Increased superfetation precedes the evolution of advanced degrees of placentotrophy in viviparous fishes of the family Poeciliidae

Abstract

The causes and consequences of the evolution of placentotrophy (post-fertilization nutrition of developing embryos of viviparous organisms by means of a maternal placenta) in non-mammalian vertebrates are still not fully understood. In particular, in the fish family Poeciliidae there is an evolutionary link between placentotrophy and superfetation (ability of females to simultaneously bear embryos at distinct developmental stages), with no conclusive evidence for which of these two traits facilitates the evolution of more advanced degrees of the other. Using a robust phylogenetic comparative method based on Ornstein–Uhlenbeck models of adaptive evolution and data from 36 poeciliid species, we detected a clear causality pattern. The evolution of extensive placentotrophy has been facilitated by the preceding evolution of more simultaneous broods. Therefore, placentas became increasingly complex as an adaptive response to evolutionary increases in the degree of superfetation. This finding represents a substantial contribution to our knowledge of the factors that have shaped placental evolution in poeciliid fishes.

1. Introduction

Non-mammalian placentas have received considerable attention during the past two decades because their natural diversity and multiple independent origins have triggered questions about the factors shaping the evolution of these astonishing reproductive structures [1]. A placenta is an organ formed through the intimate apposition of fetal and maternal tissues for physiological exchange [2]. In some groups of viviparous fishes and reptiles, the anatomical structure of their placentas varies substantially among closely related species [3,4]. Studies comparing some of these species have demonstrated that the amount of nutrients that females actively transfer to their developing embryos after fertilization by means of their placentas (known as placentotrophy [1]) is positively correlated with the degree of complexity of their placental tissues [4–6]. Thus, advanced degrees of placentotrophy are possible by means of thickened placental cells with numerous enlarged vesicles, abundant microvilli and richly supplied with capillaries (in particular of the maternal portion of the placenta; [4,5]). By contrast, lecithotrophy is a developmental pattern in which yolk of the ovum (provided by the female before fertilization) represents the main source of embryonic nutrition [1,7]. In several viviparous species, lecithotrophy is associated with simpler placentas that primarily provide the basic function of gas exchange [1,4,5]. The wide interspecific variation that occurs in some non-mammalian viviparous vertebrates in both placental complexity and relative amounts of pre-fertilization (lecithotrophy) and post-fertilization (placentotrophy) sources of nutrients for developing embryos has raised questions about the causes and consequences of the repeated evolution of complex placentas and extensive placentotrophy [8–11].

Superfetation, the capacity of females to simultaneously gestate two or more groups of embryos in different developmental stages [12], is another reproductive strategy that occurs in a few fish families (Clinidae, Poeciliidae and Zenarchopteridae [13,14]) and that has an evolutionary link with placentotrophy [11,15]. In viviparous fishes of the family Poeciliidae both reproductive modes are associated in such a way that many species that exhibit superfetation are also placentotrophic (although some exceptions exist, with a few species that exhibit one trait but not the other; e.g. [16]). Furthermore, advanced degrees of superfetation (i.e. more simultaneous broods) tend to occur in species with extensive placentotrophy [11,17]. Previous phylogenetic comparative analyses have demonstrated this evolutionary association [11,15,18], but have not addressed whether the evolution of increasingly complex placentas, and concomitant increased placentotrophy, has preceded the capacity to bear greater numbers of simultaneous broods or, alternatively, whether the evolution of advanced degrees of superfetation has caused evolutionary increases in placental complexity.

Here, we propose and test the hypothesis that relatively more complex placentas are required to differentially regulate the amount of resources that are actively transferred to embryos in distinct developmental stages and, hence, increased placentotrophy must arise first to facilitate the evolution of more simultaneous broods. Our reasoning behind this hypothesis is that the specific nutritional requirements as well as the rate at which nutrients must be transferred from the mother to embryos presumably differ between early- and late-stage embryos [19,20]. Elaborate placental tissues (richly vascularized with abundant vesicles and microvilli) are likely needed for this differential provisioning and regulation to function properly [4–6]. In addition, in poeciliid fishes more simultaneous broods result in higher total fecundities [21], which impose a greater demand on mothers for nutrients and oxygen [22]. Increased placentotrophy by means of complex placentas should allow females to provide greater amounts of both nutrients and oxygen, thereby fulfilling the requirements of all the embryos across all developmental stages. Therefore, the preceding evolution of advanced placentotrophy must have been necessary to facilitate the subsequent evolution of increased superfetation. To address this hypothesis, we focused on viviparous fishes of the family Poeciliidae because the degrees of both placentotrophy and superfetation vary substantially among species, with ample evidence of dependent evolution between these two reproductive modes [9,11,15,18]. To date, however, we still do not know if greater degrees of one of these two traits must arise first to facilitate the evolution of advanced degrees of the other.

2. Methods

From the literature, we compiled a dataset of 36 poeciliid species (electronic supplementary material, table S1). We focused on species-specific values of two variables: degree of superfetation, quantified as the average number of simultaneous broods, and the matrotrophy index (MI). The latter is calculated by dividing dry mass of offspring at birth by the dry mass of the recently fertilized egg and, hence, provides an estimate of the amount of nutrients that females transfer to their developing embryos after fertilization [23]. The MI is a traditional way to quantify degree of placentotrophy [7]. Instead of qualitatively classifying species as either placentotrophic or lecithotrophic and as either superfetating or non-superfetating, as is typically done in analyses of these traits [11,15,18], we analysed here the wide continuous variation in both traits observed among species. In our dataset, degree of superfetation varies between 1.03 and 7 average simultaneous broods. In turn, MI varies between 0.5 (indicating strict lecithotrophy) and 117 (indicating extensive placentotrophy) (electronic supplementary material, table S1). We log-transformed both variables before analysis.

We used a powerful phylogenetic comparative method [24] that is a multivariate extension of the adaptation–inertia model of Hansen et al. [25] and is based on Ornstein–Uhlenbeck (OU) models of adaptive evolution. This method, implemented in the R package mvSLOUCH [24,26], allowed us to test for coadaptation between placentotrophy and superfetation as well as to examine if one of these two traits evolved in response to the other. We used a dated molecular phylogeny of the family Poeciliidae [27], which we pruned to include only our 36 species (figure 1) because quantification of the degree of both traits is currently available only for these 36 species. Additional details on the phylogeny can be found in the electronic supplementary material, appendix S1a. We fitted three types of models that differ in their assumptions about trait coevolution: first, a multivariate Brownian motion (BM) model that assumes no stabilizing selection (i.e. no evolution toward an optimal state) and, hence, both placentotrophy and superfetation accumulate variation over time around the root value, in such a way that coevolution between the two traits is due to correlated random oscillations; second, a joint Ornstein–Uhlenbeck–Brownian-motion (OUBM) model in which one of the two traits is modelled as an OU evolutionary process, adapting to an optimum that is influenced by the other trait, which in turn is modelled as randomly following a BM process; third, an Ornstein–Uhlenbeck–Ornstein–Uhlenbeck (OUOU) model in which both traits experience adaptive evolution toward optimal values and that allows us to identify if the optimum of one of the two traits is pulled by the adaptive changes that occur in the other trait. We considered the measurement error (intraspecific variation calculated as per [28–30]) in both placentotrophy and superfetation during model implementation (electronic supplementary material, appendix S1b).

Figure 1.

Time-calibrated phylogeny of 36 fish species of the family Poeciliidae, modified from [27]. Bars represent mean values of placentotrophy (quantified by the matrotrophy index) and superfetation for each species.

The parameters of these models are estimated by means of maximum-likelihood routines, and their interpretation (which we summarize below) provides compelling information about potential coadaptation between placentotrophy and superfetation and, most importantly, about which of the two traits has influenced the primary optimum of the other (as presented by [24]). We focused on the A (drift) matrix which includes the rates of adaptation toward trait-specific optima (diagonal values) and coadaptation between traits (off-diagonal values), as well as on the Σ_yy (diffusion) matrix which includes the random (non-adaptive) trait-specific variances (diagonal values) and covariances between traits (off-diagonal values). Under this setup, an A matrix with a non-zero value above the diagonal (upper triangular matrix) indicates that superfetation has affected the optimum of placentotrophy. By contrast, an A matrix with a non-zero value below the diagonal (lower triangular matrix) indicates that placentotrophy has affected the optimum of superfetation. A diagonal A matrix (non-zero values exclusively in the diagonal) indicates no coadaptation between traits. Notice that the particular parameterization of the A matrix that provides the best fit to the data allows to identify which of the two traits has caused evolutionary changes in the other. In the case of Σ_yy, a diagonal matrix indicates no covariation between the random (non-adaptive) components of the two traits, whereas a Σ_yy matrix with non-zero off-diagonal values (non-diagonal matrix) indicates a non-adaptive correlation between the random perturbations of both traits [24]. Additional mathematical details of the multivariate adaptation–inertia models can be found in the electronic supplementary material, appendix S1c.

Using mvSLOUCH, we fitted one BM model as well as OUBM and OUOU models with all six possible combinations of A (upper triangular, lower triangular, and diagonal) and Σ_yy (diagonal and non-diagonal) matrices. Each combination was implemented for both OUBM and OUOU models, yielding 12 models, plus one BM model for a total of 13 competing models. To examine model convergence and consistency in parameter estimates, each combination of A and Σ_yy matrices for OUBM and OUOU models was run from five different starting points. We assessed model fit by means of sample-size-adjusted Akaike information criterion (AICc; [31]). Additionally, mvSLOUCH allowed us to calculate an evolutionary regression, which provides an estimate of the effect of one of the two traits on the other, accounting for phylogenetic relationships among species. We estimated this regression from the evolutionary model that provided the best fit to the data. A 95% confidence interval for the slope of this evolutionary regression was calculated by means of a parametric bootstrap procedure (electronic supplementary material, appendix S1d). The R code to implement mvSLOUCH analyses is available in the electronic supplementary material.

Finally, we tested for robustness of the results by examining the potential influence on our findings of considering (1) non-superfetating species with moderate and extensive placentotrophy (electronic supplementary material, table S2 and figure S1), and (2) exclusively species in which superfetation evolved before placentotrophy (according to [18]). We fitted all 13 competing models to these two additional sets of species and compared the results with those obtained from our original analysis with 36 species (electronic supplementary material, appendix S1e).

3. Results

We found evidence of coadaptation between placentotrophy and superfetation, with a clear causality pattern. The model with strongest support in the data was an OUOU model that included an upper triangular A matrix and a diagonal Σ_yy matrix (table 1). The five different starting points for this particular combination of A and Σ_yy matrices had almost identical AICc scores. Thus, regardless of the initial conditions, this OUOU model converged to the same solution (electronic supplementary material, table S3). An upper triangular A matrix indicates that the primary optimum of placentotrophy has been affected by the degree of superfetation. The estimated slope from the evolutionary regression was positive (1.71, 95% CI = 0.00003–2.77) and, thus, evolutionary increases in superfetation led to increases in the amount of placentotrophy (figure 2). A diagonal Σ_yy matrix indicates that the random perturbations of placentotrophy and superfetation are not correlated. Therefore, the evolutionary dependency between these two traits has arisen exclusively from their coadaptation.

Table 1.

Model selection results for our examination of the evolutionary association between placentotrophy and superfetation. The model with strongest support is highlighted in bold.

model	AICc	ΔAICc	type of drift (A) matrix	type of diffusion (Σyy) matrix
OUOU	163.91	0	upper triangular	diagonal
OUOU	167.80	3.89	diagonal	non-diagonal
OUOU	168.12	4.21	upper triangular	non-diagonal
OUOU	170.25	6.34	lower triangular	non-diagonal
BM	171.03	7.12	—	—
OUOU	177.79	13.88	lower triangular	diagonal
OUBM	178.28	14.37	upper triangular	non-diagonal
OUBM	178.37	14.46	lower triangular	non-diagonal
OUBM	178.37	14.46	lower triangular	diagonal
OUBM	178.38	14.47	upper triangular	diagonal
OUBM	178.38	14.47	diagonal	non-diagonal
OUBM	178.39	14.48	diagonal	diagonal
OUOU	179.17	15.26	diagonal	diagonal

Figure 2.

Statistical effect of degree of superfetation (number of simultaneous broods) on the amount of placentotrophy (quantified by the matrotrophy index). The line depicts the evolutionary regression estimated from the top model in table 1.

The rates of adaptation (diagonal values in the A matrix) indicate that superfetation adapts faster to its optimum compared with placentotrophy (table 2). The rate of coadaptation between traits (value above the diagonal in the A matrix) was negative (table 2), suggesting that as superfetation approaches its optimum, it pulls placentotrophy away from its own primary optimum. This influence of superfetation, which opposes a centralizing tendency toward an optimal MI value, partially explains the slower rate of adaptation that we observed in placentotrophy.

Table 2.

Drift (A) and diffusion (Σ_yy) matrices estimated from the top model in table 1.

	A matrix		Σyy matrix
	placentotrophy	superfetation	placentotrophy	superfetation
placentotrophy	2.439	−21.479	0.678	0
superfetation	0	10.135	0	2.131

Our results were robust to different sets of species. The two additional analyses that considered non-superfetating species with moderate and extensive placentotrophy and exclusively species in which superfetation evolved before placentotrophy yielded qualitatively identical results to those obtained from our main analysis of 36 species (electronic supplementary material, appendix S2, table S4, figures S2 and S3).

4. Discussion

Most previous studies on placentotrophy and superfetation of viviparous fishes of the family Poeciliidae have examined whether the presence of one of these two traits is associated with the presence of the other (i.e. treating both traits as categorical: presence or absence) and have provided solid evidence of their dependent evolution [9,11,15,18,32]. In particular, Furness et al. [11] addressed this question under a Bayesian framework using a large dataset (136 species) and concluded that both traits are likely to appear together. However, their analysis, which treated both traits as dichotomous, indicated that either of these two traits could facilitate the evolution of the other, with no clear evidence of which one was more likely to arise first. Here, we have gone one step further by using the wide continuous variation observed among species in both traits to test the hypothesis that the evolution of extensive placentotrophy (and hence of complex placentas [4,5]) is necessary to facilitate the evolution of an increased number of simultaneous broods. Intriguingly, we found the opposite pattern: the evolution of more simultaneous broods has preceded the evolution of increased placentotrophy and the optimum of this latter trait has been affected by evolutionary changes in the degree of superfetation. A key implication of this pattern of causality is that placental complexity has arisen as an indirect consequence of the evolutionary forces that have promoted the occurrence of many broods developing simultaneously, presumably contributing to optimize the nutrition and regulation of embryos at different developmental stages. Intriguingly, very little is known about the physiological mechanisms underlying superfetation and partitioned provisioning.

Furness et al. [11] also analysed the continuous variation in both traits, including species with and without superfetation, and found a significant positive relationship. However, unlike our multivariate adaptation–inertia models [24], the phylogenetic generalized least-squares analysis that they conducted was designed neither to explicitly test for coadaptation nor to examine which of the two traits preceded the evolution of the other [33].

Our results are consistent with previous across-species evidence of no effects of several environmental variables on the degree of placentotrophy. According to recent phylogenetic comparative analyses within the family Poeciliidae, neither altitude, geographical range, environmental temperature, precipitation, seasonality, water velocity nor predation risk appears to have evolutionary relationships with placentotrophy [11]. This indicates that, at the macroevolutionary scale, no selective forces associated with particular ecological factors have had clear effects on the extensive variation that exists among species in the degree of post-fertilization maternal provisioning. Rather, according to our evidence, the evolution of advanced placentotrophy and increasingly complex placentas has been facilitated by the preceding evolution of more simultaneous broods. A handful of studies conducted at the intraspecific (microevolutionary) scale in different poeciliid species have shown that higher degrees of superfetation likely evolved as a response to fast water velocity or increased predation risk (because superfetation reduces the abdominal distention of pregnant females, which allows them to exhibit streamlined bodies and efficient swimming performance), suggesting that natural selection has had strong effects on this reproductive trait [9,12,34–36]. This is in fact consistent with the faster rate of adaptation to its primary optimum that we detected in superfetation compared with placentotrophy. Taken together, these lines of evidence, from both macro- and microevolutionary perspectives, suggest that the evolution of advanced superfetation has occurred predominantly by means of adaptation to environmental conditions, whereas the evolution of extensive placentotrophy has been caused, at least partially, by an internal functional requirement for more complex placentas as the degree of superfetation increases.

Our study sheds light on one of the main processes underlying the evolution of complex placentas and extensive placentotrophy in poeciliid fishes. However, we do not contend that other potential causal factors should be disregarded. For instance, some poeciliid species exhibit moderate and extensive placentotrophy without superfetation (e.g. most Phalloceros and Pamphorichthys species; [37]). In these taxa, the evolution of advanced placentotrophy is hardly explained by the preceding evolution of advanced degrees of superfetation, because they all lack superfetation. Here is where other selective mechanisms must have been the main driving forces of placentation. In particular, elaborate placentas provide females with greater control over the amount and type of nutrients that are allocated to each embryo, which is critical in the context of intergenomic (parent–offspring) conflict [15,18,38]. This is the central idea behind the viviparity-driven conflict hypothesis, which also attempts to explain the evolution of placentotrophy [39,40]. Both intergenomic interactions and the selective pull that superfetation exerts on placentotrophy should be jointly considered in future explorations of the potential causes of placental evolution in poeciliid fishes.

Data accessibility

The dataset supporting this study and the R code to perform mvSLOUCH analyses are available in the electronic supplementary material [41].

Authors' contributions

K.N.G.-C: data curation, formal analysis, investigation, validation, writing—original draft, writing—review and editing; J.A.F.-G.: formal analysis, investigation, methodology, validation, writing—review and editing; N.S.-S: data curation, validation, writing—review and editing; J.P.: formal analysis, investigation, methodology, software, writing—review and editing; J.J.Z.-V.: conceptualization, formal analysis, investigation, methodology, project administration, supervision, writing—original draft, writing—review and editing.

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

Conflict of interest declaration

We declare we have no competing interests.

Funding

Consejo Nacional de Ciencia y Tecnología from México awarded PhD fellowships to K.N.G.-C. (grant no. 547333) and N.S.-S. (grant no. 596700).