You are on page 1of 1

PHYLOGENETIC

ANCOVA:
ASSESING THE EFFECTS OF MULTIPLE VARIABLES IN THE
EVOLUTION OF CONTINUOUS EVOLVING TRAITS

Jesualdo A. Fuentes1 *, Elizabeth Housworth1, Annie Murphy2, Ashley Weber3, Emlia P. Mar ns1

Department of Biology, Indiana University Bloomington; 2 Department of Mathema cs, Clarkson University; 3 Department of Mathema cs, University of Michigan
INTRODUCTION

The inuence of factors (e.g. environment, diet, behavior) on


traits is an important issue in evolu onary biology; however,
these inuences rarely happen in isola on.
Phylogene c compara ve methods model the evolu onary
rela onship between two con nuous traits or between one
categorical factor and one con nuous trait explicitly. But what
do we do when we want to combine categorical and
con nuous eects?
Example: Are feeding preferences associated with an
evolu onary shi in skull robustness in extant canids?
Robustness can be studied through skull width, but to do so we
need to account for allometric rela onships (Fig.1).

Model ng (Butler et al. 2000, Lavin et al. 2008):

Y = 0 + 1X + Z + 2XZ +
~ N (0 , V)

V is the phylogene c covariance matrix, which can


incorporate dierent evolu onary processes
(e.g. BM,

OrnsteinUhlenbeck: OU).
Model comparison in terms of specic hypotheses;
parameter es mates are condi onal on the hypotheses.

Phenotypic radia on (OMeara et al. 2006): A er a key


innova on, shi s may occur in both the means and the
variances (Fig.3).

Short-eared fox
Crab-eating fox
Sechuran fox
Hoary fox
Darwin's fox
Pampas fox
Culpeo
Chilla
Maned wolf
Bush dog
Grey wolf
Coyote
Golden jackal
Ethiopian wolf
African wild dog
Dhole
Black-backed jackal
Side-striped jackal
Indian fox
Fennec fox
Blandford's fox
Pale fox
Cape fox
Corsac fox
Rppell's fox
Red fox
Tibetan fox
Artic fox
Kit fox
Swift fox
Bat-eared fox
Raccoon dog
Gray fox

NEW QUESTIONS

2A

2B

Adapta oniner a (Hansen 1997, Hansen et al. 2008):


Eect of a par cular historical context:

Y = 0 + 1C + Z +
~ N (0 , V)
C is the weighted, summed, me each species has
evolved under a par cular diet regime:

2B

2A

Width

Assessing the eect of categorical and


con nuous traits simultaneously, considering
causal rela onships, allowing for unequal
phenotypic variance, and accoun ng for
phylogene c uncertainty:

3.
4.

Iden fy likely star ng points and cri cal


ques ons for the par cular case.
Fit ini al model for means, with dierent
evolu onary processes.
Compare one vs. mul ple variances.
Rees mate means with the new variance
structure.
Repeat and iterate the process un l
convergence is achieved.

= 0 + 1X + Z

~ N (0 , V)

Issues: Data are not independent, variance might dier between feeding
preferences, not all parameters are needed to explain the data.

Predictors

Nonphylogene c transforma ons: Indices and residuals from


linear regressions (Fig.2).
Simula ons (Garland et al. 1993): a) Simulate the evolu on of
cranial width along a phylogeny under a par cular model (e.g.
Brownian mo on: BM). b) Compute Fra o for each simula on
to obtain empirical null distribu ons. c) Contrast null
distribu on with real data for hypothesis tes ng.
Phylogene c transforma ons (Garland et al. 1992, Revell
2009): a) Phylogene c regression (Independent contrasts or
generalized least squares: GLS). b) Use transformed data in
compara ve analyses.

ANCOVA ng

Op miza on

X vs C
Standard model vs.
adapta oniner a
Do diet values or
selec ve regimes
predict skull
robustness be er?

Evolu onary
process

Phenotypic
radia on

Phylogene c
uncertainty

Full model

Null model
Single
mean (1)

Forward
Backward

Intercepts (2)
Slopes (2)
Interac ons (1)

How many parameters do we need to


explain the evolu onary eects on
skull robustness?

Methods
Powells algorithm
MCMC
Grid search
Sensi vity simula ons
What parameter
values t the skull
data the best?

TIPS
BM
OU

Vij =

Vij = 2ta
Vij = 2exp[ tij ]

How the data behave under


dierent evolu onary
models?

2
2A2B

Do we need a
phenotypic variance
for each diet regime?

Bayesian extension
Distribu
on of trees,
weighted by the
probability of each
tree being correct

What happens if we
have more than one
canid phylogeny?

Width/Length
r=WidthLength*b

Width

2.

5.

FORMER APPROACHES

1.

Length

Y = 0 + 1X + Z +
~ N (0 , V)

Procedure:

Fig.1. Le : Canid skull (Canis lupus) in


palatal view showing palatal width at the
level of the fourth premolar (Purple) and
cranial length (Red) (Photo: Joao Muoz
Durn). Right: Palatal width versus cranial
length; dierent feeding preferences are
depicted as indicated in the legend (Data
from MuozDurn & Fuentes 2012).

A GENERAL FRAMEWORK

Omnivory
Carnivory

Fig.3. Canid phylogeny with ancestral reconstruc on of feeding preferences. Cranial width may increase or decrease
() a er an evolu onary shi in the diet regime (A: Omnivory; B: Carnivory); but this shi could also be associated
with higher or lower variability in skull robustness (2).

Diet

Length

Fig.2. Le : Rela ve palatal width ra o compared between diet categories; this index can be used as a new variable in further
sta s cal analyses (e.g. ttest). Right: Regression of width on length; the residuals (r) can be used in a similar way as the indices.

FUTURE DIRECTIONS
Adding a second covariate (Hohenlohe & Arnold 2008).
Bayesian extension: Uncertainty in the ancestral state
reconstruc on (stochas c character mapping),
es ma on of the rate of adapta on (posterior
distribu on of the parameter).

AKNOWLEDGMENTS
This work was supported by fellowships to Jesualdo Fuentes from
COLCIENCIAS, The Indiana University Center for the Integra ve Study of
Animal Behavior (CISAB) and Department of Biology. Funding was also
provided by the NSFfunded Math REU program at Indiana University.
Data and photos: Joao MuozDurn.
Design and tle picture: Jorge Atehorta.

REFERENCES
Butler MA, Schoener TW, Losos JB. 2000. The rela onship between sexual size dimorphism and habitat use in
Greater An llean Anolis Lizards. Evolu on 54, 259272.
Garland T, Harvey PH, Ives AR.1992. Procedures for the analysis of compara ve data using phylogene cally
independent contrasts. Systema c Biology 41:1832.
Garland T, Dickerman AW, Janis CM, Jones JA.1993. Phylogene c analysis of covariance by computer simula on.
Systema c Biology 42, 265292.
Hansen TF. 1997. Stabilizing selec on and the compara ve analysis of adapta on. Evolu on 51, 13411351.
Hansen TF, Pienaar J, Orzack SH. 2008. A compara ve method for studying adapta on to a randomly evolving
environment. Evolu on 62, 19651977.
Lavin SR, Karasov WH, Ives AR, Middleton KM, Garland T.2008. Morphometrics of the avian small intes ne
compared with that of nonying mammals: A phylogene c approach. Physiological and Biochemical Zoology 81,
526550.
MuozDurn J, Fuentes JA. 2012. Evolucin de la socialidad, estrategias alimentarias y anatoma crneodental en
la subfamilia Caninae. Acta Biolgica Colombiana 17(1), 173200.
O'Meara BC, An C, Sanderson MJ, Wainwright PC. 2006. Tes ng for dierent rates of con nuous trait evolu on
using likelihood. Evolu on 60, 922933.
Revell LJ. 2009. Sizecorrec on and principal components for interspecic compara ve studies. Evolu on 63,
32583268.

jearfuen@indiana.edu

You might also like