University of Groningen
Better together
Groenewoud, Frank
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2018
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Groenewoud, F. (2018). Better together: Cooperative breeding under environmental heterogeneity.
University of Groningen.
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Chapter 3
Spatio-temporal resource variation,
group formation and the benefits of
cooperative breeding in
the Seychelles warbler
Frank Groenewoud, Sjouke A. Kingma, Martijn Hammers, Terry Burke, David S. Richardson & Jan Komdeur
ABSTRACT
Recent comparative studies suggest that environmental variability predicts the evolution of group living and cooperative breeding. Mechanisms that may explain this pattern in-clude (1) group formation driven by spatio-temporal variation in food availability and (2) selection for cooperative breeding because it improves fecundity under adverse conditions, but these hypotheses are seldom tested in concert. Using fine-scale, long-term data on the facultative cooperatively breeding Seychelles warbler Acrocephalus sechellensis, we find evi-dence that spatio-temporal variation in food availability facilitates group formation, and that groups with multiple females had higher fecundity and lower fecundity variance than other groups. However, contrary to prediction, this reduction in fecundity variance was not due to subordinate females buffering the effects of adverse food availability on reproduc-tion. Thus, spatio-temporal variation in food availability favours group formation (setting the stage for cooperative breeding), but does not favour helping behaviour per se in this species.
INTRODUCTION
Ecology has been invoked as an important driver of cooperative breeding (e.g. Stacey & Koe-nig 1990; Hatchwell & Komdeur 2000; Pen & Weissing 2000; KoeKoe-nig & Dickinson 2016) – a breeding system in which individuals forgo independent reproduction to help rear non-de-scendant young within a group (Cockburn 1998; Hatchwell 1999). Recent comparative stud-ies have shown that cooperative breeding in both birds (Rubenstein & Lovette 2007; Jetz & Rubenstein 2011) and mammals (Lukas & Clutton-Brock 2017) is positively associated with variability in environmental conditions. However, the mechanism underlying this pattern is unclear, since variability in environmental conditions can affect multiple components of cooperative breeding, such as the costs and benefits of delayed dispersal, the fecundity benefits of breeding cooperatively, or both (Griesser et al. 2017; Shen et al. 2017).
The evolution of cooperative breeding is generally considered a two-step process, whereby the formation of groups precedes the possibility of helping (Emlen 1991; Hatchwell 1999; Ligon & Burt 2004; Jennions & Macdonald 2007; Griesser et al. 2017). Constraints on dis-persal and/or independent breeding may lead to higher fitness for individuals that join a group as a subordinate (Selander 1964; Brown 1974; Emlen 1982). Such constraints could arise through temporally fluctuating environmental conditions leading to increased costs of independent breeding and/or improved benefits of group living, under adverse condi-tions (i.e. “hard life hypothesis”; Emlen 1982; Koenig et al. 2011). Additionally, sexually ma-ture offspring should forego dispersal if the benefits in their resident territory (e.g. food, protection) exceed the benefits of dispersal (“benefits of philopatry hypothesis”; Stacey & Ligon 1991), which may be particularly important when there is considerable and consis-tent spatial variation in environmental conditions. Therefore, both temporal and spatial environmental variation can play an important role in determining the costs and benefits of joining a group as a subordinate instead of dispersing and attempting to breed inde-pendently.
Temporal and spatial environmental variation may also promote cooperative breeding when it affects how subordinates improve group fecundity. The “bet-hedging hypothesis” – which has received a lot of attention recently – proposes that the benefits of cooperative breeding can arise through subordinates reducing fecundity variance, instead of increas-ing mean fecundity (e.g. Rubenstein 2011; Kennedy et al. 2018). Theoretical investigations have demonstrated that reducing fecundity variance can contribute as much to fitness as improving mean fecundity, especially under the conditions that are often experienced by cooperative breeders, i.e. small population size and high kin structure (Gillespie 1977; Tul-japurkar 1990; Lehmann & Balloux 2007; Sæther & Engen 2015). Consequently, the
preva-lence of cooperatively breeding in regions with high climatic uncertainty has often been explained in the context of bet-hedging (Rubenstein & Lovette 2007; Jetz & Rubenstein 2011; Lukas & Clutton-Brock 2017). One commonly suggested way by which subordinates can reduce fecundity variance is by improving reproductive success under adverse condi-tions (i.e. “temporal variability hypothesis”; Rubenstein & Lovette 2007; Shen et al. 2017), but this has only seldom been tested (Magrath 2001; Koenig et al. 2011) Additionally, previ-ous studies investigating cooperative breeding as a bet-hedging strategy (e.g. Rubenstein 2011; Koenig & Walters 2015) did not consider the impact of environmental conditions on delayed dispersal and group formation, which is an important prerequisite for cooper-ative breeding in many species (Griesser et al. 2017). Understanding how environmental variation affects group formation and fecundity benefits in cooperative breeders could prove vital in explaining why cooperative breeding has evolved in some species, but not in others.
Here we investigate how spatio-temporal variation in food availability affects delayed dis-persal and the fecundity benefits of cooperative breeding, using 17 years of data from a pop-ulation of facultatively cooperative Seychelles warblers Acrocephalus sechellensis (SW). Our study population is confined to the small island of Cousin. We therefore have the extensive, detailed long-term information on food availability, individual dispersal and fecundity that are necessary to investigate how spatio-temporal variation in food availability affects group formation and the fecundity benefits of cooperative breeding. In SWs, territories contain a dominant breeding pair and a variable number (1-5) of subordinates of either sex, which are mostly offspring from previous breeding attempts (Kingma et al. 2016a). Subordinates often help with the provisioning of young and improve the reproductive success of the dominant pair (Richardson, Burke & Komdeur 2002; Brouwer, Richardson & Komdeur 2012). Female subordinates generally help more than subordinate males (female subordinates provision about 80% more than subordinate males; Komdeur 1991, 1994a; Richardson et al. 2003b), including in incubation (males do not incubate) and unlike males, often co-breed with the dominant female in the territory by laying an egg in the same nest (Richardson et al. 2001; Hadfield, Richardson & Burke 2006). Therefore, female subordinates can prevent nest fail-ure at multiple breeding stages, and are probably in a better position to improve fecundity, and reduce fecundity variance, than male subordinates.
Our study has four aims. First, we quantify how food availability varies across space and time. Second, we test how this spatio-temporal variation in food availability affects delayed dispersal and the formation of social groups. Third, we test how mean fecundity and fecun-dity variance change with food availability and the presence of male and female subordi-nates in a group. Fourth, we investigate whether the impact of male or female subordisubordi-nates
on fecundity depends on food availability. These results will provide a better understanding of how cooperative breeding can be associated with fluctuating environments and provide valuable insights into the role of environmental variation in the evolution of group living and cooperative breeding.
METHODS
Study species and population monitoring
The data were collected on the Cousin Island population (04º20’S, 55º40’E; ca 29 ha) of SW from 1996 to 2016. SWs live on territories that are defended year-round and stable between years (territory number: mean = 112, range = 103-123; this study). Breeding vacancies are lim-ited and offspring of both sexes often delay dispersal after reaching sexual maturity (Kom-deur 1992; Eikenaar et al. 2007; Kingma et al. 2016a). As a result, on average 41% of territories contain subordinate group members (> 5 months old; mean number of subordinates per territory ± SE = 0.59 ± 0.02, range = 1-5, n = 2,578 “group years”; this study). Migration to or from the island is virtually absent (Komdeur et al. 2004a) and resighting probabilities between years are very high: 0.98 for birds over two years of age and 0.92 for younger indi-viduals (Brouwer et al. 2010).
Although SWs can breed year-round, there are major (June-September) and minor (Janu-ary-March) breeding seasons, following semi-annual monsoon rains (Komdeur 1996b). For the current study, we used 17 years of data from the main breeding seasons in 1996-1999, 2003-2004, and 2006-2016. The years 2000-2002 and 2005 were excluded, because data on insect abundance (see below) were not available for these years. During fieldwork periods, birds were observed on their territories at least weekly for nesting activity and, once found, nests were visited and observed every 3-4 days to determine the onset of incubation, hatch-ing success and fledghatch-ing success. SWs on Cousin Island typically raise only a shatch-ingle offsprhatch-ing on their territory (91% of clutches contain a single egg; Komdeur 1996b; Richardson et al. 2001), and offspring can receive food from their parents and other group members for up to five months after fledging (Komdeur 1991). Unringed individuals were caught either at the nest or using mist-nets within the natal territory, and 60% of unringed individuals were caught before 5 months of age. Individuals were ringed using a unique combination of co-lour rings and a metal ring (British Trust for Ornithology). Over 96% of adult birds (subordi-nates and dominants) have been ringed in every year since 1997 (Hadfield et al. 2006; Ham-mers et al. 2015). All ringed individuals were blood sampled through brachial venipuncture for molecular sexing (Richardson et al. 2001).
Food availability
SWs feed on arthropods (mainly insects), taken from the underside of leaves (Komdeur 1991). Food availability can therefore be determined by estimating the density of arthropod prey using the methods described by Komdeur (1992) and Brouwer et al. (2009). In short, ar-thropod density was determined monthly during each main breeding season at 13 locations on the island by counting the number of arthropods on the underside of 50 leaves of each main tree species (5 leaves per tree, 10 trees per species). Vegetation cover was determined once during the main breeding season at 20 locations within each territory, by estimating the presence (>50% cover), or absence (<50% cover) of vegetation at different strata (0-0.75 m, 0.75-2 m and at 2-m intervals thereafter). In each main breeding season, arthropod den-sity was calculated per territory as ∑ x=1(cx ix) where cx is the relative cover for species x and
ix is the mean arthropod abundance per unit leaf area for species x. The resulting measure
of food availability (i.e. the mean number of arthropods per unit leaf area) was log-trans-formed before analyses.
Statistical analyses
Spatio-temporal variation in arthropod density
To estimate the spatio-temporal distribution of arthropods over the island, we fitted arthro-pod density as a response variable in generalized additive mixed models (GAMMs) in the package “mgcv” version 1.8-16 (Wood 2011), assuming a Gaussian error distribution. Specif-ically, we fitted two nested models representing the following hypotheses about how ar-thropods are distributed in space and time: (i) arthropod density changes over time, but spatial variation is consistent, or (ii) both arthropod density and the spatial distribution of arthropod densities vary over time. We used a tensor product smooth with two-dimension-al isotropic thin-plate regression splines to describe spatitwo-dimension-al variation, and cubic regression splines to describe temporal variation, in arthropod density (Wood 2006). All models were fitted using maximum-likelihood approximation and checked for model assumptions such as normality of residuals, homogeneity of residual variance and adequate number of knots (k; i.e. the “wiggliness” of smoothing terms) for basis construction (Wood 2006). Fitted models were compared using AIC, and AIC weights were calculated to identify the model that makes the best out-of-sample predictions. Additionally, we estimated the repeatability of arthropod density for territories by fitting a linear mixed-effects model in the package “brms” version 1.5.1 (Bürkner 2017), which is a front end for STAN (Carpenter et al. 2017). We used variance estimates to calculate the proportion of variance explained by between-terri-tory differences, relative to within-terribetween-terri-tory differences (i.e. repeatability or intra-class cor-relation) according to Nakagawa and Schielzeth (2010). This model also contained varying intercepts for field periods to account for differences in arthropod density between years. To visualise the variability in arthropod density in different territories, and investigate the
possibility that variability differed across the island, we fitted the coefficient of variation (CV = standard deviation/mean) as a standardized measure of variation in arthropod densi-ty for each territory in an additive mixed model with a two-dimensional thin-plate regres-sion spline, as explained above. CV arthropod density estimates were weighted according to how many observations (i.e. years) contributed to a particular data point (not all territories were present in all years as some disappeared and some new territories were founded; me-dian = 13 years, range = 2–17 years).
Based on the spatio-temporal patterns that we detected using the approach described above, we partitioned arthropod density into temporal and spatial components by fitting arthro-pod density as a response variable in a random regression model with the package “brms” (Bürkner 2017), with year and territory as nested random effects. Estimates of these varying intercepts for years and territories were then used to obtain values for (i) between-year ar-thropod density (henceforth “annual food availability”), and (ii) within-year deviance in arthropod density (henceforth “territory food availability”). These estimates were used in all subsequent analyses.
Food availability and group formation
We fitted a generalized linear mixed model (GLMM) with a binomial error structure to es-timate the likelihood that offspring of different sex born in a given main breeding season had dispersed from the natal territory before the end of the next main breeding season (i.e. when they were approximately one year of age), using the package “brms” (Bürkner 2017). Individuals were only included if they were caught and ringed before three months of age (i.e. as a nestling or dependent fledgling) and were alive one year later (N = 193 of 339 individuals). We hypothesize that offspring are more likely to disperse (to attempt to breed independently) when annual food availability is higher, and external constraints on dispersal and independent reproduction are therefore lower. However, if benefits obtained in the natal territory are the main factor driving delayed dispersal then we predict that sub-ordinates are more likely to disperse from territories with low food availability compared to territories with high food availability. We included sex in the model to account for po-tential differences in dispersal between male and female offspring and we fitted five models representing different hypotheses about the relationship between sex, annual island food availability and territory food availability (Table S3.1). We then obtained model averaged predictions based on WAIC (“Widely Applicable Information Criterion”; Watanabe 2010) by resampling predicted values of each model according to its Akaike weight. Territory ID and
year ID were included as random effects in all models. To test the possibility that patterns of
delayed dispersal were due to differential survival (individuals that died within a year were not included in the previous analysis), we also analysed whether an individual’s survival to
the next breeding season was associated with annual island food availability, territory food availability or sex, in a generalized linear mixed model with a binomial error structure.
Ter-ritory ID and year ID were included as random effects. We fitted five models, similar to the
analysis of delayed dispersal (Table S3.2), and assessed models based on WAIC values. As an additional line of evidence establishing the link between group stability and food availability, we investigated whether the proportion of cooperative breeding groups de-pended on food availability. Breeding groups were classified based on their group composi-tion as having no subordinates (N = 924), having at least one female subordinate (N = 342), having at least one male subordinate (N = 186), or having at least one male and one female subordinate (N = 143). We fitted the group composition as a response variable in a multino-mial logistic regression analysis, and included mean annual island food availability and the annual deviance from this mean value as predictors. We included Territory ID and year ID as random effects. These and all other models were run with 3 chains leading to 15,000 posteri-or samples (7,000 iterations with 2,000 warm-up iterations per chain). We used weakly
reg-ularizing normal priors for intercepts (μ = 0, σ2 = 5) and beta-coefficients (μ = 0, σ2 = 2) and
half-Cauchy priors for variance components (χ = 0, γ = 1; Gelman 2006). Model convergence and assumptions were checked by visual inspection of chains, Gelman–Rubin diagnostics and posterior predictive checks (McElreath 2015). All parameter estimates are reported as posterior means with 95% credible intervals (2.5–97.5%).
Mean fecundity and fecundity variance
To explore whether cooperative breeding may act as a bet-hedging strategy in the SW, we tested how (per capita) mean fecundity and fecundity variance changed with sociality and with food availability. We fitted the number of fledglings produced (i.e. the number of offspring that fledged the nest) per breeding season for each breeding group as a re-sponse variable in a GLMM with a Poisson error distribution and log link function in the package “brms” (Bürkner 2017). Group composition, annual food availability and territory food availability were included as predictors. We fitted five models representing different hypothesis about the relationship between group composition, food availability and the number of fledglings produced (Table S3.3). For models assessing per capita reproductive success, we changed the Poisson exposure by including log group size as an offset in the different models. Year ID and territory ID were included as random effects.
We assessed whether fecundity variance changed with group composition and with annual food availability by parsing data according to year and group composition, and calculating the coefficient of variation for the per capita number of fledglings produced (CV young fledged = s.d. per capita young fledged / mean per capita young fledged). This was fitted as a
response in a generalized linear mixed model with a lognormal error distribution and year
ID as a random effect. To assess the within-year change in fecundity variance, we included
group composition and annual food availability as predictors, and fitted two separate mod-els, one with, and one without, an interaction between the predictor variables. We
calculat-ed Akaike weights, based on model WAICs, and made averaged predictions by resampling
posterior predictions from each model according to its Akaike weight (McElreath 2015).
Food availability and the benefits of cooperative breeding during different breeding stages
We investigated whether annual and territory food availability, group composition, or the interaction between these predicted nest success during different breeding stages, namely (i) clutch initiation (i.e. whether an egg was laid), (ii) the probability of hatching (i.e. the proportion of nests that reached the hatching stage after an egg was laid), and (iii) the prob-ability of fledging (i.e. the proportion of nest that reached the fledgling stage after hatch-ing). We fitted group composition, mean annual food availability and relative territory food availability as predictors, and included year ID and territory ID as random effects. For each response variable, we fitted five models to represent different hypotheses about the inter-action between group composition, annual food availability, and territory food availability (see Table S3.4). We calculated Akaike weights based on WAIC and made averaged predic-tions by resampling predicted values of each model according to its weight. All analyses were performed in R version 3.3.0 (R Core Team 2016).
RESULTS
Spatio-temporal variation in arthropod abundance
Food availability (i.e. arthropod density) was highly variable between years and between territories (Figs. 3.1A-C). The spatial distribution of arthropods over the island varied
be-tween years, as indicated by the significant tensor product interaction (F187.83 = 2.77, p <
0.001) and the large difference in AIC between the model with and without the interaction
(ΔAIC = 641.5; AICweight for interaction model = 1). The average repeatability of arthropod
density in territories across years was low (R = 0.13), but different from zero (95% CI = 0.09, 0.18). The variability in arthropod density was not consistent across the island, as indicated by a significant spatial component when modelling the coefficient of variation for each ter-ritory (F16.92 = 2.78, p < 0.001; Fig. 3.1C). Between-year differences explained almost four times
as much variation as differences among territories within years (σ2
year: mean = 0.34, 95% CI =
0.21, 0.53; σ2
Mean spatial variation Between-year variability −400 −200 0 100 200 300 response 1 1.1 1.2 1.2 1.3 1.3 1.3 1.4 1.4 1.5 1.5 1.5 1.6 1.6 1.6 1.6 1.7 1.7 1.7 1.8 1.8 1.8 1.8 1.9 1.9 1.9 2 2
C
B
1995 2000 2005 2010 2015 1 2 3 YearLog arthropod density (mean ± 95% CI)
A
−400 −200 0 100 300 −400 −200 0 100 300 response Easting.C No rthing.C 0.22 0.24 0.26 0.28 0.28 0.3 0.3 0.32 0.32 0.34 0.34 0.36 0.36 0.36 0.38 0.38 0.4 0.42 0.44 0.44 0.46 200mN
N
200mFIGURE 3. 1 (A) Shows the mean arthropod density (± 95% CI) on Cousin Island during the main Seychelles warbler breeding
seasons over time (1996-2016, but no data were available for the years 2000-2003 and 2005). In (B) the mean spatial distribution of arthropod density over Cousin Island over 17 years is shown (yellow colours indicate high values, green colours are intermedi-ate and blue colours are low for mean arthropod abundance). In (C) the between-year variability (i.e. coefficient of variation) of arthropod density over the same time period is shown (purple, white and blue indicate high, intermediate and low between-year variability in arthropod numbers, respectively).
Food availability and group formation Delayed dispersal
WAIC-values indicated no meaningful interactions between predictors and the evidence ratio of the highest ranked (additive) model was more than two times higher than the
sec-ond best ranked model (WAICweight = 0.43 vs 0.19; Table S3.1). Offspring were more likely to
disperse when mean annual island food availability was higher (βannual food availability = 1.25, 95%
CI = 0.21, 2.29; Fig. 3.2A). Dispersal by offspring was less likely in territories with higher food availability, but posterior estimates overlapped with zero (βterritory food availability = -0.48, 95% CI =
-1.18, 0.19; Fig. 3.2B). The likelihood of dispersal did not differ between the sexes (βsex = -0.09,
95% CI = -0.70, 0.53). Territory food availability was also positively associated with survival, and this effect was stronger in years with high island food availability (β annual food availability * territory food availability = 0.95, 95% CI = 0.02, 1.92). We found no relationship between survival and annual
food availability overall (β annual food availability = -0.04, 95% CI = -0.91, 0.89), and there were no
dif-ferences in survival between the sexes (βsex= -0.04, 95% CI = -0.91, 0.89).
Changes in group composition
The total proportion of cooperatively breeding groups (all groups with subordinates) vs non-cooperatively breeding groups (all groups without subordinates), decreased with in-creasing annual island food availability (βcooperative vs non-cooperative = -1.14, 95% CI = -2.15, -0.12; Fig.
3.2C) and this decrease was strongest for groups with female subordinates only (βfemale vs no
male and female subordinates (βmale vs no subordinates = -0.43, 95% CI = -1.07, 0.21; male and female
subordinates: βfemale + male vs no subordinates = -0.16, 95% CI = -0.73, 0.42). The total proportion of
coop-eratively breeding groups did not change with territory food availability (βcooperative vs
non-coopera-tive = 0.47, 95% CI = -0.16, 1.14; Fig. 3.2D). There was no change in the proportion of groups with
females (female vs no subordinates = -0.05, 95% CI = -0.31, 0.20), but the proportion of groups with male subordinates and both male and female subordinates tended to increase with increasing territory food availability (male subordinates: βmale vs no subordinates = 0.23, 95% CI = -0.10, 0.56; male and female subordinates: βfemale + male vs no subordinates = 0.29, 95% CI = -0.06, 0.66).
Mean fecundity and fecundity variance Mean fecundity
Mean fecundity was higher in groups with female subordinates and with both female and male subordinates (βfemale vs no subordinates = 0.44, 95% CI = 0.27, 0.60; βfemale + male vs no subordinates = 0.43,
95% CI = 0.20, 0.66; Figs. 3.3A, B), but not in groups with only male subordinates (βmale vs no
subordinates = 0.11, 95% CI = -0.13, 0.35).
A
B
C
D
54 35 15 13 13 14 9 6 5 12 17 0.0 0.2 0.4 0.6 0.8 1.0 Probability of dispersal (mean ± CI) −0.5 0.0 0.5Annual food availability
0.0 0.2 0.4 0.6 0.8 1.0 Propo rtion of groups −0.5 0.0 0.5 −2 −1 0 1 −1.0 0.0 0.5 1.0
Territory food availability-0.5
Annual food availability Territory food availability none female male both
FIGURE 3.2 The upper two panels show the mean probability (± 95% CI) that Seychelles warbler offspring had dispersed at one
year of age in relation to (A) mean annual island food availability and (B) relative territory food availability (i.e. territory devi-ance from the annual mean) at one year of age. Only individuals that were caught as a fledgling (< 3 months of age) and were still alive at one year of age were used (N = 193). Circles in (A) indicate the proportion of individuals dispersing and numbers in circles indicate sample sizes for that year. In the lower two panels, the model predicted mean proportion of groups of different
composi-There was no effect of annual food availability or territory food availability on mean fecun-dity (βannual food availability = 0.11, 95% CI = -0.49, 0.68; βterritory food availability = 0.04, 95% CI = -0.10, 0.17;
Figs. 3.3.A, B), and WAIC values indicated no meaningful interactions between food avail-ability and group composition. There was some support for a model with an interaction
be-tween annual food availability and territory food availability (WAICweight = 0.41; Table S3.3).
Estimates for this interaction, however, overlapped with zero (βannual food availability x territory food avail-ability = 0.15, 95% CI = -0.15, 0.45).
Per capita fecundity
Compared to pairs, the mean per capita fecundity was lower in groups with only male
sub-ordinates (βmale vs no subordinates = -0.31, 95% CI = -0.55, -0.07) and in groups with both male and
female subordinates (βfemale + male vs no subordinates = -0.28, 95% CI = -0.52, -0.05), but not in groups
with only female subordinates (βfemale vs no subordinates = -0.01, 95% CI = -0.19, 0.15; Figs. 3.3C, D).
There was no relationship between annual food availability, or territory food availability and per capita mean fecundity (βannual food availability = 0.12, 95% CI = -0.45, 0.67; βterritory food availability = 0.03, 95% CI = -0.10, 0.17; Figs. 3.3C, D). WAIC values indicated no interactions between food availability and group composition, but showed some support for a model including the
in-teraction between annual food availability and territory food availability (WAICweight = 0.36;
Table S3.3). However, estimates for this interaction overlapped with zero (βannual food availability x
territory food availability = 0.15, 95% CI = -0.16, 0.45).
Fecundity variance
As predicted by the bet-hedging hypothesis, compared to pairs, fecundity variance was low-er for groups with female subordinates (βfemale vs no subordinates = -0.33, 95% CI = -0.49), -0.16; Fig.
3.4, and for groups with female and male subordinates (βfemale + male vs no subordinates = -0.31, 95% CI
= -0.47, -0.15; Fig. 3.4) but not groups with only male subordinates (βmale vs no subordinates = -0.07,
95% CI = -0.24, 0.09; Fig. 3.4). Fecundity variance did not depend on annual food availability (βannual food availability = -0.01, 95% CI = -0.40, 0.39).
Food availability and the benefits of cooperative breeding during different breeding stages Clutch initiation
Model weights indicated (WAICweight = 0.79; Table S3.4) that group composition affected nest
initiation depending on annual food availability. Female subordinates had a larger effect on the likelihood that groups initiated a clutch when annual food availability was high, then when it was low compared to groups without subordinates (βfemale vs no subordinates x annual food
availability = 1.47, 95% CI = 0.27, 2.79), but such effects were not present for group with males (βmale vs no subordinates x annual food availability = 1.00, 95% CI = -0.58, 2,71) and were, possibly, reversed in groups
-2.78, 0.23; Fig. 3.5A). Territory food availability was not related to the likelihood that a group initiated a clutch (βterritory food availability = -0.02, 95% CI = -0.35, 0.32).
none Subordinates presentfemale male both
0.0 0.2 0.4 0.6
Mean young fledged
0.0 0.1 0.2 0.3 −0.5 0.0 0.5 −2 −1 0 1 Per capita
mean young fledged
Annual food availability Territory food availability
A B
C D
FIGURE 3.3 The mean total number of fledglings produced (A, B) and mean per capita number of fledglings produced (C, D) in
relation to annual island food availability and territory food availability by groups with different group compositions, consisting of either no subordinates (N = 924, black line), at least one female subordinate (N = 342, red line), at least one male subordinate (N = 186, blue line), or at least one female and one male subordinate (N = 143, green line). Lines indicate model averaged predicted mean effects based on WAIC values.
0.5 1.0 1.5 2.0 CV young fledged
none female male both Subordinates present
β = -0.33* β = -0.07
β = -0.31*
FIGURE 3.4 Within-year fecundity variance (CV young fledged) for groups with either no subordinates (N = 924), at least one female
subordinate (N = 342), at least one male subordinate (N = 186), and both male and female subordinates (N = 143). Beta coefficients refer to the difference with groups without subordinates and estimates that do not overlap with zero are followed by an asterisk.
There was some evidence that groups were more likely to initiate a clutch overall in years with higher food availability (βannual food availability = 0.85, 95% CI = -0.07, 1.78; Fig. 3.5A). Only in
groups with female subordinates did the likelihood of clutch initiation increase with annu-al food availability βfemale = 2.18, 95% CI = 0.79, 3.67).
none Subordinates presentfemale male both
0.6 0.8 1.0 0.6 0.8 1.0 0.6 0.8 1.0 −0.5 0.0 0.5 −2 −1 0 1
Annual food availability Territory food availability
Mean probability of clutch initiation
Mean probability of hatching
Mean probability
of fledging
A
B
C
D
E
F
FIGURE 3.5 The relationship between group composition, and annual and territory food availability on nest success during
dif-ferent breeding stages: the probability of clutch initiation (A, B), the probability of hatching given an egg was laid (C, D), the probability of fledging given a nest contained a nestling (E, F). Lines indicate model-averaged predicted mean effects based on WAIC values. Sample sizes for each analysis are N = 1595 for clutch initiation, N = 1446 for hatching success and N = 1009 for fledging success.
Hatching success
There was no association between the likelihood that a clutch would hatch and annual or territory food availability (βannual food availability = 0.02, 95% CI = -0.66, 0.68; βterritory food availability = 0.08, 95% CI = -0.11, 0.28; Figs. 3.5C, D). However, hatching success was higher when groups had a female subordinate (βfemale vs no subordinates = 0.39, 95% CI = 0.14, 0.64), but not when groups
had either male subordinates, or both male and female subordinates (βmale vs no subordinates =
0.13, 95% CI = -0.17, 0.44; βmale and female subordinates vs no subordinate = 0.30, 95% CI = -0.04, 0.66; Figs. 3.5C, D). WAIC values indicated limited support for interactions between food availability and group composition (Table S3.4).
Fledging success
The likelihood that nestlings would fledge was not associated with annual food availability or territory food availability overall (βannual food availability = -0.04, 95% CI = -0.83, 0.74; βterritory food availability = 0.01, 95% CI = -0.25, 0.27). However, WAIC values indicated support for the model
containing an interaction between group composition and annual food availability
(WAIC-weight = 0.38; Table S3.4). Fledging success improved with increasing annual food availability
in groups with female subordinates, compared to groups without subordinates (βfemale vs no
subordinates x annual food availability = 0.86, 95% CI = 0.06, 1.68), but not for groups with only male
subor-dinates, or groups with both male and female subordinates (βmale vs no subordinates = 0.46, 95% CI
= -0.57, 1.49; βboth vs none = 0.68, 95% CI = -0.37, 1.74; Fig. 3.5E, F). However, fledging success did
not improve overall with increasing annual food availability in groups with female subor-dinates (βannual food availability = -0.50, 95% CI = -0.49, 1.50), nor in groups with other group
com-positions.
DISCUSSION
Our study suggests that temporal and spatial variation in food availability play a key role as drivers of delayed dispersal and group formation in Seychelles warblers. Consistent with the bet-hedging hypothesis for cooperative breeding, we found that the presence of sub-ordinate females, but not males, in a group, reduces fecundity variance. While spatio-tem-poral variation in food availability had no effect on the number of fledglings that were produced in a territory, female subordinates improved total reproductive success of their group, resulting in similar per capita fecundity as groups without subordinates. Converse-ly, male subordinates had no effect on fecundity, and consequentConverse-ly, per capita fecundity was lower in groups with male subordinates compared to group without subordinates.
Food availability and group formation
We show that, in the Seychelles warbler, high annual food availability increases the likeli-hood that offspring disperse. This is arguably due to the decreased costs of dispersal and/ or independent breeding (Emlen 1982; Kingma et al. 2016a, b). The idea of increased costs of breeding in poorer conditions is supported by our finding that overall clutch initiation tends to decline with lower annual food availability (Fig. 3.4A). These results highlight the importance of environmental variability in the formation of groups, which is often a pre-requisite for the evolution of cooperative breeding (Griesser et al. 2017). Furthermore, in-creased dispersal during years with high food availability leads to fewer cooperative groups (Fig. 3.2C), and smaller groups in years with high food availability (Fig. S3.1B). This means that the interpretation of associations between reduced fecundity variance and average group size sensu Rubenstein (2011) and Koenig and Walters (2015), depends on the relation-ship between group size and annual food conditions. If there is a negative relationrelation-ship be-tween group size and environmental conditions (as in this study), large groups will tend to form under poorer conditions, which can mask, or even reverse, any mean effect of either group size, or conditions, on fecundity variance. However, if the opposite is true, and large groups tend to be associated with good conditions, than any average effect of group size on fecundity variance could be the result of improved food conditions. Temporal fluctuations in the costs of independent breeding have also been shown to lead to group formation in other cooperatively breeding species (Canario, Matos & Soler 2004; Covas, Doutrelant & du Plessis 2004; Koenig et al. 2011), which suggests that such effects are common. Future re-search to investigate the effect on of cooperative breeding on fecundity variance should thus be aware of such correlations, because of the potential covariance between environ-mental conditions and group size.
Evidence that subordinates were more likely to delay dispersal in high-quality territories than in low quality territories – which would be expected based on previous findings in the Seychelles warbler (Komdeur 1992) – was weak (Fig. 3.2B). This result could have been par-tially confounded by differential mortality risk, with subordinates from territories with low food availability being more likely to have disappeared before the next season. The fact that spatial variance in territory quality seems to play a lesser role in this study could also be due to territory quality differences becoming less pronounced as a result of vegetation recov-ery across the island since 1992 (Komdeur & Pels 2005; Eikenaar et al. 2010). Interestingly, a decrease in consistent spatial variation, in combination with increasing temporal fluctua-tions, could mean that the cooperative breeding system of the Seychelles warbler is shifting from being mainly benefits driven, to being mainly constraints driven. However, another analyses revealed that consistent spatial variation in food availability was positively cor-related with group size, confirming that territories with high mean food availability had
larger groups on average (Fig. S3.1A). This result corroborates the earlier results of Komdeur (1992) and supports the benefits-of-philopatry hypothesis as a driver of delayed dispersal in this species. Both constraints (i.e. the costs of dispersal and/or independent breeding) and the benefits that can be obtained in the natal territory (e.g. access to food) thus play a role in dispersal and, therefore, in group formation in the Seychelles warbler.
Mean fecundity and fecundity variance
If per capita fecundity increased with group size, then cooperative breeding would be more parsimoniously explained by such increases in per capita fecundity than by reducing fecun-dity variance (Starrfelt & Kokko 2012; Shen et al. 2017). In the Seychelles warbler, the per cap-ita fledgling output is lower in groups that contain any male subordinates, but not groups with female subordinates, compared to groups without any subordinates (Figs 3.3C, D). Overall, group fecundity increases in groups with only female subordinates and in groups with both male and female subordinates (Figs 3.3A, B). This difference in fecundity between groups with male and female subordinates is probably because female subordinates help more, and more often than males, and also co-breed (Richardson et al. 2001; Richardson, Burke & Komdeur 2003a).
The bet-hedging hypothesis for cooperative breeding predicts that fecundity variance should decrease with increasing sociality. Here we show that while male subordinates had no effect on fecundity variance, there was a significant decrease in fecundity variance when female subordinates were present in SW groups (Fig. 3.4). Sociality has been linked to reductions in fecundity variance in superb starlings Lamprotornis superbus (Rubenstein 2011), and acorn woodpeckers Melanerpes formicivorus (Koenig & Walters 2015), but only the latter study investigated how within-year fecundity variance is reduced by groups of dif-ferent composition. One notable difference is that, in the acorn woodpecker, subordinate (co-breeding) males show the highest decrease in fecundity variance, while our study shows the highest decrease for groups with female subordinates. This difference is likely due to dif-ferences in group composition and helping behaviour, since the subordinate sex ratio and helping are male biased in the acorn woodpecker (Koenig, Walters & Haydock 2016) and female biased in the Seychelles warbler (Komdeur 1996a; Richardson et al. 2002). Koenig and Walters (2015) were able to obtain fitness estimates based on the assumption that there was a linear relationship between reproductive success and the acorn crop (Frank & Slatkin 1990) and, based on these estimates, argued that the reductions in fecundity variance were probably insufficient to compensate for the overall decrease in per capita reproductive suc-cess. Due to the more complicated set of conditions encompassing reproductive success in the Seychelles warbler (e.g. density-dependence, Brouwer et al. 2009) and fecundity being expressed at the group level, we cannot deduce the effect on fitness in our current dataset.
However, we point out that, for groups containing female subordinates in the Seychelles warbler, per capita fecundity does not decline in groups with female subordinates, and that the decrease in fecundity variance is stronger compared to the acorn woodpecker. This would argue in favour of bet- hedging being potentially more important as a benefit of co-operative breeding in the Seychelles warbler than in the acorn woodpecker.
Food availability and the benefits of cooperative breeding during different breeding stages
A commonly proposed way in which subordinates can reduce fecundity variance is by im-proving fecundity during adverse food conditions (i.e. “temporal variability hypothesis”; Rubenstein & Lovette 2007), presumably by preventing nestling starvation (Rubenstein 2011). However, subordinates could also reduce nest failure during other breeding stages. In the Seychelles warbler, subordinate females most consistently prevented nest failure both during hatching and fledging (Figs. 3.5C, E). In addition, only in groups with female subor-dinates was there a positive relationship between the likelihood of clutch initiation and annual food availability (Fig. 3.5A). Female subordinates did not improve fecundity more during adverse food conditions, in contrast with the temporal variability hypothesis, and a commonly assumed mechanism of bet- hedging. Subordinate male SWs did not appear to reduce nest failure, although subordinate males did improve the likelihood of clutch initiation.
Interestingly, groups with subordinate females had a lower probability of hatch failure. Egg predation – an important cause of nest failure in the Seychelles warbler – is significantly reduced by female subordinate incubation (Komdeur & Kats 1999; Kingma et al. forthcom-ing). Whether subordinates could reduce fecundity variance by reducing nest predation, re-mains unexplored. This is an interesting possibility because nest predation potentially has a stronger effect on reproductive variance than nestling starvation through reduced food availability, because it often leads to the loss of whole broods, while starvation only leads to brood reduction (Ricklefs 1969a). Alternatively, female subordinates could buffer against other forms of stochasticity on a smaller scale than we have measured here. For example, short periods of heavy rain during the breeding season often make it difficult for Seychelles warblers to provide their offspring with enough food. Such small-scale stochastic effects are difficult to incorporate in this study, but the presence of one or more subordinates could mean the difference between breeding successfully or not.
Conclusion
In conclusion, our study shows that in the Seychelles warbler, spatio-temporal variation in food availability is associated with delayed dispersal, a key step to cooperative
breed-ing. The presence of female subordinates reduces reproductive variance, and per capita fe-cundity does not decline in groups with female subordinates, which suggests that reduced fecundity variance does not have to compensate for a large loss in per capita fecundity to provide fitness benefits. However, the reduction in reproductive variance by subordinate females is not due to subordinates improving reproductive success under adverse condi-tions, as is commonly suggested by studies invoking bet hedging. In the SW, subordinate females reduce nest failure by improving the likelihood of clutch initiation, hatch success and fledging success. The extent to which reproductive variance contributes to fitness in the Seychelles warbler now requires further investigation.
So far, few studies have investigated reduced fecundity variance as a consequence of cooper-ative breeding (Reed & Walters 1996; Rubenstein 2011; Koenig & Walters 2015), however, data on mean reproductive success and reproductive variance are readily available for many co-operative, and non-coco-operative, breeding species. It would be worthwhile to test whether species living in the same habitats and differing only in social system show differences in fecundity variance as a result of environmental fluctuations (Jetz & Rubenstein 2011; Corn-wallis et al. 2017). In addition to changing mean temperatures, climate change also affects the strength and frequency of weather extremes, thereby increasing environmental fluc-tuations (Garcia et al. 2014). If cooperative breeding is indeed a bet-hedging strategy that has evolved because it allows individuals to better cope with environmental perturbations, we would expect cooperatively breeding species to be less affected by changing climatic patterns than non-cooperatively breeding species. Investigating these patterns on a larger scale thus provides another important way to assess the importance of changing environ-ments on the evolution of sociality and cooperative breeding.
Acknowledgments
We thank Nature Seychelles for facilitating the long-term Seychelles warbler project and the Seychelles Bureau of Standards and Department of Environment for permission for sam-pling and fieldwork. We thank everyone who has helped in the field, with laboratory work and with database management, and the overall Seychelles Warbler Research Group for dis-cussions. The long-term data collection was funded by two Natural Environment Research Council (NERC) grants (NE/F02083X/1 and NE/K005502/1) and by NERC (NER/I/S/2002/00712) and Marie Curie fellowships (HPMF-CT-2000-01074) to DSR, NERC grants to TB and by grants from the Netherlands Foundation for the Advancement of Tropical Research (WOTRO, 84-519) and the Netherlands Organisation for Scientific Research (NWO) to JK (NWO-TOP, 854.11.003). SAK and MH were funded by NWO-VENI fellowships (863.13.017 and 863.15.020).
SUPPLEMENTARY INFORMATION
We investigated whether food availability predicted group size by testing whether territo-ries with higher mean food availability have larger groups, and whether annual changes in food ability were associated with changes in group size. We fitted the number of subor-dinates as a response variable in a generalized linear model with a Poisson error distribu-tion and a log link funcdistribu-tion, and fitted mean territory-level food availability and the annu-al deviance from this mean vannu-alue as predictors. We fitted four different models (see Table S3.1) and obtained averaged predictions based on WAIC (“Widely Applicable Information Criterion”; Watanabe 2010) by resampling predicted values of each model according to its Akaike weight.
Group size was positively correlated with mean territory food availability (β = 0.48, 95% CI = 0.24, 0.72; Fig. S3.1A). Furthermore, group size declined with higher annual island food availability (β = -0.14, 95% CI = -0.24, -0.03; Fig. S3.1B). In other words, groups became smaller in years when food availability was relatively high and larger in years when food availability was relatively low, which is in agreement with our results on dispersal, where dispersal in-creased with mean annual food availability.
1.5 2.0 2.5 3.0 3.5 Mean territory food availability 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 Number of subordinates (mean ± 95% CI) −1.5 −0.5 0.0 0.5 1.0 1.5 Annual deviance in food availability 0.0 0.2 0.4 0.6 0.8 1.0 Number of subordinates (mean ± 95% CI) −1.0
A
B
FIGURE S3.1 Changes in group size (i.e. the number of subordinates) in relation to mean territory food availability (A) and the annual
deviance in food availability from the territory mean (B). Solid and dashed lines represent model predicted mean values and 95% CI, respectively.
We determined the age of all breeding females and breeding males in each year on the basis of the breeding season in which they were born with 6 month increments (i.e. individuals born during the minor breeding season were classified as being 6 months old in the next major breeding season, and one year old in the following major breeding season etc.). To investigate potential changes in breeder age as a result of patterns of dispersal, we fitted
dominant breeder age for females and males in a multivariate response model with annual food availability and group composition as predictor variables.
4 5 6 7
Dominant breeder age (
years) −0.5 0.0 0.5 none female male both
Mean annual food availability (between-year)
A
Dominant femaleB
Dominant male−0.5 0.0 0.5
Mean annual food availability (between-year)
FIGURE S3.2 Changes in the average age of dominant breeding females (A) and males (B) in relation to mean annual food
availabil-ity for groups containing no subordinates (black), only female subordinates (red), only male subordinates (blue) or both female and male subordinates (green).
WAIC values indicated minimal support for the model with an interaction between group composition and annual food availability on breeder age (WAIC weight = 0.05). There was
a positive residual correlation between breeding female age and breeding male age (σ2 =
0.15, 95% CI = 0.10, 0.20). Breeder age decreased with lower annual food availability, for both female and male breeders (breeding females: β = -0.73, 95% CI = -1.09, -0.37; breeding males: β = -0.54, 95% CI = -0.89, -0.19; Fig. S3.2A, B), and groups with subordinate females and with
both subordinate males and females had older dominant breeders (breeding females: βfemale
vs no subordinates = 0.96, 95% CI = 0.58, 1.34; βfemale + male vs no subordinates = 0.76, 95% CI = 0.22, 1.30; breeding
males: βfemale vs no subordinates = 0.49, 95% CI = 0.11, 0.86; ; βfemale + male vs no subordinates = 0.82, 95% CI = 0.29,
1.35; Fig. S3.2A, B)
TABLE S3.1. Comparisons for models investigating the relationship between the likelihood of dispersal, and food availability.
Probability of dispersal (N = 193)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βAF + βTF + βSex 263.22 0.43
α + βTF * βSex + βAF 264.77 1.55 0.19 2.24
α + βAF * βSex + βTF 264.82 1.6 0.19 2.30
α + βAF * βTF + βSex 264.87 1.65 0.17 2.55
α + βAF * βTF * βSex 269.91 6.69 0.01 30.35
TABLE S3.2. Comparisons for models investigating the relationship between the likelihood of survival, food availability offspring sex.
Survival (N =339)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βAF * βTF + βSex 465.48 0.54
α + βAF + βTF + βSex 467.55 2.07 0.19 2.82
α + βAF * βSex + βTF 468.81 3.33 0.10 5.30
α + βTF * βSex + βAF 468.89 3.41 0.10 5.52
α + βAF * βTF * βSex 469.6 4.12 0.07 7.85
AF = annual island food availability; TF = territory food availability; Sex = sex
TABLE S3.3. Comparisons for models investigating the relationship between the number of young fledged per breed group in each year,
food availability and group composition.
Young fledged (N = 1595)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βGC + βAF + βTF 2682.31 0.45
α + βGC + βTF * βAF 2682.52 0.21 0.41 1.11
α + βGC * βAF + βTF 2685.62 3.31 0.09 5.22
α + βGC * βTF + βAF 2686.73 4.42 0.05 9.11
α + βGC * βAF + βGC * βTF 2690.55 8.24 0.01 61.70
GC = group composition; AF = annual island food availability; TF = territory food availability
Per capita young fledged (N = 1595)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βGC + βAF + βTF 2671.84 0.513891
α + βGC + βTF * βAF 2672.54 0.7 0.361738 1.420616
α + βGC * βAF + βTF 2675.71 3.87 0.074132 6.932106
α + βGC * βTF + βAF 2676.81 4.97 0.042919 11.97344
α + βGC * βAF + βGC * βTF 2680.34 8.5 0.00732 70.19767
TABLE S3.4. Comparisons for models investigating the relationship between the probability of clutch initiation, hatching and fledging,
in relation to food availability and group composition.
Probability of clutch initiation (N = 1595)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βGC * βAF + βTF 936.02 0.79
α + βGC * βAF + βGC * βTF 939.64 3.62 0.13 6.11
α + βGC + βAF + βTF 941.49 5.47 0.05 15.37
α + βGC + βTF * βAF 943.21 7.19 0.02 36.39
α + βGC * βTF + βAF 945.33 9.31 0.01 105.02
GC = group composition; AF = annual island food availability; TF = territory food availability
Probability of hatching (N = 1446)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βGC + βAF + βTF 2220.81 0.64
α + βGC + βTF * βAF 2222.94 2.13 0.22 2.90
α + βGC * βAF + βTF 2224.54 3.73 0.10 6.45
α + βGC * βTF + βAF 2226.83 6.02 0.03 20.28
α + βGC * βAF + βGC * βTF 2230.88 10.07 0.00 153.31
GC = group composition; AF = annual island food availability; TF = territory food availability
Probability of fledging (N = 1009)
Model structure WAIC ΔWAIC Weight Evidence-ratio
α + βGC + βAF + βTF 1328.67 0.39
α + βGC * βAF + βTF 1328.72 0.05 0.38 1.02
α + βGC + βTF * βAF 1330.67 2 0.14 2.72
α + βGC * βAF + βGC * βTF 1331.85 3.18 0.08 4.89
α + βGC * βTF + βAF 1375.99 47.32 0.00 1.88E+10
