Age-dependent genetic variation in aggression

Abstract

Understanding the extent to which behavioural variance is underlain by genotypic, environmental and genotype-by-environment effects is important for predicting how behavioural traits might respond to selection and evolve. How behaviour varies both within and among individuals can change across ontogeny, leading to differences in the relative contribution of genetic and environmental effects to phenotypic variation across ages. We investigated among-individual and among-genotype variation in aggression across ontogeny by measuring, twice as juveniles and twice as adults, both approaches and attacks against a three-dimensional-printed model opponent in eight individuals from each of eight genotypes (N = 64). Aggression was only significantly repeatable and heritabile in juveniles. Additionally, how aggression changed between juvenile and adult life-history stages varied significantly among individuals and genotypes. These results suggest that juvenile aggression is likely to evolve more rapidly via natural selection than adult aggression and that the trajectory of behavioural change across the lifespan has the potential to evolve. Determining when genetic variation explains (or does not explain) behavioural variation can further our understanding of key life-history stages during which selection might drive the strongest or swiftest evolutionary response.

1. Introduction

The study of consistent among-individual differences in behaviour, animal personality, has revealed how traits such as aggression, boldness and exploratory behaviour vary (and covary) temporally, both within and across contexts [1–4]. Understanding the relative contributions of genotype, environment and genotype-by-environment interactions to animal personality can illuminate the extent to which behavioural traits might evolve in response to selection [5]. Such evolutionary inferences, however, might depend on the age at which behaviour is assessed. As male crickets age, they become less bold [6], an ontogenetic shift in behaviour that could emerge from changes in both internal and external environments. During reproductive maturation in vertebrates, steroid hormone levels change, resulting in broad changes to the phenotype, including behaviour [7,8]. In many species, juveniles and adults also occupy different habitats [9,10] which can promote age-dependent phenotypic variation, e.g. salinity differences across ontogeny can drive variation in aggression [11].

Given the sensitivity of behaviour to environmental change [12], how much behavioural variation can be attributed to genetic factors (i.e. heritability) might also shift across ontogeny. Heritability for juvenile hormone esterase enzyme activity is lower in adult than juvenile crickets under constant environmental conditions [6]. Also, bandwidths and peak frequencies of marmot alarm calls are not heritable in juveniles but are strongly heritable in adults [13]. Such age-related changes in heritability can help pinpoint life-history stages at which selection might drive pronounced evolutionary responses [14–17]. We investigated how genetic variance (broad sense) contributes to among-individual variance in aggression over a life-history transition from juvenile to adult in self-fertilizing hermaphroditic fish, Kryptolebias marmoratus (rivulus). Rivulus can produce isogenic lineages by selfing exclusively for many generations. However, males exist at low frequencies, and outcrossing between hermaphrodites and males generates new genotypes. The capacity to produce isogenic lineages permits isolation of environmental effects on phenotypic traits, such as aggression, independent of genetic effects and without complicated breeding designs. Aggression can have important fitness benefits, including increased mating opportunities or resource acquisition [18–22], but can also be costly, e.g. hyperaggressive individuals have lower mating success [23].

We tested the following hypotheses: (i) individuals and genotypes will vary in aggression; (ii) among-individual and among-genotype variation in aggression will differ between juveniles and adults; (iii) average aggression will differ between juveniles (80–87 days post hatch (dph)) and young adults (173–180 dph hereafter ‘adults’) and (iv) there will be among-genotype and among-individual variation in how aggression changes between juvenile and adult life stages (i.e. developmental reaction norm slope). We predicted that (i) aggression will vary among individuals (repeatability) and genotypes (heritability) because previous research has uncovered genotypic differences in aggression but did not explicitly assess heritability [24,25]; (ii) adults will exhibit greater among-individual and among-genotype variation in aggression than juveniles (i.e. repeatability/ heritability higher in adults [26], see [27]); (iii) adults will be more aggressive on average than juveniles because, in many organisms, young adults have higher levels of testosterone, which can potentiate aggression [28–30] and (iv) developmental reaction norm slopes for aggression will be repeatable and heritable.

2. Methods

(a) . Study species

Rivulus were raised in the laboratory from eight field-derived lineages (F3 and F4 generation, electronic supplementary material, table S1), selected based on their low propensity to change sex from hermaphrodite to male; we focus only on hermaphrodite behaviour. Eight individuals from each genotype were tested (N = 64). Fish were housed, from hatching, individually in ventilated, 1.2 l Rubbermaid® TakeAlongs® Deep Square containers filled with 700 ml, 25 ppt saltwater (Instant Ocean® salt and aged tap water) and kept in a temperature-controlled room (26.13 ± 0.02°C) with a 12 h light : 12 h dark photoperiod. Individuals were fed 2 ml brine shrimp (Artemia) nauplii suspended in 25 ppt water (approx. 1000 nauplii) daily between 09.00 and 17.00 h.

(b) . Data collection

We quantified individual aggression using a model opponent, ensuring measurement of intrinsic aggression levels rather than responses to feedback by live or mirror-image opponents; motivation to fight (e.g. attack latencies) in model tests also reflects behaviour in real contests [31]. Individuals were placed in identical square tanks (10.16 × 10.16 × 10.16 cm) lined with black corrugated plastic on three sides and bottom. Tanks were filled with 25 ppt water, and an opaque black plastic barrier divided the tank in half diagonally to prevent focal fish from seeing the three-dimensional-printed model during 1 h acclimation (see [32] for three-dimensional-printing details). Models and focal fish were visually matched for size and models were attached to the tank via fishing line. To begin the trial, the opaque barrier was removed, and focal animals could interact with the model for 30 min. Animals were tested four times, twice as juveniles (80, 87 dph) and twice as adults (173, 180 dph); hermaphrodites reach reproductive maturity at approximately 100 dph [33] and no eggs were laid in the animals' home tubs during the juvenile phase. Each 30 min trial was recorded using a video camera facing the tank. In each trial, the number of approaches (focal animal moved, in directed fashion, to within one body length of model) and attacks (focal animal physically hit the model with any part of its body). Aggression scores were the sum of approaches and attacks.

(c) . Statistical analysis

Statistical analyses were conducted with RStudio (R v. 3.4). Hypotheses were tested using general linear mixed models (GLMM) with aggression as the response variable (all models listed in table 1). Aggression was cube-root-transformed to achieve normality. Fixed effects in all models were age (juvenile, adult) and time (80, 87, 173 and 180 dph) nested within age. To determine repeatability, the random effect of individual was included to partition among- and within-individual (i.e. residual) variances; in some models, individual × age was included as a random effect to explore among-individual variation in how aggression changed with age (i.e. developmental reaction norm slopes). To determine heritability, the random effect of genotype was included to partition variance into among- and within-genotype (i.e. residual) variance; some models included genotype × age as a random effect to explore genotypic variation in how aggression changed with age.

Table 1.

Statistical models used for data analysis.

model 1	aggression = age + age/time
model 2	aggression = age + age/time + (1|individual)
model 3	aggression = age + age/time + (1|individual) + (1|age:individual)
model 4	aggression = age + age/time + (1|genotype)
model 5	aggression = age + age/time + (1|genotype) + (1|age:genotype)

We analysed whether aggression varied among individuals irrespective of age by comparing model 1 (no random effects) with model 2 (individual as a random effect) (table 1). To compare these (and subsequent) models, we conducted a likelihood ratio test (−2*[log-likelihood_Model2 – log-likelihood_Model1]) to derive χ² and p-values. We then subset the data by age and re-ran comparisons of model 1 and model 2 to determine whether aggression varied among individuals in each age class. Comparing model 3 (now including individual × age random effect) with model 2 assessed among-individual variation in developmental reaction norms. Similar analyses ascertained variation among genotypes in aggression, except with genotype and genotype × age as random effects.

We used rptGaussian from the R package rptR to calculate repeatability and heritability (± s.e.). We estimated 95% confidence intervals using 1000 parametric bootstrapping iterations. Statistical significance threshold was set at 0.05. In this study, we can only calculate broad-sense heritability, i.e. proportion of phenotypic variation explained by total genetic variation. Calculating narrow-sense heritability would require that different isogenic lineages mate, which rarely occurs (approx. 6% laboratory-outcrossing success; [34–37]).

3. Results

Juveniles were, on average, more aggressive towards the models than adults (figure 1; electronic supplementary material, tables S2–S5) and only juveniles showed consistent among-individual differences (i.e. repeatability) in aggression (figure 2c,d, table 2; electronic supplementary material, tables S2, S4 and S6). Significant variation in aggression among genotypes (i.e. heritability) also was evident only in juveniles (figure 2a, table 2; electronic supplementary material, tables S3, S5 and S6). The extent to which aggression changed between juvenile and adult life-history stages (i.e. developmental reaction norm slopes), also varied significantly among individuals and genotypes (figure 2, table 2; electronic supplementary material, tables S4–S6)

Figure 1.

Differences in aggression (cube-root) between juveniles and adults: (a) pooled time points and (b) across all time points measured.

Figure 2.

Slopes of (a) genotypic reaction norms, (b) effect of juvenile aggression on developmental reaction norm slopes; (c) individual reaction norms and (d) subsection of individual reaction norms highlighting aggression scores from 0 to 20 to resolve individual variation.

Table 2.

Repeatability and heritability estimates.

	repeatability	heritability
all age groups	0.133 (0.062)*	0.060 (0.044)*
juveniles	0.432 (0.102)**	0.136 (0.088)*
adults	0.106 (0.107)	0.042 (0.048)
individual/genotype×age	0.437 (0.070)**	0.128 (0.068)**

* = p-value < 0.01 ** = p-value < 0.001, s.e. shown in parentheses.

4. Discussion

We found support for our hypothesis of age-related differences in aggression but not in the predicted direction; juveniles were more aggressive than adults. From an ultimate perspective, such age-related behavioural variation could indicate that the strength and/or type of selection on aggression varies across ontogeny [38]. One such selection pressure could be predators that differentially attack certain life stages [39]. Individual variation in aggression, a highly conspicuous behaviour, can lead to differential predation [40,41]. Our field observations indicate that rivulus juveniles are eaten by other fish and dragonfly larvae, ambush predators that elicit a very strong chemically mediated avoidance response in juvenile rivulus (but not adults) (C-Y Li & R Earley 2022, personal communication). Adults, however, are more susceptible to predation by water snakes and birds, both of which are active hunters. Vision is probably the only reliable cue for predator detection in adults (e.g. chemical cues might misrepresent the location of active hunters), and vision is constrained during intense combat [40,41], which could exert strong selection on lower aggression in adults. Chemically mediated predator avoidance in juveniles, on the other hand, could relax selection on aggression at this life stage, resulting in the maintenance of behavioural variation, even highly aggressive phenotypes. The best evidence for selection operating differently across ontogeny in wild populations (from which the study animals were derived) is that the vast majority of juveniles converged upon a single, low-aggression phenotype as young adults regardless of their early-life aggression levels (figure 2).

We found consistent among-individual differences in aggression in juveniles but not for adults, which did not support our hypothesis that aggression would vary consistently among individuals at both life stages; aggression thus constitutes at least one aspect of personality in juvenile rivulus [25,43]. Species that inhabit mangroves, like rivulus, experience extraordinary spatial variation in virtually every respect [44]. Such spatial heterogeneity enables diverse selection pressures (e.g. specific microhabitats) to maintain considerable behavioural variation among individuals within a population [45,46], resulting in significant repeatability of aggression for juveniles.

Repeatability was considerably higher in juveniles than in adults. One explanation could be that all juveniles habituate [47] to the model similarly across time points thus, maintaining behavioural differences among individuals (electronic supplementary material, figure S1). Non-significant repeatability in adults could reflect a combination of lower behavioural variation (e.g. floor effect) and no habituation at this life stage or that individuals habituate at different rates. Visual inspection of the data supports the latter explanation (electronic supplementary material, figure S1). Variation in the capacity for and rates of habituation across ontogeny has been documented in several species [48–51]. In a meta-analysis of repeatability, Bell et al. [52] found no differences between juveniles and adults in repeatability but, only three datasets within that analysis directly compared juvenile and adult behaviours. Thus, additional data are needed before firm conclusions or generalizations (if any) about ontogenetic shifts in repeatability can be drawn.

We hypothesized that aggression would be heritable to different extents in juveniles and adults. Supporting our hypothesis, we found consistent differences among genotypes (i.e. heritability) in aggression. However, the data did not support our prediction that adult aggression would be heritable; only juvenile aggression was heritable. Being responsive to maturation-related internal cues is important for a major transition from juvenile to adulthood [53] involving changes in sex steroid hormones, neurobiological function and reproductive status, which alters how individuals react to stimuli [54,55]. A likely proximate explanation for age-related differences in heritability is that behavioural variance in adults is driven to a greater extent by internal or external environmental sources and genotype-by-environment interactions than juveniles [56].

Considerable variation in juvenile aggression collapsed to little variation in adults, leading to striking variation among individuals and genotypes in reaction norm slopes (figure 2). Heritability of developmental reaction norms ultimately provides insight into whether and how ontogenetic changes in animal personality might evolve [57]. We devised two hypotheses to explain such changes: age-dependent selection hypothesis (ADSH) and flexibility–aggression coevolution hypothesis (FACH). In the former, ADSH, we propose that selection culls adults exhibiting high aggression constitutively or flexibly, resulting in canalization of adult aggression to a low-aggression phenotype. Selection against high aggression in adults could occur due to elevated predation risk (see above) or if increased energy allocation to reproduction in adults reduces energy available to engage in (and recover from) combat [58]. Another explanation for significant heritability of developmental reaction norms is that considerable genetic variation in aggression exists for juveniles due to relaxed selection on aggression at this life stage (see above; [59]). This, coupled with lower variance for aggression in adults, would cause considerable among-genotype variation in the developmental reaction norm (i.e. strong heritability).

In the FACH, we propose that selection culls high-aggression juveniles lacking the ability to respond appropriately to environmental change. Such flexibility would be advantageous in fluctuating environments [57,60], like the focal species’ mangrove habitats, that show extreme spatio-temporal variation in intra- and interspecific competition. Thus, individuals may experience very different selection pressures across ontogeny. Highly aggressive juveniles lacking flexibility in aggression may experience reduced fitness, e.g. constantly investing in aggression could detract from other fitness-related activities [61,62]. Thus, selection should favour behavioural flexibility across the life-history transition from juvenile to adulthood, and that change should be more pronounced in juveniles that are highly aggressive, i.e. coevolution of juvenile aggression and development reaction norm slopes.

We conducted a post hoc analysis investigating whether juvenile aggression predicted developmental reaction norm slopes, which would provide support for the FACH. We constructed a GLMM with developmental reaction norm slopes as the response variable, juvenile aggression as a fixed effect and genotype as a random effect (electronic supplementary material, table S7). Juvenile aggression predicted developmental reaction norm slopes, which supports the FACH. Aggressive juveniles had steep negative reaction norm slopes while low-aggression juveniles generally had shallow reaction norm slopes (figure 2b). This indicates that selection might have favoured heritable covariance between behavioural flexibility and aggression with highly aggressive juveniles showing more flexibility across the life-history transition. However, support of one hypothesis does not eliminate the alternative, the FACH and ADSH may not be mutually exclusive; selection might act concurrently to cull high-aggression adults and juveniles that lack flexibility, as all individuals are low-aggression adults.

Overall, we found evidence for consistent among-individual differences in aggression and significant heritability only in juveniles. These results suggest that only juvenile aggression has the potential to evolve (at any considerable rate) via natural selection. Environmental variance and genotype-by-environment interactions might thus play a larger role in mediating adult behaviour. Repeatability and heritability of developmental reaction norm slopes suggests that the trajectory of behavioural change across the lifespan also has the potential to evolve. Determining the age(s) at which genetic variation explains (or does not explain) behavioural variation can further our understanding of key life-history stages during which selection can cause the most significant behavioural evolution.

Ethics

This research was approved by The University of Alabama IACUC. Protocol Number: 18-0100898.

Data accessibility

Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.k0p2ngfcd [63]. https://datadryad.org/stash/share/iKj9uWKG7CPTcKscIgjkTU2GjkpTbqce7giYWXp_XK4

Supplementary material is available online [64].

Authors' contributions

J.A.F.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, visualization, writing—original draft and writing—review and editing; R.L.E.: conceptualization, formal analysis, methodology, supervision and writing—review and editing.

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

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by Sigma Xi (grant no. G2018031596030532).