and conservation of many forest ecosystems depend on the distribution and diversity of the com- 24
25 munity of rodents that consume and disperse seeds. The habitat preferences and interactions are especially rele- 25
26 vant in alpine systems where such granivorous rodents reach the southernmost limit of their distribution and are 26
27 especially sensitive to global warming. We analyzed the community of granivorous rodents in the Pyrenees, one 27
28 of the southernmost mountain ranges of Europe. Rodent species were identified by DNA with particular atten- 28
29 tion to the Apodemus species, which are prominent seed-dispersing rodents in Europe. We confirmed for the first 29
30 time the presence of the yellow-necked mouse, Apodemus flavicollis, in central Pyrenees, a typical Eurosiberian 30
31 species that reaches its southernmost distribution limit in this area. We also found the wood mouse, Apodemus 31
32 sylvaticus, a related species more tolerant to Mediterranean environments. Both rodents were spatially segre- 32
33 gated by altitude. A. sylvaticus was rare at high altitudes, which might cause the genetic differentiation between 33
34 populations of the different valleys reported here. We also found other seed consumers like dormice, Elyomis 34
35 quercinus, and voles, Myodes glareolus, with marked habitat preferences. We suggest that population isolation 35
36 among valleys may increase the genetic diversity of rodents, like A. sylvaticus. We also highlight the potential 36
37 threat that global warming may represent for species linked to high-altitude refuges at the southern edge of its 37
38 distribution, like Apodemus flavicollis. Finally, we discuss how this threat may have a dimension in the conser- 38
39 vation of alpine forests dispersed by these rodent populations. 39
40 40
41 Key words: DNA analysis, Pyrenees, rodents, seed dispersal, vegetation structure 41
42 42
43 43
44 44
45
INTRODUCTION 45
46 46
47 Granivorous rodents have a strong influence on plant 47
48 Correspondence: Alberto Muñoz, Departamento de Didáctica regeneration patterns as prominent seed consumers and 48
49 de las Ciencias Experimentales, Facultad de Educación, dispersers, thus playing an essential role in ecosystem 49
50 Universidad Complutense de Madrid, Madrid, 28040 Spain. function (Vander Wall 1990; Sunyer et al. 2015). Many 50
51 Email: albemuno@ucm.es authors have suggested their potential as ecological bio- 51

J. Urgoiti et al.
1 indicators of sustainable forest management and conser- ter-hoarding rodents belonging to the Apodemus genus: 1
2 vation (Pearce & Venier 2005; Avenant 2011). Hence, the yellow-necked mouse, Apodemus flavicollis(Melchi- 2
3 analysis of their diversity and distribution may help to or, 1834), a typical Eurosiberian species that has been 3
4 better understand and preserve the structure and func- found in the western Pyrenees (Castién & Gosàlbez 4
5 tioning of many forest ecosystems. 1994), although its presence in central and oriental Pyr- 5
6 The distribution, patterns of habitat selection and enees is yet to be confirmed. This species is considered 6
7 population dynamics of granivorous rodents are strong- to overlap, in terms of ecological niche, with the wood 7
8 ly conditioned by climate, predation risk and food avail- mouse, Apodemus sylvaticus (Linnaeus, 1758), a more 8
9 ability (Mazurkiewicz 1994; McCain 2005) and can be Mediterranean species with which it shares morpholog- 9
10 approached from different spatial scales, ranging from ical similarities, rendering their identification by means 10
11 macrohabitat and a landscape context, to microhabi- of molecular techniques (DNA sequencing) the most re- 11
12 tat (Bailey et al. 1996; Boyce et al. 2003). A given ro- liable method for species determination (Arrizabalaga 12
13 dent species will basically fit a given area depending on et al. 1999; Michaux et al. 2001; Ancillotto et al. 2017). 13
14 whether: (i) the environmental requirements are met; (ii) DNA sequencing provides further advantages that go be- 14
15 there is shelter against predators; and (iii) food availabil- yond species determination. At the intra-specific level, 15
16 ity is sufficient. As most habitats often hold more than a it allows identifying landscape and environmental fea- 16
17 single rodent species, their coexistence and community tures that affect inter-population gene flow, thus shaping 17
18 assemblage may also depend on the partitioning of food genetic structure (Giordano et al. 2007). In the case of 18
19 resources (e.g. seeds), variation in daily seasonal activ- rodents, some mitochondrial markers, such as the con- 19
20 ity and microhabitat segregation (Kotler 1984; Mitch- trol-region (d-loop), show a high mutation rate that per- 20
21 ell et al. 1990; Kotler et al. 1991; Ostfeld 1992; Canova mits detection of population structure at a scale of just 21
22 1993; Hille & Mortelliti 2010). According to the concept a few kilometers (Hirota et al. 2004). With these mark- 22
23 of phylogenetic niche conservatism (Wiens & Graham ers it is possible to estimate the degree of connection be- 23
24 2005; Losos 2008; Wiens et al. 2010), close relative tween populations located in different habitats, some- 24
25 species are likely to possess similar niches, which theo- thing that would not otherwise be possible. 25
26 retically would result in stronger competition, whereas The present study aims to: (i) assess the communi- 26
27 species more separated phylogenetically are more likely ty structure and the patterns of space use by granivorous 27
28 to coexist by exploiting different resources. rodents in different forest systems of the central Pyr- 28
29 enees, from the macrohabitat context to a microhabi- 29
Alpine and sub-alpine ecosystems are attractive sce-
30 tat perspective; and (ii) test the genetic population struc- 30
narios for analyzing small rodent distributions, as the
31 ture and thereby estimate the degree of connectivity 31
steep gradients of physical characteristics and climat-
32 and movement between populations at a landscape scale 32
ic variables result in contrasting habitats that may lead
33 for scatter-hoarding rodents of Apodemus spp. Final- 33
to changes in species richness of small mammals over
34 ly, we use these results to evaluate how the distribution 34
short distances (Heaney 2001; McCain 2004, 2005).
35 of seed-eating rodents in these high mountain ranges 35
This heterogeneity of alpine and sub-alpine habitats may
36 may influence forest regeneration in a scenario of global 36
also promote species diversity (Tews et al. 2004; Hortal
37 warming. 37
et al. 2009) and shape population genetic structure at the
38 38
intra-specific level (Trizio et al. 2005). Moreover, high
39 39
40
mountains are among the most sensitive regions to cli- MATERIALS AND METHODS 40
mate change, where predictions suggest a higher warm-
41 41
ing than the global average (Kohler & Maselli 2009). Study area
42 42
43 The Pyrenees is among the most emblematic alpine The study was conducted in the National Park of 43
44 systems of southern Europe that has acted as refuge for Aigüestortes i Estany de Sant Maurici (PNAESM), in 44
45 species of northern latitudes and as a place of specia- north-east Spain (Catalan Pyrenees; 42°34′N, 1°0′E), 45
46 tion after the end of the last cold period of Pleistocene which holds the most representative sub-alpine and al- 46
47 (Blondel & Aronson 1999). Thus, the Pyrenees repre- pine habitats of the central Pyrenees. The climate is 47
48 sent the southern distributional limit for many mam- montane and characterized by a variable number of mi- 48
49 mals with biogeographical origins in northern latitudes, croclimates, influenced by the altitude (between 1200 49
50 and the point of intersection with species from other and 3000 m) and topography of the valleys and slopes. 50
51 geographic areas. This is especially the case for 2 scat- Annual rainfall (also in the form of snow) is approxi- 51

Seed-dispersing rodents in central Pyrenees
1 mately 1500 mm (Torbidoni et al. 2005) and the mean m from other trapping stations of the same habitat, so 1
2 annual temperature is 4.6 °C (mean maximum of 9.6 °C that trap distribution allowed us to make an assessment 2
3 and mean minimum of −0.5 ºC; Arnán et al. 2014). Most at 3 spatial scales: (i) at the scale of habitat (macrohab- 3
4 sub-alpine systems in the Park are dominated by conif- itat) to analyze the distribution of small rodents along 4
5 erous forests of black pine (Pinus uncinata) and silver the different habitat types and the changes in commu- 5
6 fir (Abies alba), but Scots pine (P. sylvestris), birch (Bet- nity structure at a wider spatial scale; (ii) at the scale of 6
7 ula pendula) and beech (Fagus sylvatica) are also pres- the trapping station (mesohabitat) to analyze the extent 7
8 ent at lower altitudes. Besides closed forests, the Park of spatial segregation between different species within 8
9 holds open areas as meadows with different tree species a shared habitat type; and (iii) at the scale of individual 9
10 and screes due to the typical process of frost weathering trap (microhabitat) to find out preferences among differ- 10
11 that occurs in alpine and sub-alpine systems. ent vegetation structures. 11
12 Traps were baited with a mixture of flavor and tuna 12
13
Sampling of granivorous rodents 13
in oil, and an apple piece to avoid dehydration. All the
14 The main trapping campaigns were performed along traps also contained a handful of hydrophobic cotton 14
15 3 sampling points of the “Valley of Espot” in the sum- so that the captured rodents could make a nest to keep 15
16 mers of 2013 and 2014. In summer 2013, we additional- warm until they were marked and released (Muñoz et al. 16
17 ly sampled 3 sites in the “Valley of Boí,” located 5 km 2009). At each sampling point, each trapping campaign 17
18 from the “Valley of Espot,” to analyze the potential ge- had a duration of 4 nights, and in each year the traps 18
19 netic differences between the rodents captured in each were placed in the exact same places, as they were geo- 19
20 valley (Fig. 1). We sampled 4 common habitats of PNA- referenced by GPS. Each captured rodent was marked 20
21 ESM located at 4 different altitudes: (i) pinewoods of with a subcutaneous microchip inserted with a special- 21
22 Pinus sylvestris at 1600–1700 m (Boí and Espot); (ii) ized syringe. This marking method allowed us to iden- 22
23 beech (Fagus sylvatica) and oak (Quercus robur) forests tify every recaptured individual in order to calculate the 23
24 at 1700–1800 m (Boí); (iii) meadows with hazels (Co- total abundance of rodents in each habitat type (num- 24
25 rylus avellana) and birches (Betula pendula) at 1700– ber of different individuals captured) and also to esti- 25
26 1800 m (Espot); and (iv) silver-fir forests (Abies alba) at mate the survival rate from the first summer to the sec- 26
27 1800–1900 m (Boí and Espot). ond summer (Sunyer et al. 2015, 2016). 27
28 In each habitat type, we established 10 trapping sta- 28
Once marked, all individuals were released at the ex-
29 tions, each one consisting of 10 Sherman live traps (23.5 29
act point where they had been captured. Microhabi-
30 × 8 × 9 cm; H.B. Sherman Traps, Tallahassee, Flori- 30
tat structure was measured in a 1-m radius around each
31 da, USA) distributed along 2 lines of 5 traps with a dis- 31
Sherman trap. Following previous studies (Muñoz et al.
32 tance of 10 m among them (i.e. 100 traps per habitat 32
2009), we carried out a visual estimation of cover val-
33 type). Each trapping station was separated at least 50 33
ues (% projection on the ground) of: (i) tree cover; (ii)
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 Figure 1 Study area in central Pyrenees 49
50 with the location of the 2 valleys and the 3 50
51 trapping points in each valley. 51

J. Urgoiti et al.
1 high (>30) and low (<30 cm) shrub cover; (iii) high (>30 gap penalty and extension parameters as before, and 1
2 cm) and low (<30 cm) herbaceous cover; (iv) leaf litter; used to build a phylogenetic tree in which A. mystaci- 2
3 (v) rock cover; and (vi) bare soil. nus was the outgroup, as it is the closet outer species to 3
4 A. flavicollis and A. sylvaticus (Michaux et al. 2002). 4
Rodent identification
5 The phylogeny was built following Bayesian infer- 5
6 Most granivorous rodents found at the study area ence using MR BAYES 3.2 (Ronquist et al. 2012) ap- 6
7 were easily identifiable on the basis of morphologi- plying a HKY+Gamma substitution model, according 7
8 cal traits. However, due to the difficulties in identify- to the results of PARTITION FINDER 1.1.1 (Lanfear 8
9 ing the 2 potential Apodemus species (A. sylvaticus and et al. 2012). Two parallel runs of 2 million generations 9
10 A. flavicollis) with the naked eye (Arrizabalaga et al. each were conducted using one cold and 2 incremental- 10
11 1999; Reutter et al. 1999), we performed genetic analy- ly heated Markov chains (L = 0.2), sampling every 1000 11
12 ses (Michaux et al. 2001). DNA was extracted from ro- steps. We first checked the standard convergence diag- 12
13 dent tissue (tail tip or ear skin) according to the Aljanabi nostics implemented in MRBAYES and then assessed 13
14 and Martínez (1997) salt extraction protocol. We ampli- the average standard deviation of the split frequencies 14
15 fied the mitochondrial control region (d-loop) by poly- to deduce that the Markov chain had become stationar- 15
16 merase chain reaction (PCR) using the primers M15997 ity. After 500 000 generations, the average standard de- 16
17 and H16401 (Stacy et al. 1997; for the details of the viation of the split frequencies stabilized at values close 17
18 PCR conditions see Hirota et al. 2004). Sequencing was to zero (0.001). Hence, phylogenetic trees were summa- 18
19 run on a 3730XL DNA analyzer and sequences were ed- rized using the all-compatible consensus command with 19
20 ited using SEQUENCHER 4.1 (Gene Codes, Ann Arbor, 25% burn-in. These analyses enabled species to be iden- 20
21 MI, USA). The sequences were trimmed to 297 base tified by exploring the clusters in the phylogenetic tree. 21
22 pairs to reduce the proportion of missing data. We chose 22
23
Data analyses 23
this region of the mitochondrial DNA (d-loop) because
24 its mutation rate is very high and, at the same time, al- The effects of habitat type on rodent abundance were 24
25 lowed us to distinguish the 2 species of Apodemus, and analyzed by means of a generalized linear model (GLZ) 25
26 it permitted estimation of the patterns of inter-popula- with the error modeled as a Poisson distribution. We 26
27 tion gene flow and genetic structure at the intraspecific tested the effects of habitat type and microhabitat struc- 27
28 level (Hirota et al. 2004). First, we identified all Apode- tural characteristics on the occurrence of small rodents 28
29 mus to the species level by blasting the sequence of each by means of GLZ with error modeled as a binomial dis- 29
30 rodent against GenBank to assess the percentage of coin- tribution. We followed Akaike’s information criterion 30
31 cidence with the sequences of A. sylvaticus and A. flavi- (AIC) for model selection. To test whether the different 31
32 collis available at that database (number of different nu- groups of rodents were segregated in space within each 32
33 cleotides by the total number of compared nucleotides). shared habitat type we used multiple linear regression 33
34 We confirmed the identification by performing a phy- models with a stepwise procedure. We also tested at the 34
35 logenetic analysis with all the sequences of mice tenta- scale of trapping station the effects of vegetation struc- 35
36 tively assigned to 1 species or the other. We pooled all ture on the space use of each group of rodents within 36
37 the sequences and aligned them with CLUSTALW sup- each habitat, using the average values of the microhab- 37
38 plied via http://align.genome.jp (gap open and gap ex- itat variables measured in all traps of each trapping sta- 38
39 tension penalties were those provided by default by the tion. 39
40 software: 15 and 6.66, respectively). This alignment Regarding the Apodemus species, we performed a 40
41 was collapsed into its unique haplotypes using Perl Col- GLZ with the error modeled as a Poisson distribution 41
42 lapse (Chesters 2013) and these unique sequences were testing the effects of habitat type and mouse species (A. 42
43 further pooled with mitochondrial d-loop sequences of sylvaticus and A. flavicollis) on the total abundance of 43
44 A. flavicollis and A. sylvaticus, downloaded from Gen- mice. Finally, we analyzed the effects of habitat type 44
45 Bank, which were used as external reference sequences and microhabitat structural characteristics on the oc- 45
46 (accession numbers: AJ410624, AJ410626, AJ410628, currence of both Apodemus spp separately by means of 46
47 EU220767, EU220768, EF396216 and EF396220) GLZs with the error modeled as a binomial distribution. 47
48 along with that of Apodemus mystacinus (accession We used AIC for selecting the proper model for the 2 48
49 numbers AY623065 and AY623066). This new set of se- mouse species. For both A. sylvaticus and A. flavicollis, 49
50 quences was aligned using CLUSTALW with the same we additionally analyzed inter-population genetic differ- 50
51 51

1 entiation using the exact test of sample differentiation cordant with theories of DNA taxonomy, individuals 1
2 based on haplotype frequencies using a Markov chain identified as A. sylvaticus and A. flavicollis formed dis- 2
3 approach (Raymond & Rousset 1995) as implemented crete clusters, separated by substantial branch lengths 3
4 in ARLEQUIN (Excoffier et al. 2005). Gene and nucle- (Fig. 2), representing genetic discontinuities that allowed4
5 otide diversity were calculated with the same software. us to distinguish each species from the other (Herbert 5
6 and Gregory 2005), thus confirming our initial species 6
7
RESULTS determination based on raw genetic similarity and Gen- 7
8 Bank blasting. The phylogenetic tree showed that A. syl- 8
9 The density of small mammals at the study area was vaticus and A. flavicollis were more closely related than 9
10 very low. We captured 58 small rodents with a trapping with A. mystacinus or with those from other parts of Eu- 10
11 effort of 3600 traps-nights, including mice, dormice and rope downloaded from GenBank and used as references 11
12 voles. We did not recapture any rodent captured in the (Poland and Greece for A. sylvaticus and A. flavicollis 12
13 first year during trap sessions in the second year. Mice respectively; Fig. 2). In both species these non-Iberian 13
14 were the most abundant group, representing 59% of sequences were grouped in clusters with high node sup- 14
15 captures and occurring in all habitat types. The genet- port (over 0.8 Bayesian posterior probability; Fig 2). 15
16 ic analysis of mice tissues identified the wood mouse Our results showed that A. sylvaticus and A. flavi- 16
17 (Apodemus sylvaticus) and the yellow-necked mouse collis shared 3 forest types: pinewoods (1600–1700 m), 17
18 (A. flavicollis). After blasting the sequences of uniden- oak forests (1700–1800 m) and silver firs (1800–1900 18
19 tified specimens against GenBank, all of them could be m). A. sylvaticus was 1significantly more abundant than 19
20 assigned to 1 species or the other according to their ge- 20
A. flavicollis (χ21 = 5.3, P = 0.021), except in silver firs,
21 netic similarity (greater than 97% in all cases) with ref- where their abundance did not differ significantly (χ21 21
22 erence sequences available from that database. Con- 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 Figure 2 DNA phylogeny of 1 mitochondrial gene (control region d-loop). Tree topology was inferred using Bayesian inference 48
49 (HKY + gamma substitution model). Support for each node is represented by the Bayesian probability value. The tips correspond to 49
50 the unique haplotype of either Apodemus sylvaticus or Apodemus flavicollis sequenced at our study site and the reference sequences 50
51 downloaded from GenBank. The species membership and the geographical origin of each cluster are indicated. 51

J. Urgoiti et al.
1 = 0.3, P = 0.594; Fig. 3). Both species showed a segre- and A. flavicollis (Table 1). 1
2 gation at the scale of habitat type (χ22 = 8.4, P = 0.015), The intra-specific population analyses at a landscape 2
3 as the abundance of each species was inversely relat- scale revealed a total of 13 haplotypes in A. sylvati- 3
4 ed to the abundance of the other species (Fig. 3), and in cus and 5 in A. flavicollis, with a high number of single 4
5 meadows A. sylvaticus was the only mouse species. This ones. Due to the low number of individuals per popula- 5
6 segregation coincided with changes in altitude so that tion, we could not perform any population genetic anal- 6
7 A. sylvaticus was comparatively well represented in the yses on A. flavicollis. For A. sylvaticus, we compared 7
8 lowlands and A. flavicollis in the highlands. At a “meso- population genetics between the valleys of Espot and 8
9 scale” level (i.e. trapping station) our results pointed to Boí (Fig. 1), by pooling all individuals captured in the 3 9
10 a strong spatial segregation, because both species were populations of each valley. According to the high num- 10
11 always captured in different trapping stations within the ber of single haplotypes (Table 2), the gene and nucleo- 11
12 shared habitats. The analysis of microhabitat preferenc- tide diversity of A. sylvaticus was high in both sites (0.83 12
13 es showed that the probability of occurrence of A. syl- ± 0.08 and 0.010 ± 0.006 in Espot; 0.97 ± 0.06 and 0.012 13
14 vaticus was marginally associated with both shrub cover ± 0.008 in Boí). In addition, the exact test of sample dif- 14
15 (positively) and rock cover (negatively), whereas no as- ferentiation based on haplotype frequencies (20 000 15
16 sociation was found between the microhabitat variables Markov steps) showed that there was a significant ge- 16
17 netic differentiation between the A. sylvaticus popula- 17
18 tions inhabiting the 2 valleys (P = 0.014). 18
19 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
Figure 3 Mean (± SE) values of capture success of Apodemus
35 35
flavicollis and Apodemus sylvaticus captured in shared habitat Figure 4 Mean (± SE) capture success of mice, dormice and
36 36
types. voles in different habitat types of central Pyrenees.
37 37
38 38
39 Table 1 Summary statistics for the generalized linear models with binomial distribution applied to analyze the effects of microhab- 39
40 itat variables on the probability of occurrence of Apodemus sylvaticus (Model AIC: 198.5) and Apodemus flavicollis (Model AIC: 40
41 99.7) 41
42 42
Apodemus sylvaticus Apodemus flavicollis
43 43
44 Variable Parameter SE Z P Parameter SE Z P 44
45 Shrub 0.015 0.008 1.875 0.054 0.007 0.024 0.294 0.779 45
46 Herb — — — — 0.004 0.023 0.172 0.848 46
47 47
Tree 0.004 0.006 0.667 0.477 −0.002 0.011 −0.175 0.842
48 48
Litter — — — — 0.009 0.022 0.409 0.693
49 49
50 Rock −0.020 0.011 −0.185 0.063 0.005 0.020 0.245 0.811 50
51 —, These variables are not included in the final models. 51

1 Besides these 2 typical scatter-hoarding rodents, we shared habitat types), mice were not spatially aggregated 1
2 also captured other seed-eating rodents such as dor- or segregated with voles or dormice; they were random- 2
3 mice, Elyomis quercinus (Linnaeus, 1766) (24% of all ly captured across the different trapping stations of the 3
4 captures) and voles, Myodes glareolus (Schreber, 1780) same habitat. Instead, the abundance of voles and dor- 4
5 (17%), which showed a marked habitat selection. Dor- mice were significantly associated with the trapping sta- 5
6 mice were not present in beech-oak forests and were tions in silver firs (t1,27 = 3.54, P = 0.01), suggesting that, 6
7 nearly absent in meadows. Conversely, voles were not at the mesohabitat scale, they preferred the same envi- 7
8 present in beech-oak forests and pinewoods but were ronments within this habitat type. However, we found 8
9 well represented in both meadows and silver firs (Fig 4). no associations between the abundance of each rodent 9
10 Even so, all habitats were shared by at least 2 different group and the average of vegetation variables at the 10
11 groups of rodents except beech-oak forests, where mice scale of the trapping station. In contrast, at the micro- 11
12 were the only rodent captured. Pinewoods were shared habitat scale (i.e. individual trap), mice and vole abun- 12
13 by mice and dormice, whereas in silver firs and mead- dance were significantly associated with shrub cover, 13
14 ows we found mice, dormice and voles. whereas dormice were associated with litter and rock 14
15 At the mesohabitat scale (i.e. trapping station within cover (Table 3). 15
16 16
17 17
DISCUSSION
18 18
19 Our DNA analysis confirms, for the first time, the 19
Table 2 Frequency of each Apodemus sylvaticus haplotype in
20 the valleys of Espot and Boí
presence of the yellow-necked mouse (A. flavicollis) in 20
21 the central Pyrenees, suggesting that this area covers 21
22 Haplotype 1 3 part of its southern distribution, potentially connecting 22
23 Haplotype 2 1 it with other populations found in the Catalonian coastal 23
24 Haplotype 3 1 mountain ranges and the Montseny massif in the east (see 24
25 Arrizabalaga et al. 1999; Arrizabalaga & Torre 2002; 25
Haplotype 4 4
26 Torre et al. 2015). A. flavicollis records based on DNA 26
Haplotype 5 1 identification were available only for the western part
27 27
28 Haplotype 6 1 of the southern slope of the Pyrenees (Michaux et al. 28
29 Haplotype 7 1 2005), but its presence in the central and oriental sectors 29
30 Haplotype 8 1 of this mountain range is not clear yet, due to the limit- 30
31 ed studies in this zone (Gosálbez & López-Fuster 1985; 31
Haplotype 9 2
32 Arnán et al. 2014) and the difficulty of differentiating it 32
33 Haplotype 10 1 1 from the wood mouse (A. sylvaticus) using morphologi- 33
34 Haplotype 11 1 cal traits (Arrizabalaga & Torre 2002; Ancillotto 2017). 34
35 Haplotype 12 1 The phylogenetic tree confirmed that A. flavicollis 35
36 Haplotype 13 1 and A. sylvaticus are closely related species, and coin- 36
37 37
38 38
39 Table 3 Summary statistics of the generalized linear models with binomial distribution applied to determine the effects of habitat 39
40 type and microhabitat structural characteristics on the probability of occurrence of small mammal species 40
41 Variable Mice Dormice Vole 41
42 42
Parameter SE Z P Parameter SE Z P Parameter SE Z P
43 43
44 Habitat −4.257 0.774 −5.50 0.661 −6.200 1.189 −5.214 0.395 −7.465 2.134 −3.498 0.253 44
45 Shrub 0.014 0.007 2.059 0.046 −0.091 0.087 −1.046 0.295 0.037 0.017 2.189 0.028 45
46 Herb — — — — — — — — 0.022 0.019 1.176 0.237 46
47 47
Tree 0.004 0.047 0.851 0.429 — — — — 0.014 0.013 1.120 0.253
48 48
Litter — — — — 0.030 0.015 2.069 0.039 0.012 0.020 0.612 0.552
49 49
50 Rock 0,015 0.008 1.786 0.076 0.032 0.012 2.689 0.007 0.019 0.019 1.016 0.306 50
51 —, These variables are not included in the final models. 51

J. Urgoiti et al.
1 cident with phylogenies based on nuclear and other mi- The rarity and arguably lower performance of A. syl- 1
2 tochondrial markers. In fact, both species are included vaticus at high altitudes may be due to the extreme en- 2
3 in the same subgenus (Sylvaemus) within the Apodemus vironmental conditions in winter, with abundant snow 3
4 genus (Michaux et al. 2002). The tree also revealed the cover and extremely low temperatures, which limit food 4
5 genetic distinctiveness of the Iberian haplotypes in both availability and may account for the high rodent mor- 5
6 species with respect to those from other areas of Europe tality. Thus, the mortality of small mammals in gener- 6
7 (Greece and Poland for A. flavicollis and A. sylvaticus, al is probably high during winter in this area, as sug- 7
8 respectively). A. flavicollis had a glacial refuge in the gests the lack of recaptures from the first to the second 8
9 Balkans and, from there, it recolonized the remainder year of our study. In fact, this harsh alpine environment 9
10 of Europe, including the Iberian Peninsula. These dif- may also explain the generally low population densi- 10
11 ferences between Iberia and Greece have been observed ties compared to other nearby temperate ecosystems at 11
12 in phylogeographic studies based on other mitochondri- lower altitudes (Sunyer et al. 2016). Similar low den- 12
13 al markers (cytochrome b: Michaux et al. 2005). In the sities of small mammals have been documented in this 13
14 case of A. sylvaticus, Iberia was the refuge from which and other alpine areas (e.g. Bertolino et al. 2001; Arnán 14
15 it recolonized north-central Europe after glaciation (Mi- et al. 2014). These severe environmental conditions due 15
16 chaux et al. 2005). According to this, the differences be- to altitude may be, in fact, the cause of the genetic dif- 16
17 tween the Iberian haplotypes and those from Poland, in- ferentiation between the populations of A. sylvaticus in- 17
18 cluded as reference sequences, should not be as high. habiting different valleys, because to move from Boí to 18
19 However, despite their higher genetic proximity, Iberian Espot it is necessary to cross high alpine habitats. The 19
20 haplotypes are more closely related than with those from intra-specific population genetic analyses showed that 20
21 north-central Europe, which is also reflected in phylog- some haplotypes that were relatively frequent in Espot 21
22 enies based on cytochrome b (Michaux et al. 2003). were not present in Boí. The high number of haplotypes 22
23 Moreover, in our study, based on a mitochondrial mark- (including many single haplotypes) is common in stud- 23
24 er with a higher mutation rate (the control region d-loop), ies using the mitochondrial control region as a genetic 24
25 these differences are expected to be higher between geo- marker (see Trizio et al. 2005). The control region is not 25
26 graphical regions. expressed functionally so that mutations are not elimi- 26
27 Our study points identify a spatial segregation be- nated by natural selection and with rodent short genera- 27
28 tween A. sylvaticus and A. flavicollis at the scale of for- tion time, mutations are likely to accumulate in the mi- 28
29 est type, as an increase in the abundance of one species tochondrial genome (Li et al. 1996). When movements 29
30 was associated with a decrease in the other. This seg- between populations are constrained, these new haplo- 30
31 regation becomes more dramatic at a “mesoscale” (i.e. types resulting from local mutations are not spread re- 31
32 trapping station), because both species were always cap- sulting in inter-population genetic differentiation (Hi- 32
33 tured in different trapping stations despite the shared rota et al. 2004). In our study areas, this occurred at a 33
34 habitat types. Part of this segregation may be explained relatively small spatial scale (the distance between the 34
35 by altitude (see Saint-Girons 1973), as A. sylvaticus 2 valleys is only 5 km). Accordingly, other studies have 35
36 abundance decreased with altitude. demonstrated that such a genetic differentiation can take 36
37 place at similar distances when the habitat between 2 37
Interestingly, in the valley of Espot all captures of A.
38
flavicollis came from silver-fir forests (the habitat locat- populations is not suitable for dispersal (Hirota et al. 38
39 2004). However, we need to be cautious with these re- 39
ed at the highest altitude), where A. sylvaticus was ab-
40 sults because mitochondrial DNA is only inherited ma- 40
sent. A. sylvaticus is a generalist mouse that has been
41
found along a significant variety of habitat types in the ternally (Gyllesten et al. 1985) and, thus, just reflects fe- 41
42 male interchange between the 2 valleys. Regardless, we 42
Iberian Peninsula, from coastal dunes to northern mature
43 do not believe that male crossings from Boí to Espot are 43
forests (Torre et al. 2002; Muñoz et al. 2009; Sunyer et
44 likely, because we did not observe a sex-biased altitudi- 44
al. 2016), although our results suggest its worst perfor-
45
mance may be at high altitudes. In contrast, A. flavicol- nal distribution in A. sylvaticus. 45
46 We found significant changes in the distribution of 46
lis shows more strict forest requirements and is more
47 other small rodents among sub-alpine habitat types. 47
linked to Euro-Siberian environments (Arrizabalga &
48 The garden dormouse (E. quercinus) was present in all 48
Torre 2002), which seem to be fulfilled by most sub-al-
49 sub-alpine habitats of the central Pyrenees except in 49
pine habitats of the Central Pyrenees, the southern limit
50 beech-oak forests, consistent with its widespread distri- 50
of its range in Europe.
51 51

1 bution in the Palaearctic region along a variety of habi- The habitats monitored in in this study are dominat- 1
2 tat types (Amori et al. 1994). In contrast, voles (Myodes ed by tree species producing seeds (pine nuts, hazels, 2
3 glareolus) were present just in meadows and silver firs, beechnuts, acorns) that are regularly consumed by the 4 3
4 which are characterized by deeper soils with higher hu- rodent species reported. Two of these species are strict 4
5 midity, making them permeable for voles (Van Appel- seed predators, the bank vole (Myodes glareolus) and 5
6 doorn et al. 1992; Schweingruber 1993). the garden dormouse (Elyomis quercinus), with seeds 6
7 We found that most sub-alpine habitats of the cen- representing >90% of their diet (Kuipers et al. 2012; 7
8 tral Pyrenees were shared by more than one group of Bonachi et al. 2017). They consume an important pro- 8
9 small rodents (mice, dormice and voles). At the meso- portion of the seed crops in Western Europe (Ouden et 9
10 scale of the trapping station (i.e. approximately 500 m2), al. 2005), especially in oak forests, but seed losses by 10
11 we did not find any clear pattern of spatial segregation these species are not expected in beech-oak forests in 11
12 among mice, dormice or voles within the shared habi- the Pyrenees because they were not found in these habi- 12
13 tats. Rather, they seem to be randomly distributed at this tat types. 13
14 scale within each habitat type, or even preferring the However, they may cause a reduction in tree repro- 14
15 same habitat, as in the case of dormice and voles with- duction in habitats with hazel trees (i.e. meadows) where 15
16 in silver fir forests. However, at the microhabitat scale, both species are well represented and where large num- 16
17 the capture probability of mice and voles (the smaller of bers of hazels are usually found partially consumed or 17
18 the 2 species) was higher in traps with high shrub cover, damaged by rodents (pers. obs.). 18
19 which may provide protection against predators (Nor- 19
Most importantly, the 2 species of mice captured (A.
20 dyke & Buskirk 1991; Marsh & Harris 2000; Bertolino 20
sylvaticus and A. flavicollis) are prominent seed dispers-
21 2007): that is, red foxes, mustelids, martens, European 21
ers that may influence the forest recruitment dynamics,
22 wild cats, owls and boreal owls in central Pyrenees (Ca- 22
especially in the case of oaks, hazels and beech (Jen-
23 nut 2002; Gosàlbez 2002; Raspall et al. 2007). In addi- 23
sen & Nielsen 1986; Kollman & Schill 1996; Ouden et
24 tion, some shrubs in our study area, such as Rhododen- 24
al. 2005; Sunyer et al. 2015; Zwolak et al. 2016). Given
25 dron ferrugineum and Vaccinium myrtillus, may also 25
that patterns of seed selection differ between the 2 Apo-
26 provide food (i.e. seeds) for some rodent species (Pornon 26
demus species (García-Castaño et al. 2006), the increase
27 et al. 1997; Selas 2000). By contrast, the capture prob- 27
of global temperatures, favoring the presence of A. syl-
28 ability of dormice, at the microhabitat scale, was pos- 28
vaticus at the detriment of A. flavicollis, may have cas-
29 itively associated with rock and herbaceous cover, in 29
cading effects on the patterns of seed dispersal and re-
30 agreement with previous studies linking dormouse pres- 30
cruitment in the forest habitats of the Pyrenees.
31 ence to rockslides penetrating into forests (Gosàlbez & 31
32 Thus, these alpine ecosystems, hosting low densities 32
López-Fuster 1985). Rocky cavities may be inaccessible of small mammals that have an important role in for-
33 to large predators and may also offer a better chance of 33
34 est tree recruitment, can be especially sensitive to the ef- 34
escaping from small predators (Bertolino 2007). fects of global warming. Further research, carried out at
35 35
The present study provides a first assessment of the a wider scale, should be focused on rodent population
36 36
composition of small mammals in central Pyrenees dynamics and genetics in the long term, as well as ana-
37 37
along a succession of habitat types from a macrohabitat lyzing the patterns of seed dispersal by different rodents.
38 38
context to a microhabitat perspective. The climatic con- This knowledge will improve our understanding on the
39 39
ditions of high mountains are expected change dramati- structure and functioning of alpine and sub-alpine eco-
40 40
cally due to the effects of global warming, and it will be systems, and help to preserve biodiversity in the context
41 41
challenging to predict its consequences for the conserva- of global warming.
42 42
tion of rodent communities in alpine and sub-alpine sys-
43 43
tems. We point to the role of population isolation among
44 ACKNOWLEDGMENTS 44
valleys of high-altitude mountain chains in increasing
45 45
the genetic diversity of rodents, like A. sylvaticus. We are grateful to Pau Sunyer for his assistance
46 46
However, our study suggests that the elevation of during field work. This work was supported by the proj-
47 47
global temperatures might favor the presence of A. syl- ects IMPACTBOAR (394/2011) and PLAGANADO
48 48
vaticus in higher areas, probably to the detriment of A. AGL2014-54739-9. R.B. was funded by a contract of
49 49
flavicollis, which may ultimately experience limits to its the Atracción de Talento Investigador Programme (Go-
50 50
southern European distribution. bierno de Extremadura TA13032). We are grateful to the
51 51

J. Urgoiti et al.
1 personnel of the Parc Nacional d’Aigüestortes i Estany Bonacchi A, Bartolommei P, Gasperini S, Manzo E, 1
2 de Sant Maurici for their assistance during this research Cozzolino R (2017). Acorn choice by small mammals 2
3 project. All small mammal trappings were performed in a Mediterranean deciduous oak forest. Ethology, 3
4 with a license from the National Park. Ecology and Evolution 29, 105–18. 4
5 Boyce MS, Mao JS, Merril EH et al. (2003). Scale and 5
6 REFERENCES heterogeneity in habitat selection by elk in Yellow- 6
7 stone National Park. Ecoscience 10, 421–31. 7
8 Aljanabi SM, Martínez I (1997). Universal and rap- Canova L (1993). Resource partitioning between the 8
9 id salt-extraction of high quality genomic DNA for bank vole Clethrionomys glareolus and the wood 9
10 PCR- based techniques. Nucleic Acids Research 25, mouse Apodemus sylvaticus in woodland habitats. 10
11 4692–3. Bolletino di Zoologia 60, 193–8. 11
12 Amori G, Cantini M, Rota V (1994). Distribution and Canut J (2002). Avifauna. In: Canseco, ed. Parque Na- 12
13 conservation of Italian dormice. Hystrix, the Italian cional de Aigüestortes i Estany de Sant Maurici. Tala- 13
14 Journal of Mammalogy 6, 331–6. vera de la Reina, pp. 187–222. 14
15 Ancillotto L, Mori E, Sozio G, Solano E, Bertolino S, Castién E, Gosálbez J (1994). Habitat selection of Apo- 15
16 Russo D (2017). A novel approach to field identifi- demus flavicollis in a Fagus sylvatica forest in the 16
17 cation of cryptic Apodemus wood mice: Calls differ western Pyrenees. Folia Zoologica 43, 219–24. 17
18 more than morphology. Mammalian Review 47, 6–10. 18
Chesters D (2013). collapsetypes.pl
19 Arnan X, Comas L, Gracia M, Retana J (2014). Compo- 19
Excoffier L, Laval G, Schneider S (2005). Arlequin ver.
20 sition and habitat use of small mammals in old- 20
3.0: An integrated software package for population
21 growth mountain forests. Journal of Natural History 21
genetics data analysis. Evolutionary Bioinformatics
22 48, 481–94. Online 1, 47–50.
22
23 Arrizabalaga A, Torre I, Catzeflis F, Renaud F, San- 23
24
García-Castaño JL, Jordano P, Kollmann J (2006). Spa- 24
talla F (1999). Primera citació d’Apodemus flavi- tial variation of post-dispersal seed removal by ro-
25 collis (Melchior, 1834) al Montseny. Determinació 25
dents in highland microhabitats of Spain and Switzer-
26 morfològica i genètica. Trobada d’Estudiosos del 26
land. Seed Science Research 16, 213–22.
27 Montseny 3, 193–5. 27
28 Giordano AR, Ridenhour BJ, Storfer A (2007). The in- 28
Arrizabalaga A, Torre I (2002). Ratón leonado (Apode- fluence of altitude and topography on genetic
29 mus flavicollis Melchior, 1834). In: Palomo LJ, 29
30
structure in the long-toed salamander (Ambystoma 30
Gisbert J eds. Atlas de los Mamíferos terrestres de macrodactulym). Molecular Ecology 16, 1625–
31 España. Dirección General de Conservación de la 31
37.
32 Naturaleza-SECEM-SECEMU, pp. 400–403. 32
33 Gosàlbez J (2002). Mamíferos. In: Canseco, ed. Parque 33
Avenant N (2011). The potential utility of rodents and Nacional de Aigüestortes i Estany de Sant Maurici.
34 other small mammals as indicators of ecosystem ‘in- 34
35 Talavera de la Reina, pp. 223–56. 35
tegrity’ of South African grasslands. Wildlife Re-
36 Gosàlbez J, López-Fuster MJ (1985). The natural com- 36
search 38, 626–39.
37 munities of small mammals (insectivores and rodents) 37
Bailey DW, Gross JE, Laca EA et al. (1996). Mecha- of Catalonia (Spain). Miscelanea Zoologica 9, 375–
38 38
nisms that result in large herbivore grazing distribu- 87.
39 39
tion patterns. Journal of Range Management 49, 386–
40 Gyllensten U, Wharton D, Wilson AC (1985). Maternal 40
400.
41 inheritance of mitochondrial DNA during backcross- 41
Bertolino S (2007). Microhabitat use by garden dormice ing of two species of mice. Journal of Heredity 76,
42 42
during nocturnal activity. Journal of Zoology 272, 321–24.
43 43
176–82. Heaney LR (2001). Small mammal diversity along ele-
44 44
45 Bertolino S, Viano C, Currado I (2001). Population dy- vational gradients in the Philippines: An assessment 45
46
namics, breeding patterns and spatial utilisation of the of patterns and hypotheses. Global Ecology Biogeog- 46
garden dormouse (Eliomys quercinus) in an Alpine raphy 10, 15–39.
47 47
habitat. Journal of Zoology 253, 513–21. Hille SM, Mortelliti A (2010). Microhabitat partitioning
48 48
49 Blondel J, Aronson J (1999). Biology and Wildlife of of Apodemus flavicollis and Myodes glareolus in the 49
50 the Mediterranean Region. Oxford University Press, sub-montane Alps: A preliminary assessment. Hystrix 50
51 USA. Italian Journal of Mammalogy 21, 157– 63. 51

1 Hirota T, Hirohata T, Mashima H, Satoh T, Obara Y McCain CM (2004). The mid-domain effect applied to 1
2 (2004). Population structure of the large Japanese elevational gradients: Species richness of small mam- 2
3 field mouse, Apodemus speciosus (Rodentia: Muri- mals in Costa Rica. Journal of Biogeography 31, 19– 3
4 dae), in suburban landscape, based on mitochondrial 31. 4
5 D-loop sequences. Molecular Ecology 13, 3275–82. McCain CM (2005). Elevational gradients in diversity 5
6 H o r t a l J , Tr i a n t i s K A , M e i r i S , T h é b a u l t E , of small mammals. Ecology 86, 366–72. 6
7 Sfenthourakis S (2009). Island species richness in- 7
Michaux JR, Kinet S, Filippucci MG, Libois R, Besnard
8 creases with habitat diversity. American Naturalist 8
A, Catzeflis F (2001). Molecular identification of
9 174, 205–17. 9
three sympatric species of wood mice (Apodemus syl-
10 10
Jensen TS, Nielsen OF (1986). Rodents as seed dispers- vaticus, A. flavicollis, A. alpicola) in western Europe
11 ers in a heath — Oak wood succession. Oecologia 11
(Muridae: Rodentia). Molecular Ecology Notes 1,
12 70, 214–21. 12
260–3.
13 13
Kollman J, Schill H-P (1996). Spatial patterns of disper- Michaux JR, Chevret P, Filippucci MG, Macholan M
14 14
sal, seed predation and germination during coloniza- (2002). Phylogeny of the genus Apodemus with a
15 15
tion of abandoned grassland by Quercus petraea and special emphasis on the subgenus Sylvaemus using
16 16
Corylus avellana. Vegetatio 125, 193–205. the nuclear IRBP gene and two mitochondrial mark-
17 17
18 Kotler BP (1984). Risk of predation and structure of ers: Cytochrome b and 12S rRNA. Molecular Phylo- 18
19 desert rodent communities. Ecology 65, 689–701. genetics and Evolution 23, 123–36. 19
20 Kotler BP, Brown JS, Hasson O (1991). Factors affect- Michaux JR, Magnanou E, Paradis E, Nieberding C, Li- 20
21 ing gerbil foraging behaviour and rates of owl preda- bois R (2003). Mitochondrial phylogeography of the 21
22 tion. Ecology 72, 2249–60. woodmouse (Apodemus sylvaticus) in the Western 22
23 Kohler T, Maselli D (2009). Mountains and Climate Palearctic region. Molecular Ecology 12, 685–97. 23
24 Change – From Understanding to Action. Geographi- Michaux JR, Libois R, Filippucci MG (2005). So close 24
25 ca Bernensia, Bern. and so different: comparative phylogeography of two 25
26 Kuipers L, Scholten J, Aiss JBM et al. (2012). The diet small mammal species, the yellow-necked fieldmouse 26
27 of the garden dormouse (Eliomys quercinus) in the (Apodemus flavicollis) and the woodmouse (Apode- 27
28 Netherlands in summer and autumn. Lutra 55, 17–27. mus sylvaticus) in the Western Palearctic region. He- 28
29 Lanfear R, Calcott B, Ho SYW, Guindon S (2012). redity 94, 52–63. 29
30 Partitionfinder: Combined selection of partitioning Mitchell WA, Abramsky Z, Kotler BP, Pinshow B, 30
31 schemes and substitution models for phylogenetic an- Brown JS (1990). The effect of competition on for- 31
32 alyzes. Molecular Biology and Evolution 29, 1695– aging activity in desert rodents: Theory and experi- 32
33 701. ments. Ecology 71, 844–54. 33
34 34
Li WH, Ellsworth DL, Krushkal J, Chang BH, Muñoz A, Bonal R, Díaz M (2009). Ungulates, rodents,
35 35
Hewett-Emmett D (1996). Rates of nucleotide sub- shrubs: Interactions in a diverse Mediterranean eco-
36 36
stitution in primates and rodents and the genera- system. Basic and Applied Ecology 10, 151–60.
37 37
tion-time effect hypothesis. Molecular Phylogenetics Nordyke KA, Buskirk SW (1991). Southern redbacked
38 38
and Evolution 5, 182–7. vole, Clethrionomys gapperi, populations in relation
39 39
40
Losos JB (2008). Phylogenetic niche conservatism, phy- to stand succession and old growth character in the 40
41
logenetic signal and the relationship between phylo- central Rocky Mountains. Canadian Field Naturalist 41
42
genetic relatedness and ecological similarity among 105, 330–34. 42
species. Ecology Letters 11, 995–1007. Ostfeld RS (1992). Small mammal herbivores in a
43 43
44 Marsh A, Harris S (2000). Partitioning of woodland hab- patchy environment: Individual strategies and popula- 44
45 itat resources by two sympatric species of Apodemus: tion responses. In: Hunter MD, Ohgushi T, Price PW, 45
46 lessons for the conservation of the yellow-necked eds. Effects of Resource Distribution on Animal– 46
47 mouse (A. flavicollis) in Britain. Biological Conser- plant Interactions. Academic Press, New York, pp. 47
48 vation 92, 275–83. 65–116. 48
49 Mazurkiewicz M (1994). Factors influencing the distri- Ouden J den, Jansen PA, Smit R (2005). Jays, mice and 49
50 bution of the bank vole in forest habitats. Acta Theri- oaks: Predation and dispersal of Quercus robur and 50
51 ologica 39, 113–26. Q. petraea in North-western Europe. In: P.M.Forget 51

J. Urgoiti et al.
1 et al. eds. Seed Fate: Predation, Dispersal and Seed- Sunyer P, Muñoz A, Mazerolle MJ, Bonal R, Espelta JM 1
2 ling Establishment, CABI Publishing, Wallingford (2016). Wood mouse population dynamics: Interplay 2
3 pp. 223–40. among seed abundance seasonality, shrub cover and 3
4 Pearce JL, Venier LA (2005). Small mammals as bio- wild boar interferance. Mammalian Biology 81, 372– 4
5 indicators of sustainable boreal forest management. 9. 5
6 Forest Ecology and Management 208, 153–75. Tews J, Brose U, Grimm V et al. (2004). Animal species 6
7 driven by habitat heterogeneity/diversity: The impor- 7
Pornon A, Escaravage N, Hill-Bottraud I, Doche B
8 tance of keystone structures. Journal of Biogeogra- 8
(1997). Variation of reproductive traits in Rhododen-
9 phy 31, 79–92. 9
dron ferrugineum L. (Ericaceae) populations along a
10 10
successional gradient. Plant Ecology 130, 1–11. Torbidoni EF, Grau RH, Camps A (2005). Trail prefer-
11 11
Raymond M, Rousset F (1995). An exact test for popu- ences and visitor characteristics in Aigüestortes I Es-
12 12
lation differentiation. Evolution 49, 1280–3. tany de Sant Maurici National Park, Spain. Mountain
13 13
Raspall A, Comas L, Boet O (2007). Mamiférs. In: Or- Research and Development 25, 51–59.
14 14
15 ganismo Autónomo de Parques Nacionales, ed. Fau- Torre I, Arrizabalaga A, Díaz M (2002). Ratón de cam- 15
16 na salvatge del Parc Nacional D’Aigúestortes i Esta- po (Apodemus sylvaticus Linnaeus, 1758). Galemys 16
17 ny de Sant Maurici, OAPN, Madrid, pp. 64. 14, 1–26. 17
18 Reutter BA, Hausser J, Vogel P (1999). Discriminant Torre I, Fernández L, Arrizabalaga A (2015). Using barn 18
19 analysis of skull morphometric characters in Apode- owl Tyto alba pellet analyzes to monitor the distribu- 19
20 mus sylvaticus, A. flavicollis and A. alpicola (Mam- tion patterns of the yellow-necked mouse (Apodemus 20
21 malia; Rodentia) from the Alps. Acta Theriologica 44, flavicollis Melchior 1834) in a transitional Mediterra- 21
22 299–308. nean mountain. Mammal study 40, 133–42. 22
23 Ronquist F, Teslenko M, van der Mark P et al. (2012). Trizio I, Crestanello B, Galbusera P et al. (2005). Geo- 23
24 MrBayes 3.2: Efficient Bayesian phylogenetic infer- graphical distance and physical barriers shape the ge- 24
25 ence and model choice across a large model space. netic structure of Eurasian red squirrels (Sciurus vul- 25
26 Systematic Biology 61, 539–42. garis) in the Italian Alps. Molecular Ecology 14, 26
27
Saint-Girons MC (1973). Les mammiféres de France et 469– 81. 27
28 Van Apeldoorn RC, Oostenbrink WT, Vanwinden A, 28
du Benelux. Paris. Doin.
29 Vanderzee FF (1992). Effects of habitat fragmenta- 29
30
Schweingruber FH (1993). Trees and Wood in Dendro- 30
chronology: Morphological, Anatomical, and Tree- tion on the bank vole, Clethrionomys glareolus, in an
31 agricultural landscape. Oikos 65, 265–74. 31
32
ring Analytical Characteristics of Trees Frequently 32
Used in Dentrochronology. Springer–Verlag. Berlin. Vander Wall SB (1990). Food Hoarding in Animals. The
33 33
Selas V (2000). Seed production of a masting dwarf University Chicago Press, Chicago, Illinois.
34 34
35 shrub, Vaccinium myrtillus, in relation to previous re- Wiens JJ, Ackerly, Allen AP et al. (2010). Niche conser- 35
36 production and weather. Canadian Journal of Botany vatism as an emerging principle in ecology and con- 36
37 78, 423–563. servation biology. Ecology Letters 13, 1310–24. 37
38 Stacy JE, Jorde PE, Steen H, Ims RA, Purvis A, Jakob- Wiens JJ, Graham CH (2005). Niche conservatism: Inte- 38
39 sen KS (1997). Lack of concordance between mtD- grating evolution, ecology, and conservation biology. 39
40 NA gene flow and population density fluctuations in Annual Review of Ecology Evolution and Systematics 40
41 the bank vole. Molecular Ecology 6, 751–9. 36, 519–97. 41
42 Sunyer P, Boizadera E, Muñoz A, Bonal R, Espelta JM Zwolak R, Bogdziewicz M, Wróbel A, Crone EE (2016). 42
43 (2015). The interplay among acorn abundance and Advantages of masting in European beech: timing of 43
44 rodent behaviour drives the spatial pattern of seed- granivore satiation and benefits of seed caching sup- 44
45 ling recruitment in mature Mediterranean oak forests. port the predator dispersal hypothesis. Oecologia 180, 45
46 PLoS ONE 10, e0129844. 749–58. 46
47 47
49 49
Urgoiti J, Muñoz A, Espelta JM, Bonal R (2018). Distribution and space use of seed-dispersing rodents in central
50 50
Pyrenees: implications for genetic diversity, conservation and plant recruitment. Integrative Zoology 13, 307–18.
51 51

2 2
3 3
4 4
5 5
6 Environmental variation shifts the relationship between trees 6
7 7
8 and scatterhoarders along the continuum from mutualism to 8
9
9
10
11
antagonism 10
11
12 12
13 13
14 14
15 Gina M. SAWAYA,1 Adam S. GOLDBERG,1 Michael A. STEELE2 and Harmony J. DALGLEISH1 15
16 1
16
Department of Biology, College of William and Mary, Williamsburg, Virginia, USA and 2Department of Biology, Wilkes
17 17
18 University, Wilkes-Barre, Pennsylvania, USA 18
19 19
20 20
21 21
22 22
23
Abstract 23
24 The conditional mutualism between scatterhoarders and trees varies on a continuum from mutualism to antag- 24
25 onism and can change across time and space, and among species. We examined 4 tree species (red oak [Quer- 25
26 cus rubra], white oak [Quercus alba], American chestnut [Castanea dentata] and hybrid chestnut [C. dentata 26
27 × Castanea mollissima) across 5 sites and 3 years to quantify the variability in this conditional mutualism. We 27
28 used a published model to compare the rates of seed emergence with and without burial to the probability that 28
29 seeds will be cached and left uneaten by scatterhoarders to quantify variation in the conditional mutualism that 29
30 can be explained by environmental variation among sites, years, species, and seed provenance within species. 30
31 All species tested had increased emergence when buried. However, comparing benefits of burial to the probabil- 31
32 ity of caching by scatterhoarders indicated a mutualism in red oak, while white oak was nearly always antago- 32
33 nistic. Chestnut was variable around the boundary between mutualism and antagonism, indicating a high degree 33
34 of context dependence in the relationship with scatterhoarders. We found that different seed provenances did 34
35 not vary in their potential for mutualism. Temperature did not explain microsite differences in seed emergence 35
36 in any of the species tested. In hybrid chestnut only, emergence on the surface declined with soil moisture in the 36
37 fall. By quantifying the variation in the conditional mutualism that was not caused by changes in scatterhoarder 37
38 behavior, we show that environmental conditions and seed traits are an important and underappreciated compo- 38
39 nent of the variation in the relationship between trees and scatterhoarders. 39
40 40
41 Key words: Castanea dentata, conditional mutualism, Quercus alba, Quercus rubra, temperate forest 41
42 42
43 43
44 44
45 INTRODUCTION 45
46 46
47 The relationship between scatterhoarders and hard- 47
48 Correspondence: Harmony J. Dalgleish, Department of wood trees varies along a continuum from mutualism to 48
49 Biology, College of William and Mary, 540 Landrum antagonism (Theimer 2005; Zwolak & Crone 2012). In 49
50 Drive, Williamsburg, VA 23185, USA. this interaction, scatterhoarders benefit from having a 50
51 Email: hjdalgleish@wm.edu non-perishable food source throughout periods of food 51

G. M. Sawaya et al.
1 scarcity, whereas trees potentially benefit from seed dis- population in some years followed by years of very low 1
2 persal and seedling establishment when caches are not seed production (Sork 1993). Hence, seed numbers can 2
3 recovered. However, the stability of this relationship vary considerably, causing the seed : scatterhoarder ratio 3
4 varies considerably. High relative numbers of scatter- to fluctuate substantially. The Zwolak and Crone (2012) 4
5 hoarders can lead to low seedling recruitment as a result model illustrates how the abiotic environment and the 5
6 of high rates of seed predation; however, in contrast, characteristics of the seeds also influence this interac- 6
7 scatterhoarders can also benefit trees through disper- tion. Yet, we have little information on the role of envi- 7
8 sal and subsequent abandonment of the caches when ronmental and seed characteristics in shaping this rela- 8
9 seed abundance is high. Dispersal is critical for allow- tionship. 9
10 ing seeds to escape the shadow of the parent tree, where In this study, we investigated 2 potential sourc- 10
11 density-dependent mortality is highest due to a variety es of variation in the Es/Ec ratio: (i) the provenance, 11
12 of factors (Janzen 1970; Connell 1971; Howe & Miri- or source, of the seed; and (ii) different abiotic condi- 12
13 ti 2000; Liu et al. 2012). Therefore, understanding how tions to which seeds were exposed during germination 13
14 the scatterhoarder–tree relationship shifts between a mu- and establishment. Different seed origins may have dif- 14
15 tualistic and antagonistic outcome is central to under- ferent responses to abiotic conditions because of local 15
16 standing the process of animal-mediated seed dispersal. adaptation to various conditions. Evidence of local ad- 16
17 Theimer (2005) describes scatterhoarders as “condi- aptation of oak seeds and seedlings to varying environ- 17
18 tional mutualists,” whose effect on trees can vary over mental conditions (e.g. frost) has been documented in 18
19 time and space as the ratio of seeds to scatterhoard- several sites in Europe (Deans & Harvey 1996; Jensen 19
20 ers varies. Zwolak and Crone (2012) propose a simple 2000), although a similar study in North America pro- 20
21 mathematical model to quantify the nature of this rela- duced more ambiguous results (Huang et al. 2016) 21
22 tionship between scatterhoarders and trees using only 3 22
The second source of variation in the Es/Ec ratio that
23 variables. The first, Es, is the rate of emergence of seeds 23
we investigated includes abiotic conditions such as soil
24 from the soil surface (i.e. those that are not cached by 24
moisture, leaf litter and temperature. Soil moisture in-
25 scatterhoarders). The second, Ec, is the emergence rate 25
creases seed survival and germination rates (Garcia et
26 of seeds when cached. The last variable, Pc, is the prob- 26
27 ability that cached seeds remain uneaten (i.e. the proba- 27
28 bility that seeds will be cached and left uneaten). When 28
29 the probability is higher than the ratio of seedling emer- 29
30 gence on the surface to emergence in a cache, the seed– 1 30
Pc – probability of seed
31 scatterhoarder relationship is mutualistic. If the ratio of 31

survival when cached
32 the emergence of the seeds from the surface to the emer-

Mutualism 32
33 gence of the seeds in the cache is higher than the proba- 33
34 bility that seeds will be cached and left uneaten, the re- 34
35 lationship is antagonistic (Fig. 1). 35
36 Antagonism 36
Most studies on the scatterhoarder–tree relationship
37 37
have focused on understanding factors that influence Pc,
38 38
which is directly related to the ratio of seeds to scatter- 0
39 0 1 39
hoarders. When there is a low seed : scatterhoarder ra-
40 40
tio (e.g. few seeds are produced in a year), virtually all Es/Ec – Emergence on surface/
41 41
of the seeds will be handled by a scatterhoarder, and the Emergence from cache
42 42
vast majority of seeds will be eaten and Pc will be low.
43 43
When there is a high seed to scatterhoarder ratio (e.g. in Figure 1 Graphical model of the relationship between scatter-
44 44
a year of high seed production) the scatterhoarders cache hoarders and trees from Zwolak and Crone (2012). The y-axis
45 is Pc, the proportion of seeds that survive when cached by scat- 45
more seeds than they require, which results in a high-
46 terhoarders. The x-axis is Es/Ec, the emergence rate of seeds 46
er Pc (Garcia et al. 2002; Jansen et al. 2004; Zwolak &
47 on the surface divided by the emergence rate from the cache. 47
Crone 2012; Lichti et al. 2015; Zwolak et al. 2016; but
48 When the Es/Ec ratio is high (i.e. high surface emergence com- 48
see Xiao et al. 2013). The number of seeds produced in
49 pared to emergence from the cache), the relationship leans to- 49
a population of trees is strongly influenced by “masting,”
50 wards antagonism. When Es/Ec is low, the relationship leans 50
the synchronous production of seeds by most trees in a
51 towards mutualism. 51

Variation in conditional mutualism
1 al. 2002) by preventing desiccation (Lopez-Barrera MATERIALS AND METHODS 1

2 & Gonzalez-Espinosa 2001), improving frost resis- 2
3 tance (Esteso-Martinez & Gil-Pelegrin 2004) and po- Study species 3
4 tentially increasing “germinative energy” (number of 4
Study species include red oak (Q. rubra), white oak,
5 days required to reach 50% germination), although this 5
(Q. alba), American chestnut (C. dentata) and a hybrid
6 latter effect is still unclear (Huerta-Paniagua & Rodri- 6
American/Chinese chestnut (C. dentata × C. mollissi-
7 guez-Trejo 2010). Indeed, leaf litter and burial have 7
ma). The oak species occur throughout the eastern US
8 been shown to induce germination in acorns of oak 8
deciduous forests with ranges that extend as far south as
9 (Houle 2002) and in pine seeds (Briggs et al. 2009). We 9
Texas and as far north as Canada (Bourdeau 1954). The
10 therefore hypothesize that warmer temperatures in buri- 10
primary seed disperser for these tree species is the east-
11 al sites may explain a key benefit of caching. 11
12 ern gray squirrel (Sciurus carolinensis), with eastern 12
The goal of this study was to quantify the variation chipmunk (Tamias striatus) and mice (Peromyscus spp.)
13 in the hardwood tree seed: the scatterhoarder relation- 13
14 also dispersing seeds. 14
ship using the model presented by Zwolak and Crone
15 Red oak acorns require 2 growing seasons to ma- 15
(2012). Specifically, we focus on 2 sources of variation
16 ture (Straub 2016) and a dormancy period between seed 16
in the Es/Ec ratio: seed provenance and abiotic condi-
17 fall and germination the following spring (Licthi et al. 17
tions, both in and outside the cache. To test these, we es-
18 2014). Red oak acorns in eastern deciduous forests typ- 18
tablished seed plots with predator exclusion cages. We
19 ically have high concentrations of tannins (Smallwood 19
established plots in 5 different sites across the Central
20 et al. 2001): chemical deterrents that can have negative 20
Hardwoods Region to expose the seeds to varying en-
21 physiological effects on granivores (Chung-MacCoubrey 21
vironmental conditions, and at 1 site (Virginia) we also
22 1997). In addition, red oak acorns have higher fat lev- 22
varied seed source to test its effect on seed survival.
23 els, which serves as the primary source of energy in the 23
We use seeds from red oak (Quercus rubra), white oak
24 acorn. Because red oak seeds have a long dormancy pe- 24
(Quercus alba), American chestnut (Castanea dentata),
25 riod, granivores prefer to cache red oak seeds because of 25
and a hybrid American-Chinese chestnut (C. dentata × C.
26 their low perishability (Hadj-Chikh et al. 1996; Small- 26
mollissima).
27 wood et al. 2001). 27
We tested 3 predictions. First, we predicted that Es/Ec
28 Acorns of white oak have lower fat and tannin con- 28
values will correspond to seed dormancy requirements
29 tent, and they germinate in the autumn shortly after seed 29
because the microsite characteristics within a cache will
30 fall (Hadj-Chikh et al. 1996; Smallwood et al. 2001). 30
prevent desiccation and enhance survival during dor-
31 Squirrels tend to eat white oak acorns and immediate- 31
mancy. Acorns of red oak, for example, should have
32 ly cache red oaks because of these differences in perish- 32
lower Es/Ec ratios (higher potential for mutualism) be-
33 ability; when they do cache white oak acorns they often 33
cause of their longer dormancy period; white oaks
34 excise the embryos, preventing germination (Hadj-Chikh 34
acorns will have a higher Es/Ec ratio (higher potential
35 et al. 1996). 35
for antagonism) because they show no dormancy; and
36 The historic range of the American chestnut ex- 36
chestnuts are somewhere in between due to their inter-
37 tends from Maine to Georgia (Paillet 2002). Between 37
mediate dormancy period. Second, we predict that abi-
38 1900 and 1925, chestnut populations plummeted due to 38
otic conditions will have a stronger effect on Es/Ec than
39 the introduced fungal pathogen, chestnut blight (Cry- 39
seed source. We predict that we will see more variation
40 phonectria parasitica). The social and economic impor- 40
in Es/Ec across sites and years (differing abiotic con-
41 tance of the tree prompted breeding efforts to create a 41
ditions), than when comparing seed sources within the
42 blight resistant tree (Jacobs et al. 2013). When present- 42
same site and year. Finally, we predict that temperature
43 ed with chestnuts, and red oak and white oak acorns, 43
will explain much of the variation in the Es/Ec ratio. We
44 squirrels prefer to eat white oak acorns; likewise, in the 44
expect sites with lower temperatures to have lower Es/
45 presence of only chestnuts and white oak acorns, white 45
Ec ratios (more potential for mutualism) because these
46 oak acorns are selectively eaten (Lichti et al. 2014; Bly- 46
seeds will benefit more from the extra protection and
47 the et al. 2015). 47
warmth within a cache. We predict that seeds planted in
48 48
northern sites will have a more mutualistic relationship Study sites
49 49
(lower Es/Ec) with scatterhoarders than those planted in
50 We conducted experiments over 3 years in 5 hard- 50
the south.
51 51

G. M. Sawaya et al.
1 Table 1 Summary of the provenance study design in Virginia wood forests: Atkinson Grove, Maine; Harvard Forest, 1
2
Year Species Provenances planted in Virginia Massachusetts; Martell Forest, Indiana; Dorrance Town- 2
3 ship, Pennsylvania; and the College Woods, Virginia 3
4 2012 Red Oak Virginia, Pennsylvania, Maine
(Tables 1 and 2). Maine has warm summers, with tem- 4
5 2013 Illinois, Pennsylvania peratures reaching 26 °C, followed by 4 to 8 months of 5
6 2015 Virginia temperatures as low as −12 °C. Similarly, Massachu- 6
7 2012 White Oak Virginia setts has an equally short summer, capped at 26 °C, with 7
8
2013 Illinois, Pennsylvania
equally cold winters: −11 °C in Barre. The Martell For- 8
9 est site near West Lafayette, Indiana has a winter low of 9
2015 Virginia
10 −8 °C and a summer high of 29 °C, but with moderate 10
11 2012 Hybrid Maine temperatures from April to October. Northeastern Penn- 11
12 2013 Chestnut Massachusetts, Maine sylvania has an even milder winter, with a low of −6 °C, 12
13 2015 (B3F2) Maryland followed by 5 to 7 months of summer, peaking at 30 °C. 13
14 The College of William and Mary’s College Woods of 14
2012 American Indiana, Maine
15 Williamsburg, Virginia experiences minimal winter con- 15
16 2013 Chestnut Indiana, Virginia ditions, with a winter low of −1 °C and a summer high 16
17 of 30 °C. 17
18 18
19 19
20 20
Table 2 Summary of the cross-site study design
21 21
22 Site Coordinates Mean Annual Mean Annual Dominant canopy Years as Species planted at 22
23 Temperature Precipitation study site this site (and their 23
24 provenance) 24
25 Atkinson 45°9′N, 69°3′W 4.6°C 9.4 cm Castanea dentata, 2013 Red Oak (VA), 25
26 Grove, ME Quercus rubra, White Oak (VA), B3F2 26
27 Abies balsamea (ME) 27
28 College 37°16′N, 76°42′W 14.8°C 10.4 cm Quercus alba, 2013, 2014, Red Oak (VA, IL), White 28
29 Woods, VA Liriodendron tulipifera, 2015 Oak (VA, IL), B3F2 29
30 Pinus taeda, (ME, MA, MD) 30
31 Fagus grandifolia, 31
32 Quercus falcata, 32
33 Acer rubrum, 33
34 Cornus florida 34
35 35
Harvard 42°31′N, 72°11′W 7.2°C 9.7 cm Acer rubrum, 2013, 2015 Red Oak (VA),
36 36
Forest, MA Quercus rubra, White Oak (VA),
37 37
Pinus strobus, Tsuga B3F2 (ME, MD)
38 38
sieboldii,
39 39
Picea abies,
40 40
Pinus resinosa
41 41
42 Martell 40°26′N, 86°54′W 11.1°C 7.9 cm Quercus alba, 2013 Red Oak (VA), 42
43 Forest, IN Carya glabra, White Oak (VA), 43
44 Quercus velutina, B3F2 (ME) 44
Quercus rubra,
45 45
Acer saccharum
46 46
47 Dorrance 41°13′N, 75°98′W 9.6°C 97.2 cm Quercus rubra, 2014, 2015 Red Oak (VA, IL), White 47
48 Township, PA Acer rubrum, Oak (VA, IL), 48
49 Quercus alba B3F2 (MA, MD) 49
50 50
51 IL, Illinois; ME, Maine; PA, Pennsylvania; VA, Virginia. 51

1 Seed collection and field methods The cages were secured with landscaping staples in the 1
2 apron and on the corners of the cages to minimize pil- 2
3 We hand-collected seeds from a variety of loca- ferage by small mammals (Barnett 1977). At each site, 3
4 tions in all the years of study except for 2014 when a each year, we relied on 15 replicate plots (1350 seeds 4
5 mast failure in Virginia prompted us to purchase red per plot). 5
6 oak seeds from Illinois. Upon collection, each seed was 6
visually inspected and float-tested to eliminate seeds To investigate the effect of provenance on the Es/Ec
7 ratio, we tested seeds from several sources in Virgin- 7
8 with insect infestations. Virginia and Pennsylvania seed 8
sources were collected from the same sources in differ- ia in 2 different years (Table 1). Each provenance was
9 replicated in 15 plots for a total of 450 seeds per prove- 9
10 ent years. Hybrid chestnuts were donated from 3 differ- 10
ent provenances (Table 1). nance per year. To investigate the effects of cache sites
11 (i.e. abiotic conditions) on the Es/Ec ratio, we planted 11
12 Each plot contained a 9 × 10 array in which each col- a single provenance of red oak, white oak and chestnut 12
13 umn of the array contained the seeds of a single spe- across 2–4 different sites over 3 years (Table 2). Each 13
14 cies, and included a total of 3 rows of each species per species was replicated in 15 plots/site for a total of 450 14
15 exclosure (30 seeds/species; 90 seeds/cage). All rows seeds per species per site per year. 15
16 in the plot alternated between cached and non-cached 16
17 seeds, for a total of 10 rows. Buried seeds were cached Analysis 17
18 so that the bottom of the seed was 2 cm below the soil 18
We found high rates of pilferage at some of our cag-
19 surface. Non-cached seeds remained on the surface of 19
es and these cages were removed from the analysis. We
20 the soil. We pushed the leaf litter aside at each seed site 20
calculated the percent emergence from the surface and
21 and then replaced the leaf litter immediately after plac- 21
the percent emergence from the cache for each plot.
22 ing the seed. Along with the seeds, each plot had 2 iBut- 22
Most plots had an individual value for the Es/Ec ratio
23 ton temperature loggers: 1 placed on the surface and 1 23
and we used these plot-level values to test for differenc-
24 buried like the seeds, recording temperatures every 4 h. 24
es between species and provenances using ANOVA (R
25 Plots were then covered with cages made of hardware 25
core development Team). However, some plots at some
26 cloth that measured 1.2 m × 1.2 m × 17.8 cm and were 26
sites did not have sufficient seeds remaining on the sur-
27 attached to wire aprons measuring 15.2 cm in width. 27
28 28
29 a 29
1.5
30 30
●
31 31
32 32
b
33 ● 33
34 34
35 ●
● 35
36 c 36
1.0
37 ●● 37
●
38 38
Es/Ec
39 39
40 ●●●
40
●
41 ● 41
42 ●
42
0.5
43 43
●
44 ●
● ● 44
●●
45 Figure 2 Comparison of the Es/Ec ratio in
●
●
45
● ●
46 3 species. All 3 species were significant- ● ● ● ● 46
● ●
47 ly different from each other with red oak ●
● ● ●
47
● ● ●
48 having the lowest Es/Ec ratio (mutualistic), 48
0.0
49 white oak having the highest (antagonistic)

●
●●
●● ●●
●
● ●
●●●●
●
●●
●●●
●●
●
●● ●
●
●●●●●
● ●●
●
49
50 and hybrid American chestnut intermedi- 50
AC RO WO
51 ate. 51
tree species

G. M. Sawaya et al.
1 face due to pilferage. We therefore took an average of than white oak (F2,83 = 8.435, P = 0.0005, Fig. 2). Amer- 1
2 the emergence from the surface and the emergence from ican chestnuts were intermediate. 2
3 the cache over all plots within a site and used these av- 3
How does seed provenance affect the Es/Ec
4 erages to calculate the Es/Ec ratio for each species/site/ 4
5 year combination. ratio? 5
6 We quantified the potential for a mutualistic versus 6
Multiple provenances of all 3 species were test-
7 antagonistic relationship between each of our 3 tree spe- 7
ed within the same site within a single year. Compar-
8 cies and scatterhoarders using the model of Zwolak and 8
ing provenances of the same species within a single
9 Crone (2012). To do this, we needed estimates of Pc, 9
year showed no differences among provenances of the
10 the proportion of cached seeds that remain uneaten, for 10
same species (P > 0.10, Fig. 3). However, when com-
11 all 3 of our species. We took these values from Lichti et 11
paring the same provenance across 2 years, differenc-
12 al. (2014), who tracked the fate of marked seeds for red 12
es emerged. For white oak, American chestnut and B3F2
13 oak, white oak and American chestnut across 2 years 13
hybrid chestnut provenances, the Es/Ec ratio differed
14 and 3 sites. In addition, each seed species was paired 14
significantly between years (Fig. 4, P < 0.02). In con-
15 with both of the other 2 species in different trials. Thus, 15
trast, neither red oak provenance differed between years
16 we had 8 estimates of Pc for each seed species (see sur- 16
(Figs 3 and 5, P > 0.5).
17 vival/removal in Fig. 2 in Lichti et al. 2014). We calcu- 17
18 lated the mean, minimum and maximum Pc values re- How do abiotic conditions affect the Es/Ec ratio? 18
19 ported by Lichti et al. (2014) across both years and all 19
20
Across sites and years 20
sites. We plotted all Es/Ec ratios against the Pc values
21 from Lichti et al. (2014). We analyzed the relationship We found that red oak had the least amount of varia- 21
22 between Es/Ec values and temperature as well as the re- tion in the Es/Ec ratio across all sites and years. Red oak 22
23 lationship between surface emergence rates and tem- also fell exclusively within the mutualistic range across 23
24 perature using Pearson’s correlation. all sites and years given average survival when cached 24
25 (Fig. 5a). The 2 site-year combinations with the highest 25
26 Es/Ec ratio for red oak could be within the antagonistic 26
27
RESULTS range during years with low survival when cached (Pc, 27
28
Differences among species Fig. 5a). 28
29 White oak and hybrid chestnut were much more vari- 29
30 We tested for species differences using the data able than red oak, and both these species had some 30
31 from Virginia only as this site had all 3 tree species years that fell within the mutualistic range and others 31
32 planted in the same location in each of the 3 years. that fell within the antagonistic range (Fig. 5b,c). For 32
33 All 3 species had higher emergence rates when cached hybrid chestnut, 5 of the 8 points fell within the mutual- 33
34 than when left on the surface. Red oak acorns had the istic range as the 95% confidence interval for the proba- 34
35 lowest emergence on the surface (0.008 ± 0.03, mean bility of survival when cached on these points all over- 35
36 ± SD) with white oak acorns the highest (0.10 ± 0.14, lap the 1:1 line (Fig. 5b). The 3 points with the highest 36
37 mean ± SD) and hybrid chestnuts intermediate (0.06 ± Es/Ec ratios would be near the mutualistic range only in 37
38 0.07, mean ± SD). In contrast, hybrid chestnuts had the years of exceptionally high cache survival. 38
39 highest emergence from a cache (0.40 ± 0.25, mean ± 39
For white oak, only 3 of the 8 points fell within the
40 SD) with red oak and white oak only slightly less when 40
mutualistic range, with at least 1 of these points falling
41 cached (0.31 ± 0.25, 0.30 ± 0.16, respectively, mean ± 41
in the antagonistic range in a year of lower than aver-
42 SD). 42
age seed survival when cached (Pc, Fig. 5c). The other
43 Despite the apparent benefit that all seeds experi- 43
5 points were not in the mutualistic range given the 95%
44 ence when cached (i.e. all seeds increased emergence 44
confidence intervals of Pc values. Indeed, even the high-
45 when cached), comparisons between the Es/Ec ratio and 45
est observed survival within a cache (Pc), would place
46 Pc values illustrate that in some situations it might not 46
only 2 of these 3 points within the mutualistic range (Fig.
47 be enough to make the interaction mutualistic. The Es/ 47
5c).
48 Ec ratio was significantly lower for red oak than white 48
49 Effects of temperature and soil moisture 49
oak, supporting our hypothesis that red oak acorns have
50 a higher propensity for mutualism with scatterhoarders Across all sites, caching increased the average tem- 50
51 51

1 a b 1
2 2
1.5
1.5
● 2012
2013
3 3
4 4
5 a a 5
6 6
b b
1.0
1.0
7 7
8 8
Es/Ec
Es/Ec
a a
9 9
10 ● 10
11 11
0.5
0.5
12 12
13 13
●
14 ●
14
●
●
15 15
0.0
0.0
●
●●
●●
● ●
●
●●
●
●●●
●
16 16
17 IN 2012 ME 2012 IN 2013 VA 2013 MA 2013 ME 2013
17
18 18
19 19
20 c d 20
1.5
1.5
2012
21 21
●
2013
22 a a 22
23 23
24 24
25 25
1.0
1.0
26 26
Es/Ec
Es/Ec
27 27
28 28
29 29
a a a a a
0.5
0.5
30 30
31 31
32 ● ● 32
● ●
33 33
34 34
0.0
0.0
●
●
●
●●
●
● ●●
●●
●●
●● ●
●●
●●
●
●
●●●
35 35
36 VA 2012 PA 2012 ME 2012 IL 2013 PA 2013 IL 2013 PA 2013 36
37 Figure 3 The Es/Ec ratio did not differ among provenances planted in the same location within the same year for any of the species 37
38 tested. American chestnut had significant differences between years, while red oak was more consistent. (a) American chestnut, (b) 38
39 hybrid chestnut, (c) red oak and (d) white oak. 39
40 40
41 41
42 42
43 perature experienced by seeds during cold months. thermal environment experienced by the seeds, we 43
44 From November to February, the average temperatures found no correlation between average temperature and 44
45 in a cache were up to 2 °C warmer than temperatures on the Es/Ec ratio at the microclimate scale (temperatures 45
46 the soil surface at all 5 sites (Appendix I). In April and measured at each individual plot) in red oak (P = 0.69), 46
47 May, surface temperatures were on average 1 °C warm- white oak (P = 0.54), or American chestnut (P = 0.70) 47
48 er than the cache temperature. On average, there was (Appendix I, Fig. 3). Similarly, we found no relation- 48
49 no difference between cache and surface temperature in ship between average or minimum temperatures and the 49
50 March. Es/Ec ratio at the site level. However, warmer sites had 50
51 Despite these consistent effects of caching on the higher frequency of surface emergence in white oak: 51

G. M. Sawaya et al.
1 1
1.5
2 ● 2012 ● 2
3 2013 3
2015
4 4
5 5
6 6
7 7
8 8
1.0
9 9
10 ● 10
11 11
Es/Ec
12 ●
● 12
13 ●
●
● 13
14 14
15 0.5
15
16 ●
16
17 ● ●● 17
●
18 ●
● ●
18
19 ● ●
●
19
●
20 Figure 4 The Es/Ec ratio was significantly ● ●
20
●●
21 different between years for the same prov- ●●
21
22 enance with the exception of red oak. IN, 22
0.0
●
●
●●
●● ●● ●●
●
●●
●●
● ●
●
●●
●
●● ●●
●
23 Indiana; MD, Maryland; ME, Maine; VA, 23

24 Virginia. IN American chestnut ME hybrid chestnut PA red oak VA red oak VA white oak
24
25 25
26 26
a b c
27 27
1.0
1.0
1.0
28 2012
28
●
2013
29 29
2015
0.8
0.8
0.8
VA
30 ME
IN
30
31 31
0.6
0.6
0.6
MA
PA
Pc
32 32
0.4
0.4
0.4
33 33
34 ●● ● 34
0.2
0.2
0.2
35 ● ● ● ●
● ● ● 35
36 36
0.0
0.0
0.0
37 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 37
38 Figure 5 Sites and years plotted on the continuum from mutualism to antagonism: (a) red oak, (b), hybrid chestnut and (c) white 38
39 oak. The Pc values were taken from Lichti et al. (2014). The upper dotted line is the maximum Pc value observed by Lichti et al. 39
40 (2014). The lower dotted line is the minimum Pc value observed. The black lines encompass the 95% confidence interval for Pc 40
41 values for the species. 41
42 42
43 43
44 44
57% emergence from the surface in Virginia, 36% in ginia, 64% in Massachusetts and 36% in Maine.
45 45
Massachusetts and 15% in Maine. We did not see this Soil moisture measured in the fall was not correlated
46 46
pattern in red oak, which had very low frequency of sur- with the Es/Ec ratio for any species (P > 0.3). In hybrid
47 47
face emergence in Virginia (5%) and Maine (8%), but American chestnut, emergence from the soil declined
48 48
higher surface emergence in Massachusetts (40%). In with increasing fall soil moisture (ρ = −0.30, P = 0.01).
49 49
American chestnut, we saw a higher, and more similar, A similar pattern was observed in the Es/Ec ratio for hy-
50 50
frequency of surface emergence at all sites: 55% in Vir- brid American chestnut, although marginally significant
51 51

1 (ρ = −0.23, P = 0.07). We found no relationship between rate when on the surface versus buried. The Zwolak 1
2 hybrid chestnut emergence when buried and soil mois- and Crone (2012) model provides the added insight that 2
3 ture. In addition, there was no relationship between soil American chestnut, rather than benefiting the most from 3
4 moisture and surface or buried emergence in either red the interaction as suggested by the differences in emer- 4
5 oak or white oak. gence rates, actually has the potential for the environ- 5
6 ment to easily shift the interaction outcome because its 6
7
DISCUSSION Es/Ec ratios are so near the border between antagonism 7
8 and mutualism. 8
9 Although seed burial increased emergence in all 3 Although all species that we studied exhibited varia- 9
10 species, the Zwolak and Crone (2012) model, combined tion in their Es/Ec ratios, none of that variation was at- 10
11 with multiple environmental measures across sites and tributable to differences in seed provenance. When we 11
12 years, allowed us to tease apart significant differenc- tested different provenances in the same location in the 12
13 es among our 3 species in the potential for the benefit same year, we found no evidence for differences among 13
14 of caching to create a mutualistic relationship between seed provenances in the Es/Ec ratio, suggesting that 14
15 trees and scatterhoarders. Es/Ec ratios appeared relat- seeds from different provenances have similar potential 15
16 ed to dormancy requirements, which supports our first for mutualism and antagonism under the same environ- 16
17 prediction. By comparing the Es/Ec ratio to the Pc val- mental conditions. These results are similar to those of 17
18 ues, we found that buried red oak acorns nearly always Huang et al. (2016), who found that the emergence rates 18
19 resulted in mutualism, whereas acorns of white oak of white oak acorns were more strongly explained by 19
20 consistently fell in the antagonistic range, across near- planting site than seed provenance. They did, however, 20
21 ly all sites and years. Hybrid chestnuts were intermedi- find significant differences among provenances of white 21
22 ate, suggesting that environmental context could have a oak in seedling height and diameter (Huang et al. 2016). 22
23 much larger role in shaping the outcome of the interac- 23
To date, conceptual models and explanations of the
24 tion. Our second predication was also supported: abiot- 24
conditional nature of the seed : scatterhoarder relation-
25 ic conditions had a stronger effect on Es/Ec ratios than 25
ship have focused on the impact of variable seed pro-
26 seed provenance. However, our final predication was 26
duction (i.e. mast years) and the seed : scatterhoarder
27 not well supported: we found that temperature explained 27
ratio (Theimer 2005; Lichti et al. 2015; Xiao & Krebs
28 little of the variation we observed in the Es/Ec ratio. Al- 28
2015). Theimer (2005) notes that in addition to the per
29 though white oak experienced the more antagonistic re- 29
capita seed abundance, environmental conditions (i.e.
30 lationship among the 3 tree species studied, we found 30
microsites) are not fixed and that the variation across
31 that it benefited the most from the warmer tempera- 31
space and years could also shift the relationship. Sim-
32 tures at the most southern site, Virginia. White oak ex- 32
ilarly, Jorge and Howe (2009) found that forest frag-
33 perienced a steady increase in the frequency of surface 33
mentation altered seed removal and dispersal. Our study
34 emergence from north to south. No such trend was ob- 34
is among the first to quantify the role of environmen-
35 served for either red oak or chestnut. 35
tal variation by focusing on the Es/Ec ratio. While the
36 Our finding that burial increased seed emergence in 36
seed : scatterhoarder ratio will influence the amount of
37 all 3 species is consistent with literature on many oth- 37
variation observed in the probability that seeds will be
38 er species and is often taken as evidence that scatter- 38
cached and left uneaten (Pc), how a species responds to
39 hoarders are mutualists (Theimer 2005 and references 39
differing conditions of environmental microsites across
40 therein). However, the presence of a measurable benefit 40
space and time also influences the amount of variation
41 alone cannot determine the net effect of the interaction. 41
observed in the Es/Ec ratio. Zwolak and Cone (2012)
42 In a “conditional mutualism” such as scatterhoarding, 42
present an analysis of 15 studies from the literature that
43 the costs of seed predation as well as the relative benefit 43
report values of surface (Es) or cache (Ec) emergence.
44 of caching compared to the benefit obtained in the ab- 44
They found that most species tended toward mutualism,
45 sence of scatterhoarders is, indeed, key (Theimer 2005; 45
but most studies did not quantify environmental impact
46 Zwolak & Crone 2012; Xiao & Krebs 2015). However, 46
on the Es/Ec ratio. In our study, both American chestnut
47 simply comparing the increase in emergence as a result 47
and white oak had wide-ranging Es/Ec values. Indeed,
48 of seed burial would lead us to conclude that American 48
the amount of variation among Es/Ec ratios for these
49 chestnut experienced the greatest benefit from scatter- 49
species was larger than the 95% confidence intervals of
50 hoarding, as it had the greatest difference in emergence 50
Pc reported by Lichti et al. (2014). Although masting al-
51 51

G. M. Sawaya et al.
1 ters the seed : scatterhoarder ratio and can strongly in- a significant and underappreciated role in shifting the 1
2 crease the probability of seed survival within caches, seed : scatterhoarder relationship along the mutualism– 2
3 our data illustrate that the traits of the seed, combined antagonism continuum. Understanding the variation in 3
4 with abiotic conditions, can be equally important in de- both the seed : scatterhoarder ratio and the Es/Ec ratio is 4
5 termining where species fall on the continuum from mu- necessary to quantify the mutualism between seeds and 5
6 tualism to antagonism. scatterhoarders. 6
7 Our data support that caching can provide an insulat- 7
8 8
ing effect; buried seeds were 2 °C warmer during cold ACKNOWLEDGMENTS
9 months and 1–2 °C cooler in the spring. The cooler tem- 9
10 peratures in spring may prevent seeds from germinating This work was supported by a Charles Center Honors 10
11 too early or succumbing to late spring frosts. Despite Fellowship to ASG and Ferguson Fund Awards for Un- 11
12 these measurable effects of caching on temperature, this dergraduate Research to GMS and ASG. MAS recog- 12
13 variable was not significantly related to the Es/Ec ratio nizes the support of a Bullard Fellowship from Harvard 13
14 or to emergence rates of seeds placed on the soil surface Forest, Harvard University, current support from the U.S. 14
15 (i.e. seeds ignored by scatterhoarders). Many studies in- National Science Foundation (DEB 15556707), and the 15
16 dicate that burial and leaf litter can increase seed survi- H. Fenner Research Endowment of Wilkes Universi- 16
17 vorship due to an insulation effect of the soil and leaf ty. We thank the Indiana, Maine, Maryland and Virginia 17
18 (Garcia et al. 2002; Briggs & Vander Wall 2004). Garcia Chapters of the American Chestnut Foundation for sup- 18
19 et al. (2002) found that buried red oak acorns and those plying seed and Colonial Williamsburg for permission 19
20 covered by litter had greater moisture content and were to collect acorns from their oaks. 20
21 significantly less likely to be eaten by deer (Odocoileus 21
22 virginianus), suggesting that moisture and protection 22
23
REFERENCES 23
could be the driving factors in the benefits of burial in-
24 stead of temperature. Barnett R (1977). The effect of burial by squirrels on 24
25 germination and survival of oak and hickory nuts. 25
Our study focuses on what Xiao and Krebs (2015)
26 The American Midland Naturalist 98, 319–30. 26
call the “burial effect” of scatterhoarders. They argue
27 27
that the “dispersal effect” may be more important for Blythe RM, Lichti NI, Smyser TJ, Swihart RK (2015).
28 28
understanding the benefits of scatterhoarders. To mea- Selection, caching, and consumption of hardwood
29 29
sure this, one compares the emergence of buried and seeds by forest rodents: Implications for restoration
30 30
surface seeds underneath the parent tree and away from of American chestnut. Restoration Ecology 23, 473–
31 31
the parent tree. Because we did not take into account 81.
32 32
whether seeds in our experiment were placed beneath
33
trees of the same or a different species, we cannot ful- Blythe RM, Smyser TJ, Swihart RK (2015). Tree squir- 33
34 rel seed predation patterns may influence American 34
ly tease apart a burial versal dispersal effect in our data.
35 chestnut restoration success. New Forests 46, 593– 35
In addition, we are only focusing on 1 stage of the plant
36 600. 36
life cycle (Zwolak & Crone 2012; Xiao & Krebs 2015).
37 37
Our interpretation of the potential for mutualism versus Bourdeau P (1954). Oak seedling ecology determining
38 38
antagonism also could vary if seedling growth and sur- segregation of species in Piedmont oak-hickory for-
39 39
vival were enhanced with burial more than is measured ests. Ecological Monographs 24, 297–320.
40 40
by simple seedling emergence. By using artificial caches
41 Briggs JS, Vander Wall SB (2004). Substrate type af- 41
and exclusion cages, we removed other potential effects
42 fects caching and pilferage of pine seeds by chip- 42
of scatterhoarder behavior, such as escape from dis-
43 munks. Behavioral Ecology 15, 666–72. 43
tance-dependent or density-dependent mortality (Hirsch
44 Briggs JS, Vander Wall SB, Jenkins SH (2009). For- 44
et al. 2012) or providing directed dispersal into particu-
45 45
larly suitable microsites (Yi et al. 2013). est rodents provide directed dispersal of Jeffrey pine
46 46
Although the ratio of seeds to scatterhoarders is im- seeds. Ecology 90, 675–87.
47 47
48 portant for quantifying the mutualism between seeds Chung-MacCoubrey AAL, Hagerman AE, Kirkpatrick 48
49 and scatterhoarders, our data indicate the importance of RL (1997). Effects of tannins on digestion and detox- 49
50 variable Es/Ec ratios as a factor to quantify the continu- ification activity in gray squirrels (Sciurus carolinen- 50
51 um. Our results illustrate that the abiotic conditions play sis). Physiological Zoology 70, 270–77. 51

1 Connell JH (1971). On the role of natural enemies in Jensen JS (2000). Provenance variation in phenotypic 1
2 preventing competitive exclusion in some marine traits in Quercus robur and Quercus petraea in Dan- 2
3 mammals and in rain forest trees. In: Den Boer PJ, ish provenance trials. Scandinavian Journal of Forest 3
4 Gradwell GR eds. Dynamics of Populations, Centre Research 15, 297–308. 4
5 for Agricultural Publishing and Documentation, Wa- 5
Jorge MLSP, Howe HF (2009). Can forest fragmenta-
6 geningen, The Netherlands. pp. 298–310. 6
7
tion disrupt a conditional mutualism? A case from 7
Deans J (1996). Frost hardiness of 16 European prove-
8 central Amazon. Oecologia 161, 709–18. 8
nances of sessile oak growing in Scotland. Forestry
9 69, 5–12. Lichti NI, Steele MA, Zhang H, Swihart RK (2014). 9
10
Esteso-Martinez J, Gil-Pelegrin E (2004). Frost resis- Mast species composition alters seed fate in North 10
11 tance of seeds in Mediterranean oaks and the role of American rodent-dispersed hardwoods. Ecology 95, 11
12 litter in the thermal protection of acorns. Annals of 1746–58. 12
13 Forest Science 61, 81–6. 13
Lichti NI, Steele MA, Swihart RK (2015). Seed fate and
14 14
García D, Bañuelos MJ, Houle G (2002). Differen- decision-making processes in scatter-hoarding ro-
15 15
tial effects of acorn burial and litter cover on Quer- dents. Biological Reviews 92, 474–504.
16 16
cus rubra recruitment at the limit of its range in east- Liu X, Liang M, Etienne RS, Wang Y, Staehelin C, Yu
17 17
ern North America. Canadian Journal of Botany 80,
18 S (2012). Experimental evidence for a phylogenetic 18
1115–20.
19 Janzen-Connell effect in a subtropical forest. Ecology 19
20 Hadj-Chikh L, Steele M, Smallwood P (1996). Cach- Letters 15, 111–8. 20
21 ing decisions by grey squirrels: A test of the handling 21
time and perishability hypothesis. Animal Behavior Lopez-Barrera F, Gonzalez-Espinosea M (2001). In-
22 22
52, 941–8. fluence of litter on emergence and early growth of
23 23
Hirsch BT, Kays R, Pereira VE, Jansen PA (2012). Di- Quercu rugosa: A laboratory study. New Forests 21,
24 24
rected seed dispersal towards areas with low conspe- 59–70.
25 25
26 cific tree density by a scatter-hoarding rodent. Ecolo- Paillet FL (2002). Chestnut: History and ecology of a 26
27 gy Letters 15, 1423–9. transformed species source. Journal of Biogeography 27
28 Howe HF, Miriti MN (2000). No question: Seed disper- 29, 1517–30. 28
29 sal matters. Trends in Ecology & Evolution 15, 434– R Core Team (2014). R: A language and environment 29
30 6. for statistical computing. R Foundation for Statistical 30
31 Huang Y, Zhang H, Rogers S, Coggeshall M, Woeste K Computing, Vienna, Austria. Available from URL: 31
32 (2016). White oak growth after 23 years in a three- 32
http://www.R-project.org/.
33 site provenance/progeny trial on a latitudinal gradient 33
34 in Indiana. Forest Science 62, 99–106. Smallwood PD, Steele MA, Faeth SH (2001). The ulti- 34
35 mate basis of the caching preferences of rodents, and 35
Huerta-Paniagua R, Rodríguez-Trejo DA (2011). Efecto
36 the oak-dispersal syndrome: Tannins, insects, and 36
del tamano de semilla y la temperatura en la germina-
37 cion del Quercus rugosa Née. Revista Chapingo Serie seed germination. American Zoologist 41, 840–51. 37
38 Ciencias Forestales Y Del Ambiente 17, 179–87. Sork VL (1993). Evolutionary ecology of mast-seeding 38
39 in temperate and tropical oaks (Quercus spp.). Vege- 39
Jacobs DF, Dalgleish HJ, Nelson CD (2013). A concep-
40 tatio 107/108, 133–47. 40
tual framework for restoration of threatened plants:
41 41
The effective model of American chestnut (Castanea Straub JN, Kaminski RM, Leach AG, Ezell AW,
42 42
dentata) reintroduction. New Phytologist 197, 378– Leininger T (2016). Acorn yield and masting traits of
43 43
93. red oaks in the lower mississippi river alluvial valley.
44 44
Janzen DH (2009). The American Society of Naturalists Forest Science 62, 18–27.
45 45
Herbivores and the number of tree species in tropical
46 Theimer TC (2005). Rodent scatterhoarders as condi- 46
forests. American Naturalist 104, 501–28.
47 tional mutualists. In: Lambert JE, Hulme PE, Vander 47
48 Jansen PA, Bongers F, Hemerik L (2004). Seed mass 48
Wall SB, eds. Seed Fate: Predation, Dispersal, and
49 and mast seeding enhance dispersal by a neotropical 49
Seedling Establishment. CABI Publishing, Oxford-
50 scatter-hoarding rodent. Ecological Monographs 74, 50
569–89. shire, UK and Cambridge, MA, pp. 284–95.
51 51

G. M. Sawaya et al.
1 Xiao Z, Zhang Z, Krebs CJ (2013). Long-term seed sur- of plant–granivore interactions. Oikos 121, 20–7.
1
2 vival and dispersal dynamics in a rodent-dispersed 2
Zwolak R, Bogdziewicz M, Wrobel A, Crone EE (2016).
3 tree: Testing the predator satiation hypothesis and 3
Advantages in European beech: Timing of granivore
4 the predator dispersal hypothesis. Journal of Ecology 4
satiation and benefits of seed caching support the
5 101, 1256–64. 5
predator dispersal hypothesis. Oecologia 180, 749–
6 Xiao Z, Krebs CJ (2015). Modeling the costs and ben- 58. 6
7 efits of seed scatterhoarding to plants. Ecosphere 6, 7
8 1–8. 8
9
SUPPLEMENTARY MATERIALS 9
Yi X, Liu G, Steele MA, Shen Z, Liu C (2013). Directed
10 Additional supporting information may be found in 10
dispersal by a scatter-hoarding rodent: The effects of
11 the online version of this article. 11
soil water content. Animal Behaviour 86, 851–7.
12 Appendix I 12
13 Zwolak R, Crone EE (2012). Quantifying the outcome 13
14 14
15 15
17 Sawaya GM, Goldberg AS, Steele MA, Dalgleish HJ (2018). Environmental variation shifts the relationship be- 17
18 tween trees and scatterhoarders along the continuum from mutualism to antagonism. Integrative Zoology 13, 18
19 319–30. 19
20 20
21 21
22 22
23 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Effects of rodent abundance and richness on cache pilfering 6
7
8 8
9 9
10 10
11 Jacob W. DITTEL1 and Stephen B. VANDER WALL2 11
12 1
12
13 Department of Fisheries and Wildlife, Oregon State University, Corvallis, Oregon, USA and 2Department of Biology and the 13
14 Program in Ecology, Evolution, and Conservation Biology, University of Nevada, Reno, Nevada, USA 14
15 15
16 16
17 17
18 18
19 Abstract 19
20 Scatterhoarding is a common behavioral strategy to conserve food during periods of scarcity, but this type of 20
21 food storage is vulnerable to theft or pilferage. A variety of environmental factors and cache characteristics in- 21
22 fluence the rate of pilferage. Here we investigate 2 environmental factors, which heretofore have not received 22
23 much attention: the abundance and species richness of scatterhoarding animals in the vicinity of scatterhoarded 23
24 seeds. We measured the rate of cache pilferage at 7 sites that differed in the number and species composition of 24
25 granivorous rodents in western Nevada using local native seeds and sunflower seeds. We found that there was 25
26 no difference between the pilferage rate of native seeds and sunflower seeds, but that sites with different rodent 26
27 abundances had different pilferage rates. Pilferage rates were proportional to the abundance of scatterhoarding 27
28 rodents. Scatterhoarding rodents removed seeds at the rate of 1.3%/day/rodent individual. Species richness of 28
29 scatterhoarding rodents was not correlated with rates of pilferage. These results suggest that density-dependent 29
30 competition for scatterhoarded seeds is a strong determinant of pilferage rates. 30
31 31
32 Key words: cache pilfering, competition, food hoarding, heteromyids, rodent abundance, scatter hoarding, 32
33 sciurids 33
34 34
35 35
36 36
37 INTRODUCTION der. Hoarded food can play a critical role in maintaining 37
38 body condition and improving survival and/or reproduc- 38
Many species of mammals (mostly rodents) and tive success (Price & Jenkins 1986; Wauters et al. 1995;
39 39
many species of birds (mostly corvids) scatterhoard Kuhn & Vander Wall 2008). However, these undefend-
40 40
food to conserve it for future use (Vander Wall 1990). ed food stores are subject to pilfering from a variety
41 41
Scatterhoarding is the preferred means of food storage of competitors. For animals that store seeds to be con-
42 42
when the hoarder is not able to defend a large collection sumed weeks or months later, pilfering rates are typical-
43 43
of food items (i.e. larder) and when the foraging meth- ly 2–30% per day (Vander Wall & Jenkins 2003). These
44 44
od is not conducive to spending most time near the lar- high rates of pilferage occur despite the fact that ani-
45 45
46 mals have evolved a variety of behavioral strategies to 46
47 reduce pilferage (Dally et al. 2006; Zhang et al. 2013, 47
48 2014; Steele et al. 2013, 2015). Managing pilferage is a 48
49 Correspondence: Jacob W. Dittel, Department of Fisheries and key component of the adaptive strategy of scatterhoard- 49
50 Wildlife, Oregon State University, Corvallis, OR 97331, USA. ing animals. 50
51 Email: Jacob.dittel@oregonstate.edu An aspect of scatterhoarding that makes this means 51

J. W. Dittel & S. B. Vander Wall
1 of food storage work is that pilfered items are not im- in many species of plants that are dispersed by scat- 1
2 mediately eaten but recached, often repeatedly, until terhoarding animals (Vander Wall 2002; Jansen et al. 2
3 consumption (Vander Wall & Joyner 1998; Jansen et al. 2004). In mast years, animals store more seeds than they 3
4 2004, 2012; Perea et al. 2011) or germination (Vander can consume and many cached seeds are neglected and 4
5 Wall 2008; Jansen et al. 2012). In one intensive study 5
eventually germinate. Finally, rates of pilferage are like-
6 of cache pilfering (Jansen et al. 2012), 57% of caches 6
ly to rise with increases in the number of animals that
7 were recovered or pilfered within 1 week, with 99% of 7
are foraging for hidden food (i.e. increased competition
8 the caches being recovered or pilfered within the study 8
9
among granivores). This latter potential factor does not 9
period of 1 year. Seeds were often recached multiple appear to have been studied.
10 times (up to 36 times), sometimes being moved several 10
11 Given that pilferage of caches is a universal and im- 11
times in a single day. Most removals (84%) were pilfer-
12 portant feature of this food storing strategy, we need 12
age events (i.e. by an animal other than the cacher) and
13
despite being frequently handled most seeds were re- to understand it thoroughly. The goal of this study was 13
14 to investigate the effect of rodent abundance and rich- 14
cached rather than eaten (Jansen et al. 2012).
15 ness on pilfering rates. We predict that increased ro- 15
16 There is a variety of factors that influence how quick- 16
dent abundance will increase the rate of cache pilferage.
17 ly scatterhoarded items are pilfered, including those 17
arising from the characteristics of the cache and those With more individuals, the demand for a limited food
18 18
that are features of the environment. Important cache resource will increase. Most work has focused on the
19 19
characteristics include the strength of odor emanating behavior of individuals, and by not fully examining pil-
20 20
21 from seeds. Some species of seeds have stronger odors fering as a community process (Dittel et al. 2017), we 21
22 than others. From a seed’s perspective, once cached it are likely missing important aspects of the mechanisms 22
23 is best not to be found, so there is strong natural selec- that determine rates of pilferage. 23
24 tion to minimize odors emanating from seeds. Native 24
25 seeds that are adapted for dispersal by scatterhoarding MATERIALS AND METHODS 25
26 animals are pilfered at a slower rate (suggesting weak- 26
27 er odor) than non-native seeds (Hollander et al. 2012) Study areas 27
28 and seeds with indurate seed coats (a coat that appears 28
This study was conducted in 7 habitat types across an
29 to restrict passage of odor molecules) are pilfered more 29
elevation gradient in western Nevada: low desert shrub,
30 slowly than seeds that lack indurate seed coats (Paulsen 30
sand dune, juniper forest, piñon forest, montane shrub,
31 et al. 2013, 2014). Attractiveness of seeds may be im- 31
Jeffrey pine forest and an alpine shrub (see Table 1 for
32 portant, with more attractive (e.g. large with thin seed 32
more details). Different habitats had differences in ro-
33 coats) seeds being removed at a faster rate that less at- 33
dent abundance, species richness and species composi-
34 tractive (e.g. small with chemical defenses) seeds. Rates 34
tion, which were important elements of the study design.
35 of pilferage are also influenced by the size and depth of 35
Sites were chosen near each other for logistical reasons,
36 a cache (Vander Wall 2000; Vander Wall et al. 2003). 36
and in the cases where sites were within the same geo-
37 37
Finally, spacing of caches is important; widely spaced graphic area (e.g. Pine Nut Range and Whittell Forest
38 38
caches discourage area-restricted search and reduces and Wildlife Area) sites were chosen that were of obvi-
39 39
cache pilferage (Stapanian & Smith 1978; Clarkson et ous different habitat and not transition zones. Within the
40 40
al. 1986). Pine Nut range, the juniper forest site was at a location
41 41
One of the most important environmental influenc- dominated by Utah juniper (Juniperus osteosperma)
42 42
es of cache pilfering is moisture. Wet soils and even and the piñon forest site was at a location dominated by
43 43
high humidity can increase pilferage rates (Vander Wall piñon pine (Pinus monophylla). The montane shrub site
44 44
1998; Downs & Vander Wall 2009). This means that the was located on the eastern slope of the Carson range
45 45
rate of cache pilferage can change with the weather. Al- in an area that burned in 1982 and was dominated by
46 46
ternative food, either hidden in the soil (i.e. cached) or greenleaf manzanita (Arctostaphylos patula ) and tobac-
47 47
available on the ground surface or on plants can influ- co brush (Ceanothus velutinus). The Jeffrey pine site
48 48
ence how diligently animals search for stored seeds. was located within an unburned forest community dom-
49 49
inated by Jeffrey pine and lodgepole pine (Pinus contor-
50 This appears to be one of the causes of masting (the 50
ta).
51 synchronous production of large seed crops) observed 51

Rodent abundance and cache pilfering
1 Table 1 Description of the seven sites used for this study. All seven sites were located within the state of Nevada, USA 1
2 2
Site Location Coordinates Elevation (m) Substrate description Native seed used Month sampled
3 3
4 Low desert shrub Nightingale Flats 39°42′5″N, 1261 Mix of fine sand and Achnatherum May 4
5 119°1′58″W gravel hymenoides 5
6 Sand dune Hot Springs 39°42′2″N, 1270 Loose, sandy Achnatherum May 6
7 Mountains 119°3′34″W hymenoides 7
8 Juniper forest Pine Nut Range 39°5′10″N, 1920 Sandy Pinus monophylla May 8
9 9
119°39′0″W
10 10
11 Piñon forest Pine Nut Range 39°5′10″N, 1920 Sandy Pinus monophylla May 11
12 119°39′0″W 12
13 Montane shrub Whittell Forest 39°15′10″N, 1975 Decomposed granite Purshia tridentata September 13
14 and Wildlife Area 119°52′35″W 14
15 15
Jeffrey pine forest Whittell Forest 39°15′10″N, 1975 Decomposed granite Purshia tridentata September
16 16
and Wildlife Area 119°52′35″W
17 17
18 Alpine shrub Mount Rose 39°18′31″N, 2800 Decomposed granite Pinus jeffreyi September 18
19 119°55′17″W 19
20 20
21 21
22 22
23 23
24 24
25 Study design (Helianthus annuus) seeds to act as a standard treatment 25
26
across all study sites. A non-native seed type was nec- 26
We determined rodent abundance and species rich- essary because there is not a native seed species found
27 ness by live-trapping. We placed 50 Sherman live ani- 27
28
across all 7 habitat types. The native seeds that we used 28
mal traps (H. B. Sherman) in a 5 × 10 trapping grid with were Indian ricegrass (low desert shrub and sand dune
29 15-m spacing between traps at each site. Traps were left 29
30
sites), piñon pine (piñon and juniper forest sites), ante- 30
open for 5 days and 5 nights for a total of 250 trapping lope bitterbrush (montane shrub and Jeffrey pine for-
31 days and nights, and checked twice daily (morning and 31
32
est sites) and Jeffrey pine (alpine shrub). The number of 32
evening). A single trap was placed at each point on the seeds per cache was chosen based on the average size of
33 grid and the trap was baited with a mixture of sunflower 33
34
rodent-created caches and adjusted so that all seed cach- 34
and millet. We identified rodents to species and marked es across sites were approximately equal in value (ed-
35 individuals with a serially numbered ear tag to prevent 35
36
ible mass). Indian ricegrass caches contained 50 seeds 36
recounting. We classified rodent species as scatterhoard- (McMurray et al. 1997), piñon caches contained 3 seeds
37 ing and non-scatterhoarding. Because the study was de- 37
38
(Vander Wall 1997), antelope bitterbrush caches con- 38
pendent on rainfall events trapping occurred in either tained 10 seeds (Vander Wall 1994), Jeffrey pine caches
39 the early spring (May) or fall (September) when rainfall 39
40
contained 5 seeds (Tomback 1982) and sunflower seed 40
events were more frequent. Table 1 includes the month caches contained 5 seeds. Cache depth was constant
41 that each site was sampled. All institutional guidelines 41
42
across sites at 10-mm deep at the top, which is within 42
for the care and use of animals were followed under the range of depths that most rodents and corvids cache
43 University of Nevada Reno IACUC A07/08-30. 43
44
seeds (Tomback 1982; Vander Wall & Joyner 1998; 44
Concurrent with trapping, we established 3 wander- Vander Wall 2000; Longland et al. 2001). We made
45 45
ing transects each consisting of 50 caches spaced ap- caches using a metal spoon to minimize contaminat-
46 46
proximately 5-m apart. The seed cache transects started ing seeds with human scent, and groomed the soil after
47 47
>15 m along one side of the trapping grid and “wan- cache placement to remove signs of digging. We marked
48 48
dered” around the trapping grid, never entering it. We caches with natural objects (e.g. sticks, cones and rocks)
49 49
made half of the caches on each transect with seeds na- in various arrangements to avoid rodents learning visual
50 50
tive to the study site, and the other half with sunflower cues of cache locations.
51 51

1 Seed cache transects were checked daily for 5 days RESULTS 1

2 starting the day after establishment. If signs of digging 2
3 were observed at cache sites, we excavated those sites The total abundance of rodents over 5 days and 3
4 to verify that seeds were removed. If seeds were not re- nights of trapping ranged from 3 (alpine shrub site) to 4
5 moved, they were reburied. Because rodents have dif- 38 (low desert shrub and Jeffrey pine forest sites; Ta- 5
6 ficulties detecting caches under dry conditions (Vander ble 2). Species richness across sites ranged from 2 (al- 6
7 Wall 2000), all trials followed significant rain events. To pine shrub site) to 5 (low desert shrub, juniper forest 7
8 ensure soil moisture conditions were conducive to cache and piñon forest sites), with an averaged of 4 species 8
9 detection, 5 soil samples were collected at day 0 and day per site. We caught 2 species of cricetids (Peromyscus 9
10 5 in Nasco Whirl-pack plastic bags. Plastic bags were 10
11 sealed and kept in a cool location until returned to the 11
12 laboratory. In the laboratory, soil samples were weighed, 12
a b
13 dried in an oven at 100°C for 48 hand then reweighed to 100 13
Caches remaining (%)

14 determine water content. 80 14
15 60 15
Data Analysis 40
16 16
20
17 All data were analyzed using R (R Development Core 0
17
18 Team 2017). Rodent capture data (number of unique in- 18
c d
19 dividuals) were used to determine species richness and 100 19
20 abundance. Differences in the rates of cache removal Caches remaining (%) 80 20
21 were analyzed using survival analysis (package surviv- 60 21
22 al; Therneau & Grambsch 2000; Therneau 2013). Be- 40 22
20
23 cause seeds were only monitored for 5 days and we do 23
0
24 not know the precise time of removal for seeds taken by 24
25 rodents, we used right censoring in models. All mod- e f 25
100
26 els were fitted with a Weibull distribution, known for its 80 26

27 robustness when modeling survival data (Mudholkar et 60 27
28 al. 1996). Each site was analyzed separately comparing 40 28
rates of removal among the 3 transects and seed types 20
29 29
0
30 (native vs non-native). In addition, native and non-na- 30
31 tive caches (sunflower seeds) were analyzed to deter- 100
g 0 1 2 3 4 5 31
32 mine if there were differences in removal rates across 80

Time (days)
32
33 sites using parametric survival regressions. Rodent 60 33
34 abundance and species richness were regressed using 40 34
35 linear mixed models with mean removal rate to deter- 20
35
0
36 mine if there were any effects of abundance or richness 36
37 on removal rates. Models were run using the lmm func- 0 1 2 3 4 5 37
38 tion within the lme4 package (Bates et al. 2015). Be- Time (days)
38
39 cause this R package does not offer any P-values we Figure 1 Removal rates of native and non-native (sunflower) 39
40 performed Type II Wald χ2-tests using the Anova func- seeds at each site over the 5-day trial period. The x-axis rep- 40
tion in the car package to obtain P-values. For all mod- resents time (days) and the y-axis represents the percentage of
41 41
els, the dependent variable was the cache pilferage rate, caches remaining. Dashed lines represent native seeds, where-
42 42
and the independent variables were seed type with ei- as solid lines represent sunflower seeds. Letters correspond to
43 43
ther rodent abundance, rodent richness or scatterhoard- site location; (a) is low desert shrub, (b) is sand dune, (c) is ju-
44 44
ing rodent abundance. In each model, we used transect niper forest, (d) is piñon forest, (e) is montane shrub, (f) is Jef-
45 fery pine first, and (g) is alpine shrub. There is no difference in 45
46 (1–3) at each site as random effects. Because we were 46
the rates of removal in 4 of the 7 sites. At the juniper (c) and
47 interested in determining the strength of the relation- 47
piñon forest sites (d), native piñon pine seeds were removed
48 ship between our variables we calculated and are report- 48
faster than sunflower seeds (χ2 = 25.6, df = 1, P < 0.001 and
49 ing R2 values for mixed models following Nakagawa 49
χ2 = 15.36, df = 1, P < 0.001, respectively). At the low desert
50 and Schielzeth (2013) using the sem.model.fits function shrub site (a), sunflower seeds were removed faster (χ2 = 11.85, 50
51 within the piecewiseSEM package (Lefcheck 2016). df = 1, P < 0.001) than the native Indian ricegrass seeds. 51

1 Table 2 Abundance (number of individuals and number of scatterhoarding individuals) and richness (number of species) at each 1
2 study site arranged from low to high elevation 2
3 Low desert shrub Sand dune Juniper forest Piñon forest Montane shrub Jeffrey pine forest Alpine shrub 3
4 4
Total rodent 38 17 9 16 21 38 3
5 5
6 abundance 6
7 Scatterhoarding 30 17 9 16 19 38 3 7
8 rodent abundance 8
9 Species richness 5 4 5 5 3 4 2 9
10 10
11 11
12 12
13 13
drop below 1.5% during the study. Linear mixed mod- 14
Pilferage rate (caches/day/individual)
14 0.6
15 Sunflower seeds els showed negligible effects of transect within each site 15
16 Native seeds (residual variance = 0.007, standard deviation = 0.08). 16
17
0.4 Overall, the rates of removal of sunflower versus na- 17
18
tive seed caches was not significantly different across 18
19 0.2 all sites (χ2 < 0.01, df = 1, P =0.93). However, at 3 sites 19
20
there were within-site differences in sunflower and na- 20
21 0.0
tive seed removal. At the juniper forest (χ2 = 25.6, df = 21
22
1, P < 0.001) and piñon forest (χ2 = 15.36, df = 1, P < 22
23
0.001) sites, rodents removed piñon pine seeds faster 23
−0.2 and at the low desert shrub site, rodents removed Indi- 24
24
25
an ricegrass slower than sunflower seeds (χ2 = 11.85, df 25
26
5 10 15 20 25 30 35 = 1, P < 0.001, Fig. 1). The low desert shrub site experi- 26
27
Rodent abundance enced the fastest rate of cache removal (mean ± SD; 43 27
28 Figure 2 Predicted values for the linear mixed models of mean ± 22% per day) followed by sand dune (26% ± 14% per 28
29 daily removal rates for sunflower (solid line) and native seeds day), Jeffrey pine forest (26% ± 8% per day), montane 29
30 (dashed line) as a function of scatterhoarding rodent abundance shrub (12% ± 7% per day), piñon forest (11% ± 7% per 30
31 (χ2 = 55.34, df = 1, P < 0.001). Gray polygons surrounding the day), juniper forest (7% ± 3% per day), and the alpine 31
32 lines represent the 95% confidence intervals. The effect size of shrub site with the lowest removal rate (2% ± 0.5% per 32
33 scatterhoarder abundance was 1.3% caches pilfered/individual/ day). 33
day. Seed removal rates were significantly positively cor- 34
34
35 related with total rodent abundance (χ2 = 32.93, df = 1, 35
36 P <0.001). The mean removal rate per individual was 36
37 1.1% per day. Mean rates of removal were not correlat- 37
38
trueii and Peromyscus maniculatus), 6 species of sciu- ed with rodent species richness (χ2 = 1.6, df = 1, P = 38
39
rids (Ammospermophilus leucurus, Callospermophilus 0.43). When re-running the analysis including only the 39
40
lateralis, Otospermophilus beecheyi, Tamias amoenus, effect of scatterhoarding rodent abundance (i.e. exclud- 40
41
Tamias minimus and Tamias quadrimaculatus) and 7 ing larder-hoarding ground squirrels) the correlation be- 41
42
species of heteromyids (Chaetodipus formosus, Di- tween pilfering and rodent abundance was improved (χ2 42
43
podomys merriami, D. ordii, D. panamintinus, Microdi- = 55.34, df = 1, P < 0.001, Fig. 2; R2All = 0.54, R2scatter- 43
44
podops megacephalus, Perognathus longimembris and hoarders = 0.65) and there was a significant interactive ef- 44
45
P. parvus). All species were identified as scatterhoarders fect between scatterhoarder abundance and seed type (χ2 45
46
except C. lateralis and O. beecheyi, which are known = 4.53, df = 1, P <0.03). Confining the analysis to scat- 46
47
larder hoarders (Saigo 1969; Vander Wall et al. 2009). terhoarder abundance increased the daily pilferage rates 47
48
These 2 species only accounted for 10 individuals of the slightly (1.3%/day/individual). The abundance of sciu- 48
49
total 142 individuals caught. rid rodents (the only family of scatterhoarding rodents 49
50 Soil moisture at all 7 sites was above 4% at the time fount at every site) did not have a significant effect on 50
51 that we established seed cache transects and did not mean daily pilferage rate (χ2 = 0.33, df = 1, P = 0.56). 51

1 DISCUSSION er soil moisture during the study period than others. De- 1
2 spite differences in rainfall, the minimum soil moisture 2
3 Pilferage rates did not significantly differ between required to promote pilfering was met for all 7 sites. 3
4 sunflower seeds and native seeds across the 7 sites. Ro- The relationship between soil moisture and seed detec- 4
5 dent abundance was the only factor that had a signifi- tion is a step function, with a threshold at approximate- 5
6 cant effect on seed pilferage rates in this study. The sites ly 0.5% soil moisture (Vander Wall 2000, 2003). Above 6
7 with the highest rodent abundance experienced the high- the threshold seeds become detectable to rodents, and 7
8 est mean daily pilferage rates over the 5-day trials. Ex- the amount of moisture above that level (within limits) 8
9 cluding non-scatterhoarding rodents from the analy- does not really matter. We were also unable to quanti- 9
10 sis improved the correlation between rodent abundance fy the amount of fruit and seed available to rodents at 10
11 and pilfering rates from explaining 54% of the variation our sites. Seed abundance may have influenced pilfering 11
12 in pilferage rates with all rodents to explaining 65% of rates through density-dependent factors, but our results 12
13 the variation when considering only scatterhoarding ro- were well within recorded pilfering rates (Vander Wall 13
14 dents. The non-scatterhoarding rodents (golden-mantled & Jenkins 2003) and none of the sites were experienc- 14
15 ground squirrels and California ground squirrels) only ing mast years during the study (personal observations). 15
16 accounted for 7% of the total individuals captured and Thus, we contend that differences seen in pilfering rates 16
17 occurred at only 2 sites. Our results suggest that on av- were primarily due to differences in rodent abundance. 17
18 erage, a single seed-pilfering rodent contributes 1.3% to 18
The alpine shrub site happened to overlap a cach-
19 the daily pilferage rates. The larger effect size of rodent 19
ing area used by Clark’s nutcrackers (Nucifraga colum-
20 abundance probably occurred because rodents that do 20
biana). These birds cache pine seeds on open slopes at
21 not scatterhoard seeds appear not to be very good pilfer- 21
higher elevations, often many kilometers away from and
22 ers (Vander Wall et al. 2009). This is not surprising as 22
over 1000 m above the forests where these seeds had
23 scatterhoarding species are proficient pilferers (Vander 23
been collected (Vander Wall & Balda 1977; Hutchins &
24 Wall et al. 2009). Non-scatterhoarding rodents likely 24
Lanner 1982; Tomback 1982; Tomback & Linhart 1990;
25 pilfer infrequently, if at all, and their presence in a com- 25
Lorenz et al. 2011). Our results help to explain why
26 munity has little effect on pilferage rates. 26
Clark’s nutcrackers invest so much energy to transport
27 Rodent species richness had no effect on mean daily seeds to these high elevation sites. Pilferage rates at the 27
28 pilferage rates. The lack of a correlation between rodent alpine shrub site were low. In fact, nutcrackers often se- 28
29 richness and seed removal rates suggests that the num- lect caching areas more sparsely vegetated than our site, 29
30 ber of pilfering rodents has a larger impact on remov- including cliff faces and rock ledges where seed-cach- 30
31 al rates than the species mix (e.g. ratio of chipmunks to ing rodents are virtually absent. In the forests where the 31
32 kangaroo rats). This is due, in part, to the fact that there seeds were collected (i.e. Jeffrey pine and piñon pine 32
33 was little variation in species richness across sites (max forests), pilferage rates are more than on order of mag- 33
34 = 5, minimum = 2) with only 2 of the sites (montane nitude greater. Furthermore, snow covers the ground for 34
35 shrub and alpine shrub) having fewer than 4 species much longer periods (1–2 months longer) at higher ele- 35
36 (Table 2). The notion that cache pilfering in rodent com- vations sites, and during periods of snow cover, pilfer- 36
37 munities is often reciprocal (Price & Mittler 2003, 2006; age rates would approach 0% per day. Therefore, cach- 37
38 Vander Wall & Jenkins 2003) suggests that the iden- es at higher elevation sites with few rodents would be 38
39 tity of the species might not matter as long as the ro- available to pilferers for shorter periods of time than 39
40 dent caches seeds. If any 1 species of pilferer were able at lower elevation sites where the seeds are harvested. 40
41 to disproportionately pilfer from heterospecifics, they Greatly reduced rates of pilferage may explain why nut- 41
42 would likely extirpate that species because they would crackers invest so much energy transporting seeds to 42
43 lose seeds faster than they could replace them. At the these remote sites. 43
44 Jeffrey pine forest site, Dittel et al. (2017) showed that 44
This study is, to our knowledge, the first to show that
45 pilfering was reciprocal among the most common scat- 45
there is a positive relationship between rodent abun-
46 terhoarding species and this is likely the case at the oth- 46
dance and cache pilfering rates. On average, the addi-
47 er study sites. 47
tion of a single scatterhoarding individual increases pil-
48 Natural rainfall events were sufficient to promote 48
fering by 1.3% per day. Pilfering rates are not influenced
49 cache pilfering during this study. Of course, not all 49
by species richness as long as those species are scatter-
50 rain events were equal so some sites experienced high- 50
hoarders. Per capita rates of pilferage are similar across
51 51

1 habitat types and elevations. As long as the rodents scat- zopsis hymenoides: Are desert granivores mutualists 1
2 terhoard seeds, the composition of the rodent commu- or predators? Ecology 82, 3131–48. 2
3 nity did not seem to matter much. With regard to pilfer- McMurray MH, Jenkins SH, Longland WS (1997). Ef- 3
4 age, these results suggest that different scatterhoarding fects of seed density on germination and establish- 4
5 rodent species play a similar role in different ecological ment of a native and an introduced grass species dis- 5
6 communities. persed by granivorous rodents. American Midland 6
7 Naturalist 138, 322–30. 7
8 8
ACKNOWLEDGMENTS Mudholkar GS, Srivastava DK, Kollia GD (1996). A
9 9
generalization of the Weibull distribution with appli-
10 We would like to thank C. Moore, K. Hefty and B. 10
cation to the analysis of survival data. Journal of the
11 Beebe for assistance in the field, as well as M. Matocq, C. 11
American Statistical Association 91, 1575–83.
12 Feldman, W. Longland, S. Mensing and anonymous re- 12
13 viewers who provided comments on a previous draft of Paulsen TR, Colville L, Kranner I et al. (2013). Physical 13
14 the manuscript. dormancy in seeds: A game of hide and seek? New 14
15 Phytologist 198, 496–503. 15
16 REFERENCES Paulsen TR, Högstedt G, Thompson K, Vandvik V, Eli- 16
17 assen S (2014). Conditions favouring hard seeded- 17
18 Bates D, Mächler M, Bolker B, Walker S (2015). Fitting ness as a dispersal and predator escape strategy. Jour- 18
19 linear mixed-effects models using lme4. Journal of nal of Ecology 102, 1475–84. 19
20 Statistical Software 67, 1–48. Perea R, Miguel AS, Gil L (2011). Acorn dispersal by 20
21 Clarkson K, Eden SF, Sutherland W, Houston AI (1986). rodents: The importance of re-dispersal and distance 21
22 Density dependence and magpie food hoarding. Jour- to shelter. Basic and Applied Ecology 12, 432–9. 22
23 nal of Animal Ecology 55, 111–21. Price MV, Jenkins SH (1986). Rodents as seed consum- 23
24 Dally JM, Clayton N, Emery NJ (2006). The behaviour ers and dispersers. In: Murray DR, ed. Seed Disper- 24
25 and evolution of cache protection and pilferage. Ani- sal. Academic Press, San Diego, CA, pp. 191–235. 25
26 mal Behaviour 72, 13–23. Price MV, Mittler JE (2003). Seed-cache exchange pro- 26
27 Dittel JW, Perea R, Vander Wall SB (2017). Reciprocal 27
motes coexistence and coupled consumer oscilla-
28 pilfering in a seed-caching rodent community: impli- 28
tions: A model of desert rodents as resource proces-
29 cations for species coexistence. Behavioral Ecology 29
sor. Journal of Theoretical Biology 223, 215–31.
30 and Sociobiology 71, 147. 30
31
Price MV, Mittler JE (2006). Cachers, scavengers, 31
Downs CJ, Vander Wall SB (2009). High relative hu- thieves: A novel mechanism for desert rodent coexis-
32 32
midity increases pilfering success of yellow pine tence. American Naturalist 168, 194–206.
33 33
chipmunks. Journal of Mammalogy 90, 796–802. R Development Core Team (2017). R: A language and
34 34
Hutchins HE, Lanner RM (1982). The central role of environment for statistical computing. R Foundation
35 35
Clark’s nutcracker in the dispersal and establishment for Statistical Computing, Vienna, Austria.
36 36
of whitebark pine. Oecologia 55, 192–201.
37 Saigo BW (1969). The relationship of non-recovered ro- 37
38 Jansen PA, Bongers F, Hemerik L (2004). Seed mass dent caches to the natural regeneration of ponderosa 38
39 and mast seeding enhance dispersal by a neotropical pine (Master’s thesis). Oregon State University, Cor- 39
40 scatter-hoarding rodent. Ecological Monographs 74, vallis. 40
41 569–89. 41
Stapanian MA, Smith CC (1978). A model for seed scat-
42 Jansen PA, Hirsch BT, Emsens W-J, Zamora-Gutierrez V, terhoarding: Coevolution of fox squirrels and black 42
43 Wikelski M, Kays R (2012). Thieving rodents as sub- walnuts. Ecology 59, 884–96. 43
44 stitute dispersers of megafaunal seeds. Proceedings 44
Steele MA, Contreras TA, Hadj-Chikh LZ, Agosta SJ,
45 of the National Academy of Sciences 109, 12610–5. 45
Smallwood PD, Tomlinson CN (2013). Do scatter
46 Kuhn KM, Vander Wall SB (2008). Linking summer hoarders trade off increased predation risks for low- 46
47 foraging to winter survival in yellow-pine chipmunks er rates of cache pilferage? Behavioral Ecology 25, 47
48 (Tamias amoenus). Oecologia 157, 349–60. 206–15. 48
49 Longland WS, Jenkins SH, Vander Wall SB, Veech 49
Steele MA, Rompré G, Stratford JA, Zhang H, Sucho-
50 JAJA, Pyare SS (2001). Seedling recruitment in Ory- 50
ki M, Marino S (2015). Scatterhoarding rodents favor
51 51

1 higher predation risks for cache sites: The potential seeds: A model based on the behavior of pesticides in 1
2 for predators to influence the seed dispersal process. soil. Journal of Mammalogy 84, 1089–99. 2
3 Integrative Zoology 10, 257–66. Vander Wall SB, Balda RP (1977). Coadaptations of the 3
4 Therneau T (2013). A package for survival analysis in Clark’s nutcracker and the piñon pine for efficient 4
5 S. R package version 2.37-4. [Accessed Jul 2015.] seed harvest and dispersal. Ecological Monographs 5
6 Available from URL: http://CRAN. R-project. org/ 47, 89–111. 6
7 package=survival. 7
Vander Wall SB, Joyner JW (1998). Recaching of Jef-
8 8
Therneau TM, Grambsch PM (2000). Modeling survival frey pine (Pinus jeffreyi) seeds by yellow pine chip-
9 9
data: Extending the Cox model. Springer Science & munks (Tamias amoenus): Potential effects on plant
10 10
Business Media, New York, USA. reproductive success. Canadian Journal of Zoology
11 11
Tomback DF (1982). Dispersal of whitebark pine seeds 76, 154–62.
12 12
13 by Clark’s nutcracker: A mutualism hypothesis. Jour- Vander Wall SB, Jenkins SH (2003). Reciprocal pilfer- 13
14 nal of Animal Ecology 51, 451–67. age and the evolution of food-hoarding behavior. Be- 14
15 Tomback DF, Linhart YB (1990). The evolution of havioral Ecology 14, 656–67. 15
16 bird-dispersed pines. Evolutionary Ecology 4, 185– Vander Wall SB, Beck MJ, Briggs JS et al. (2003). Inter- 16
17 219. specific variation in the olfactory abilities of granivo- 17
18 Vander Wall SB (1994). Seed fate pathways of antelope rous rodents. Journal of Mammalogy 84, 487–96. 18
19 bitterbrush: Dispersal by seed-caching yellow pine Vander Wall SB, Enders MS, Waitman BA (2009). 19
20 chipmunks. Ecology 75, 1911–26. Asymmetrical cache pilfering between yellow pine 20
21 Vander Wall SB (1997). Dispersal of singleleaf pinon chipmunks and golden-mantled ground squirrels. An- 21
22 pine (Pinus monophylla) by seed-caching rodents. imal Behaviour 78, 555–61. 22
23 Journal of Mammalogy 78, 181–91. Wauters LA, Suhonen J, Dhondt AA (1995). Fitness 23
24 consequences of hoarding behaviour in the Eurasian 24
Vander Wall SB (1998). Foraging success of granivo-
25 red squirrel. Proceedings of the Royal Society B 262, 25
rous rodents: Effects of variation in seed and soil wa-
26 277–81. 26
ter on olfaction. Ecology 79, 233–41.
27 27
Vander Wall SB (2000). The influence of environmental Zhang H, Gao H, Yang Z, Wang Z, Luo Y, Zhang Z
28 28
conditions on cache recovery and cache pilferage by (2014). Effects of interspecific competition on food
29 29
yellow pine chipmunks (Tamias amoenus) and deer hoarding and pilferage in 2 sympatric rodents. Be-
30 30
mice (Peromyscus maniculatus). Behavioral Ecology haviour 151, 1579–96.
31 31
32
11, 544–9. Zhang M-M, Shen Z, Liu G-Q, Yi X-F (2013). Seed 32
33 Vander Wall SB (2002). Masting in animal-dispersed caching and cache pilferage by 3 rodent species in a 33
34 pine facilitates seed dispersal. Ecology 83, 3508–16. temperate forest in the Xiaoxinganling Mountains. 34
35 Vander Wall SB (2003). How rodents smell buried Dongwuxue Yanjiu 34, E13. 35
36 36
37 37
38 38
39 39
40 Dittel JW, Vander Wall SB (2018). Effects of rodent abundance and richness on cache pilfering. Integrative Zool- 40
41 ogy 13, 331–8. 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51

2 2
3 3
4 4
5 5
6
7
Phenotypic trait matching predicts the topology of an insular 6
7
8
9 plant–bird pollination network 8
9
10 10
11 11
12 12
13
Matthew BIDDICK and Kevin C. BURNS 13
14 14
15 School of Biological Sciences, Victoria University of Wellington, Wellington, New Zealand 15
16 16
17 17
18 18
19 19
20 Abstract 20
21 Conceptualizing species interactions as networks has broadened our understanding of ecological communities. 21
22 However, the factors shaping interaction patterns among species and, therefore, network structure remain un- 22
23 clear. One potentially important factor is the matching of phenotypic traits. Here, we tested for trait matching in 23
24 a bird–flower visitation network from New Zealand. We first quantified the overall network structure and test- 24
25 ed whether flower size could account for differences in the visitation rates of flowering plants. We then explored 25
26 the relationship between the flower size and bill size. The results showed that the interaction network is nest- 26
27 ed. Plant species with large flowers received more visits from birds than plant species with small flowers. More- 27
28 over, plant species with large flowers were visited more frequently by birds with large bills, while species with 28
29 smaller flowers were visited more frequently by birds with small bills. Overall, the interaction patterns between 29
30 birds and flowering plants could be predicted by their morphology, suggesting that phenotypic trait matching is 30
31 an important predictor of network structure. 31
32 32
33 Key words: bird–flower visitation network, flower morphology, plant–animal interactions, size coupling 33
34 34
35 35
36 36
37
INTRODUCTION communities (Bascompte 2009). However, the mech- 37
38 anisms shaping species interaction patterns and, there- 38
39 The development of network theory has been crucial fore, overall network structure, remain unclear. 39
40 to our understanding of ecological communities. Con- 40
One potential mechanism shaping network topolo-
41 ceptualizing the myriad of species interactions that oc- 41
gy is the matching of phenotypic traits. Species mor-
42 cur in nature as networks has provided ecologists with 42
phology modulates network structure by both permit-
43 a more holistic approach to understanding ecological 43
ting and restricting possible interactions (Krishna et al.
44 44
2008; Olesen et al. 2010; Sankamethawee et al. 2011;
45 45
Zhang et al. 2015). For example, the body size of preda-
46 46
tors can determine the size of prey they consume (Cohen
47 47
Correspondence: Matthew Biddick, School of Biological et al. 1993; Emmerson & Raffaelli 2004; Owen-Smith
48 48
Sciences, Victoria University of Wellington, P.O. Box 600, & Mills 2008). Similarly, the gape size of frugivores
49 49
Wellington, New Zealand.
determines the size of fruits they consume and pre-
50 50
dicts the interaction patterns of frugivores with fruit-
51 Email: matt.biddick@vuw.ac.nz 51

M. Biddick & K.C. Burns
1 ing plants (Wheelwright 1985; Rey et al. 1997; Burns serve is surrounded by a unique fence designed to ex- 1
2 2006, 2013). Phenotypic trait matching can also shape clude introduced mammalian predators, which have led 2
3 the topology of plant–insect pollination networks (Stang to the extinction of nearly half of New Zealand’s avifau- 3
4 et al. 2006, 2007, 2009; Ibanez 2012). However, less is na (Wilson 2004). In the absence of intensive mammal 4
5 known about plant–bird pollination networks (although control, New Zealand forests typically support marginal 5
6 see Maruyama et al. 2014; Maglianesi et al. 2015a,b). populations of native bird species (Innes et al. 2010). 6
7 We quantified a bird–flower visitation network by Many rare bird species have been translocated into 7
8 observing birds visit flowers in a temperate rainforest the reserve following the construction of the fence (Ta- 8
9 on the North Island of New Zealand for 8 consecutive ble 1), including the bellbird (Anthornis melanura Spar- 9
10 years. We then evaluated whether species morphology rman, 1786), the kaka (Nestor meridionalis Gmelin, 10
11 (i.e. bill and flower size) could predict network struc- 1788), the saddleback (Philesturnus carunculatus 11
12 ture by testing 3 hypotheses: (1) plant species with large Gmelin, 1789), the hihi (Notiomystis cincta Du bus de 12
13 flowers receive greater total number of visits from birds Gesignies, 1839) and the whitehead (Mohoua albicil- 13
14 than plant species with small flowers; (2) plant species la Lesson, 1830). Many native species that were already 14
15 with large flowers are visited more frequently by birds present within the reserve prior to the construction of 15
16 with large bills; and (3) plant species with small flowers the fence have also benefited from its construction (e.g. 16
17 are visited more frequently by birds with small bills. the tui, Prosthemadera novaeseelandiae Gmelin, 1788). 17
18 The Waxeye (Zosterops lateralis Latham, 1802) is a 18
19 frugivore originally native to Australia that colonized 19
20
MATERIALS AND METHODS 20
New Zealand in the recent past (<200 years ago), whose
21 All observations were made in Zealandia, a man- numbers have also increased. 21
22 aged forest reserve on the North Island of New Zea- 22
Bird–flower visits were quantified by observing wild
23 land (41°18.3′S, 174°44.8′E). Zealandia experiences a 23
birds foraging for nectar along a 4.6-km reserve trail
24 mild, temperate climate and supports evergreen forest 24
that took approximately 1 h to traverse. Observations
25 composed of trees, tree ferns and lianas. The 225-ha re- 25
were made approximately every 5 days from January
26 26
27 27
28 28
29 29
Table 1 List of species comprising a bird–flower interaction network in New Zealand
30 30
31 Plant species Stamen length (mm) Relative abundance Flower color 31
32 A Fuchsia excorticata 22.99 29 Purple 32
33 B Pseudopanax arboreus 4.11 32 White 33
34 C Phormium tenax 50.97 7 Red 34
35 35
D Sophora microphylla 34.72 3 Yellow
36 36
37 E Cytisus proliferus 14.81 4 White 37
38 F Geniostoma lingustrifolium 0.81 51 Green 38
39 G Schefflera digitata 3.4 13 Green 39
40 40
H Berberis darwinii 5.57 1 Yellow
41 41
Bird species Bill length (mm) Relative abundance
42 42
43 A Prosthemadera novaeseelandiae 23.93 49.2 43
44 B Zosterops lateralis 10.35 12.5 44
45 C Anthornis melanura 14.29 11.8 45
46 46
D Notiomystis cincta 15.97 23.4
47 47
48 E Nestor meridionalis 51.15 32 48
49 F Mohoua albicilla 10.22 103.2 49
50 Species labels correspond to those of Figure 1. Stamen and bill lengths are species means. Relative species abundances for birds 50
51 were gleaned from Bell (2015). Flower color is described as it appears to the human eye. 51

1 2006 to December 2013. Given the erratic movements ber of occurrences in the matrix (the matrix fill) (Greve 1
2 of foraging birds in the cover of dense foliage, accu- & Chown 2006). A null model is then used to generate 2
3 rate estimates of the total number of interactions made a distribution of expected values, against which the ob- 3
4 in a foraging bout could not be made. Consequently, we served value is tested. The network was assessed against 4
5 quantified bird–plant interactions more generally fol- 100 null model simulations with fixed row and col- 5
6 lowing Snow and Snow (1988) and Burns (2006). A vis- umn totals (as suggested by Ulrich & Gotelli 2007) and 6
7 it was scored when a bird approached a plant and fed the independent swap algorithm to fill matrices (Gotel- 7
8 from at least 1 flower. To minimize pseudoreplication, li 2000, 2001). That is, row and column totals in null 8
9 observations were ceased after each recorded visit, and model simulation never exceeded that of the observed
9
10 then resumed 10 m down the trail. Note that, through- 10
network.
11 out, birds are referred to sparingly as pollinators regard- 11
12 Negative co-occurrences were quantified using the 12
less of their effectiveness.
13 χ-score index. This is calculated by finding the (normal- 13
To quantify network structure, we calculated connec- ized) mean checkerboard units for each species pair:
14 14
tance, nestedness and negative co-occurrences using the cu = (ni – Nij)(nj – Nij), where ni and nj are the row to-
15 15
R package bipartite v. 2.08 (Dormann et al. 2008). Be- tals (number of occurrences) of species i and j, and Nij is
16 16
17
cause connectance values alone are sensitive to sam- the number of co-occurrences of both species (Stone & 17
18 pling effort (and, therefore, network species richness), Roberts 1990). The observed χ-score is then compared 18
19 weighted connectance is more appropriate for making to a distribution of expected χ-scores under chance 19
20 comparisons between habitats. Weighted connectance is alone. That is, the positions of entries within the matrix 20
21 calculated by dividing linkage density by network spe- are randomized by computer simulation. Again, the net- 21
22 cies richness (Tylianakis et al. 2007). work was assessed against 100 simulations with fixed 22
23 The Brualdi and Sanderson (1999) discrepancy index row and column totals and the independent swap algo- 23
24 (hereinafter ‘BR’) (Brualdi & Sanderson 1999) counts rithm for matrix fill. A χ-score greater than what is ex- 24
25 the number of discrepancies (absence or presence) that 25
pected under the null model would support negative
26 must be erased to produce a perfectly nested matrix. 26
co-occurrences within the network.
27 This is then standardized by dividing by the total num- 27
28 To ensure that the mean size bill that foraged on flow- 28
29 ering species was estimated accurately, we omitted bird 29
30 species that were observed less than 10 times across the 30
31 observation period. Likewise, plant species that regis- 31
32 tered fewer than 10 visits across the observation peri- 32
Table 2 Table of bird species and the length of their bills taken
33 od were also omitted for their effect on the mean size of 33
from skin specimens housed in the National Museum of New
34 flowers visited by birds. Most omitted species were only 34
Zealand (Te Papa Tongawera, Wellington, New Zealand)
35 observed a handful of times and are, therefore, unlikely 35
36 Species Bill length (mm) Mean (mm) to have an appreciable effect on results. 36
37 Prosthemadera ♂ = 25.22 (n = 10) 23.94 37
To estimate the mean bill size of each bird species,
38 novaeseelandiae ♀ = 22.86 (n = 10) 38
we measured specimens housed in the National Muse-
39 Zosterops lateralis ♂ = 10.55 (n = 10) 10.35 39
um of New Zealand (Te Papa Tongarewa, Wellington,
40 ♀ = 10.15 (n = 10) 40
New Zealand). The distance from the tip of the bill to
41 Anthornis melanura ♂ = 15.35 (n = 9) 14.29 41
the anterior edge of the skull was measured using a pair
42 ♀ = 13.22 (n = 10) 42
of digital callipers. Between 11 and 20 specimens for
43 43
Notiomystis cincta ♂ = 16.30 (n = 10) 15.97 each species were measured from the collection of skins
44 44
♀ = 15.64 (n = 7) available, including both males and females (see Table
45 45
46 Nestor meridionalis ♂ = 51.63 (n = 8) 51.15 2). 46
47 ♀ = 50.67 (n = 3) To characterize flower size, 4 aspects of flower mor- 47
48 Mohoua albicilla ♂ = 10.35 (n = 10) 10.22 phology were measured. Stamen length was measured 48
49 ♀ = 10.08 (n = 6) as the distance from the base of the nectary to the distal 49
50 Measurements are broken down by sex and sample size (in pa- tip of the anther. Pistil length was measured as the dis- 50
51 rentheses). 51

1 tance from the base of the nectary to the distal tip of the Both axes were log-transformed to conform to model 1
2 style. Tube length was measured from base of the nec- assumptions. 2
3 tary to the distal point of the corolla. Tube width was To explore the relationship between bill and flow- 3
4 measured as the widest internal distance transecting 4
er size, we ran 2 linear model regressions: 1 of flower-
5 the corolla. One flower from each of 30 individuals per 5
ing species stamen length against the mean bill length
6 plant species was measured. Species means for stamen 6
of visitors, and 1 of bird species’ bill length against the
7 length, pistil length, tube length and width were then 7
mean stamen size of flowers they visited. Bill length
8 8
correlated against each other in a 4-way correlation ma- was log-transformed to conform to model assumptions.
9 9
trix and the respective correlation and Pearson P-values By using mean stamen and bill size, we are able to vi-
10 10
were calculated using the “Hmisc” package in the R-en- sualize the extent to which interactions are skewed. Re-
11 11
vironment. Stamen size was used as a proxy for over- gression analyses were conducted in the R environment
12 12
13
all flower size because of its known scaling relationship (R Core Team 2017). 13
14 with other flower traits (see Table 3), and because it is 14
easy to measure consistently. Flower color was also re-
15 RESULTS 15
16 corded as it appears to the human eye. 16
17 To estimate relative local abundances of plant spe- A total of 1216 bird–flower visits were observed over 17
18 cies, a 2-m pole was placed at regular intervals, alter- an 8-year period involving 8 plant species and 6 bird 18
19 nating left and right either side of the trail. All vascular species (Fig. 1). After excluding seldom observed spe- 19
20 cies, total interactions (visits) for plants ranged from 15– 20
plant species immediately touching the pole or its ver-
21 640 and from 11–882 for birds. Of 48 possible species 21
tically projected plane were recorded. This was repeat-
22 interactions, 32 were realized, giving a network connec- 22
ed 100 times along the trail. Relative abundance was
23 tance value of 67%, significantly less than expected un- 23
recorded as the number of observations out of 100. Rel-
24 der null model simulation (81.5% ± 0.7% mean ± SD; 24
ative abundance estimates for birds were gleaned from T = 40.357, P < 0.01). When weighted by network spe-
25 Bell (2015). 25
26 cies richness, this value drops to 20.6%, again less than 26
27
To explore the relationship between flower size and expected under null model simulation (24.5% ± 0.03% 27
28 the number of visits received from birds, we ran lin- mean ± SD; T = 532.9, P < 0.01). 28
29 ear model regression of stamen length against visitation Network-level analysis of interactions found signif- 29
30 rate. Because we were interested in the effect of mor- icant evidence for nestedness. The observed discrepan- 30
31 phology, and not the effect of abundance, the visitation cy index (BR = 2) was statistically greater than the dis- 31
32 rate for a species was calculated as the total number of tribution of expected values generated under null model 32
33 visits divided by its relative abundance. This way, dif- simulation (BR = 1.22 ± 0.188, mean ± SD; T = –8.12, 33
34 ferences in visitation rates could be assessed without the P < 0.01). However, the results indicated no support for 34
35 confounding effect of differences in relative abundance. negative co-occurrence patterns. The observed χ-score 35
36 36
37 37
38 38
39 Table 3 A correlation matrix of morphological flower traits of 8 plant species comprising a bird–flower visitation network in New 39
40 Zealand 40
41 Tube length Tube width Pistil length Stamen length 41
42 42
Tube length r = 0.97 r = 0.95 r = 0.95
43 43
P = 0.001 P = 0.003 P = 0.003
44 44
45 Tube width r = 0.91 r = 0.91 45
P = 0.011 P = 0.013
46 46
47 Pistil length r = 0.99 47
48 P < 0.001 48
49 Stamen length 49
50 Each cell denotes the relative (r) value and Pearson P-value between 2 morphological traits. Analysis was run in the R-environment 50
51 using the “Hmisc” package. 51

1 (C = 0.171) was statistically indistinguishable from Moreover, small-billed birds tended to visit small flow- 1
2 scores expected under null model simulation (C = 0.171 ers and large billed birds tended to visit large flowers. 2
3 ± 0.032; mean ± SD; T = −0.006, P = 0.995). Species of greater stamen lengths received significantly 3
4 Flower visitation rates were positively associated more visits from birds of larger bill sizes (T = 4.178, P = 4
5 with stamen length (Fig. 2). After accounting for dif- 0.006, r2 = 0.702) (Fig. 3). Likewise, larger-billed birds 5
6 ferences in relative abundance, species with larger sta- were more frequently observed interacting with spe- 6
7 men sizes and red or yellow color consistently showed 7
8 greater visitation rates (T = 6.220, P < 0.001, r2 = 0.843). 8
9 9
10 10
11 11
12 12
13 13
14 14
15 15
16 16
17 17
18 18
19 19
20 20
21 21
22 22
23 23
24 24
25 Figure 2 Stamen length is associated with the amount of vis- 25
26 its flowering species receive from birds. Visitation rate is cal- 26
27 Figure 1 Matrix of interactions between 8 flowering plant spe- culated as the total number of observed bird visits divided by 27
28 cies (x-axis) and 6 bird species (y-axis) located on the North Is- the plant species’ relative abundance. Plant species with long 28
29 land of New Zealand. Species are ranked by their total interac- stamens (i.e. big flowers) consistently receive more visits from 29
30 tion frequencies, which are denoted in bold and represented by birds than plant species with small stamens, irrespective of 30
31 histograms above and right. their relative abundance in the community. 31
32 32
33 33
a b
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 Figure 3 There is a positive size coupling relationship between bill and flower size. (a) The mean length of bills of observed visit- 49
50 ing flowering species is plotted against their stamen length. (b) Likewise, the mean stamen length of flowers visited by birds is plot- 50
51 ted against their bill length. Birds disproportionately visit flowers of similar size to their bills. Open circles represent single species. 51

1 cies with greater stamen lengths (T = 3.068, P = 0.037, and moths (Haber & Frankie 1989; Agosta & Janzen 1
2 r2 = 0.627). These results suggest that trait matching (i.e. 2005). Little is known about the role species traits play 2
3 flower and bill size) is a significant predictor of network in plant–bird pollinator network topology. However, 3
4 topology in plant–bird pollination networks. bill and flower size might also shape interaction pat- 4
5 terns within these communities (Maruyama et al. 2014; 5
6 Maglianesi et al. 2015a,b). 6
DISCUSSION
7 Birds in this study foraged disproportionately on 7
8 Observed interactions between birds and flowers 8
large, red or yellow flowers. Flower size is usually cor-
9 were non-random. The interaction network was hierar- 9
related with nectar volume (Thompson & Plowright
10 chically nested. Plant species with large, showy flowers 10
1980; Harder et al. 1985; Harder & Cruzan 1990; Fen-
11 received greater total visits from birds than plant species 11
ster et al. 2006), which, in turn, elicits higher visita-
12 with small, inconspicuous flowers. Significant pairwise 12
tion rates (Galen & Plowright 1985; Mitchell & Was-
13 size coupling was also observed, with birds foraging 13
er 1992). By selectively foraging on large flowers, birds
14 disproportionately on flowers that were similar in size to 14
might reduce the number of visits needed to meet their
15 their bills. Overall, results suggest that trait matching is 15
energy expenditure. Alternatively, species with large
16 an important predictor of plant–bird pollination network 16
flowers might receive more visits if they were sim-
17 topology. 17
ply more abundant in the community. However, in this
18 Many ecological networks in nature are nested. In 18
study, large-flowered species received more visits than
19 nested networks, the partners of specialists form proper 19
small-flowered species regardless of their abundance.
20 subsets of the partners of generalists. Animal food webs 20
21 are often nested (i.e. “hierarchical trophic guilds” Paine Size coupling is typically interpreted as a determin- 21
22 1966; Pauly et al. 1998; Frederiksen et al. 2006). Most istic process whereby species actively seek out partners 22
23 plant–animal interaction networks are also nested (Bas- that have matching traits. Alternatively, it could also be 23
24 compte et al. 2003; Bascompte & Jordano 2013). The viewed as a passive process whereby birds forage ran- 24
25 apparent ubiquity of nestedness among communities domly on flowers that their bill morphology enables 25
26 suggests that it might be a common attribute of ecolog- them to. For example, Burns (2013) found that a simple 26
27 ical networks. However, it is unclear whether the mech- null model that included gape limitation predicted the 27
28 anisms underpinning nestedness and other attributes topology of a fruit–frugivore network at the same study 28
29 of network topology are consistent across community site. In other words, although gape width constrained 29
30 types. which fruits birds ate, birds foraged otherwise random- 30
31 ly on the fruits available to them. This study found that 31
Morphological constraints (i.e. “forbidden links”)
32 birds interact similarly with flowering plants. 32
and size coupling between interacting species pairs are
33 Bird-flowers are typically rare on islands (Crawford 33
often associated with network topology (Krishna et al.
34 et al. 2011, and references therein). While large, col- 34
2008; Olesen et al. 2010; Sankamethawee et al. 2011;
35 orful, tubular flowers are commonplace on continents, 35
Zhang et al. 2015). For example, body size limits the
36 flowers in Hawaii (Carlquist 1974), Juan Fernandez Is- 36
number of prey species that predators can exploit in
37 lands (Bernadello et al. 2001) and New Zealand (God- 37
predator–prey webs (Cohen et al. 1993; Memmott et al.
38 ley 1979; Lloyd 1985; Webb & Kelly 1993) are pre- 38
2000; Woodward & Hildrew 2002; Emmerson & Raffa-
39 dominantly small, dull and shallow. However, birds are 39
elli 2004; Woodward et al. 2005; Owen-Smith & Mills
40 often dominant vertebrates on islands (Holdaway 1989) 40
2008). Similarly, “gape limitation” has been invoked to
41 and are unappreciated pollinators (Olesen & Valido 41
explain the phenotypic matching between fruits and fru-
42 2004). Results from this study indicate that the dearth of 42
43
givores. Much like body size in animal predators, gape 43
width restricts the size of fruits that can be consumed bird-flowers in New Zealand does not result from a lack
44 of ecological selection by birds. 44
45
by frugivores. This creates a nested hierarchy that is or- 45
dered by gape width and fruit size (Wheelwright 1985; Flower color in this study was qualitatively described
46 46
Rey et al. 1997; Burns 2006, 2013). Size coupling has as it appears to the human eye (see Table 1). Other stud-
47 47
also been documented in a variety of plant-pollinator ies have utilized more quantitative methods of mea-
48 48
groups, including solitary bees (Armbruster & Guinn suring plant pigments (see Fadzley et al. 2009). Future
49 49
1989), bumblebees (Harder 1985), long-proboscid flies work exploring the relationship between flower color
50 50
(Goldblatt & Manning 2000), butterflies (Corbet 2000) and pollinator habits should employ such methods.
51 51

1 Overall, this study documented a nested plant–bird Carlquist S (1974). Island Biology. Columbia Universi- 1
2 pollination network in New Zealand. Network structure ty Press, New York. 2
3 was associated with species traits. We observed signif- Cohen JE, Pimm SL, Yodzis P, Saldara J (1993). Body 3
4 icant size coupling between birds and flowers. There- sizes of animal predators and animal prey in food 4
5 fore, phenotypic trait matching appears to be an import- webs. Journal of Animal Ecology 64, 62–78. 5
6 ant driver of nested bird–plant interactions. 6
Corbet SA (2000). Butterfly nectaring flowers: Butterfly
7 7
morphology and flower form. Entomologia Experi-
8 8
REFERENCES mentalis et Applicata 96, 289–98.
9 9
10 Agosta SJ, Janzen DH (2005). Body size distributions Crawford DJ, Anderson GJ, Bernardello G (2011). The 10
11 of large Costa Rican dry forest moths and the under- reproductive biology of islands plants. In: Bramwell 11
12 lying relationship between plant and pollinator mor- D, Caujapé-Castells, eds. The Biology of Island Flor- 12
13 phology. Oikos 108, 183–93. as. Cambridge University Press, Cambridge, pp. 11– 13
14 36. 14
Anderson SH (2003). The relative importance of birds
15 and insects as pollinators of the New Zealand flora. Dormann CF, Fründ J, Blüthgen N, Gruber B (2008). 15
16 New Zealand Journal of Ecology 27, 83–94. Indices, graphs and null models: Analyzing bipartite 16
17 ecological networks. The Open Ecological Journal 2, 17
Armbruster WS, Guinn DA (1989). The solitary bee
18 7–24. 18
fauna (Hymenoptera: Apoidea) of interior and arc-
19 tic Alaska: Flower associations, habitat use, and phe- Emmerson MC, Raffaelli D (2004). Predator–prey body 19
20 nology. Journal of Kansas Entomological Society 62, size, interaction strength and the stability of a real 20
21 468–83. food web. Journal of Animal Ecology 73, 399–409. 21
22 Fadzly N, Jack C, Schaefer HM, Burns KC (2009). On- 22
Barrett SCH, Emerson B, Mallet J (1996). The repro-
23 togenetic colour changes in an insular tree species: 23
ductive biology and genetics of island plants [and
24 signalling to extinct browsing birds? New Phytologist 24
discussion]. Philosophical Transactions of the Royal
25 184, 495–501. 25
Society of London B: Biological Sciences 315, 725–
26 26
33. Fenster CB, Cheely G, Dudash MR, Reynolds RJ (2006).
27 27
Bascompte J (2009). Disentangling the web of life. Sci- Nectar reward and advertisement in humming-
28 28
ence 24, 416–9. bird-pollinated Silene virginica (Caryophyllaceae).
29 29
Bascompte J, Jordano P (2013). Mutualistic Networks. American Journal of Botany 93, 1800–7.
30 30
31 Princeton University Press, Princeton. Frederiksen M, Edwards M, Richardson AJ, Halliday 31
32 Bascompte J, Jordano P, Melián CJ, Olesen JM (2003). NC, Wanless S (2006). From plankton to top preda- 32
33 The nested assembly of plant–animal mutualistic net- tors: Bottom–up control of a marine food web across 33
34 works. Proceedings of the National Academy of Sci- four trophic levels. Journal of Animal Ecology 75, 34
35 ences 100, 9383–7. 1259–68. 35
36 Bell BD (2015). Temporal changes in birds and bird Galen C, Plowright RC (1985). The effects of nectar 36
37 song detected in Zealandia sanctuary, Wellington, level and flower development on pollen carry-over in 37
38 New Zealand, over 2011–2015. Notornis 62, 173–83. inflorescences of fireweed (Epilobium angustifolium) 38
39 (Onagraceae). Canadian Journal of Botany 63, 488– 39
Bernardello G, Anderson GJ, Stuessy TF, Crawford DJ
40 91. 40
(2001). A survey of floral traits, breeding systems,
41 floral visitors, and pollination systems of the angio- Godley EJ (1979). Flower biology in New Zealand. New 41
42 sperms of the Juan Fernandez Islands (Chile). The Zealand Journal of Botany 17, 441–66. 42
43 Botanical Review 67, 255–308. Goldblatt P, Manning JC (2000). The long-proboscid fly 43
44 pollination system in southern Africa. Annals of Mis- 44
Brualdi RA, Sanderson JG (1999). Nested species sub-
45 souri Botanical Garden 87, 146–70. 45
sets, gaps, and discrepancy. Oecologia 119, 256–64.
46 Gotelli NJ (2000). Null model analysis of species co-oc- 46
Burns KC (2006). A simple null model predicts fruit–
47 currence patterns. Ecology 81, 2606–21. 47
frugivore interactions in a temperate rainforest. Oikos
48 48
121, 761–71. Gotelli NJ (2001). Research frontiers in null model anal-
49 49
Burns KC (2013). What causes size-coupling in fruit– ysis. Global Ecology and Biogeography Letters 10,
50 50
frugivore interaction webs? Ecology 94, 295–300. 337–43.
51 51

1 Greve M, Chown SL (2006). Endemicity biases nested- generality and body sizes in a natural food web. Jour- 1
2 ness metrics: A demonstration, explanation and solu- nal of Animal Ecology 69, 1–15. 2
3 tion. Ecography 29, 347–56. Mitchell RJ, Waser NM (1992). Adaptive significance of 3
4 Haber WA, Frankie GW (1989). A tropical hawkmoth Ipomopsis aggregata nectar production: Pollination 4
5 community: Costa Rican dry forest Sphingidae. Bio- success of single flowers. Ecology 73, 633–8. 5
6 tropica 21, 155–72. Olesen JM, Valido A (2004). Lizards and birds as gen- 6
7 eralized pollinators and seed dispersers of islands 7
Harder LD (1985). Morphology as a predictor of flower
8 plants. In: Fernández-palacios JM, Morici C, eds. 8
choice by bumble bees. Ecology 66, 198–210.
9 Ecología Insular/Island Ecology. Asociation Espano- 9
10 Harder LD, Cruzan MB (1990). An evaluation on the 10
la de Ecología Terrestre (AEET), Cabildo Insular de
11 physiological and evolutionary influences of inflo- 11
la Palma, pp. 229–49.
12 rescence size and flower depth on nectar production. 12
Functional Ecology 4, 559–72. Olesen JM, Bascompte J, Dupont YL, Elberling H, Ras-
13 mussen C, Jordano P (2010). Missing and forbid- 13
14 Holdaway RN (1989). New Zealand’s pre-human avi- 14
den links in mutualist networks. Proceedings of the
15 fauna and its vulnerability. New Zealand Journal of 15
Royal Society of London B: Biological Sciences 278,
16 Ecology 12, 11–25. rspb20101371. 16
17 Ibanez S (2012). Optimizing size thresholds in a plant– Owen-Smith N, Mills MGL (2008). Predator–prey size 17
18 pollinator interaction web: Towards a mechanis- relationships in an african large-mammal food web. 18
19 tic understanding of ecological networks. Oecologia Journal of Animal Ecology 77, 173–83. 19
20 170, 233–42. 20
Paine RT. (1966). Food web complexity and species di-
21 Innes J, Kelly D, Overton JM, Gillies C (2010). Preda- 21
versity. The American Naturalist 100, 65–75.
22 tion and other factors currently limiting New Zealand 22
23 Pauly D, Christensen V, Dalsgaard J, Froese R, Torres 23
forest birds. New Zealand Journal of Ecology 34, 86– F (1998). Fishing down marine food webs. Science
24 114. 24
25
279, 860–3. 25
Krishna A, Guimaraes PR Jr, Jordano P, Bascompte J R Core Team (2017). R: A Language and Environment
26 26
(2008). A neutral-niche theory of nestedness in mutu- for Statistical Computing. R Foundation for Statisti-
27 27
alistic networks. Oikos 117, 1609–18. cal Computing, Vienna, Austria:
28 28
29
Levin DA (1978). The evolution of isolating mecha- Rey PJ, Gutiérrez JE, Alcántara J, Valera F (1997). Fruit 29
30
nisms in flowering plants. Evolutionary Biology 11, size in wild olives: Implications for avian seed dis- 30
31
185–317. persal. Functional Ecology 11, 611–8. 31
32 Lloyd DG (1985). Progress in understanding the natural Sanderson JG, Pimm SL (2015). Patterns in Nature: 32
33 history of New Zealand plants. New Zealand Journal The Analysis of Species Co-occurrences. University 33
34 of Botany 23, 707–22. of Chicago Press, Chicago, IL, USA. 34
35 Maglianesi MA, Böhning-Gaese K, Schleuning M Sankamethawee W, Pierce AJ, Gale GA, Hardest BD 35
36 (2015a). Different foraging preferences of humming- (2011). Plant–frugivore interactions in an intact trop- 36
37 birds on artificial and natural flowers reveal mecha- ical forest in north-east Thailand. Integrative Zoology 37
38 nisms structuring plant–pollinator interactions. Jour- 6, 195–212. 38
39 nal of Animal Ecology 84, 655–64. Snow B, Snow D (1988). Birds and Berries. Poyser, 39
40 Maglianesi MA, Blüthgen N, Böhning-Gaese K, Staffordshire, UK. 40
41 Schleuning M (2015b). Functional structure and spe- Stang M, Klinkhammer GL, van der Meijden E (2006). 41
42 cialization in three tropical plant–hummingbird in- Size constraints and flower abundance determine the 42
43 teraction networks across an elevational gradient in number of interactions in a plant–flower visitor web. 43
44 Costa Rica. Ecography 38, 1119–28. Oikos 112, 111–21. 44
45 Maruyama PK, Vizentin-Bugoni J, Oliviera GM, Oliv- 45
Stang M, Klinkhammer GL, van der Meijen E (2007).
46 iera PE, Dalsgaard B (2014). Morphological and spa- 46
Asymmetric specialization and extinction risk in
47 cio-temporal mismatches shape a neotropical savanna 47
plant–flower visitor webs: A matter of morphology or
48 plant–hummingbird network. Biotropica 46, 740–7. 48
abundance? Oecologia 151, 442–53.
49 49
Memmott J, Matinez ND, Cohen JE (2000). Predators, Stang M, Klinkhammer GL, Waser NM, Stang I, van
50 50
parasitoids and pathogens: species richness, trophic der Meijden E (2009). Size-specific interaction pat-
51 51

1 terns and size matching in a plant–pollinator interac- Wheelwright, NT (1985). Fruit-size, gape width, and the 1
2 tion web. Annals of Botany 103, 1459–69. diets of fruit-eating birds. Ecology 66, 808–18. 2
3 Stone L, Roberts A (1990). The checkerboard score and Wilson, K-J (2004). Flight of the Huia. Canterbury Uni- 3
4 species distributions. Oecologia 85, 74–9. versity Press, Christchurch, New Zealand. 4
5 5
Thompson JD, Plowright RC (1980). Pollen carryover, Woodward G, Hildrew HG (2002). Body-size determi-
6 6
nectar rewards, and pollinator behavior with special nants of niche overlap and intraguild predation with-
7 7
reference to Diervilla ionicera. Oecologia 46, 68–74. in a complex food web. Journal of Animal Ecology
8 8
Tylianakis JM, Tscharntke T, Lewis OT (2007). Habitat 71, 1063–74.
9 9
10 modification alters the structure of tropical host–par- Woodward G, Ebenman B, Emmerson M et al. (2005). 10
11 asitoid food webs. Nature 455, 202–5. Body size in ecological networks. Trends in Ecology 11
12 Ulrich W, Gotelli NJ (2007). Null model analysis of and Evolution 20, 402–9. 12
13 species nestedness patterns. Ecology 88, 1824–31. Zhang H, Wang Z , Zeng Q , Chang G, Wang Z, Zhang 13
14 Webb CJ, Kelly D (1993). The reproductive biology of Z (2015). Mutualistic and predatory interactions are 14
15 the New Zealand flora. Trends in Ecology & Evolu- driven by rodent body size and seed traits in a ro- 15
16 tion 8, 442–7. dent-seed system in warm-temperate forest in north- 16
17 ern China. Wildlife Research 42, 149–57. 17
18 18
19 19
20 20
22 Biddick M, Burns KC (2018). Phenotypic trait matching predicts the topology of an insular plant–bird pollination 22
23 network. Integrative Zoology 13, 339–47. 23
24 24
25 25
26 26
27 27
28 28
29 29
30 30
31 31
32 32
33 33
34 34
35 35
36 36
37 37
38 38
39 39
40 40
41 41
42 42
43 43
44 44
45 45
46 46
47 47
48 48
49 49
50 50
51 51


Integrative Zoology 2018 Symposium Plant Animal Interaction

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Integrative Zoology 2018 Symposium Plant Animal Interaction

Uploaded by

Copyright:

Available Formats

Volume 13 Issue 3 May 2018 The official journal of the International Society of Zoological Sciences

Original Articles 297 Factors shaping life history traits of two

Volume 13 Issue 3 May 2018

Special issue: Plant–Animal Interactions

For full journal content,

Editorial Advisory Board

COPYRIGHT AND COPYING

© 2018 International Society of Zoological Sciences, Institute of Zoology/ 225

226 © 2018 International Society of Zoological Sciences, Institute of Zoology/

1 REFERENCES Dittel JW, Vander Wall SB (2018). Effects of rodent 1

© 2018 International Society of Zoological Sciences, Institute of Zoology/ 227

228 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 229

230 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 231

232 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 233

234 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 235

236 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 237

238 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 239

240 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 241

1 ferences between treatments (C, T1, T2 and NC) were RESULTS 1

242 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 243

244 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 245

246 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 247

248 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 249

250 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 251

1 INTRODUCTION Robertson et al. 1990; Yi & Zhang 2008; Sage et al. 1

252 © 2017 International Society of Zoological Sciences, Institute of Zoology/

1 oaks. minating (Smallwood et al. 2001). The other 3 species 1

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 253

254 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 255

1 For control acorns, seed mass was a significant pos- 70 1

Percent cotyledon loss

in biomass (ADB) of the control acorns of

256 © 2017 International Society of Zoological Sciences, Institute of Zoology/

Percentage of seeds surviving damage

17 Figure 3 The survival of each species in 9

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 257

258 © 2017 International Society of Zoological Sciences, Institute of Zoology/

Mean ADB ± SE (mg/day)

Mean ADB ± SE (mg/day)

in biomass (ADBs; ± SE) for the 6 species 14

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 259

20 Figure 6 The mean seed size of the 6 spe- 0.8

260 © 2017 International Society of Zoological Sciences, Institute of Zoology/

Relative survival (treatment/control)

Relative survival (treatment/control)

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 261

262 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 263

264 © 2017 International Society of Zoological Sciences, Institute of Zoology/

© 2017 International Society of Zoological Sciences, Institute of Zoology/ 265

266 © 2017 International Society of Zoological Sciences, Institute of Zoology/