Ecology Letters, (2020) 23: 831–840 LETTER Andres Hagmayer,1 Andrew I. Furness,2,3 David N. Reznick,4 Myrthe L. Dekker1 and Bart J. A. Pollux1* The peer review history for this article is available at https://publons.c om/publon/10.1111/ele.13487 doi: 10.1111/ele.13487 Predation risk shapes the degree of placentation in natural populations of live-bearing fish Abstract The placenta is a complex life-history trait that is ubiquitous across the tree of life. Theory proposes that the placenta evolves in response to high performance-demanding conditions by shifting maternal investment from pre- to post-fertilisation, thereby reducing a female’s reproductive burden during pregnancy. We test this hypothesis by studying populations of the fish species Poeciliopsis retropinna in Costa Rica. We found substantial variation in the degree of placentation among natural populations associated with predation risk: females from high predation populations had significantly higher degrees of placentation compared to low predation females, while number, size and quality of offspring at birth remained unaffected. Moreover, a higher degree of placentation correlated with a lower reproductive burden and hence likely an improved swimming performance during pregnancy. Our study advances an adaptive explanation for why the placenta evolves by arguing that an increased degree of placentation offers a selective advantage in high predation environments. Keywords Life-history, live-bearing, matrotrophy, placenta, placentotrophy, Poeciliidae, predation, superfetation, Trexler–DeAngelis, viviparity. Ecology Letters (2020) 23: 831–840 Understanding the origin and elaboration of complex traits is of fundamental interest to evolutionary biologists. Placentas are complex organs that have independently evolved many times throughout the animal kingdom in widely divergent lineages (Wourms 1981; Wooding & Burton 2008; Blackburn 2015; Wake 2015; Ostrovsky et al. 2016). This repeated evolution points to a possible adaptive advantage (Losos et al. 1998), however, the nature of this advantage remains elusive. To date, two adaptive hypotheses (termed the resource allocation and locomotor cost hypotheses) provide potential explanations for why the placenta evolves (i.e. its selective advantage). Both hypotheses treat the placenta as a life-history adaptation that evolves in response to ecological selection pressures (Thibault & Schultz 1978; Trexler & DeAngelis 2003). Furthermore, both hypotheses posit that the ancestral state is for all nutrients to be prepackaged in the form of egg yolk (lecithotrophy) and that the derived state is for nutrients to be supplied to embryos throughout pregnancy via a placenta (placentotrophy). This is presumably achieved via a gradual shift in the timing of nutrient provisioning from preto post-fertilisation. If true, then the evolution of the placenta should coincide with a reduction in egg mass at fertilisation without affecting the total investment per neonate at birth (Reznick et al. 2002; Trexler & DeAngelis 2003; Pollux et al. 2009). The live-bearing fish family Poeciliidae is a useful system in which to test these ideas, because it contains species that span a continuum of pre- versus post-fertilisation nutrient provisioning, ranging from strictly lecithotrophic to highly placentotrophic. In addition, placentotrophy has evolved independently numerous times in this family (Reznick et al. 2002; Pollux et al. 2014; Furness et al. 2019). The resource allocation hypothesis (Trexler & DeAngelis 2003) posits that the evolution of the placenta permits females to produce greater brood sizes and hence attain higher fecundity than lecithotrophic females when sufficient resources are available to carry the developing embryos to term. Moreover, placental females rely on a steady nutritional supply to provision embryos throughout pregnancy. Trexler & DeAngelis (2003) therefore argued that the ability to selectively abort embryos and recycle this investment, should resource conditions suddenly deteriorate, is a crucial preadaptation for the evolution of placentation. Studies in the Poeciliidae have not shown that they are able to do this, suggesting that the conditions under which the placenta might be favoured by natural selection are restricted to environments characterised by high and stable resource conditions (Reznick et al. 1996; Trexler 1997; Banet & Reznick 2008; Banet et al. 2010; Pollux & Reznick 2011). The locomotor cost hypothesis (Thibault & Schultz 1978; Pollux et al. 2009; Pires et al. 2011) postulates that the placenta evolves to offset some of the locomotor costs associated with a live-bearing mode of reproduction. The physical and physiological burden of pregnancy negatively affects a female’s locomotor performance in a broad range of live- 1 3 INTRODUCTION Department of Animal Sciences, Wageningen University, 6708 WD, Wagenin- gen, Netherlands 2 Department of Ecology and Evolutionary Biology, University of California, Irvine, CA 92697, USA Department of Biological and Marine Sciences, University of Hull, HU6 7RX, Hull, UK 4 Department of Biology, University of California, Riverside, CA 92521, USA *Correspondence: E-mail: bart.pollux@wur.nl © 2020 The Authors. Ecology Letters published by CNRS and John Wiley & Sons Ltd This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 832 A. Hagmayer et al. bearing animals (Seigel et al. 1987; Plaut 2002; Noren et al. 2011; Fleuren et al. 2019). The evolution of the placenta should allow females to attain higher fitness, because the production of smaller eggs at fertilisation reduces a female’s reproductive burden during pregnancy. This improves body streamlining and locomotor performance, notably without sacrificing reproductive output. The presumed benefit of a higher degree of placentation is that the improved swimming performance offers a selective advantage to females in performance-demanding (e.g. high predation) environments because it enhances survival probability (Pollux et al. 2009; Pires et al. 2011). Whereas the resource allocation hypothesis has already been the focus of several empirical studies (Pires et al. 2007; Banet & Reznick 2008; Banet et al. 2010; Pollux & Reznick 2011; Bassar et al. 2014), the locomotor cost hypothesis has not yet been subject to similar systematic investigation. Here, we quantify the degree of placentation in high and low predation populations of the placental live-bearing fish species Poeciliopsis retropinna (family Poeciliidae, Regan 1908) in Costa Rica. We test a key prediction of the locomotor cost hypothesis, which is that placentation evolves in performancedemanding (high predation) environments. If the placenta is favored under high predation conditions, we should find a higher degree of placentation in high predation populations. Theory further predicts that this higher degree of placentation should be due to a smaller egg size at fertilisation without affecting the number, size, or quality of offspring at birth, translating to a lower reproductive burden for females during pregnancy (Thibault & Schultz 1978; Pollux et al. 2009; Pires et al. 2011). Finally, we discuss the potential fitness advantage of evolving a higher degree of placentation in high predation populations. Letter location/snorkeler) at different positions along each river. In one location (Rio Conte), water visibility was low and predator community was assessed using seine and cast nets. We found 17 ‘high’ predation localities where piscivorous predator species were present, and 12 ‘low’ predation localities where predators were absent (Table S1). At each location, 5– 37 adult females were collected, euthanised with an overdose of MS-222, and preserved in 5% formaldehyde. Laboratory measurements Maternal standard length and the proportion of body fat were measured using established protocols (Supporting Information Methods 1.1). 28 of the 29 sampled populations contained pregnant females (Table S1), therefore all subsequent anatomical and statistical analyses were carried out only with females from 28 populations (npreg = 463). The ovaries were dissected to count the total number of embryos (fecundity), regressors (aborted embryos), broods at different developmental stages (superfetation), embryos in a brood (brood size), and to determine the developmental stage and average dry mass of embryos in a brood (Table 1). The developmental stages are based on morphological criteria described in Haynes (1995) and range from 0 (eggs at fertilisation, no development) to 45 (fully developed embryos). Fecundity was calculated by excluding stage 0 embryos, because it was difficult to assess if 0-staged eggs were fertilised or not. Instead, embryos at developmental stage 2, rather than 0, were defined as ‘eggs at fertilisation’. Quantification of the degree of placentation The Matrotrophy Index (MI), calculated as the ratio of offspring mass at birth to egg mass at fertilisation, was used as MATERIAL AND METHODS Study species and collection sites Poeciliopsis retropinna is found in freshwater streams of varying water velocity and predation pressure in Costa Rica and Panama (Bussing 2002). They are observed in habitats with low predation risk (no strongly piscivorous species present), or co-occurring with one or more piscivorous predator species: Parachromis dovii, Eleotris picta, and Gobiomorus maculatus. During gestation, P. retropinna females transfer nutrients to developing embryos via a placenta. The degree of post-fertilisation maternal provisioning in this species is extensive, with offspring increasing in dry mass more than 100-fold during gestation (MI = 117) (Reznick et al. 2002). Moreover, P. retropinna has superfetation, the ability to carry several broods at different developmental stages. During February and March 2013, 2017, and 2018, P. retropinna were collected at 27 different locations in the Rio Terraba and Rio Coto drainages in the province of Puntarenas, Costa Rica (Table S1). Two of the locations were repeatedly sampled resulting in 29 study populations. When water visibility was high (i.e. in all but one location), the occurrence of piscivorous predator species was most effectively detected using underwater visual census by three independent snorkelers. Censuses took place during daytime (c. 3–5 h per Table 1 Summary of maternal life-history traits Maternal life-history traits Egg mass at fertilisation Offspring mass at birth Proportion egg fat Proportion offspring fat Reproductive allotment Absolute reproductive allotment Brood size Fecundity Superfetation Abortion incidence © 2020 The Authors. Ecology Letters published by CNRS and John Wiley & Sons Ltd Dry mass of eggs at fertilisation (i.e. developmental stage 2) Dry mass of fully developed embryos (i.e. developmental stage 45) Egg fat at fertilisation divided by dry mass of eggs at fertilisation Offspring fat at birth divided by offspring dry mass at birth Proportion of the mother’s dry mass allocated to reproduction (i.e. embryo dry mass, regressor dry mass, and placental dry mass) Total dry mass allocated to reproduction (i.e. embryo dry mass, regressor dry mass, and placental dry mass) Number of embryos in a given brood Number of embryos carried by a female counted across all broods excluding stage 0 embryos Number of broods at different developmental stages Number of regressors (i.e. aborted embryos) divided by the sum of the number of regressors and embryos Letter Predation drives placental evolution 833 an unbiased measure of the degree of placentation (Reznick et al. 2002). Some live-bearing species allocate all resources to eggs prior to fertilisation in the form of large fully-yolked eggs (termed lecithotrophy). These embryos lose dry mass over the course of gestation due to metabolic processes, leading to an MI < 1. Other species allocate nutrients to the developing offspring post-fertilisation throughout pregnancy (termed matrotrophy). Such species have an MI > 1, indicating embryos gain dry mass during pregnancy. Placentotrophy represents one specific type of matrotrophy that is achieved through a follicular placenta, roughly an analog to the mammalian placenta (Pollux et al. 2009). Because the MI is determined by both egg mass at fertilisation and offspring mass at birth, an increase in the degree of placentation can be brought about by an increase in offspring mass at birth and/or a decrease in egg mass at fertilisation. The MI for P. retropinna was estimated in relation to (1) a given population in a specific year, and (2) high and low predation risk by using the Bayesian programming environment JAGS (Plummer 2003) in R v 3.5 (R Core Team 2019) (Supporting Information Methods 1.2, 1.3). In (1), ln-transformed embryo dry mass was fitted as a function of the developmental stage of embryos (stage), stage2, proportion of maternal body fat (BF), maternal standard length (SL), and BF 9 stage. For each population, the model estimates year-specific intercepts and slopes on stage. This allows for the prediction of MI for a given population in a specific year. In addition, the model includes mother identity as additional intercept to allow for variation among females that is not accounted by maternal BF and SL. The population-specific MI’s were subsequently calculated by dividing offspring mass at birth (stage 45) by egg mass at fertilisation (stage 2) that were predicted for a given population in a specific year and for a female of overall average SL and BF. Since the MI’s are predicted for a female of the same SL and BF, the resulting MI’s are independent of these traits (Fig. S1). In (2), ln-transformed embryo dry mass was fitted as a function of stage, stage2, predation, BF, SL, BF 9 stage and predation 9 stage. For each population, the model estimates yearspecific intercepts to account for systematic differences among populations within a year. Moreover, the model includes mother identity as additional intercept to allow for variation among females that is not accounted for by maternal BF and SL. The MI for high and low predation females was subsequently calculated by dividing offspring mass at birth by egg mass at fertilisation that were predicted for a given predation regime and for a female of overall average SL and BF. Quantification of life-history variation among predation regimes The effects of predation on life-history traits (egg mass at fertilisation, offspring mass at birth, proportion of egg and offspring fat, reproductive allotment, brood size, fecundity, superfetation, and abortion incidence) were analysed by fitting each trait as a function of (1) high and low predation risk, and (2) predator community in a series of (generalised) linear mixed effect models in R v 3.5 (R Core Team 2019), using the package lme4 (Bates et al. 2015). Considering Parachromis dovii (P), Eleotris picta (E), and Gobiomorus maculatus (G) as predator species, we observed six different predation categories in nature: low (no strongly piscivorous species present), G, E, EG, PG, and P. Except in the case of reproductive allotment, additional fixed effects included the proportion of maternal body fat (BF) and standard length (SL). Reproductive allotment is defined as the proportion of the mother’s dry mass allocated to reproduction, and hence accounts for female dry mass, rather than BF and SL. Maternal BF predictably responds to experimental manipulation of food availability in the laboratory (Reznick et al. 1996; Banet & Reznick 2008; Banet et al. 2010; Pollux & Reznick 2011), and is believed to be a good indicator of fish condition (Leips et al. 2013). Therefore, accounting for maternal BF may enable us to partly decouple the effects of predation risk and food availability on life-history traits. The association of life-history traits with maternal BF and SL is reported and discussed in the Supporting Information (Table S2–S10; Fig. S2). In the case of reproductive allotment, fecundity and superfetation, the developmental stage of the most-developed brood was fitted as an additional fixed effect to account for females early in the reproductive cycle. Population, year, river, and population 9 year were fitted as random intercepts accounting for spatio-temporal non-independence of observations. Likewise, in the case of brood size, mother identity was fitted as additional random intercept to correct for pseudoreplication, because brood size is measured multiple times in females with superfetation. To optimise normality and homoscedasticity of model residuals, reproductive allotment, abortion incidence, proportion of egg and offspring fat and maternal BF were arcsine squareroot transformed. Fecundity, superfetation, and brood size were fitted in generalised linear mixed effect models using a log link for the Poisson-distributed responses. Path analysis Differences in reproductive allotment (RA) among populations could be due to effects on several life-history traits. For instance, RA could be decreased by reducing brood size or superfetation, which in turn decreases the number of embryos (fecundity). Alternatively, RA diminishes when producing smaller eggs at fertilisation or offspring at birth. We used a path analysis to determine the contribution of each of these life-history traits to differences in RA among predation regimes. In total, three (generalised) linear mixed effect models, implemented in MCMCglmm (Hadfield 2010), were used to estimate all paths (Supporting Information Methods 1.4). Each model was refitted as a function of an intercept only (null model) to compare the deviance information criterion (DIC) of the full model against that of the null model (DDIC). In the first model, egg mass at fertilisation, offspring mass at birth, average brood size for a given mother, and superfetation were fitted in a multivariate model as a function of high and low predation risk allowing for the covariance between the residuals of all responses (DDIC =
272.75). In the second model, maternal fecundity was fitted as a function of superfetation and average brood size for a given mother, as changes in both brood size and superfetation will affect fecundity (DDIC =
388.449). The third model subsequently © 2020 The Authors. Ecology Letters published by CNRS and John Wiley & Sons Ltd 834 A. Hagmayer et al. Letter predicts absolute RA as a function of fecundity, egg mass at fertilisation, and offspring mass at birth (DDIC =
135.094). All three models included the proportion of maternal body fat and standard length as fixed effects. In the case of superfetation and fecundity, the developmental stage of the most-developed brood was an additional fixed effect (see above). To aid convergence, we did not fit a random year effect (which was effectively zero), but otherwise the random effect structure was the same as above. Furthermore, all continuous input variables were z-standardised to obtain standardised partial regression coefficients (b*) that take values between
1 and 1 (Schielzeth 2010). In the case of Poisson-distributed responses (fecundity and superfetation), b* was obtained retrospectively by dividing the estimated slope (b) by the phenotypic standard deviation of the response variable (rY). However, rY was indirectly estimated using the variance of the predicted model fits on the link scale (r2log(Ŷ)) and the pseudo-R-squared (R2) (Menard 2011): rY ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r2log Y^ þ R2 : ð Þ The effect of predation on RA, mediated through a specific maternal life-history trait, is then given by the direct effect of predation on the life-history trait and its contribution to the RA. This effect is quantified by multiplying b* of predation on the life-history trait with b* of the life-history trait on RA. RESULTS There was a more than two-fold range in the estimated degree of placentation among the 28 natural populations of P. retropinna (MI ranging from 14.87 to 32.34; Table S1). Consistent with the locomotor cost hypothesis, we found that P. retropinna females from high predation (HP) localities exhibit a significantly higher degree of placentation (MIHP: mean = 22.86, 95% CI