The adaptive value of morphological, behavioural and life-history traits in reproductive female wolves



  1. Reproduction in social organisms is shaped by numerous morphological, behavioural and life-history traits such as body size, cooperative breeding and age of reproduction, respectively. Little is known, however, about the relative influence of these different types of traits on reproduction, particularly in the context of environmental conditions that determine their adaptive value.
  2. Here, we use 14 years of data from a long-term study of wolves (Canis lupus) in Yellowstone National Park, USA, to evaluate the relative effects of different traits and ecological factors on the reproductive performance (litter size and survival) of breeding females.
  3. At the individual level, litter size and survival improved with body mass and declined with age (c. 4–5 years). Grey-coloured females had more surviving pups than black females, which likely contributed to the maintenance of coat colour polymorphism in this system.
  4. The effect of pack size on reproductive performance was nonlinear as litter size peaked at eight wolves and then declined, and litter survival increased rapidly up to three wolves, beyond which it increased more gradually.
  5. At the population level, litter size and survival decreased with increasing wolf population size and canine distemper outbreaks.
  6. The relative influence of these different-level factors on wolf reproductive success followed individual > group > population. Body mass was the primary determinant of litter size, followed by pack size and population size. Body mass was also the main driver of litter survival, followed by pack size and disease. Reproductive gains because of larger body size and cooperative breeding may mitigate reproductive losses because of negative density dependence and disease.
  7. These findings highlight the adaptive value of large body size and sociality in promoting individual fitness in stochastic and competitive environments.


Knowledge about the adaptive value of traits is fundamental to understanding the biology of natural systems and anticipating species response to environmental change. A key challenge lies in understanding a trait's contribution to fitness relative to environmental conditions (e.g. competition and disease). Indeed, the extent to which a trait mitigates the impact of environmental stress on fitness is perhaps the most robust gauge of its adaptive value. Studies of reproductive success highlight several potentially adaptive traits including body size (Clutton-Brock 1988; Hamel et al. 2008), genetic heterozygosity (Slate et al. 2000; Zedrosser et al. 2007), cooperative breeding (Cockburn 1998; Clutton-Brock et al. 2001; Silk 2007) and age-specific performance (Clutton-Brock 1988; Rebke et al. 2010). Little is known, however, about the relative importance of these traits for reproductive success, particularly in the context of environmental conditions that determine whether a trait is favoured or penalized by natural selection.

Environmental conditions that impact reproduction include disease prevalence, resource availability and population density. Disease is a top-down influence that reduces reproduction directly via offspring mortality (Kissui & Packer 2004; Almberg et al. 2009) and/or indirectly through complex gene–environment interactions that generate fitness trade-offs between reproduction and survival (Graham et al. 2010). Resource abundance and population density impact reproduction through bottom-up, density-dependent processes (e.g. intraguild competition; Creel, Spong & Creel 2001; Kissui & Packer 2004; Watts & Holekamp 2008, 2009). For cooperatively breeding species, group-size-related increases in reproduction conceivably offset population-size-related decreases in vital rates, especially in group-territorial species where larger groups dominate smaller groups in the battle for limited resources (Mosser & Packer 2009). However, the relationship between density dependence across multiple levels of biological organization remains largely unexplored.

Here, we use 14 years of morphological, life history, demographic and ecological data from wolves (Canis lupus) in Yellowstone National Park (YNP) to assess the relative influence of different traits on the reproductive success of females under varying levels of environmental stress. Wolves are social carnivores that live in territorial, kin-structured packs in which they cooperate to raise young, defend resources from competitors and hunt (Mech 1970). They have a fast life history relative to other large carnivores including a short generation time (c. 4–5 years), early first reproduction (2 years old), high fecundity (5–6 pups per litter), rapid development (80% of adult body size acquired by the end of their first year) and brief life span (c. 5 years) (Peterson et al. 1998; Fuller, Mech & Cochrane 2003; MacNulty et al. 2009a). Although wolves are among the most-studied mammals in the world (Mech & Boitani 2003; Musiani, Boitani & Paquet 2010), surprisingly, little is known about which traits drive reproduction in this charismatic top predator.

We used multivariate mixed effects models to assess the simultaneous influence of multiple traits and ecological conditions on the annual reproductive performance (litter size and litter survival) of 55 individually known female wolves and tested various predictions. With respect to female traits, we expected (i) that because of weakening selection for reproductive performance later in life (Hamilton 1966; Charlesworth 1980), the age-specific reproductive profile should follow a concave-down pattern with a peak (i.e. onset of reproductive senescence) near age 5 (median life span of YNP wolves; MacNulty et al. 2009b); (ii) a positive correlation between female body mass and reproductive success given that larger body size in mammals can indicate higher quality individuals (Hamel et al. 2008) or be a pre-requisite for social dominance (Russell et al. 2003) with reproductive advantages; (iii) a positive correlation between multilocus heterozygosity estimates and reproductive success given enhanced success demonstrated in outbred individuals of some species (Amos et al. 2001); (iv) differential reproductive performance between black and grey-coloured females (determined by the K locus, CBD103, a β-defensin gene that has two alleles) because recent studies have suggested fitness differences among coat colours in North American wolves (Musiani et al. 2007; Anderson et al. 2009; Coulson et al. 2011); and (v) a positive correlation between a female's pack size and her reproductive success because of the role of auxiliary adults as helpers. With respect to ecological conditions, we expected reproductive success to decline (i) with increasing population size because of density-dependent resource competition and (ii) during disease outbreaks because of spikes in offspring mortality (Almberg et al. 2009).

We used a multilevel analysis that statistically controls for the effects of multiple factors, along with a sensitivity analysis, to measure the relative strength of factors driving wolf reproduction across biological levels. Because ecologically important phenotypic traits (e.g. morphology, life history, behaviour) are thought to represent adaptive responses to selection pressures from competitive and stochastic environments, we expected reproductive gains from these traits to compensate for losses because of environmental pressures. Specifically, we predicted group size to be most influential given the central role of sociality in wolf behavioural ecology (Mech & Boitani 2003). Consequently, we discuss our results in terms of the adaptive importance of multiple traits to fitness in this large carnivore.

Materials and methods

Study Site and Population

We analysed data from the initial 14 years (1996–2009) of a long-term study of wild wolves in YNP. This 8991-km2 protected area is located mainly in north-western Wyoming, USA. A full description of the study area is given elsewhere (e.g. Smith et al. 2004). Wolves in this study were either members or descendants of a population of 41 radio-marked wolves reintroduced to YNP in 1995–1996 (Bangs & Fritts 1996). Each winter (January–February) after the reintroduction, we captured and radio-marked 20–30 wolves, including 30–50% of pups born in the previous year (Smith et al. 2004). Each wolf was aged, weighed, identified as having a black or grey coat and genetically sampled by drawing whole blood. Our capture and handling protocols were approved by the National Park Service and are in accordance with recommendations from the American Society of Mammalogists (Sikes, Gannon & The Animal Care Use Committee of the American Society of Mammalogists 2011). Including both radio-marked (N = 316) and unmarked wolves, we monitored over 1000 individuals from 38 packs during our study. Each year, an average of 40% of the total population census size (21–174 wolves; Smith et al. 2010) was radio-marked, including the female breeder(s) of each pack.

Breeders were identified by observing a female's breeding behaviour, physical evidence of pregnancy (extended abdomen) and/or denning behaviour (localization at or inside dens), followed by observations of nursing pups and/or genetically verified maternity (see below). A female breeder produced only one litter per year, and reproduction was typically monopolized by a single female that was behaviourally dominant to same-sex pack members. Subordinate females sometimes produced litters, but the majority of litters in our data set (81%) came from socially dominant females (i.e. female parents or siblings to other potential breeders). A total of 55 female breeders were individually identifiable by combination of radio frequency, pelage colour, body shape and/or size.

Data Collection

The reproductive performance of each female breeder was measured annually for 1–7 years, and 31 females were measured in multiple years. Measurements came from year-round monitoring of all YNP packs (2–15 packs per year). Field personnel observed female breeders and their pack mates daily in early (mid-November to mid-December) and late winter (March) and about weekly throughout the remainder of the year. Observations were recorded from the ground and fixed-wing aircraft (see Smith et al. 2004 for details).

We measured two components of reproductive performance for each female breeder: litter size and litter survival. Litter size was the maximum number of pups observed in a litter in the weeks following den emergence (10–14 days following parturition). This was a minimum estimate given that some pups may have died prior to den emergence. Pups are generally weaned at 5–9 weeks of age then fed by various pack members via meat regurgitation (Mech 1970). Litter survival was the number of pups that survived until 31 December of each year (8 months old). At this age, wolves are approaching functional independence with respect to their ability to hunt, disperse and breed (Medjo & Mech 1976; Mech & Boitani 2003). Eight-month-old pups were identifiable by virtue of their size and behaviour relative to older age classes. Litter size and survival were scored as 0 (N = 4 for litter size; N = 30 for litter survival) if a female was first identified as a breeder and subsequently had total reproductive failure (see Appendix S1, Supporting information).

Age determination

We calculated the age of female breeders as the number of years after their birth year. Females < 1 year old were classified as age zero. The birth year of most females (85%) was determined by marking them as pups, which provided an exact measure of age. Tooth wear and cementum annuli were used to estimate the birth year of some live and dead adults, respectively (Gipson et al. 2000). Wolves not captured as pups were sometimes caught as adults and considered known-aged if individually recognized from birth via distinct morphological features (e.g. pelage markings, colour, body shape and size). We assigned a birth year to non-captured wolves only if first observed as pups and individually identifiable as adults.

Body mass

We used estimates of individual, age-specific body mass (kg) to assess the effects of female body size on reproductive performance because it was not feasible to annually weigh each female. We derived these estimates from an updated version of the wolf growth model described in MacNulty et al. (2009a; see Table S1, Supporting Information). Estimates correspond to a female's mass on the average parturition date in YNP (April 15). The growth model was fit to body mass data from 172 females that were measured during annual capture. Twenty-six per cent of these wolves were caught in multiple years to replace damaged radio tags and so were weighed more than once. Body mass was recorded using a 0–100 kg Pesola spring scale (Rebmattli, Baar, Switzerland). Wolves were placed in a weighing tarp attached to the scale and hoisted aloft until clear of the ground. We did not estimate and subtract stomach-content mass from body mass, so our measurements are likely maximum estimates.

Maternity analysis

We analysed litter size and survival for only litters of known maternity. We used field observations to assign maternity of litters to females, based on the obvious behaviour and appearance of breeding females (see above), and confirmed these putative mother–offspring relationships with genetic data where available (75% of all litters). Genetic analysis included cases where maternity of litters was uncertain from field observations or in packs, where both dominant and subordinate breeders existed (see Appendix S2, Supporting information). In these cases, we assigned maternity according to a population pedigree constructed for the YNP wolves where individual genotypes (N = 337) were based on 26 domestic dog microsatellite loci and PCR amplification as described in vonHoldt et al. (2008). Because we restricted analyses to litters of known maternity, only a subset (52%) of total observed litters from packs with both dominant and subordinate breeders could be analysed. Consequently, the sample of litters from subordinate breeders was too small (N = 25 litters) to determine how social status affected reproduction.

Data analysis

To understand the adaptive value of different morphological, behavioural and life-history traits, we examined their relative effects on female reproductive performance under varying levels of environmental stress. This involved a separate analysis of the annual number of pups born and survived per litter. Analyses were conducted with generalized linear mixed models (GLMMs) with a Poisson error distribution after verifying that the data were not over-dispersed. Such models can account for correlation between the observations taken on the same female in multiple years and on multiple females in the same year. We fitted individual identity and year as crossed random intercepts, such that the random intercept for individual i is shared across all years for a given individual i, whereas the random intercept for year j is shared by all individuals in a given year j. Note that the random effect for year accounts for unmeasured year-related effects on reproduction, including prey abundance and climate. All models included a compound symmetric correlation structure, which assumed that all observations within individuals and years were, on average, equally correlated (Weiss 2005). Models were estimated with Laplacian approximation, with parameters estimated from maximum likelihood, and significance of effects determined by an approximate z-test.

Fixed effects

We fitted fixed effects corresponding to different female traits, including age, body mass, coat colour, internal relatedness (IR) and pack size. Because age-related reproductive patterns are potentially confounded by the selective (dis)appearance of different quality individuals, we also fitted individual age at first and last reproduction as fixed effects (van de Pol & Verhulst 2006; Nussey et al. 2008). IR reflects a quantity measured between parental alleles that weights allele sharing by the frequencies of the alleles involved. This estimate of heterozygosity gives more weight to homozygotes involving rare alleles and reflects parental similarity more effectively than commonly used heterozygosity indices (Amos et al. 2001; Zedrosser et al. 2007). We calculated IR for 47 genotyped females following Amos et al. (2001). With respect to coat colour, all genotyped grey-coloured female breeders were homozygous at the K locus, while all but one black female were heterozygous at the K locus (Anderson et al. 2009). Homozygous black females are rare in the YNP population (Coulson et al. 2011).

Pack size is an index of sociality, and we measured it differently for each analysis. In the analysis of litter size, pack size was the mode total group size recorded over 3 months prior to parturition (January–March). We used this measure because it encompasses the potential influence of pack members on breeder condition during breeding season and offspring development in utero. In the analysis of litter survival, pack size was the mode adult (≥1 year old) group size recorded from June through December, as this measures how pack members may affect pup provisioning, development and protection. Breeding females were included in all pack size counts.

To assess how environmental stress shapes the relationship between female traits and reproductive performance, we fitted fixed effects for disease and wolf population size. Canine distemper virus (CDV) has been linked to poor pup survival in YNP wolves (Almberg et al. 2009), and individuals in our study experienced three CDV outbreaks (1999, 2005, and 2008). Our fixed effect for disease was therefore a dummy variable for the occurrence of a CDV outbreak in a given year. We used annual counts of the YNP wolf population (Smith et al. 2010) to track changes in the intensity of the competitive environment given that territorial aggression between wolf packs increases with population size (Yellowstone Wolf Project, unpublished data). Counts of population size at April 1 and December 31 were used in the analyses of litter size and survival, respectively. We used unadjusted population counts because resighting rates were high (0·96) and constant across years (Coulson et al. 2011). The probability of a CDV outbreak tended to increase with wolf population size (logistic regression: β = 0·03 ± 0·02, = 0·15, = 15 years), but this association was not strong (pseudo-r= 0·20). In general, our fixed effects were not highly collinear (< 0·30).

We used piecewise linear splines to test for nonlinear effects of age and pack size (see Appendix S3, Supporting information). Specifically, we tested for whether reproductive performance declined at advanced ages and in large packs because of physiological senescence and intrapack competition, respectively. We created variables containing a linear spline for age and pack size with the MKSPLINE command in STATA 11.0. The variables were constructed so that the estimated coefficients measure the slopes for the segments before and after a segment break.

Model selection

We performed a series of model selection procedures to identify the most parsimonious GLMMs of litter size and survival. First, we used information-theoretic statistics (Burnham & Anderson 2002) to determine the best functional form of age- and pack-size-specific effects on reproduction. For each effect, we separately evaluated a set of saturated GLMMs of litter size and survival that contained all the other fixed effects along with different linear and nonlinear terms for age or pack size. Nonlinear terms were spline variables with a single knot at 3–6 years old (age) or 3–10 wolves (pack size). We selected knots a priori in accordance with guidelines for the efficient use of knots (Eubanks 1984; Seber & Wild 2003). By definition, knots selected a priori are fixed (i.e. not random variables) and are therefore not estimated as parameters in models. The best age and pack size models were the ones with the lowest Akaike Information Criterion (adjusted for small sample, AICc) and smallest ∆AICc. ∆AICc equals the AICc for the model of interest minus the smallest AICc for the set of models being considered. The best model has a ∆AICc of zero, and models with ∆AICc < 2 are plausibly the best.

Next, we added our best-fit terms for age and pack size to the saturated GLMMs of litter size and survival and reduced these models with a backward stepwise procedure. We dropped non-significant terms (> 0·10) one at a time until a likelihood ratio test indicated that the fit of the reduced model was significantly worse than that of the full model containing the dropped term. After the initial stepwise search, we refit the reduced model using the selected terms and reconsidered inclusion of the omitted terms by adding them to the model one at a time and testing for significance.

We then verified the functional form of the age- and pack-size-specific effects remaining in the reduced models. For each effect, we used AICc to separately evaluate a set of reduced GLMMs of litter size and survival that included different linear and nonlinear terms for age and pack size as described above. We tested for pairwise interactions between terms in the best-fit, reduced GLMMs by adding interaction terms to the models one at a time and testing for significance.

We calculated population-averaged fitted values from the final models by deriving marginal expectations of the responses averaged over the random effects but conditional on the observed covariates. We separately refitted certain fixed effects as categorical factors and plotted the associated fitted values to illustrate the distribution of the underlying data after controlling for individual and annual heterogeneity and the fixed effects of other variables. We interpreted Poisson regression coefficients in terms of incidence-rate ratios (IRRs), which we obtained by exponentiating the coefficients (Rabe-Hesketh & Skrondal 2008). IRR minus 1 equals the percentage change (+/−) in the dependent variable (e.g. litter size) for each one unit increase in a continuous variable (e.g. body size) or when comparing one group to another in a categorical variable (e.g. disease vs. non-disease years) while holding all other variables in the model constant. IRRn rescales the unit change to n units. Throughout, means are reported with standard errors.

Sensitivity analysis

To evaluate the relative strength of different effects in the final models of litter size and survival, we performed a sensitivity analysis that allowed comparison of effects across a common scale. First, we calculated the predicted number of pups born (or survived) with continuous variables set to observed means and categorical variables for coat colour and disease set to ‘black coat’ and ‘CDV outbreak’, respectively. Next, we separately perturbed each model parameter by 10%, recalculated the prediction and computed the difference between the initial and perturbed prediction. A large difference corresponds to a high sensitivity, and parameters with the highest sensitivity had the greatest effect on reproductive performance. We report absolute sensitivities and sum those for spline parameters to show the overall influence of a nonlinear effect and to facilitate comparison between linear and nonlinear effects.


We recorded 140 observations of annual female reproduction from 55 female breeders (30 grey, 25 black) in 32 different packs across YNP, 1996–2009. An observation included information on whether a female breeder had produced a litter, the size of the litter and the number of pups surviving to the end the year. Most observations involved litters (= 125), and most of these included information on litter size and survival (= 119). Observations that included litter survival but not litter size (= 3) or vice versa (= 3) occurred because litter size at den emergence or subsequent pup survival, respectively, was undocumented. Thus, the total sample differed somewhat between the analyses of litter size (= 126 observations, 51 females) and litter survival (= 136 observations, 54 females).

Most litter observations were from packs with a single breeding female (77% of 125). Some were from packs with > 1 female breeder, and these litters were produced by socially dominant (13%) and subordinate (10%) females. Female breeders were 2–9 years old with mean (±SE) age of primiparity at 2·7 (±0·1) years. Female breeders lived 5·4 (±0·4) years on average, and 50% died by age 5 (median life span). We estimated age-specific body mass of female breeders from a model of female growth (Table S1, Supporting information). According to this best-fit model (ΔAICc = 0·00), females exhibited three growth phases: rapid growth to 0·75 years old, moderate growth from 0·75–2·75 years old and no growth beyond 2·75 years old (Fig. 1). Other top-scoring models (ΔAICc < 2·00; Table S2, Supporting information) suggest that growth levelled-off from 1·75–3·75 years old. The predicted body mass of female breeders from the best-fit growth model was 26–53 kg.

Figure 1.

Growth profile of female wolves (= 172) in Yellowstone National Park, 1995–2010. Points are observed age-specific weights (= 238), and the solid line represents the population-averaged fitted growth curve from the best-fit mixed effects growth model (Table S1, Supporting information). Model selection results are in Table S2 (Supporting information).

Litter size

Variation in litter size at den emergence (range = 0–11 pups; mean 4·74 ± 0·21 pups) was attributable to individual-, group- and population-level factors. The final GLMM of litter size includes terms for female age and body mass, adult pack size, population size and disease (Table 1). The terms for age and pack size are spline variables that were identified in initial model selection (Table S3–S4, Supporting information) and confirmed in follow-up analyses (Table S5–S6, Supporting information). There were no significant pairwise interactions between terms in the final GLMM model (results not shown).

Table 1. Best-fit GLMM model for the effects of individual-, group- and population-level factors on the number of pups in a female's litter at den emergence (litter size). Female's age-1 and age-2 are the effects of a female's age before and after she reaches 4 years old, respectively. Pack size-1 and pack size-2 are the effects of pack size when a female's pack includes less than or greater than eight wolves (including breeding females), respectively. Disease year refers to years with canine distemper outbreaks. Incident rate ratios (IRR) are the exponentiated Poisson regression coefficients. Model selection results are in Tables S3–S4 (Supporting information)
ParameterIRRβSEzP95% confidence interval for β
Intercept 0·7230·4041·790·073−0·0681·515
Female age-11·0720·0700·0710·990·323−0·0690·208
Female age-20·926−0·0760·037−2·040·041−0·149−0·003
Female body mass (kg)1·0140·0140·0091·620·104−0·0030·032
Pack size-11·0940·0910·0253·68<0·0010·0420·139
Pack size-20·909−0·0950·045−2·100·036−0·184−0·006
Population size0·997−0·0030·001−2·510·012−0·006−0·001
Disease year0·772−0·2580·116−2·220·027−0·486−0·030

The spline variable for age indicates that litter size changed little from 2–4 years old (= 0·323), whereas it decreased by 7% (incidence-rate ratio [IRR] = 0·93 ± 0·04, = 0·041) for each year beyond 4 (Fig. 2a). Note the strength of this effect is uncertain given that a similar model with a simple linear term for age (β = −0·03 ± 0·02, = 0·180) fit the data nearly as well (ΔAICc = 0·06; Table S5, Supporting information). But fitting age as a categorical factor corroborated the apparent age-related decline because 7- and 9-year-old females produced 33% and 53% fewer pups, respectively, than did 4-year-old females (< 0·040). This age-related decline was not an artefact of the selective (dis)appearance of different quality individuals as age at both first and last measurement was not retained in the final model.

Figure 2.

Effects of (a) age, (b) body mass, (c) total pack size and (d) population size on the number of pups in a female's litter at den emergence (pups born) in Yellowstone National Park, 1996–2009. The number of wolves and individual litters included in this analysis were 51 and 126, respectively. Solid lines are population-averaged fitted values from the best-fit GLMM model (Table 1) with dotted lines indicating pointwise 95% confidence intervals. Points are predicted means with standard errors obtained by separately refitting covariates (a–d) as categorical factors in the best-fit model, with sample size indicated above each point.

The main effect for female body mass suggests that bigger females produced bigger litters (Fig. 2b). Specifically, litter size tended to increase by 15% for each 10-kg increase in female body mass, although this was statistically insignificant (IRR = 1·15 ± 0·09, = 0·104).

The spline variable for pack size (total pack size prior to parturition; range = 2–15 wolves; mean 5·19 ± 0·24 wolves) indicates a threshold at which the effect of pack size on litter size suddenly changed (Fig. 2c). Below 8 wolves, each additional wolf increased litter size by 10% (IRR = 1·10 ± 0·03, < 0·001). But for each wolf beyond 8, litter size decreased by 9% (IRR = 0·91 ± 0·05, = 0·036). Note the confidence set of models (ΔAICc < 2·00; Table S6, Supporting information) indicates the threshold adult group size may have occurred at seven or nine wolves.

Population size and disease were important extrinsic constraints on litter size. Wolf population size in YNP at female parturition time (April) was 33–165 wolves (mean = 95 ± 10 wolves), and litter size decreased with increasing population size (Fig. 2d). For every 10 additional wolves in the population, litter size decreased by 3% (IRR = 0·97 ± 0·01, = 0·012). There were 23% fewer pups in a litter at den emergence in years with canine distemper virus outbreaks than in years without CDV (IRR = 0·77 ± 0·09, = 0·027).

Litter survival

As with litter size, variation in litter survival (range = 0–9 pups; mean 3·07 ± 0·20 pups) was attributable to individual-, group- and population-level factors. The final GLMM of litter survival includes female age, body mass, coat colour, pack size, population size and disease (Table 2). Terms for age and pack size are spline variables that were identified in initial model selection (Table S7–S8, Supporting information) and confirmed in follow-up analyses (Table S9–S10, , Supporting information). There were no significant pairwise interactions between terms in the final GLMM model (results not shown). Overall, litter survival was positively related to litter size at den emergence (Pearson correlation = 0·65, < 0·001, N = 119 litters).

Table 2. Best-fit GLMM model for the effects of individual-, group- and population-level factors on the number of pups in a female's litter surviving until 8 months old (litter survival). Female's age-1 and age-2 are the effects of a female's age before and after she reaches 5 years old, respectively. The reference coat colour is black. Pack size-1 and pack size-2 are the effects of adult (≥1 year old) pack size when a female's pack includes less than or greater than three adults (including breeding females), respectively. Disease year refers to years with canine distemper outbreaks. Incident rate ratios (IRR) are the exponentiated Poisson regression coefficients. Model selection results are in Tables S5–S6 (Supporting information)
ParameterIRRβSEzP95% confidence interval for β
Intercept −1·0220·705−1·450·148−2·4040·361
Female age-1 1·0290·0290·0620·470·637−0·0920·150
Female age-20·843−0·1710·071−2·410·016−0·309−0·032
Female body mass (kg)1·0330·0330·0122·820·0050·0100·056
Female coat colour0·751−0·2870·145−1·980·047−0·571−0·003
Pack size-11·4520·3730·1632·290·0220·0540·693
Pack size-21·0470·0460·0133·400·0010·0190·072
Population size0·995−0·0050·003−1·820·069−0·0100·000
Disease year0·471−0·7540·180−4·20<0·001−1·106−0·402

The spline variable for age indicates that litter survival changed little from 2–5 years old (= 0·637), whereas it decreased by 16% (IRR = 0·84 ± 0·07, = 0·016) for each year beyond 5 (Fig. 3a). Although a similar model with a simple linear term for age (β = −0·06 ± 0·03, = 0·06) scored well (ΔAICc = 0·79), the apparent age-related decline is corroborated by fitting age as a categorical factor: 7- and 9-year-old females produced 39% fewer (= 0·040) and 59% fewer (= 0·065) surviving pups, respectively, than did 5-year-old females. As indicated by the confidence set of models (ΔAICc < 2·00; Table S9, Supporting information), the threshold for reproductive decline conceivably occurred at age 4 or 6. This age-related decline was not an artefact of the selective (dis)appearance of different quality individuals as age at both first and last measurement was not retained in the final model.

Figure 3.

Effects of (a) age, (b) body mass, (c) adult pack size and (d) population size on the number of pups in a female's litter surviving until 8 months old (pups survived) in Yellowstone National Park, 1996–2009. The number of wolves and individual litters included in this analysis were 54 and 136, respectively. Solid lines are population-averaged fitted values from the best-fit GLMM model (Table 2) with dotted lines indicating pointwise 95% confidence intervals. Points are predicted means with standard errors obtained by separately refitting covariates (a)-(d) as categorical factors in the best-fit model, with sample size indicated above each point.

The effect for female body mass indicates that pups born to bigger females have higher survival to independence (Fig. 3b). Specifically, litter survival increased by 39% for each 10-kg increase in female body mass (IRR = 1·39 ± 0·12, = 0·005). Moreover, the effect of coat colour indicates that the K locus is linked to reproductive performance. Specifically, black-coloured females (all heterozygous at K locus, but one) had 25% fewer surviving pups annually than grey-coloured females (IRR = 0·75 ± 0·15, = 0·047).

Pack size (adult pack size during pup rearing to independence; range = 0–26 wolves; mean 6·76 ± 0·36 wolves) positively influenced litter survival throughout its range, but the spline variable indicates a threshold at which this effect diminishes (Fig. 3c). Below three wolves, each additional adult increased litter survival by 55% (IRR = 1·45 ± 0·16, = 0·022), but for each additional adult beyond 3, litter survival increased by only 5% (IRR = 1·05 ± 0·01, = 0·001). Note the confidence set of models (ΔAICc < 2·00; Table S10, Supporting information) indicates the threshold adult group size may have occurred at four or five wolves. Pack size of 0 (N = 3) were cases where all adult pack members (including breeding female) died or disappeared during late pup rearing season, resulting in zero pup survival to independence in each case.

Constraints on litter survival were driven by both population size and disease. YNP wolf population size (excluding pups) at the time pups approached independence (December) was 28–115 wolves (mean = 84 ± 2·2 wolves), and litter survival decreased with increasing population size (Fig. 3d). Each 10 additional wolves in the population decreased litter survival by 5%, but this effect was statistically weak (IRR = 0·95, = 0·069). Finally, pup survival was greatly reduced in CDV outbreak years, with 53% fewer surviving pups in a litter than in non-disease years (IRR = 0·47 ± 0·18, < 0·001).

Relative influence of main effects on reproduction

Our sensitivity analysis revealed the relative importance of a female's morphological, behavioural and life-history traits under the conditions of population density and disease outbreaks observed during the study. The sensitivity analysis indicates that individual-level traits were most influential on female reproductive success, followed by group effects, then population-level factors. Using total sensitivity scores (below in parentheses) to rank in the order of largest effect, litter size was most influenced by female body mass (0·27), pack size (0·20), population size (0·14), female age (0·12) and disease (0·11), respectively (Fig. 4). For litter survival, female performance was most influenced by female body mass (0·21), followed by pack size (0·19), disease (0·10), population size (0·06), coat colour (0·04) and age (0·02), respectively (Fig. 4).

Figure 4.

Relative influence of individual-, group- and population-level effects on a female's litter size at den emergence (black bars) and litter survival until 8 months old (white bars) in Yellowstone National Park, 1996–2009. Each bar represents a sensitivity value generated by taking the difference between initial and perturbed (10%) predicted values for each parameter identified in Tables 1 and 2. The greater the sensitivity value, the more influential that parameter is on reproductive performance. Sensitivities for age and pack size are sums of the sensitivity of each variable's spline parameters.


Our comprehensive analysis of wolf reproduction demonstrates that heterogeneity in female performance was significantly influenced by several individual-, group- and population-level factors (summarized in Table S11, Supporting information). Importantly, we demonstrate that the strength and direction of effects on litter size and litter survival were not uniform and sometimes contrasting as reproductive losses at one level were offset by gains at another. By simultaneously evaluating the effects and comparing their relative influence, our analysis indicates that individual traits (body size) and behaviour (sociality) had the greatest effect on reproduction and helped counter the impact of competition and disease. Below, we discuss each effect and its importance to reproductive performance in Yellowstone wolves.

Individual-level factors influencing reproduction

Improved reproductive performance with increased female body mass is common in mammals (e.g. ungulates – Albon, Guinness & Clutton-Brock 1983; sciurids – King, Festa-Bianchet & Hatfield 1991; pinnipeds – Iverson et al. 1993), but hitherto undocumented in wild wolves. The reproductive benefit of larger size can be manifested in many ways including improved conception (Boyd 1984), foraging ability (MacNulty et al. 2009a), lactation (Bowen et al. 2001) and offspring mass (Iverson et al. 1993). Our analysis of YNP wolves revealed that the effect of body mass on female reproduction exceeded that of other factors including pack size. Although the effect body mass on litter size was not statistically significant, it was qualitatively similar (Fig. 2b) to its highly significant effect on litter survival (Fig. 3b). The stronger effect on litter survival may reflect greater maternal investment in pup survival (e.g. lactation, offspring development) vs. litter size (e.g. gestation). The reproductive importance of body mass is further highlighted by close correspondence between the ages of primiparity (2·7 years) and maximum body mass (2·75 years). By contrast, female ungulates and ursids start reproducing long before reaching maximum size (Zedrosser et al. 2009; Martin & Festa-Bianchet 2011), perhaps because they exhibit a comparatively slower life history and less competitive social environment. In wolves, the reproductive benefits of large size combined with early age at first reproduction are likely strong selective pressures for rapid neonatal growth as illustrated in our growth model (Fig. 1). This growth pattern suggests that early life experiences may strongly influence lifetime fitness, as has been shown in other cooperative breeders (meerkats Suricata suricatta, Russell et al. 2003; red wolf, Canis rufus, Sparkman et al. 2011).

Reproduction's sensitivity to body mass, combined with rapid neonatal growth, effectively constant adult body mass with increasing age and short life span, likely explains the moderate effects of age on female reproduction. Age-specific reproductive performance is typically attributed to either increased experience or investment in resources (Curio 1983; Clutton-Brock 1988) or senescence (Nussey et al. 2008), or alternatively, can be masked by the selective (dis)appearance of individuals of varying quality (Reid et al. 2003; de Pol & Verhulst 2006). With respect to the latter, we found no effect of age at first or last reproduction, indicating that age-related performance is not being masked by differential onset of reproduction or mortality associated with individual quality. Although reproductive success did not improve with age, it did decline beyond 4–5 years old. To our knowledge, this is the first demonstration of reproductive senescence in wild wolves. This result complements recent findings that YNP wolves exhibit senescence in hunting ability (MacNulty et al. 2009b) and supports the hypothesis that natural selection is too weak to support genetic health late in life (Hamilton 1966; Charlesworth 1980). Whereas social effects may moderate ageing (Lee 2003; Bourke 2007), we found no significant interaction between age and group size, suggesting auxiliaries did not influence the rate or onset of reproductive senescence, similar to findings in other cooperative breeders (Sharp & Clutton-Brock 2010). Although senescence has been widely detected among mammals (Nussey et al. 2008), its importance to fitness in wild populations remains controversial, especially in short-lived species (Turbill & Ruf 2010). Given wolves' relatively short life spans, it is not surprising that age was relatively unimportant to reproduction compared to body mass.

That grey females were reproductively more successful than black females is intriguing, despite the moderate influence of coat colour overall. While proximate mechanisms mediating covariation between coat colour and reproduction remain unclear, our findings may represent trade-offs between reproduction and other fitness measures (e.g. survival). This is possibly due to antagonistic pleiotropy associated with the K locus (a β-defensin gene) responsible for melanism in wolves (Anderson et al. 2009) and associated with innate and adaptive immunity (Yang et al. 1999). Given that melanin-based coloration is often associated with regulatory effects on energy balance, stress and immunity in wild vertebrates (Ducrest, Keller & Roulin 2008; Gasparini et al.2009), there may be a positive association between melanism and immunocompetency that grants a survival advantage to black wolves. Indeed, preliminary investigations of YNP wolf survival rates show black females experience greater survival than grey females (Yellowstone Wolf Project, unpublished data). However, the costs of immunity may contribute to reproductive costs, a pattern demonstrated in other mammals (Graham et al. 2010). Additionally, a recent study found higher fitness for black heterozygous wolves in YNP, suggesting balancing selection (Coulson et al. 2011). Together, these patterns indicate fitness differences associated with the K locus that may help explain the maintenance of colour polymorphism in some North American wolf populations.

Although estimates of heterozygosity (internal relatedness, IR) have been correlated with the variation in reproductive performance in some species (e.g. Amos et al. 2001; Zedrosser et al. 2007), we found no significant effect of this trait on reproduction. This is likely due to high genetic variation and inbreeding avoidance in YNP wolves (vonHoldt et al. 2008), along with an insufficient range of IR values to evaluate whether relatively outbred individuals have enhanced success compared to relatively inbred ones (Amos et al. 2001).

Group-level factors influencing reproduction

The evolution of sociality in large carnivores is influenced by many factors, including territorial defence (Mech & Boitani 2003; Mosser & Packer 2009), group hunting (MacNulty et al. 2009), food defence (Creel, Spong & Creel 2001; Vucetich, Peterson & Waite 2004), kin selection (Schmidt & Mech 1997) and cooperative breeding (e.g. Mech 1970; Clutton-Brock 2002). Our results highlight the adaptive value of sociality by showing pack size as the second most important driver of reproductive success. Similar effects have been found in other canid systems (e.g. Harrington, Mech & Fritts 1983; Moehlman 1986; McNutt & Silk 2008; Sparkman et al. 2011) and are typically attributed to auxiliaries caring for pups. Yet, some canid studies have shown no correlation (e.g. Peterson, Woolington & Bailey 1984; Pletscher et al. 1997) or a negative effect of auxiliaries on reproduction, particularly when unfavourable socio-ecological conditions prevail (e.g. high intraguild competition, low prey density; Harrington, Mech & Fritts 1983; Gusset & Macdonald 2010; Sparkman et al. 2011).

Our results showing nonlinear, contrasting effects of pack size on reproduction are significant because they demonstrate that group effects are conditional on the breeder's life cycle stage and not uniform across group sizes. Few studies have demonstrated significant nonlinear effects of group size in cooperatively breeding mammals (e.g. marmots, Armitage & Schwartz 2000). The negative correlation between early litter size and larger pack sizes (> 8 wolves) draws attention to apparent costs of sociality at this stage of reproduction. Mechanisms underlying such costs may include intrapack competition for food (Harrington, Mech & Fritts 1983; Schmidt & Mech 1997) or socially induced stress from competitors during the breeding season (Creel 2001; McNutt & Silk 2008), both of which can impact maternal condition important to early components of reproduction in cooperative breeders (Russell et al. 2003; Sharp & Clutton-Brock 2010). In contrast, we found positive effects of auxiliaries on litter survival throughout all pack sizes, demonstrating that pup survival was enhanced in larger packs. In addition to having more helpers to provision young, larger groups have numerical advantages during intergroup (Mech & Boitani 2003) and intraguild (Wilmers et al. 2003; Vucetich, Peterson & Waite 2004) competition for resources (e.g. food, territory), which may contribute to offspring survival, as shown in lions (Panthera leo; Mosser & Packer 2009). Importantly, the positive influence of auxiliaries was strongest for small packs, indicating that there is a threshold below which helpers are particularly critical to breeder success. Although wolves are not considered obligate cooperative breeders, our results are consistent with an Allee effect (i.e. inverse density dependence) at the pack level where recruitment critical to group persistence depends on a minimum group size (Courchamp, Clutton-Brock & Grenfell 1999; Gusset & Macdonald 2010).

We recognize that our analysis did not examine the role of food availability as a mechanism underlying covariation between pup production and pack size. Additionally, this study did not address whether group effects on reproduction were influenced by kin-directed altruism and inclusive fitness benefits, which may occur in wolves (Harrington, Mech & Fritts 1983; Schmidt & Mech 1997). Future work aims to test these ideas by evaluating effects of pack hunting success (e.g. prey and biomass acquisition rates), pack composition (e.g. relatedness, age structure, sex ratio) and social dominance on fitness measures.

Population-level factors influencing reproduction

Although CDV was less influential overall than individual traits or group effects, it did have a pronounced, stochastic impact on female reproduction. CDV-related decreases in litter survival are concordant with earlier findings for this system (Almberg et al. 2009), with pup mortality occurring after weaning and sometimes culminating in complete loss of litters. Our results add to this knowledge by showing that CDV outbreaks also lead to reduced litter size at den emergence. CDV has been documented to cross placental barriers in domestic dogs (Canis lupus familiaris), causing abortion, weak offspring and neurological diseases (Pandher et al. 2006). Depending on CDV exposure and infection patterns in mothers, maternal condition and/or disease transmission to neonate pups may explain reduced early litter sizes during outbreak years. Our findings suggest that disease may be a strong selective force in canid systems, especially if linked to individual traits that offset its negative effects. For example, selection for traits that may be linked to maternal condition and immunocompetency (e.g. body size, coat colour) may strengthen via associations with variable environmental stresses such as disease.

Our finding of negative density-dependent effects on reproduction is consistent with many vertebrate populations, where changes in vital rates occur through behaviourally meditated competition over resources (Fowler 1981). This effect in YNP is likely due to increased competition with conspecifics under high wolf densities during our study (Yellowstone Wolf Project, unpublished data). However, our sensitivity results showing that group-level positive density dependence was more influential than population-level negative density dependence highlights a benefit of wolf sociality. Specifically, group augmentation can serve as a buffer against the negative effects of intergroup competition. These results demonstrate a more nuanced relationship between ecological conditions and sociality, as favourable conditions (i.e. high resource abundance) are thought to relax the need for cooperative behaviour, making significant group effects on offspring fitness less apparent (Gusset & Macdonald 2010). Here, we propose that under the socio-ecological conditions of high prey abundance during our study, which in turn resulted in high wolf densities, competition over territories and/or breeding opportunities strengthened the relationship between sociality and fitness. Our findings differ from other wolf studies (Harrington, Mech & Fritts 1983; Sparkman et al. 2011) where helpers were found to have either negative or no effect on pup survival under high densities. While these findings downplay the extent to which individuals benefit from group-living (Silk 2007), our results highlight how the fitness consequences of sociality are conditional upon prevailing socio-ecological conditions.


Measures of reproductive success from longitudinal studies are essential for linking individual traits to ecological and evolutionary dynamics in wild populations, especially in response to environmental change (e.g. Coulson et al. 2006, 2011). Our study identifies trait- and environmental-specific patterns of wolf reproduction that could improve models linking age structure, social structure, density and disease patterns with ecological and evolutionary dynamics in large carnivores. Additionally, our study clarifies how life history, sociality and ecological conditions interact in cooperative breeders and ranks the adaptive value of traits in promoting individual fitness in competitive and stochastic environments. Consistent with findings from a diverse array of mammalian studies, we demonstrate similar patterns in trait- and environmental-specific influences on reproduction, while uniquely evaluating the relative strengths of such factors. In wolves, it appears that individual performance is influenced more by phenotypes than environmental conditions, and it would be valuable to know if this were true in other taxa. Knowledge of traits that promote fitness in the context of environmental stress is a key to understanding how wild populations respond to global climate change, disease outbreaks, habitat alteration and human exploitation, particularly with respect to apex species, which can have a disproportionate effect on natural systems via trophic cascades (Estes et al. 2011).


We thank Erin Stahler, Debra Guernsey, Rick McIntyre, and numerous field technicians with data collection and management assistance. We also thank Roger Stradley from Gallatin Flying Service and Bob Hawkins from Hawkins and Powers, Inc. and Sky Aviation, Inc. for safe piloting. This work was supported in part by the National Science Foundation grants DEB-0613730 and DEB-1021397, University of California, Los Angeles, Yellowstone National Park, and many donors through the Yellowstone Park Foundation. We also thank significant donors to the Yellowstone Wolf Project: an anonymous donor, Annie and Bob Graham and Frank and Kay Yeager. Tim Coulson, Dan Blumstein and Joan Silk provided valuable comments on an earlier version of the manuscript.