Population age structure shapes selection on social behaviour in a long-lived insect

Abstract

Social traits are expected to experience highly context-dependent selection, but we know little about the contextual factors that shape selection on social behaviours. We hypothesized that the fitness consequences of social interactions will depend on the age of social partners, and therefore that population age structure will shape evolutionary pressures on sociality. Here, we investigate the consequences of age variation at multiple levels of social organization for both individual fitness and sexual selection on social network traits. We experimentally manipulated the age composition of populations of the forked fungus beetle Bolitotherus cornutus, creating 12 replicate populations with either young or old age structures. We found that fitness is associated with variance in age at three different levels of organization: the individual, interacting social partners, and the population. Older individuals have higher reproductive success, males pay a fitness cost when they interact with old males and females achieve lower fitness in older populations. In addition to influencing fitness, population age structure also altered the selection acting on social network position in females. Female sociality is under positive selection only in old populations. Our results highlight age structure as an understudied demographic variable shaping the landscape of selection on social behaviour.

This article is part of the discussion meeting issue ‘Understanding age and society using natural populations’.

1. Introduction

Selection regimes are shaped by ecology and demography. The strength and direction of natural selection may be affected by proximate variables such as population density or intensity of predation, as well as large-scale environmental factors such as precipitation [1]. The relationships between sexual selection and factors such as operational sex ratio [2,3]; population density, size and crowding [4–6]; and group phenotypic composition [7–9] have been well studied. However, relatively little is known about what drives variation in any form of selection on social behaviours. Due to their labile and interactive nature, social traits are expected to experience more context-dependent selection than other traits [10–13]. Here, we investigate whether population age structure is one of the contextual factors shaping selection on social behaviours.

Population age structure, which we define as the composition of individual ages within a population or deme, can depend on the time since founding, habitat connectivity and recruitment, density, climate, severe weather events, intraspecific competition and parasitism or predation [14–19]. Age structures of populations or subpopulations are thus highly variable in space and time [20]. The consequences of this variation for population dynamics have been extensively studied by ecologists. Less attention has been given to possible consequences of variation in age structure for the strength and direction of selection.

Age is associated with many factors that can impact the fitness of conspecifics with whom an individual interacts: variation in information [21,22], foraging ability and energy reserves [23,24] and immune function and infection status [25,26], as well as a variety of social behaviours [27–29], competitive ability [30,31] and mating success or extra-pair paternity [32–35]. An individual’s fitness may depend on the age of the conspecifics with whom that individual either directly interacts or shares group membership. For example, interacting with an older social partner whose accumulated experience has made them a strong competitor but who has a contagious illness due to their senescing immune system would likely have very different consequences than interacting with a young, healthy individual who is a weak competitor.

If the fitness consequences of social interactions depend on the age of social partners, then we would expect the selection on social traits to depend on the ages of all potential partners [36]—being highly social will be beneficial when interactions with partners are likely to increase fitness, but not when interacting is costly. Differences in selection on social behaviour among populations could therefore be caused by differences in age structure.

Social network position is one way of describing variation in social behaviour. Metrics of social network position describe where an individual falls within the social structure of all interactions within a population, which modulates exposure to information, parasites, disease, grooming and other positive and negative fitness consequences of social interactions [37–45]. Many of these fitness consequences of interaction are due to traits associated with age, and so the net fitness effect of sociality will depend on the age of possible partners. We hypothesized that selection on an individual’s position within the social network of a population will depend on the age structure of that population. We tested this hypothesis by studying how the age of directly interacting partners and all members of the population, shapes fitness and selective landscapes in the forked fungus beetle Bolitotherus cornutus.

Selection has been well studied in forked fungus beetles. In a classic example of demography shaping sexual selection, selection on horn length was found to be stronger in subpopulations where the density of males relative to fungal brackets in the population is low [5]. Male morphology is also under social selection: males who interact with large male social partners pay a fitness cost [9]. We have also measured sexual selection on social network traits, although its mechanisms remain unknown. Selection on social network position is more variable than that on morphology, with the strength and even direction of selection differing among wild subpopulations [12], but it is not clear what causes this variation. Work in semi-natural mesocosm populations has shown that selection on social network position is affected by experimentally altering the spatial distribution of resources [46], but wild subpopulations vary along many other physical and demographic variables other than resource distribution, including age structure ([47], PA Cook and A Medina-Valencia 2021, unpublished data).

Forked fungus beetles are long-lived. Although roughly a third of adults survive for less than a month after emergence [32], some have been recaptured as long as 5 years after initial capture (V Formica 2022, unpublished data), leading to significant age variation within populations. Age is known to be associated with variation in several behavioural traits in this species. Older males are more aggressive [48], and may therefore be more likely to win fights, as aggression is associated with contest outcomes in male–male interactions [49]. Individuals of both sexes alter their social behaviour as they age, participating in fewer non-mating social interactions and occupying more peripheral positions within the social network of all interactions within their population [50]. These changes scale up, such that the overall social network structure of a subpopulation depends on its age composition [50]. Older individuals also achieve higher reproductive success, although we do not know whether offspring quality changes with parental age [32,51].

Given the extensive age-related variation and unexplained variation in selective pressures on social behaviour documented in this system, we hypothesized that population age structure impacts both individual fitness and sexual selection through changes in the social environments experienced by individuals. We experimentally manipulated mesocosm populations to ask whether individual age, social neighbourhood age and population age explain variation in individual reproductive success and whether age structure shapes selection on sociality. This approach allows for strong experimental tests of how population demographic factors alter selection regimes; past work has shown that this design has sufficient power to detect differences in selection [52]. Controlled replicate populations are especially useful for answering otherwise intractable questions about social behaviours [53,54].

2. Material and methods

(a). Study system

Bolitotherus cornutus is a long-lived tenebrionid beetle found throughout the forests of eastern North America [47]. The metapopulation occupying a region of forest is spatially subdivided into subpopulations, each consisting of the beetles living on a single dead standing tree or fallen log. These logs provide resource patches of the wood-decaying fungi in which larvae and pupae develop and on which adults feed, mate and oviposit [55,56]. This species is not eusocial, but the concentration of resources means that beetles on the same log interact with each other frequently. Male B. cornutus use their thoracic horns in competitive interactions over access to females. Males vary substantially in both body and horn size. Larger horns allow a male to pry mate-guarding competitors off of females, and larger body size can make a male difficult to remove [57]. This species is holometabolous—after they eclose and emerge from the fungus brackets, adults do not moult or grow over time. Size is therefore not associated with adult age.

Since 2017, we have maintained a captive breeding colony of B. cornutus at Mountain Lake Biological Station (37°22'37.0"N, 80°31'17.5"W). The captive colony founders were collected from the surrounding area in the spring of 2017 and 2018. These founders have been allowed to interact and reproduce within 2.4 by 2.4 by 1.2 m screened experimental enclosures built to mimic resource patches in the forest where this species naturally occurs. Each enclosure contained 18 brackets of the B. cornutus host fungus species Ganoderma tsugae (Sharondale Mushroom Farm) growing out of bags of hardwood sawdust held on wooden shelves, which act as ‘artificial logs’ (for images, see Costello et al. and Cook et al. [46,50]. Individuals in the colony have thus been allowed to experience semi-natural conditions but are shielded from predation and prevented from emigrating. Each year at the start of the season we collect the offspring that emerged in late autumn or early spring, and we continue to search for new offspring through the field season until the end of the summer. Newly caught offspring, who may emerge any time from early spring to late autumn, are marked with a unique three-character code affixed to their elytra. Each individual is also imaged with a flatbed scanner so that even if both labels are lost they can still be identified by unique patterns on the pronotum.

By identifying newly enclosed beetles each spring, we created three age cohorts for our experiment. The founders, adults in the spring of 2017 or 2018, were therefore at least 3 years old during our experiment in the summer of 2020. Offspring collected in early 2019 are considered 2-year-olds in 2020, as they likely emerged in autumn 2018, and newly emerged adults collected in early 2020 are ‘first-year’ adults.

(b). Experimental design and data

In early August 2020, 432 beetles were placed into 12 experimental populations, six each of two different treatments: ‘young populations’ and ‘old populations’. Each of the six ‘young populations’ consisted of 18 first-years and 18 two-year-olds, and each of the six ‘old populations’ consisted of 18 two-year-olds and 18 beetles that were three or more years old. Two-year-olds were included in both treatments to keep the age structures close to those found in natural populations and to allow for comparison across treatments. Both the physical environment (enclosure size and layout; fungal resource age, size, genotype and distribution in space) and population demographics (population size and density, sex ratio and average body size) of the replicates were controlled, with the only difference between treatments being age structure (for more details of population assignment methods, see Cook et al. [50]). The number of males per resource, a 1:1 ratio, in each population was designed to match conditions under which strong sexual selection has been observed in the wild [5]. Individuals were kept in isolation for a minimum of 7 days before being placed into the enclosure to allow patterns of social interaction to ‘reset’ [58] and allowed to acclimatize to the enclosures for 36 h before we began behavioural observations.

We performed scan sampling of behaviour three times a day (0630–0930, 1430–1630 and 2130–0030) for 21 days in August of 2020, noting the behaviour and social partners of all visible, identifiable individuals. Male reproductive success was estimated as the number of successful insemination events, which can be identified from the stereotyped mate-guarding behaviour that follows [59]. Female reproductive success was estimated as the number of observed oviposition events, which is the same as the number of eggs laid because females lay only one egg at a time. Observer identity and the order in which populations were surveyed were randomized every survey period, to prevent treatment from being confounded with possible effects of time or observer identity. Observers could not be made blind to which populations had young or old age structures because the identification codes used to label individuals have progressed in a predictable sequence over the years. However, the majority of the observers were not aware of the research questions, and no a priori predictions as to the direction of effects had been made when data collection was underway.

To characterize each individual’s social network position, we recorded all social interactions during our sampling periods. Social partners were defined as individuals who were physically touching or within 5 cm of each other, which is the threshold at which beetles begin to visibly react to each others’ presence [39], but excluded courting, copulating and mate-guarding males and female pairs to allow for the measurement of selection without creating autocorrelation between our social phenotypes and fitness measures. For each individual, we calculated the average age of male social partners, weighted by the number of interactions with each partner. We chose to focus on the age of male social partners because male partners influence male fitness through competition [9], and indirect evidence suggests that interactions with males are correlated with female laying [46].

The set of all interactions within each population was converted to a weighted, unidirectional social network. For each possible pair of individuals within the population, we calculated the simple ratio index, which is defined as the proportion of all survey periods at least one of the pair was observed that they were observed together [60]. We then calculated how socially connected each individual was using the network metric strength, which measures an individual’s weighted number of interactions, using the R package tnet [61]. The calculation of strength can be tuned to adjust whether and how repeated interactions with the same partner are counted; following our past work in this system, we used a tuning parameter of 0.5, which means that the first interaction with a new partner adds 1 unit to strength and subsequent interactions 0.5 units [61]. Network strength in both-sex networks is repeatable in this species [58,62], and selection on strength is highly variable among subpopulations in the wild [12].

Twenty-three beetles died over the course of the experiment. These individuals were included in the measurement of social neighbourhoods and networks but were removed from all subsequent analyses. The full dataset, along with code for analyses, is archived on the Dryad Digital Repository [63]. Note that the same dataset was used for a previous analysis of the relationship between age and social behaviour [50].

(c). Statistical analysis

We modelled reproductive success using two generalized linear mixed models, one for males and one for females. The sexes were analysed separately because past work has shown that selection is often sex-specific [52] and because their reproductive success is estimated with different fitness components. In the lab, females of this species often lay fertile eggs after long periods of isolation, meaning that they can store sperm and therefore parentage cannot be confidently attributed to individual males. Counts of mate-guards, which indicate successful inseminations, were scored for males and counts of oviposition behaviours, which indicate the laying of a single egg, were scored for females. Additionally, these models included two covariates known to explain fitness variation in this species, the number of survey periods an individual was active and body size. Body size was estimated as the length of the elytra, measured in ImageJ [64], and was globally standardized because body size is under hard selection [12].

We included age at the following three levels of social organization: age, mean age of immediate social partner and age structure of the population. Past work has found that reproductive success is associated with age [51], so we included individual age in these models as a categorical fixed effect. To assess the impact of social partner age on fitness, we included the mean age of all males with which each individual directly interacted. To test whether population age structure impacts individual fitness, we included the age structure (either young or old) of the population, with the identity of the replicate population included as a random effect.

The last term in our mixed models tested whether selection on sociality, measured as network strength, differed between young and old populations. We included an interaction between standardized strength and population age structure. We know from past work that this design has sufficient power to detect differences in selection regimes between two conditions [52].

Counts of successful inseminations (for males) and number of eggs laid (for females) were modelled assuming a negative binomial error distribution in the R package glmmTMB [65], and model assumptions were checked visually and statistically using DHARMa [66]. Marginal means were calculated in emmeans [67]. Because social network data violate the assumptions of conventional statistical tests, we assessed the significance of these models using a permutation approach [68,69]. We used node, rather than datastream, permutations to avoid inflating our false positive rate [62,70]. Each permutation randomized all variables used in the linear mixed models without replacement across individuals, breaking any relationships between those variables and fitness measures. We repeated this process 2000 times and ran the mixed models in each of these permuted datasets. The statistical significance of our observed test statistics was determined by comparison to the distribution of those in the permuted datasets. P-values were calculated as the proportion of all test statistics greater than the observed estimate.

In addition to asking whether population age structure predicts the number of times we observed females laying eggs, we also looked at whether population age structure affected total reproductive output with a separate analysis. We counted the total number of egg scars on the surface of the fungus brackets in each population. An egg scar is a distinctive, long-lasting mark created when a female covers an egg with frass. These eggs cannot be attributed to specific mothers but include the eggs laid outside of our survey periods. The total number of all eggs laid was compared between the young and old populations using a t‐test to check that the observed behavioural data were a representative sample of all fitness events. All analyses were performed in R 4.1 [71].

3. Results

Individual age and social partner age, but not population age structure, explained the variation in reproductive success among males (table 1). Both 2- and 3-year males were observed mate-guarding females significantly more often than were first-year adults (figure 1a; post hoc pairwise contrasts t = −8.78, p < 0.0001 and t = −5.11, p < 0.0001, respectively), but 3-year-olds had lower fitness than 2-year-old males (t = 3.68, p = 0.009). Males with higher average social partner age had fewer successful inseminations than males who interacted with younger males (figure 1b), but overall population age structure did not explain additional variation in male fitness (figure 1c). Male body size and network strength were both under positive selection, and selection on social network position did not differ with population age structure (figure 1d). Activity level was not associated with male fitness.

Table 1.

Fixed effects on male reproductive success. Model estimates (slopes) are reported for continuous variables, back-transformed marginal means for categorical variables and back-transformed marginal trends for the interaction. P-values are calculated as the proportion of 2000 permuted model F-statistics greater than the observed model estimate. P-values significant at the α = 0.05 level are in bold.

	estimate	p‐value
scans active	0.01	0.167
body size (mm)	0.23	0.001
individual age	one: 0.48	<0.001
	two: 3.05
	three+: 1.85
male social partner age	−0.18	0.003
population age structure	young: 1.46	0.499
	old: 1.33
network strength	0.31	<0.001
age structure × strength	young: 0.29	0.802
	old: 0.32

Figure 1.

Selected fixed effects from the model of male reproductive success, measured as the number of mate-guarding events observed for that individual. (a) Back-transformed marginal means of reproductive success for the three age cohorts. (b) Marginal effects of average male partner age on male fitness, with raw data plotted. (c) Back-transformed marginal means of male reproductive success in the two age structure treatments. Note that this is the main effect involved in an interaction. (d) Marginal effects of individual network strength on male fitness, subset by population age structure, with raw data plotted. Lighter points represent raw data from young populations, and darker points denote old populations. Selection on male strength is positive and similar in magnitude in both treatments.

Individual age and population age explained variation in female reproductive success, but the average age of male partners did not (table 2). Two- and 3-year-old females laid more eggs than did 1-year-olds (figure 2a; post hoc pairwise contrasts t = −6.29, p ≤ 0.0001 and t = −4.28, p = 0.0001, respectively), with no difference between the two older age classes (t = 1.44, p = 0.324). Females in young populations laid more eggs than females in old populations (figure 2b), despite raw counts of all eggs showing that old populations had on average 38 more visible eggs than did young populations after the three weeks of the experiment (95% confidence interval = 14.5–60.7). Neither activity level nor body size predicted female fitness, but network strength was under positive selection (figure 2d). The interaction between network position and population age structure shows that the individual network strength of females was only under selection in the old populations (figure 2d).

Table 2.

Fixed effects on female reproductive success. Model estimates (slopes) are reported for continuous variables, back-transformed marginal means for categorical variables, and back-transformed marginal trends for the interaction. P-values are calculated as the proportion of 2000 permuted model F-statistics greater than the observed model estimate. P-values significant at the α = 0.05 level are in bold.

	estimate	p‐value
scans active	0.01	0.474
body size (mm)	−0.09	0.242
individual age	one: 0.39	<0.001
	two: 1.94
	three+: 1.56
male social partner age	−0.18	0.085
population age structure	young: 1.24	0.033
	old: 0.91
network strength	0.19	0.014
age structure × strength	young : 0.00	0.006
	old: 0.38

Figure 2.

Selected fixed effects from the model of female reproductive success, measured as the number of oviposition events observed for that individual. (a) Back-transformed marginal means of reproductive success for the three age cohorts. (b) Marginal effects of average male partner age on female fitness, with raw data plotted. (c) Back-transformed marginal means of female reproductive success in the two age structure treatments. Note that this is the main effect involved in an interaction. (d) Marginal effects of individual network strength on female fitness, subset by population age structure, with raw data plotted. Lighter points represent raw data from young populations, and darker points denote old populations. Selection regimes differ between populations with young and old age structures.

4. Discussion

Our experiment demonstrates that age impacts both individual fitness and its covariance with social network phenotypes in B. cornutus. Fitness is associated with variance in age at the following three different levels of organization: the individual, interacting social partners and the population. Older members of both sexes have higher reproductive success than first-years. Males, but not females, pay a fitness cost of interacting with older male social partners. Females, but not males, have lower fitness in populations with older age compositions. Selection on social network position in females depends on the age structure of the population. Our results suggest the age of the social environment, both within and among populations, is a source of variation in fitness, social phenotypes and the covariance between them.

(a). Group age structure shapes sexual selection regimes

Population age composition altered selection gradients on female network sociality in both-sex networks. The network strength of females was not correlated with fitness in young populations but experienced positive directional selection in old populations. We speculate that this is because older populations have sparse networks and the young populations are more densely connected [50]. The benefits of sociality are likely not linear; past a certain point, the costs of additional social interactions outweigh the benefits. Being relatively more connected might therefore only be advantageous in sparse networks and neutral (or even harmful) when the average level of sociality is high. However, the exact mechanisms of selection on social network position in this species are still unknown. Understanding how selection gradients vary in relation to multiple demographic variables will help illuminate possible mechanisms by which social network metrics impact fitness [12,52]. Importantly, we find that although network strength covaries with age [50], selection on strength in both sexes remains even after controlling for age at multiple levels of social organization, suggesting that social network traits do capture fitness-relevant variation rather than simply covarying with a variable we had previously overlooked.

Looking at multiple levels of social organization has revealed that age structure may contribute to variation in selection across a metapopulation. In wild populations, age structure can be difficult to untangle from genetic relatedness, which also depends on the time since population founding [47], and relatedness may impact selection on social behaviour [36]. However, because the populations in this study were randomly assembled from a breeding colony, our experimental design avoids these confounds. Age structure, independent of other demographic factors, contributed to variation in selection gradients among our populations. This is, to our knowledge, a novel addition to the existing body of work on the complex relationship between age demography and sexual selection [72,73].

Our results suggest that age structure may contribute to spatial and temporal variation in the selective landscape. Selection is a powerful evolutionary force, but we understand fairly little about the ecological drivers that generate variation in space and time [1,74]. We suggest that further work considers age structure as a factor shaping selection regimes, especially on social behaviours such as social network traits. As interacting phenotypes, social behaviours may be both more likely than other traits to experience variable selection and more likely to respond rapidly [12,75,76].

(b). Individual, partner and group age impact on fitness

At the individual level, we found that older males and females had significantly higher reproductive success than young individuals. One possibility is that these patterns are caused by increasing reproductive success with age, as opposed to the selective disappearance of less fecund individuals over time. Because beetles are holometabolous, such improvement cannot be due to increased size, but instead might be related to increased energy storage, higher aggression, more experience with competition or courtship behaviours or preferences among members of the other sex. The overall increase in fitness is consistent with an early study in this system that found that males increased their rates of insemination success over the first 60 days since the initial sighting [32]. However, disentangling change over time from selective disappearance will require longitudinal datasets following the same beetles over multiple years. Another possibility for further work to evaluate is whether selection depends on individual age [13,77,78].

Another question for future work is whether the sexes differ in rates of reproductive senescence. The oldest males decline in their number of successful inseminations from their peak at 2 years old, while the oldest females do not decline in number of eggs laid. Other studies of wild invertebrates have found faster senescence of males than females [79,80], possibly because males invest energy early on in their lives to be competitive [72]. Most studies of reproductive senescence in the wild have focused on females, and more studies are needed to understand whether ageing processes are sex-specific [81].

Male fitness was also negatively correlated with the age of social partners. This social selection may be because associating with older males creates a more challenging competitive environment since individual age is positively associated with aggression [48] and males have lower mating success when surrounded by high-quality competitors [9]. Spatial variation in the age of social environments may thereby create variation in the competitive environment within a single population. Because individual and partner age are correlated [50], this effect will counter the fitness advantage of older males. By contrast, we found no cost to females of interacting with older male social partners, despite evidence that female fitness is also negatively impacted by high numbers of interactions and courtship attempts from males [52]. One explanation for this lack of observed effect is that the number of interactions is what matters for female fitness, rather than the phenotypic characteristics of the partner. Another possibility is that older males are more aggressive with other males but not with potential mates; within individuals, aggression in intrasexual competition is not correlated with behaviours in intersexual mating interactions [82].

At the population level, counts of total eggs show that populations of older beetles produced more potential offspring than young populations, consistent with our finding that older individual females lay more eggs than young females. However, after controlling for individual age, our modelling results show that females oviposit less when surrounded by old conspecifics. This apparent contradiction highlights an important reason to consider multiple levels of selection: membership in a group with higher total reproductive output can lower an individual’s relative fitness. The potential benefits of being surrounded by members of the opposite sex heavily invested in reproduction are outweighed by high average fitness in these conditions.

(c). Natural history considerations

The wide range of ages within our experimental populations, from beetles who eclosed only two weeks before the study began to those at least 3 years old, was possible because of the surprisingly long lifespan of forked fungus beetles. Although social insect queens can live and reproduce for many years [83], there are few insect species in which a large fraction or even the majority of residents in the population could be mature adults more than 1 year old. Iteroparity may be more common among Coleoptera than other orders, but only a handful of such cases have been documented (reviewed in Pace [56, pp. 6−7], Danks [84, p. 178] and Promislow et al. [85]; figure 1), and of these B. cornutus appears to be one of the most extreme, with some individuals observed mating and ovipositing across five breeding seasons (V Formica 2022 , unpublished data). However, there are relatively few long-term studies of wild insect populations. There may be more insects with long life cycles than are currently recognized, as measuring invertebrate age and lifespan can be difficult [86]. The conditions favouring such strategies seem to include cool and stable climates; reliable resource availability over the winter to increase the odds of surviving to the next breeding season and therefore the fitness payoff of extended lifespan; and low-quality food that is difficult to digest [56,84,87,88]. Thick cuticles may also be associated with longer lifespans in insects [87]. Long-term studies of insect species meeting these criteria may reveal more examples of extended reproductive windows.

However, we emphasize that no one life history strategy is required to produce the patterns seen in this study. Similar results might be found in any taxon that meets two requirements. First, age structure must vary at the level of populations or subpopulations, for example, because of differences in the time since founding, histories of disturbance or source-sink dynamics. Second, individual age must be associated with a trait likely to affect conspecific fitness, such as aggression. Such an association might be created through senescence, through learning or other forms of improvement over time or by natural selection acting within a generation such that individuals surviving to old age represent a non-random sample of the initial cohort. Whenever these conditions are met, we expect that age structure will create variation in the selection acting on social traits. The ecological and demographic factors that influence age structure will therefore be linked with the evolutionary processes shaping social behaviours.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The full dataset is archived at [63].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

P.A.C.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, validation, visualization, writing—original draft, writing—review and editing; R.A.C.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, validation, visualization, writing—review and editing; E.D.B.: conceptualization, funding acquisition, investigation, methodology, supervision, writing—review and editing; V.F.: conceptualization, formal analysis, funding acquisition, methodology, software, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interests

We declare we have no competing interests.

Funding

Funding for this work was provided by an NSF Graduate Research Fellowship for P.A.C., a University of Virginia Double Hoo research award to P.A.C., NSF IOS-1355029 to V.A.F. and IOS-1355003 and DEB-1911485 to E.D.B.III and NSF REU grant 1461169 to Mountain Lake Biological Station.