How selection acts to drive trait evolution at different stages of divergence is of fundamental importance in our understanding of the origins of biodiversity. Yet, most studies have focused on a single point along an evolutionary trajectory. Here, we provide a case study evaluating the strength of divergent selection acting on life-history traits at early-versus-late stages of divergence in Brachyrhaphis fishes. We find that the difference in selection is stronger in the early-diverged population than the late-diverged population, and that trait differences acquired early are maintained over time.
Comparative studies of recently diverged populations (i.e. nascent species) provide valuable insight into the forces that drive trait evolution and speciation . However, the relationship between trait evolution at early-versus-late stages of divergence is not well understood. Specifically, little is known about how the strength of divergent selection changes at different stages of speciation, despite a long-standing theoretical framework suggesting that the strength of selection should change as diversification occurs . For example, Fisher's fundamental theorem , for which evidence from the wild is uncommon (but see [3–6]), posits that selection strength should increase as variance in fitness increases. Thus, if recently diverged populations, which have yet to reach adaptive optima, have experienced more recent gene flow relative to more established species pairs, the strength of divergent selection acting on those populations should be higher than that experienced by more divergent and presumably better-adapted species pairs. Increased variance in fitness early in divergence is expected in part owing to increased genetic variation upon which selection can act.
Life-history traits (e.g. age/size at maturity) are of particular importance when considering how traits change in response to selection, because they translate directly into population-level demographic phenomena . In addition, they are often subjected to strong selection that can initially result in rapid evolutionary change [8–10]. Unfortunately, although variable life histories among populations from different environments are often described [11,12], the effects of this variation on population-level metrics, such as population growth rate (λ) and the strength of selection acting on these traits, are seldom addressed . Even rarer are studies evaluating these processes at early-versus-late stages of divergence.
The Central American Brachyrhaphis fishes (Poeciliidae) are a useful system for studying selection over time because this genus contains several within species population pairs and between sister species pairs occurring in similarly divergent selective regimes (figure 1; ). Populations of B. rhabdophora (BRh) occur in divergent predation environments throughout their range (e.g. presence/absence of predators and a suite of correlated factors that could influence life-history evolution), resulting in the evolution of divergent life histories  and morphologies . A strikingly similar pattern is observed between sister species B. roseni (BR) and B. terrabensis (BT) , which primarily occur in high- and low-predation streams, respectively . This species pair has evolved similar patterns of morphological  and life-history  divergence to those seen among populations of BRh from different predation environments, suggesting that each pair is found at different points along the same evolutionary trajectory . Here, we use serial mark–recapture (SMR) experiments and population matrix models (PMMs) to test whether the strength of divergent selection on life-history traits is greater at early- (e.g. within BRh) versus-late (e.g. between BR and BT) stages of life-history divergence. We predict that, in accordance to Fisher's fundamental theorem , the strength of divergent selection will be greater between recently diverged populations of BRh than between more established sister species BR and BT.
2. Material and methods
(a) Mark–recapture experiment
We conducted an SMR with BR (high-predation) and BT (low-predation). To facilitate comparisons with previously published work on BRh (in which populations occur both in high-predation ‘Javilla’ and low-predation ‘Grande’ environments; ), and to allow us to compare patterns of selection at early-versus-late stages of divergence, we followed the methods of Johnson & Zúñiga-Vega . In short, we selected two sites, one with BR and one with BT, which consisted of relatively isolated pools within streams characterized by a pool–riffle–pool structure (see the electronic supplementary material). For each location, we sampled the pool for 1–2 h until confident that we had captured most fish in the pool (at least 10 subsequent seine hauls with no captures). We anaesthetized (using MS-222), measured, sexed and marked each fish in the caudal region with a unique sub-cutaneous injection of latex paint (suspended in Ringer's solution), allowing us to recognize individual fish upon recapture. After marking, fish recovered in a poolside tank until day's end (4–6 h), at which time they were released into their pool of origin. We returned four times to each site at one-week intervals and repeated the protocol, recording and measuring recaptures and marking newly captured fish, resulting in individualized mark–recapture and growth histories for each fish over a five-week period. This design allowed us to account for the impact of migration (in/out of the pool) and incomplete sampling on recapture rates, and track growth over the five-week period (electronic supplementary material). In total, we marked 223 BR and 266 BT. Marking mortalities were extremely rare (less than 1%) and marked fish held under controlled conditions through the duration of the experiment kept their marks.
(b) Mortality estimates
We analysed recapture histories for females to estimate mortality rates using Program MARK . Mortality rates are a critical input for PMMs, which we use to create elasticity estimates (i.e. an indicator of strength of selection). We assigned each fish to one of five ontogenetic stages (three non-reproductive and two reproductive stages; electronic supplementary material). We tested 12 competing models that varied parameter constraints for mortality and recapture rates among the five stages, using Akaike Information Criterion to select the best-fit model . We used model averaging to generate our final mortality estimates for each stage, thus taking into account the relative weight of each model and providing more robust estimates than if we had only considered the top model .
(c) Demographic analyses
Using mortality estimates from MARK, we created PMMs  to estimate several demographic variables, including population growth rate (λ; electronic supplementary material), sensitivities and elasticities. These models use the following as inputs: stage-specific mortality (from MARK), growth rates (i.e. transition rates among classes) and fecundity (estimated for sampled populations using life-history dissections; electronic supplementary material). PMMs facilitate comparative evaluations of population dynamics using estimates of sensitivities (the effect on λ of changing each vital rate by the same magnitude [7,20]) and elasticities of λ (the effect on λ of changing each vital rate by the same proportion [7,20]). Elasticities allow for comparison among matrices derived from populations/species with divergent life histories because they are standardized [7,21], and represent a standardized estimate of the strength of selection acting on each component of the life-history matrix (e.g. stage-specific survival, growth and fecundity [20,22]). Thus, comparing elasticities among stages and populations allows us to identify the vital rates that are under the strongest selection at early- (among BRh populations) versus-late stages of divergence (between BR and BT). Adjustments to standard p-values were not warranted given the modest number of comparisons; nevertheless, employing multiple tests increases the probability that differences emerge by chance. Finally, we conducted a permutation analysis following the methods of Johnson & Zúñiga-Vega  to generate 95% confidence intervals (CI) for elasticities and population growth parameters (electronic supplementary material), providing an estimate of significance when comparing ranges of elasticities for each vital rate and summed elasticities for each stage.
Our SMR revealed that mortality rates were higher in BR (predators present) than in BT (predators absent). Furthermore, we found that large adult BR suffered the highest mortality (electronic supplementary material). Despite differences in mortality rates, the 95% CIs for λ for each species overlapped and spanned 1 (i.e. stable population size; table 1), indicating that population growth rates did not differ between species and were stable.
Matrix elasticities revealed that selection acted similarly on both BR and BT (CIs overlapped for all summed stages; table 1 and figure 2), with strong selection on surviving and remaining in the first four stages. However, populations of BRh experienced divergent selection on both small juveniles and large adults (non-overlapping CIs for juvenile 1 (J1), minimal overlap for adult 2 (A2); table 1 and figure 2; electronic supplementary material, figure S4). Overall selection on growth, fecundity and stasis (i.e. surviving and remaining in a life-history stage) was similar in all populations (overlapping CIs; electronic supplementary material), although stasis was under slightly stronger selection in BR and BT than in BRh (electronic supplementary material).
Our work provides additional evidence that predation environment is a common driver of life-history divergence, a pattern found both within Brachyrhaphis and across poeciliids in general (e.g. [8,11]). Indeed, our work and others' suggest that increased mortality rates, whether owing to predation or abiotic stressors , predictably drive the evolution of life histories. Furthermore, Brachyrhaphis, both at early and late stages of divergence, maintain similar population growth rates despite different mortality rates. When taken in context, these results highlight the conserved nature of not only life-history traits, but also growth rates of populations that occur in divergent predation environments (, electronic supplementary material). In short, different populations often solve the same demographic challenges using alternative life-history strategies. It should be noted that by only comparing one late-diverged with one early-diverged pair of populations, we cannot rule out factors other than divergence stage driving the differences in the patterns of selection we observe. Future work following our approach—but using multiple populations differing in divergence stage—would help confirm that divergence stage is driving the patterns we observe here.
Despite a long-standing theoretical framework, the way in which selection acts on recently diverged populations relative to established sister species remains poorly studied in the wild. Our findings suggest that divergent selection on life-history traits can be stronger during early stages of divergence (i.e. between Javilla and Grande) relative to late stages of divergence (i.e. between BR and BT). Furthermore, our results suggest that trait differences can be maintained as divergent selection weakens. Thus, although strong selection might be required to drive divergence initially, more modest selective pressure could be sufficient to maintain differences over time.
Assuming that our results are owing to differences in the stage of divergence, the observed differences in the strength of divergent selection at early-versus-late stages of divergence could be attributed to several processes. According to Fisher's fundamental theorem, divergent selection should strengthen in proportion to variance in fitness . Several lines of evidence suggest that variance in fitness could be higher in BRh relative to BR and BT. First, genetic divergence between Javilla and Grande is nearly an order of magnitude lower than between BR and BT . Previous work suggests that low-predation populations of BRh recently diverged from high-predation populations, likely as they moved among drainages along the coast and subsequently expanded their ranges upstream to reaches without predators [14,15,24]. The timing of this divergence, measured by the cessation of gene flow and subsequent genetic differentiation, appears to have occurred more recently between BRh populations than between BR and BT [14,15,24]. This pattern suggests that Grande would have only recently become subjected to a selective regime divergent to that of Javilla, with little time to move towards adaptive optima. Furthermore, morphological  and life-history traits , which are likely tightly linked to fitness, show greater within-population variance in BRh relative to BR and BT, suggesting that a recent origin and/or more recent gene flow could contribute to increased variance in fitness, providing more material upon which selection can act. By contrast, BR and BT are more likely to have neared their phenotypic optima some time ago. Given the depth of divergence between BR and BT [14,15], selection could have eroded additive genetic variance in these species relative to BRh. Thus, differential patterns of selection could possibly have been dampened owing to a lack of variance upon which selection can act . More detailed studies evaluating the amount of additive genetic variance found in these taxon pairs would help clarify the relationship between strength of selection and level of divergence. That we note differences in strength of selection in early-versus-late stages of evolutionary divergence in Brachyrhaphis fishes suggests that a single snapshot in evolutionary time may often fail to capture the process by which evolutionary diversification occurs.
This research adheres to the Association for the Study of Animal Behaviour (ASAB) guidelines and was approved by the Smithsonian Tropical Research Institute Institutional Animal Care and Use Committee (IACUC), protocol no. 2011-0616-2014-08.
Raw data have been deposited in Dryad: http://dx.doi.org/10.5061/dryad.17sq0.
Conceived and designed experiments: S.J.I. and J.B.J. Data collection: S.J.I. Data analysis: S.J.I. and J.B.J. Wrote/revised article: S.J.I. and J.B.J. All authors agree to be held accountable for the content and approve the final version of the manuscript.
We have no competing interests.
S.J.I., National Geographic Young Explorers Grant and US-National Science Foundation Graduate Research Fellowship. J.B.J., Brigham Young University Internship Grant and Mentoring Environment Grant. S.J.I. and J.B.J., Brigham Young University Department of Biology. S.J.I. and J.B.J., Brigham Young University Kennedy Center for International Studies.
We sincerely thank D. Money, J. Rehm and I. Ingley for help in the field. We thank the Smithsonian Tropical Research Institute for help with permitting and IACUC oversight. We thank J.J. Zúñiga-Vega for input on our permutation analysis, and P. Nosil, B. Adams and four anonymous reviewers for input on the manuscript.
- Received December 4, 2015.
- Accepted February 25, 2016.
- © 2016 The Author(s)
Published by the Royal Society. All rights reserved.