MEC683.

fm Page 1199 Wednesday, June 30, 1999 3:54 PM

Molecular Ecology (1999) 8, 1199 –1209

Male reproductive success in a promiscuous mammal:

Blackwell Science, Ltd

behavioural estimates compared with genetic paternity

D . W. C O L T M A N , * D . R . B A N C RO F T , † A . R O B E R T S O N , ‡ J . A . S M I T H , † T. H . C L U T T O N - B RO C K ‡
and J . M . P E M B E R T O N *
*Institute of Cell, Animal and Population Biology, University of Edinburgh, Edinburgh EH9 3JT, UK, †Department of Genetics,
University of Cambridge, Downing Street, Cambridge CB2 3EH, UK, ‡Large Animal Research Group, Department of Zoology,
University of Cambridge, Downing Street, Cambridge CB2 3EJ, UK

Abstract
Molecular techniques have enabled behavioural ecologists to reassess mating systems
from a genetic perspective. Studies of paternity frequently reveal that mating behaviour
does not always reflect parentage, and may bring to light alternative mating tactics. Here
we present a comparison of behavioural and genetic measures of male reproductive suc-
cess in a mammalian mating system in which both sexes are highly promiscuous. Rather
than having a stable harem social structure, Soay rams (Ovis aries) on the island of Hirta,
St Kilda, UK usually consort with individual oestrous ewes sequentially. Not all matings
occur between consort pairs, however, and ewes have been seen to mate with up to 10
different rams on the same day. Using locus-specific polymorphism at five protein and 10
microsatellite DNA loci, we determined paternity for 236 lambs born into three cohorts,
and compared paternity with estimates of mating success derived from more than one
census of rutting behaviour. The correlation between the number of ewes with which
each ram was witnessed in consort and the number of paternities assigned was positive
and statistically significant, and rams that were observed in consort with a ewe were 18
times more likely to have sired her offspring than other candidate rams. However, most
lambs (73%) were not sired by a ram seen in consort with the oestrous mother. Many
juveniles, yearlings and some adult rams were rarely seen in consort with ewes, yet were
assigned a significant number of paternities. These results suggest that mating tactics
differ between age groups, and that alternative mating strategies among adults that do
not involve forming consorts with many females also confer mating success. For these
reasons, census-based observations of consort associations between individuals cannot
be used to accurately estimate individual male reproductive success in this population.
Keywords: mating success, mating systems, microsatellites, Ovis aries, protein polymorphism,
Soay sheep

Received 2 November 1998; revision received 17 February 1999; accepted 25 February 1999

Introduction is now commonplace to distinguish between social and

genetic mating systems (Hughes 1998). From the social
Mating systems describe the ways in which individuals
point of view, mating systems describe different forms of
of both sexes interact for the purpose of reproduction,
mate guarding or pair bonding, which can been seen as
and are best viewed as the consequence of the repro-
adaptations to the need for uni- vs. bi-parental care, and
ductive strategies of individuals rather than population
the spatial and temporal distribution of members of the
or species characteristics (Clutton-Brock 1989). Considerable
limiting sex (Emlen & Oring 1977). Genetic mating systems
progress has been made in the study of mating systems at
often differ, because many animals that form long-term
the level of the individual in the DNA profiling era, and it
pair bonds (social monogamy) produce extra-pair offspring
Correspondence: D. Coltman. Fax: 0131-667-3210; E-mail: (genetic polygamy). Studies employing molecular markers
David.Coltman@ed.ac.uk have frequently demonstrated that in many cases levels

© 1999 Blackwell Science Ltd

MEC683.fm Page 1200 Wednesday, June 30, 1999 3:54 PM
1200 D . W. C O L T M A N E T A L .
of extra-pair parentage cannot accurately be estim-
ated from behavioural observation alone (Hughes 1998).
Molecular methods, therefore, offer the opportunity to
enrich our understanding of mating systems, and may
suggest productive new avenues of behavioural and
ecological research.
In polygynous animals, male mating success has tradi-
tionally been measured from the time during which a
male has exclusive access to an oestrous female, harem
size, or from the number of observed copulations. Mat-
ing success can be measured accurately in this fashion if
females are not promiscuous (i.e. if females mate with
only one male) and associations between males and
females are stable. However, assessing female promiscu-
ity from behavioural observation is problematic due to
the difficulties in observing covert mating behaviour (e.g.
extra-pair copulations and ‘sneaky’ male strategies) and
nocturnal behaviour. For these reasons, it is necessary to
use genetic markers to assess the extent to which beha-
vioural measures of male mating success reflect genetic
parentage. Previous studies of polygynous mammals
have revealed widely different results. In some cases
dominance rank or copulation frequency reflects parent-
age accurately (e.g. lions, Pantherus leo (Gilbert et al. 1991);
red deer, Cervus elaphus (Pemberton et al. 1992); man-
drills, Mandrillus sphinx (Dixson et al. 1993); baboons,
Papio cynocephalus (Altmann et al. 1996)). In these examples,
males usually guard groups of females and there may be
relatively stable dominance hierarchies among males.
In other species in which associations between males
and females are brief and transient, or there are more
frequent shifts in dominance rank, there may be a lower
correlation (grey seals, Halichoerus grypus (Amos et al.
1993); Japanese macaques, Macaca fusculata (Inoue et al.
1993); pigtail macaques, M. nemestrina (Gust et al. 1996);
sooty mangabeys, Cercocebus torquatus (Gust et al. 1998)).
Therefore, the use of behavioural vs. genetic methods
for assessing male mating success depends both on the
study population, social structure and the ultimate study
aim. If behavioural measures do not reflect genealogical
relationships exactly, genetic methods provide the only
means to determine parentage for studies of heritability,
for example.
In this study we present a genetic analysis of paternity
in free-living Soay sheep (Ovis aries) on the island of
Hirta, UK. Both sexes of Soay sheep are socially pro-
miscuous, and associations between rams and ewes are
dynamic. Rams do not hold harems nor do they form
stable dominance hierarchies; rather they associate with
individual ewes, mating repeatedly before attempting to
locate the next oestrous ewe (Grubb 1974). Such associ-
ations involving a single ram and ewe pair are termed
‘consorts’ (Grubb 1974). Consorts involve the co-operation
of both sexes. The ram follows the ewe and mates with
her exclusively, and the ewe makes no attempt to join
other groups of sheep for the duration of oestrus. The vast
majority of consorts are formed with mature rams aged
2 years or more, yet there is extensive variation among
mature rams in the number of consorts they establish
during the mating season (Grubb & Jewell 1973). Consorts
are not necessarily exclusive to one ram throughout oestrus,
however, and a ewe may form successive consorts and
mate multiply with several rams on the same day. Grubb
& Jewell (1973) witnessed a Soay ewe mated 47 times by
10 different consorting rams in 8.5 h. Occasionally, a ewe
may also become involved in a ‘chase’, during which she
is pursued by one or more (commonly young) rams. Chases
frequently end with copulation of the ewe by one of the
pursuing rams. This high level of promiscuity is typical of
wild sheep; Hogg (1988) noted that over 48 h of oestrous,
a Rocky Mountain bighorn ewe (O. canadensis) may be
mated 30 times by one or more consorting rams, and a
further 30 times by ‘coursing’ males. ‘Coursing’ refers to
the tactic described in bighorn sheep whereby subordin-
ate rams fight dominant rams in consort with oestrous
ewes for temporary, and brief, mating access (Hogg 1984,
1988). DNA evidence indicates that coursing bighorn
rams sire equal numbers of progeny as do dominant rams
that defend ewe consorts (Hogg & Forbes 1997).
In this study we sought to determine the degree to
which behavioural observations reflect male mating suc-
cess determined genetically. Two hypotheses were tested.
First, we predicted a strong positive correlation between
behavioural success (estimated as the number of differ-
ent ewes a ram was sighted in consort with) and genetic
success (the number of lambs sired the following spring).
Second, we predicted that the genetically identified sire of
a given lamb was the ram most frequently observed in
consort with the ewe during the previous rut. Evidence
supporting these hypotheses would indicate that the
most successful mating tactic across all age classes was to
form consorts with as many ewes as possible. It would
also suggest that census-based observations could be
used to predict relative success among rams and the
paternity of individual lambs. However, male Soay sheep
display marked precocial sexual maturity and participate
in the rut as juveniles at only 7 months of age (Stevenson
& Bancroft 1995). Juveniles are frequently reproductively
successful (Pemberton et al. 1996), yet rarely hold ewes in
consorts (Grubb & Jewell 1973), perhaps due to their
small body size and lack of previous breeding experience.
It is thus possible that juveniles adopt a reproductive
strategy that does not involve forming consorts with
ewes. In this case, their reproductive success will be
underestimated by behavioural observation of consort
associations. We therefore also compared behavioural
and genetic measures of paternity between and within
age classes.
© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199 –1209
MEC683.fm Page 1201 Wednesday, June 30, 1999 3:54 PM

G E N E T I C P A T E R N I T Y I N A P R O M I S C U O U S M A M M A L 1201

Table 1 Characteristics of loci used in

Individuals Number of Expected paternity analysis of Soay lambs conceived
Locus typed alleles heterozygosity in the 1987–1989 ruts
Proteins
Adenosine deaminase 741 2 0.382
Beta haemoglobin 673 2 0.498
Glutamate oxaloacetate transaminase 750 2 0.327
Isocitrate dehydrogenase 696 3 0.497
Transferrin 827 7 0.783
Microsatellites
DRB 3* 744 8 0.825
MAF 18† 843 3 0.542
MAF 35‡ 841 4 0.563
MAF 45§ 833 6 0.743
MAF 65¶ 820 4 0.491
OarFCB 304** 806 4 0.614
OarCP 26†† 764 5 0.711
OarRM 106‡‡ 795 4 0.457
OarVH 34§§ 800 6 0.564
RBP 3¶¶ 828 3 0.616

*Schwaiger et al. (1993).

†Crawford et al. (1990).
‡Swarbrick et al. (1991).
§Swarbrick et al. (1992).
¶Buchanan et al. (1992).
**Buchanan & Crawford (1993).
††Ede et al. (1995).
‡‡Kossarek et al. (1993).
§§Pierson et al. (1993).
¶¶MacHugh (1997).

Materials and methods
The study population
Soay sheep on St Kilda have been the focus of biological
research since the early 1960s ( Jewell et al. 1974; Clutton-
Brock et al. 1991, 1992). An unmanaged population has
existed on the island of Hirta (638 ha), the largest island
in the St Kilda group, since their introduction in 1932
from the adjacent island of Soay (99 ha) where they have
existed for centuries. The analysis described here is based
on individually marked animals that reside in a 170-ha
study area centred on Village Bay. The Village Bay popula-
tion comprises approximately 30% of the island total. Since
1985, over 95% of lambs born in the study area have been
sampled for genetic analyses shortly after birth. Genetic
samples have also been obtained by catching sheep resident
in the Village Bay area each August, and by tranquilliz-
ing immigrant rams from other parts of the island during
the rut. All individuals sampled for genetic analyses have
been marked with ear tags for permanent identification.
the entire study area was censused four to five times per
day (approximately once every 2 h of daylight). During
each census, the entire study area was surveyed and the
identity of each oestrous ewe seen in association with one
or more attending rams was noted. The occurrence of
oestrus in the Soay ewe is marked by persistent associ-
ation with rams and the display of overt sexual behaviour
by rams, and is confirmed by her receptivity (Grubb
& Jewell 1973). A ram–ewe association was defined as a
‘consort’ if the pair were observed in close proximity
(typically within 5 m), but isolated from other sheep.
Consorts are further defined by the behaviour of the ram.
In a consort, the ram will follow the ewe if she moves
away, and will challenge another ram if he approaches
the pair. So defined, consorts exclude chases and associ-
ations of oestrous ewes with more than one ram during
the same census. A ‘consort observation’ is a recorded
observation of the identities of each member of a consort-
ing pair in a census. Particular consorts lasted from one to
eight consort observations.

Observations during the rut
The mating season of Soay sheep, the rut, lasts from late
October until early December. During the 1987–1989 ruts,
Paternity analysis
Eight hundred and forty-nine sheep were typed for five
polymorphic protein and 10 microsatellite DNA loci
(Table 1). The methods used to screen these polymorphisms

© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199–1209

MEC683.fm Page 1202 Wednesday, June 30, 1999 3:54 PM
1202 D . W. C O L T M A N E T A L .
and the genetic characteristics of these loci are described
in detail elsewhere (Bancroft 1993; Bancroft et al. 1995;
Smith 1996). No locus showed significant genotypic
disequilibrium, nor was there evidence for allelic dis-
equilibrium between any pair of loci (tests implemented
in genepop 3.0 (Raymond & Rousset 1995), data not
shown).
We assigned paternity using a likelihood-based approach
described in Marshall et al. (1998). The simulation pro-
gram within cervus 1.0 (Marshall et al. 1998) was used to
estimate the critical difference in log-likelihood scores
between the most likely and second most likely candid-
ate male for assignment of paternity at a level of 80%
confidence in each rut separately. For each rut, all sam-
pled rams, including lambs born in the spring preceding
the rut (henceforth referred to as juveniles), known to
be alive at the time of the rut were considered as candid-
ate males. We assumed that 80% of candidate males
were sampled, 95% of loci were typed, and a 1% rate of
typing error.
The analyses presented here are based on paternities
assigned at 80% statistical confidence. Although we may
expect one in five assignments to be incorrect at this
confidence level, incorrect assignments are unlikely to be
biased in relation to the behavioural data. Three times as
many assignments could be made at the 80% confidence
level than at 95%. The 80% confidence data therefore
offered greater statistical power for analyses in which
paternity was the response variable. We observed similar
qualitative results using paternities assigned at 95% con-
fidence to those reported here (data not shown).
Comparison of behavioural and genetic measures of male
reproductive success
We analysed the number of paternities assigned to each
candidate ram in relation to: (i) the number of different
ewes with which each candidate ram was observed in
consort; and (ii) the total number of consort observa-
tions in which a ram was observed. Because many rams
were considered in more than 1 year, and the distribution
of paternity among rams is known to differ between
years (Stevenson & Bancroft 1995; Pemberton et al. 1996),
correlation analyses are presented separately for each
year.
To ascertain whether consort observations may predict
the paternity of specific lambs, we restricted our ana-
lysis to lambs for which paternity had been determined
and whose mother was observed in consort with one or
more rams in the previous rut. Furthermore, we only
considered consort observations that occurred 150 ± 7
days prior to the date of birth of that lamb (gestation
period ± one standard deviation; Grubb & Jewell 1973).
This restricted the behavioural data to observations that
were likely to span the oestrous period during which con-
ception occurred. We then estimated the proportion of
lambs expected to be sired by a ram seen in consort with
the ewe, assuming that all sampled candidate rams were
equally likely to be the sire. Finally, to address the question
whether a sire seen in consort with the ewe was the most
frequently seen consort ram, we compared our observed
results with the estimated proportion of lambs expected
to be sired by the most commonly seen consort ram,
assuming all rams seen in consort with a ewe were equally
likely to be the sire.
Statistical analyses
Consort observations and paternities are discrete count
data, and thus were analysed using generalized linear
models (GLMs) with Poisson error structure and a log
link function (Crawley 1993). Initially tests were per-
formed using all data; however, many individuals were
present in more than 1 year. Therefore, to assure statist-
ical independence when comparing across years one
observation per individual was randomly selected for re-
analysis. The results of both tests are shown. Correlations
between nonnormal data were analysed using Spearman’s
rank method.
GLMs with binomial error structure were used to
analyse factors contributing to the probability that the sire
was either seen in consort at least once, or seen most
frequently in consort, with the mother. Variables tested
included: rut year (factor), total number of consort observa-
tions in which the mother was observed (continuous
variable), age class of the ram most frequently seen in
consort (factor: lamb/yearling vs. 2 year old plus), number
of consort observations with the most frequently seen
consort ram (continuous) and the age class of the sire
(factor: lamb/yearling vs. 2 year old plus). GLMs were
constructed and the significance of predictor variables
tested using standard techniques (Crawley 1993) in splus
(version 4.5, MathSoft).
Results
Consort observations
A total of 1141 consort observations were recorded dur-
ing rut censuses from 1987 to 1989 (Table 2). The distri-
butions of consort observations, and of the number of
different individuals observed in consort, were strongly
right skewed for both rams and ewes in all years (Fig. 1).
Many individuals were observed only once or twice in
consorts, while a smaller number of individuals were
observed in five or more consorts (Fig. 1). Significantly
more consort observations per marked ewe and ram
were recorded in 1987, and marked rams were seen in
© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199 –1209
MEC683.fm Page 1203 Wednesday, June 30, 1999 3:54 PM

G E N E T I C P A T E R N I T Y I N A P R O M I S C U O U S M A M M A L 1203

Table 2 Summary of consort observations recorded in the 1987, 1988 and 1989 ruts

Total consort Marked Consort Number of Marked Consort Number of

observations ewes seen observations rams seen in rams seen observations ewes seen in
Year recorded* in consorts marked ewe consort ewe in consorts marked ram consort ram

1987 504 132 3.63 ± 0.26 2.17 ± 0.11 51 8.96 ± 1.60 5.31 ± 0.76
1988 403 140 2.70 ± 0.17 2.06 ± 0.11 80 4.36 ± 0.64 3.36 ± 0.39
1989 234 79 2.86 ± 0.27 1.73 ± 0.12 31 5.58 ± 1.34 3.58 ± 0.69
Mean 380 117 3.09 ± 0.14 2.03 ± 0.07 54 6.04 ± 0.66 4.02 ± 0.34

All data
∆ deviance† 20.3 5.0 105.0 29.9
(P < 0.001) (P = 0.08) (P < 0.001) (P < 0.001)
Independent data‡§
∆ deviance† 21.2‡ 6.7‡ 97.6§ 36.9§
(P < 0.001) (P = 0.036) (P < 0.001) (P < 0.001)

*Includes observations of consorts in which one individual was unmarked.

†Differences between years were tested by a generalized linear model assuming Poisson error structure. The significance test is the
change in deviance explained by the inclusion of year, which is compared to χ2 with 2 d.f.
‡Some individuals were represented in more than 1 year. When this occurred, one observation from each ewe was selected at random to
ensure statistical independence.
§Some individuals were represented in more than 1 year. When this occurred, one observation from each ram was selected at random to
ensure statistical independence.

A B Fig. 1 Frequency distributions of the total

consorts observed and the number of
50 marked individuals observed in consorts
1987 for ewes (A and B) and rams (C and D) in
40 1988 the 1987, 1988 and 1989 ruts.
Number of ewes

1989

30

20

10

0
1 2 3 4 5 6 7 8 9 10+
Number of consort observations
per marked ewe
C D
1 2 3 4 5 6 7 8 9 10+
Number of different rams
per marked ewe seen in consort

30
Number of rams

20

10

0
1 2 3 4 5 6 7 8 9 10+ 1 2 3 4 5 6 7 8 9 10+
Number of consort observations Number of different ewes
per marked ram per marked ram seen in consort

© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199–1209

MEC683.fm Page 1204 Wednesday, June 30, 1999 3:54 PM

1204 D . W. C O L T M A N E T A L .

Table 3 Summary of paternity analysis of lambs conceived in the 1987, 1988 and 1989 ruts

Rut Village Bay Number of Number of lambs born in Number of paternities

year population size candidate rams the following spring assigned (% of tests)

1987 311 132 154 97 (63.0%)

1988 457 191 81 36 (44.4%)
1989 290 86 130 93 (71.5%)
Total 365 226 (61.9%)

consort with a greater number of different ewes in 1987, 1.0

on average (Table 2). This was due in part to a greater 1987, mean = 0.74, var = 1.40
censusing effort in 1987. However, fewer different 1988, mean = 0.19, var = 0.19
0.8 1989, mean = 1.52, var = 6.79
rams were seen in consort with each ewe in 1989 due to
their being fewer rams alive following a year of high

Frequency
0.6
male-biased over-winter mortality (Clutton-Brock et al.
1991).
The number of consort observations recorded for each 0.4
marked ram and the number of different ewes with
which each marked ram was seen in consort were highly 0.2
correlated in each year (ρ = 0.996, 0.998 and 0.997 in 1987,
1988 and 1989, respectively). Therefore, all subsequent
0.0
analyses of consort observations use the number of ewes 0 1 2 3 4 5 6 7 8+
seen in consort. Similar results would have been obtained Number of paternities per ram
using the number of consort observations in place of the
Fig. 2 Distributions of paternities assigned per ram, sired in the
number of ewes seen in consort. 1987, 1988 and 1989 ruts.

Paternity analysis
Sires for 226 of 365 sampled lambs were identified at 80%
statistical confidence over the three ruts investigated
(Table 3). Fewer lambs were born following the 1988
rut due to very high over-winter mortality in 1988–1989
(Clutton-Brock et al. 1991). The paternity assignment
rate differed significantly between years (χ2d.f.=2 = 15.67,
P < 0.001) with a lower success rate in the cohort con-
ceived during the 1988 rut. This was a consequence of the
large number of candidate rams in the 1988 rut. The dis-
tribution of paternities among all sampled rams was right
skewed in each year (Fig. 2), with many rams assigned

Table 4 Consorts observed and offspring

Rut Ram age Mean number of different Mean number of sired by age category for all marked rams
year class N* ewes seen in consort with (± SE) paternities assigned (± SE) in the 1987, 1988 and 1989 ruts. Differences
between age classes were tested by a gener-
1987 Adult (2+) 31 5.7 ± 1.2 1.4 ± 0.3 alized linear model (GLM) assuming Poisson
Yearling 36 2.3 ± 0.4 0.9 ± 0.2 error structure
Juvenile 65 0.2 ± 0.1 0.3 ± 0.1
∆ dev.† 323.2, P < 0.001 33.6, P < 0.001
1988 Adult (2+) 64 3.4 ± 0.5 0.3 ± 0.1
Yearling 61 0.8 ± 0.2 0.11 ± 0.04
Juvenile 66 0.05 ± 0.03 0.11 ± 0.04
∆ dev. 305.6, P < 0.001 11.4, P = 0.003
1989 Adult (2+) 36 2.0 ± 0.6 2.8 ± 0.6
Yearling 6 1.0 ± 0.8 0.8 ± 0.5
Juvenile 44 0.5 ± 0.2 0.6 ± 0.1
∆ dev. 39.8, P < 0.001 61.8, P < 0.001

* Total number of rams known to be alive and using the Village Bay study area, therefore
included in the paternity analysis.
† Significance tested by comparison of change in deviance with χ2d.f.=2.
SE, Standard error.

© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199 –1209

MEC683.fm Page 1205 Wednesday, June 30, 1999 3:54 PM

G E N E T I C P A T E R N I T Y I N A P R O M I S C U O U S M A M M A L 1205

Table 5 Total number of rams considered,

Juveniles Yearlings Adult rams Total ewes observed in consort, and paternities
assigned to rams of different age classes
N 175 (42.8%) 103 (25.2%) 131 (32%) 409 combined over the 1987, 1988, and 1989 ruts
Ewes observed in consort 36 (5.7%) 133 (21.0%) 464 (73.3%) 633
Paternities 56 (21.2%) 43 (16.3%) 165 (62.5%) 264

paternity of none or a single lamb, and only a few rams A 10

assigned many more (maxima of eight, two and 20 in ρ = 0.517, N = 132, P < 0.01
8
each respective year).

Lambs sired
6

Consorts, age and relative male reproductive success 4

In each year, rams 2 years of age and older were seen in 2

consort with significantly more ewes than yearlings and
0
juveniles (Table 4). They were also assigned the pater-
nity of significantly more lambs in every year (Table 4). 0 5 10 15 20
Yearlings did slightly better than juveniles, both in terms B 3
of total ewes observed in consort and the number of ρ = 0.172, N = 191, P < 0.05
patern-ities assigned. Over all three seasons combined,
2
Lambs sired

juveniles were greatly under-represented in the total

number of ewes observed in consorts (5.7%) compared
with their abundance (42.8% of rams considered) and 1
mating success (21.2% of all paternities; Table 5). Adult
rams accounted for a greater fraction of both consort-
0
ships observed (73.3%) and paternities (62.5%) than their
representation in the total sample (32% of all rams consid- 0 2 4 6 8 10 12 14 16
ered were adults; Table 5). The difference between age C 25
classes in the number of consortships observed and pater- ρ = 0.296, N = 86, P < 0.01
20
nities assigned was highly statistically significant
Lambs sired

(χ2 = 49.0, 2 d.f., P < 0.001). 15

Figure 3 illustrates the correlation between the number
of ewes each ram was sighted in consort with and the 10
number of paternities assigned to each ram in each rut 5
across all age classes. The correlation was significant and
positive in each year. However, in each year a single ram 0
with both relatively high numbers of ewes in consort 0 2 4 6 8 10 12 14 16
and assigned paternities appeared to have great leverage. Number of ewes sighted in consort with
This was the same adult ram in 1987 and 1989. When
Points represent:
these points were removed the correlations in 1987 and
1989 remained statistically significant (ρ = 0.503, n = 131,
P < 0.01; ρ = 0.151, n = 190, P > 0.05; and ρ = 0.264, n = 85, 1 2–5 6–10 11–15 16–20 21 or more
P < 0.05, respectively). Restricting the data to observa- individuals
tions of rams greater than 2 years of age or more revealed Fig. 3 Correlation of the number of ewes sighted in consort and
similar levels of correlation (ρ = 0.784, n = 31, P < 0.01; the number of paternities assigned to rams for the 1987 (A), 1988
ρ = 0.228, n = 64, P < 0.05; and ρ = 0.233, n = 36, P < 0.05, (B) and 1989 (C) ruts.
for 1987, 1988 and 1989, respectively). When data were
combined across years, there was a significant correlation ewes observed in consort therefore explained 22.4% of
between the number of ewes observed in consort and the variation in the number of paternities assigned per
the number of paternities assigned (all data: ρ = 0.277, marked ram (GLM assuming Poisson error structure, one
n = 408, P < 0.001; one observation per ram randomly observation per ram randomly selected, χ2 = 85.44, 1 d.f.,
selected: ρ = 0.306, n = 242, P < 0.001). The number of P < 0.001).

© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199–1209

MEC683.fm Page 1206 Wednesday, June 30, 1999 3:54 PM

1206 D . W. C O L T M A N E T A L .

Table 6 Observed and expected frequencies of consort observations of sires with mothers of lambs conceived in the 1987, 1988 and 1989
ruts

Sire seen with mother in at Sire most commonly seen consort ram
Lambs least one consort observation (proportion of times when sire seen in
Year considered (proportion of lambs) the majority of consort observations)

1987 59 19 (0.32) 10 (0.53)

1988 24 3 (0.12) 1 (0.33)
1989 45 13 (0.29) 7 (0.54)
Total 128 35 (0.27) 18 (0.51)
Expected probability 0.015* 0.49†
Expected number of 1.9 17.1
lambs
Significant difference χ2d.f.=1 = 34.4, P < 0.001 χ2d.f.=1 = 0.1, P = 0.8
from expectation

*The estimated probability of a ram seen in consort being the sire under the assumption that all candidate rams were equally likely to be
the sire. This was estimated as (mean numer of rams seen in consort per ewe)/(mean number of candidate rams) = (2.03/136).
†The estimated probability that the most commonly seen consort ram is the sire under the assumption that all rams seen in consort were
equally likely to be the sire. This was estimated as 1/(mean number of rams seen in consort per ewe) = (1/2.03).

In each year, the variance among rams in the number of

Frequency with which sire observed in consort

ewes with which they were seen in consort (σ2c) exceeded 1.0
the variance in the number of paternities assigned (σ2p) 8
(σ2c = 17.9, 7.9, and 7.1; σ2p = 1.40, 0.20, 6.79; for 1987, 1988 0.8 Juveniles and yearlings 13
and 1989, respectively). However, when scaled to the Rams (2 years and older)
mean, the levels of variation were similar (coefficients
0.6
of variation for consorts = 2.06, 2.04 and 2.34; for
paternities = 1.61, 2.35, and 1.74; for 1987, 1988 and 1989, 22
respectively). 0.4
15
12
0.2 7
Exact paternity and consorts 11 11
20 9
Paternity was determined for 128 lambs for which at least 0.0
one ram was seen in consort during the mother’s recept- 1 2 3 4 5
ive period (Table 6). In 35 cases (27%), the sire was seen Number of consort observations
in consort with the ewe. If all candidate rams were Fig. 4 Frequency with which the sire was seen in consort
equally likely to be the sire, the frequency of seeing the with the mother of a lamb of identified paternity in relation to
sire in consort with a ewe could be estimated as 0.015 the total number of consort observations in which the mother
(Table 6). A ram seen in consort with a given ewe was, was seen. Data are shown for cases in which the sire was a
therefore, 18 times more likely to be the sire than a male juvenile or yearling (filled columns) or an adult ram (open
columns). The numbers at the top of the columns represent the
randomly chosen from the pool of candidate males, aver-
sample size.
aged across years (Table 6). In 18 of these cases (51%), the
sire was seen in more consort observations with the ewe
than any other ram. This was no more frequent than influenced by two factors. These were the total number
would be expected if we were to assume that all rams of consort observations in which the ewe was seen
seen in consort with a ewe were equally likely to be the (χ2d.f.=1 = 24.95; P < 0.001), and whether the sire was a
father of her lamb (Table 6). Being in consort with a ewe mature ram (2 years or greater) or younger (χ2d.f.=1 = 10.48;
therefore makes a ram a more likely candidate for the P < 0.005; Fig. 4). This model explained 23.6% of the total
paternity of her lamb. However, the most commonly seen deviance (null deviance = 150.2 with 127 d.f.). If the sire
consort ram is not more likely to be the sire than any was an adult ram, the mean probability of seeing him in
other consort ram. consort with the ewe was greater than 0.8 if the ewe was
GLMs indicated that the probability of seeing the seen in at least four consort observations (Fig. 4). The sire
sire in consort with the ewe during the rut was positively was seen less than one-third of the time if fewer consort

© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199 –1209

MEC683.fm Page 1207 Wednesday, June 30, 1999 3:54 PM
G E N E T I C P A T E R N I T Y I N A P R O M I S C U O U S M A M M A L 1207
observations were recorded, or if the sire was a juvenile
or yearling. Juvenile and yearling sires were very rarely
seen in consort, regardless of the number of consort
observations in which the ewe was recorded (Fig. 4).
Discussion
Behavioural estimates of relative reproductive success in
Soay rams
Our data suggest that the number of ewes a ram is
seen in consort with is a weak predictor of relative male
reproductive success, particularly for young rams. Even
among adult rams, while those seen in consort with many
ewes tended to be assigned the paternity of more lambs,
this was not always the case. For example, in 1987 an
adult ram seen in consort with 17 different ewes was
assigned the paternity of only one lamb, yet an adult ram
seen in consort with only two ewes was assigned three
paternities in the same year. This discrepancy is partly
due to the fact that juveniles and yearlings gain some
paternities, yet rarely form consorts. However, there was
no great improvement in the relationships when the data
were restricted to adult rams. There are several possible
explanations for this.
First, juveniles, yearlings and some adults may adopt
mating tactics that do not involve forming consorts.
Rams that attempt to mate opportunistically or predomin-
ately following chases will have their reproductive out-
put underestimated based on the number of consorts
observed. It may be difficult for young rams to defend a
consort ewe from competitors due to their small body
size and limited previous breeding experience, as com-
petitive interactions between rams for access to oestrous
ewes involve physical conflict including nudging, butt-
ing and pushing (Grubb & Jewell 1973). Similarly, adult
rams of lower physical competitive ability may find that
there is a better fitness payoff for an opportunistic mat-
ing strategy, or searching more widely for unattended
oestrous ewes, than to attempt to hold ewes in exclusive
consorts. The fact that both the number of consorts and
paternities per ram increase from juveniles to yearlings
then to adults (Tables 4 and 5) suggests that increasing
body size and breeding experience favour a shift towards
the tactic of holding individual ewes in consort for most
rams.
Second, sperm competition may play a major role in
the Soay sheep mating system. Sheep have relatively
large testes, and Soay males have the largest testicular
to body weight ratio of all sheep breeds (Lincoln 1989).
Order, frequency and/or timing of copulation with respect
to ovulation and sperm depletion may be important
determinates of paternity which are not reflected in an
association index, such as the total number of ewes
© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199–1209
consorted with. The data collected for this study are
unfortunately insufficient to test these alternative ideas.
Estimating paternity exactly from behavioural data
The fact that rams seen in consorts were 18 times more
likely to be the sire than other candidate rams means that
observing associations between individuals provides some
useful information for predicting paternity. Furthermore,
with an increasing number of recorded consort observa-
tions, the probability of seeing the sire increases (Fig. 4).
On the surface, this suggests that given more intensive
observations, some useful predictions about paternity
could be made. For example, information about rams
seen in consort could be used to modify the probability
of paternity calculations using a Bayesian approach (e.g.
Adams et al. 1992). Rams seen in consort with the mother
of a given lamb might be considered to have a higher
prior probability of paternity than other candidate rams.
This information could improve the success rate in deter-
mining paternity. However, juvenile and yearling rams
are rarely seen in consorts and may sire many offspring
(see also Stevenson & Bancroft (1995)). The probability
that the sire was seen in consort only improves with an
increasing number of behavioural observations when the
sire was an adult ram (Fig. 4). When the sire was a
juvenile or yearling, he was highly unlikely to have been
witnessed in consort with the mother (four times in 58
paternities), whereas when the sire was an adult ram he
was sighted 44.3% of the time (31 of 70 paternities). Over-
all, the sire was seen in consort with the ewe on at least
one occasion in only 35 cases of 128 assigned paternities
(27.3%). The most frequently sighted consort ram was the
father in only 18 of 128 cases (14.1%).
Taken together, these results indicate that considerably
more detailed observational data would be needed to
make reliable predictions about paternity from beha-
viour. The census-based approach used here is a form
of scan sampling which may not efficiently record
behaviours (such as chases and brief nonconsort mating)
that may occur in a shorter period of time than the inter-
val between scans. However, collecting more detailed
observational data is not a trivial task. Continuous focal
watches on oestrous ewes would be required to record
the identity and behaviour of every possible candidate
male. Alternatively, focal watches on rams could be car-
ried out to better characterize the variation in reproduct-
ive behaviour and the payoff for different mating tactics
among individuals within and between age classes. Both
of these approaches would necessitate a limited sample
size of individuals that could be monitored. The advant-
age of the census-based approach used in this study was
in the number of individuals that could be monitored
given the practical limits to observer effort.
MEC683.fm Page 1208 Wednesday, June 30, 1999 3:54 PM
1208 D . W. C O L T M A N E T A L .
The low predictive value of the behavioural observa-
tions presented here (number of ewes in consort explain-
ing 22.4% of the variation in the number of lambs sired)
contrasts with the study of Pemberton et al. (1992) which
showed that around 90% of the variance in male repro-
ductive success in red deer could be predicted from
behavioural data. This disparity is probably due to the
different levels of female promiscuity in the two mating
systems, and the frequently brief mating associations that
occur in Soay sheep. The red deer harem structure is
relatively stable. Hinds copulate only once per oestrus,
thus hind–stag combinations can be accurately monitored
by censusing techniques. In comparison, Soay sheep are
highly promiscuous, with a rapid turnover of consort
pairs. Furthermore, not all rams adopt the consort-holding
mating strategy. In particular, juveniles and yearlings
rarely form consorts (26.7% of all observed consort pairs),
yet account for a greater fraction of all paternities than
their consortship would suggest (37.5%). This study indic-
ates that the measurement of male reproductive success is
unreliably predicted by behavioural observation in this
population, necessitating the use of genetic markers to
assess paternity. Such phenomena may be more wide-
spread among polygynous species that exhibit promiscuous
mating patterns or transient associations between mate
pairs than has been anticipated (e.g. grey seals, Halichoerus
grypus (Amos et al. 1995); harbour seals, Phoca vitulina
(Coltman et al. 1998); chimpanzees, Pan troglodytes (Morin
et al. 1994)).
Acknowledgements
We thank the National Trust for Scotland and Scottish Natural
Heritage for permission to work on St Kilda and the St Kilda
Detachment of the Royal Artillery and the Royal Corps of Trans-
port for logistic support. We are grateful to L. Birchenough,
R. Clarke, F. Gulland, A. Jarvis, D. Robertson, I. Stevenson and
many other volunteers for practical assistance. D. Green provided
instrumental assistance in the first years of rut observations and
tissue sampling. I. Stevenson, K. Wilson, David Westneat, Terry
Burke and one anonymous referee provided useful comments on
this manuscript. NERC, SERC, the Wellcome Trust, the Smith-
Kline Foundation, the Cambridge Philosophical Society and the
Mammal Conservation Trust provided financial support.
References
Adams WT, Griffin AR, Moran GF (1992) Using paternity ana-
lysis to measure effective pollen dispersal in plant populations.
American Naturalist, 140, 762 – 780.
Altmann J, Alberts SC, Haines SA et al. (1996) Behavior predicts
genetic structure in a wild primate group. Proceedings of the
National Academy of Sciences USA, 93, 5797–5801.
Amos W, Twiss S, Pomeroy P, Anderson S (1993) Male mating
success and paternity in the grey seal, Halichoerus grypus, a
study using DNA fingerprinting. Proceedings of the Royal Society
of London, Series B, 252, 199 –207.
Amos B, Twiss S, Pomeroy P, Anderson S (1995) Evidence for
mate fidelity in the grey seal. Science, 268, 1897–1899.
Bancroft DR (1993) Genetic variation and fitness in Soay sheep.
PhD Thesis, University of Cambridge.
Bancroft DR, Pemberton JM, King P (1995) Extensive protein and
microsatellite DNA variability in an isolated, cyclic ungulate
population. Heredity, 74, 326–336.
Buchanan FC, Crawford AM (1993) Ovine dinucleotide repeat
polymorphisms at the OarFCB11, OarFCB 128, OarFCB 193,
OarFCB 266, and OarFCB304 loci. Animal Genetics, 24, 145.
Buchanan FC, Swarbrick PA, Crawford AM (1992) Ovine
dinucleotide repeat polymorphism at the MAF65 locus. Animal
Genetics, 23, 85.
Clutton-Brock TH (1989) Mammalian mating systems. Proceed-
ings of the Royal Society of London, Series B, 236, 339 – 372.
Clutton-Brock TH, Price OF, Albon SD, Jewell PA (1991) Per-
sistent instability and population regulation in Soay sheep.
Journal of Animal Ecology, 60, 593–608.
Clutton-Brock TH, Price OF, Albon SD, Jewell PA (1992) Early
development and population fluctuations in Soay sheep. Journal
of Animal Ecology, 61, 381–396.
Coltman DW, Bowen WD, Wright JM (1998) Male mating
success in an aquatically mating pinniped, the harbour seal
(Phoca vitulina) determined by microsatellite DNA markers.
Molecular Ecology, 7, 627–638.
Crawford AM, Buchanan FC, Swarbrick PA (1990) Ovine
dinucleotide repeat polymorphism at the MAF18 locus. Animal
Genetics, 21, 433–434.
Crawley MJ (1993) GLIM for Ecologists. Blackwell, Oxford.
Dixson AF, Bossi T, Wickings EJ (1993) Male dominance and
genetically determined reproductive success in the mandrill
(Mandrillus sphinx). Primates, 34, 525 – 532.
Ede AJ, Pierson CA, Crawford AM (1994) Ovine microsatellites
at the OarCP9, OarCP16, OarCP20, OarCP21, OarCP23 and
OarCP26 loci. Animal Genetics, 26, 129 –130.
Emlen ST, Oring LW (1977) Ecology, sexual selection and the
evolution of mating systems. Science, 197, 215 – 223.
Gilbert DA, Packer C, Pusey JC, Stephens JC, O’Brien SJ (1991)
Analytical DNA fingerprinting in lions: parentage, genetic
diversity, and kinship. Journal of Heredity, 82, 378 – 386.
Grubb P (1974) The rut and behaviour of Soay rams. In: (eds
Jewell PA, Milner G, Boyd JM) Island Survivors, pp. 200 –235.
Althone Press, London.
Grubb P, Jewell PA (1973) The rut and the occurrence of oestrus
in the Soay sheep of St. Kilda. Journal of Reproduction and Fertility,
Suppl. 19, 491–502.
Gust DA, Gordon TP, Gergits WF, Casna NJ, Gould KG,
McClure HM (1996) Male dominance rank and offspring
initiated affiliative behaviors were not predictors of paternity
in a captive group of pigtail macaques (Macaca nemestrina).
Primates, 37, 271–278.
Gust DA, McCaster T, Gordon TP, Gergits WF, Casna NJ,
McClure HM (1998) Paternity in sooty mangabeys. Inter-
national Journal of Primatology, 19, 83 – 94.
Hogg JT (1984) Mating in bighorn sheep: multiple creative male
strategies. Science, 225, 526–529.
Hogg JT (1988) Copulatory tactics in relation to sperm competition
in Rocky Mountain sheep. Behavioral Ecology and Sociobiology,
22, 49–59.
Hogg JT, Forbes SH (1997) Mating in bighorn sheep: frequent
male reproduction via a high risk ‘unconventional’ tactic.
Behavioral Ecology and Sociobiology, 41, 33 – 48.
© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199 –1209
MEC683.fm Page 1209 Wednesday, June 30, 1999 3:54 PM
G E N E T I C P A T E R N I T Y I N A P R O M I S C U O U S M A M M A L 1209
Hughes C (1998) Integrating molecular techniques with field
methods in studies of social behavior: a revolution results.
Ecology, 79, 383 – 399.
Inoue M, Mitsunaga F, Nozaki M et al. (1993) Male dominance
rank and reproductive success in an enclosed group of Japanese
macaques with special reference to post-conception mating.
Primates, 34, 503 – 511.
Jewell PA, Milner C, Boyd JM (1974) Island Survivors: the Ecology
of the Soay Sheep of St Kilda. Althone Press, London.
Kossarek I, Su MX, Grosse WM et al. (1993) Bovine dinucleotide
repeat polymorphism RM106. Journal of Animal Science, 71,
3180.
Lincoln GA (1989) Seasonal cycles in testicular activity in
Mouflon, Soay sheep and domesticated breeds of sheep: breeding
seasons modified by domestication. Zoological Journal of the
Linnean Society, 95, 137–147.
MacHugh DE (1997) Molecular biogeography and genetic struc-
ture of domestic cattle. PhD Thesis, Trinity College, Dublin.
Marshall TC, Slate J, Kruuk LEB, Pemberton JM (1998) Statistical
confidence for likelihood-based paternity inference in natural
populations. Molecular Ecology, 7, 639–655.
Morin PA, Wallis J, Moore JJ, Woodruff DS (1994) Paternity
exclusion in a community of wild chimpanzees using hyper-
variable simple sequence repeats. Molecular Ecology, 3, 469 – 478.
Pemberton JM, Albon SD, Guinness FE, Clutton-Brock TH,
Dover GA (1992) Behavioral estimates of male mating success
tested by DNA fingerprinting in a polygynous mammal.
Behavioral Ecology, 3, 66 – 75.
Pemberton JM, Smith JA, Coulson TN et al. (1996) The mainte-
nance of genetic polymorphism in small island populations:
large mammals in the Hebrides. Philosophical Transactions of the
Royal Society of London, Series B, 351, 745–752.
Pierson CA, Hanrahan V, Ede AJ, Crawford AM (1993) Ovine
© 1999 Blackwell Science Ltd, Molecular Ecology, 8, 1199–1209
microsatellites at the OarVH34, OarVH41, OarVH58, OarVH61
and OarVH72 loci. Animal Genetics, 24, 224.
Raymond M, Rousset F (1995) genepop (Version 1.2) — popula-
tion genetics software for exact tests and ecumenicism. Journal
of Heredity, 86, 248–249.
Schwaiger FW, Buitcamp J, Weyers E, Epplen JT (1993) Typing of
artiodactyl MHC-DRB genes with the help of intronic simple
repeated DNA sequences. Molecular Ecology, 2, 55 – 59.
Smith JA (1996) Polymorphism, parasites and fitness in Soay
sheep. PhD Thesis, University of Cambridge.
Stevenson IR, Bancroft DR (1995) Fluctuating trade-offs favour
precocial maturity in male Soay sheep. Proceedings of the Royal
Society of London, Series B, 262, 267–275.
Swarbrick PA, Buchanan FC, Crawford AM (1991) Ovine
dinucleotide repeat polymorphism at the MAF35 locus. Animal
Genetics, 22, 369–370.
Swarbrick PA, Schmack AE, Crawford AM (1992) MAF45, a
highly polymorphic marker for the pseudoautosomal region
of the sheep genome, is not linked to the FecX′ (Inverdale)
gene. Genomics, 13, 849–851.
This research is part of a long-term collaborative study of the
ecology, life history and genetics of Soay sheep on St Kilda
involving Professor Tim Clutton-Brock and coworkers, includ-
ing Tony Robinson (Department of Zoology, Cambridge), Dr
Josephine Pemberton and coworkers, including Dave Coltman,
David Bancroft and Judith Smith (formerly Department of
Genetics, Cambridge and now ICAPB, Edinburgh), and several
other UK researchers. Dave Coltman brought this paper to
parturition (after an extended gestation) while a post-doctoral
fellow at Edinburgh.