Why do placentas evolve? A test of the life-history facilitation hypothesis in two clades in the genus Poeciliopsis representing two independent origins of placentas

Authors

  • Ronald D. Bassar,

    Corresponding author
    1. Department of Biology, University of California, Riverside, California, USA
    Current affiliation:
    1. Department of Environmental Conservation, University of Massachusetts, Amherst, Massachusetts, USA
    Search for more papers by this author
  • Sonya K. Auer,

    1. Department of Biology, University of California, Riverside, California, USA
    Current affiliation:
    1. Department of Environmental Conservation, University of Massachusetts, Amherst, Massachusetts, USA
    Search for more papers by this author
  • David N. Reznick

    1. Department of Biology, University of California, Riverside, California, USA
    Search for more papers by this author

Summary

  1. Most of what we know about placentas comes from mammals, yet little can be learned from them about the adaptive significance of the placental mode of reproduction because they all derived their placenta from a single common ancestor that lived over 100 million years ago. We can make inferences about the adaptive significance of placentation from fish in the family Poeciliidae because there have been multiple, recent origins of placentation, affording an opportunity to compare close relatives with and without placentas and to seek properties that are common to each origin of placentation.
  2. Here, we used field collections and a common garden study to quantify the degree of placentation and related it to aspects of the life history in two clades of live-bearing fish from the genus Poeciliopsis that each contains an independent origin of placentation. Doing so enables us to test the ‘life history facilitation hypothesis’, or the proposal that the placenta evolved to facilitate the evolution of some other feature of the life history.
  3. We found that the evolution of placentation in each clade is tightly correlated with the evolution of other components of the life history, but that the nature of the association is radically different across the two clades. In the Northern Clade the magnitude of post-fertilization maternal provisioning is negatively correlated with age at maturity, mass at maturity, offspring dry mass and interlitter interval. In contrast, degree of matrotrophy in the Southern Clade is positively correlated with age at maturity, mass at maturity, offspring dry mass and inter-litter interval.
  4. There is thus no consistent relationship between the evolution of placentas and other features of the life history, which negates those proposals that the placenta evolved to facilitate the evolution of other features of the life history. However, there is a negative correlation between degree of placentation and ovary dry mass and reproductive allocation common to both clades, suggesting that placentation may be an adaptation that facilitates a reduction in reproductive allocation.

Introduction

Complex adaptations, such as the vertebrate eye or the mammalian placenta, are generally common to large groups of organisms, all of whom inherited the character from a single, ancient common ancestor. We typically have no knowledge of the circumstances in which the character evolved or any transitional states in the evolution of the character found in living descendants. This combination of circumstances means that we have little means for inferring how or why complex traits evolve. Here, we present the unusual circumstance of the evolution of placentation in the fish family Poeciliidae. We have established that there have been multiple independent origins of placentas within the Poeciliidae and that there are often close relatives who either do or do not have a placenta, plus species that vary in the extent of placentation (Reznick, Mateos & Springer 2002; Pires, Arendt & Reznick 2010; Meredith et al. 2011). Such circumstances enable us to test hypotheses for why placentas evolved and in other ways make inferences about the adaptive significance of placentation that would never be possible for mammals because they lack the necessary variation.

A broad diversity of reproductive modes, ranging from internal versus external fertilization or oviparity versus viviparity, is found throughout the natural world. The evolution of viviparity requires the prior evolution of internal fertilization. Within viviparous organisms, strategies for providing nourishment to developing offspring fall along a continuum that ranges from lecithotrophy, wherein all nourishment for growth and development of offspring is provided prior to fertilization, to matrotrophy, wherein offspring are provided with nourishment throughout their development (Wourms, Grove & Lombardi 1988; Blackburn 2000). The most well-known example of matrotrophic organisms are the placental mammals, whose post-fertilization provisioning of offspring is facilitated by a placenta, defined as an integration of maternal and embryonic tissues that are specialized for the physiological maintenance of the developing young (Mossman 1937). To these criteria, we add that the placenta must also be adapted for the transfer of nutrients from the mother to developing young. Matrotrophy as a generalized strategy is not restricted to placental mammals. It has evolved independently numerous times across a wide spectrum of organisms, including terrestrial and aquatic gastropods (Baur 1994; Von Rintelen & Glaubrecht 2005), clams (Korniushin & Glaubrecht 2003), pseudoscorpions (Makioka 1968), flies (Meier, Kotrba & Ferrar 1999), cockroaches (Williford, Stay & Bhattacharya 2004), isopods (Warburg & Rosenberg 1996), elasmobranchs (Hamlett & Hysell 1998), several groups of bony fish (Wourms, Grove & Lombardi 1988; Reznick, Mateos & Springer 2002), amphibians (Wake 1993; Greven 1998) and reptiles (Stewart 1992). Across these taxa, matrotrophy is characterized by a spectrum of complex morphological structures and physiological pathways. Some matrotrophic species have evolved the functional equivalent of the mammalian placenta.

Virtually, all hypotheses proposed for the evolution of matrotrophy are independent of the framework of life-history theory, one exception being the model proposed by Trexler & DeAngelis (2003). Most of these hypotheses are ad hoc in the sense that they are ideas suggested by the study of life histories of one or a few matrotrophic species. Many of them suggest that matrotrophy evolved to facilitate the evolution of some other feature of the life history. For example, it has been proposed that matrotrophy evolved to facilitate the evolution of larger litter size, larger offspring size at birth, improved survivorship early in life or earlier maturity (Thibault & Schultz 1978; Blackburn, Vitt & Beuchat 1984; Wourms & Lombardi 1992; Wourms 1993; Trexler 1997; Holbrook & Schal 2004; Schrader & Travis 2005; Wildman et al. 2006). We reference these hypotheses collectively as ‘life history facilitation hypotheses’ because they all share the attribute of predicting that matrotrophy evolves to facilitate the evolution of some other life-history trait.

Pires et al. (2011) tested the life-history facilitation hypothesis in six species from the Northern Clade of the fish genus Poeciliopsis. Three of these fish species lack matrotrophy and three have matrotrophy that varies from sustaining a 10% increase (P. occidentalis) to an eightfold increase (P. prolifica) in the dry mass of the developing young between fertilization and birth. A general way of evaluating the plausibility of the life-history facilitation hypothesis is to ask whether this graded increase in post-fertilization provisioning is predictably associated with the evolution of other components of the life history. Pires et al. (2011) found that the evolution of matrotrophy was tightly correlated with the evolution of earlier maturity, a smaller size at maturity, an increase in the rate of production of offspring early in life and the production of smaller offspring. These trends support those life-history facilitation hypotheses that suggest that the evolution of the placenta facilitates the evolution of earlier maturity and an increase in the rate of offspring production.

Pires et al. (2011) results also suggest a possible bridge between the evolution of matrotrophy and the more general demographic theory of life-history evolution. The constellation of life-history attributes associated with the evolution of matrotrophy in the Northern Clade of Poeciliopsis is the same that is predicted to evolve in response to exposure to high extrinsic rates of adult mortality or to high mortality rates across all age classes (Roff 1992; Stearns 1992). If the evolution of matrotrophy is indeed consistently associated with this same complex of life-history traits and if it is also associated with species that experience high extrinsic mortality rates, then we can incorporate the evolution of matrotrophy into this more general life-history framework.

A virtue of the genus Poeciliopsis is that it contains three independent origins of extensive matrotrophy (Reznick, Mateos & Springer 2002). The Southern Clade of this genus consists of six described species, four of which are lecithotrophic and two of which have extensive matrotrophy, sufficient to sustain a greater than a 30-fold increase in the dry mass of offspring between fertilization and birth (Reznick, Mateos & Springer 2002). Here, we take the next step in evaluating the generality of Pires and colleagues' results by repeating their study on the Southern Clade and comparing associations between matrotrophy and the life history among the two clades, each representing an independent origin of matrotrophy. If their results are general, then we should obtain the same associations between the evolution of matrotrophy and the evolution of the rest of the life history as seen in the Northern Clade.

Materials and methods

The genus Poeciliopsis (Cyprinodontiformes: Poeciliidae) contains 20 described viviparous species that inhabit Pacific slope drainages from southern Arizona, USA, to Colombia (Mateos, Sanjur & Vrijenhoek 2002). All species in Poeciliopsis have the ability to carry multiple, simultaneous litters (superfetation). The number of simultaneous litters varies from two to five across the genus (Turner 1937, 1947; Scrimshaw 1944; Thibault & Schultz 1978). In the placental species, resources are transferred from the mother to the developing offspring via the follicular placenta (Turner 1939, 1940, 1947; Grove & Wourms 1991, 1994; Wourms & Lombardi 1992), which is an integration of maternal tissue (the follicle) with either a modified yolk sac or externalized pericardial membrane of the embryo. Reznick, Mateos & Springer (2002) present descriptions of the pattern of maternal provisioning for all but one species in the genus and combine them with the phylogeny to make inferences about the evolution of maternal provisioning in this genus. They demonstrate that provisioning across species can range from having virtually no post-fertilization provisioning to having nearly a 120-fold increase in the dry mass of offspring between fertilization and birth. They also established that there have been three independent origins of placentation within the genus.

Study species and collection sites

In the first portion of this paper, we quantify the association between the degree of matrotrophy and the life history in species of the Southern Clade of Poeciliopsis. The Southern Clade is further subdivided into two clades, the first containing P. fasciata, P. latidens and P. baenschi and the second containing P. catemaco, P. hnilicki, P. gracilis, P. turneri, P. presidionis, P. scarlii and P. turrubarensis (Mateos, Sanjur & Vrijenhoek (2002). The Southern Clade contains only one independent origin of matrotrophy, in the common ancestor of P. presidionis and P. turneri (Reznick, Mateos & Springer 2002), so we only considered the seven species from the second clade in the current study.

We examined the association between matrotrophy and the life history in the wild and laboratory using field collections and a common garden experiment on the second generation of laboratory born offspring from wild-caught parents, respectively. Fish from the field collections of all seven species and 16 localities (up to four populations per species) were either collected by ourselves or were subsets of collections from museums (Appendix S1, Supporting information). We included only 4 of these species (the lecithotrophic species P. gracilis, P. scarllii, and the matrotrophic species P. turneri and P. presidionis) in the laboratory common garden analysis. P. scarllii was represented by two populations (Rio Tomatlan and Rio San Blas). Founders for our laboratory stocks for these populations were collected by D. Reznick, M. Pires and M. Mateos in May 2003 and January 2004 (Appendix S1, Supporting information).

In the second portion of this paper, we present combined analyses of the Northern and Southern Clades. To do so, we include data from the Northern Clade originally presented by Pires et al. (2011). Details of the collection sites for the Northern Clade can be found in Pires et al. (2011).

Dissection of field collected fish

We quantified the life histories of each collection – including the minimum and mean size of reproducing females, number of offspring per litter, degree of superfetation, ovary dry mass, mean offspring mass and reproductive allocation – and the degree of matrotrophy using similar protocols as done for the Northern Clade (Pires et al. 2011) to facilitate comparison. We determined female size by measuring standard length and weighing individual somatic dry mass. Developing offspring and associated reproductive tissues were removed from each female, litters were separated based on stage of development and the number of litters and number of offspring in each litter were quantified (Reznick 1981, 1982; Haynes 1995; Pires et al. 2011). Litter size was defined as the number of offspring in a litter of offspring of similar developmental stage. The degree of superfetation was measured as the number of distinct litters a female was carrying at the time of dissection. Ovary dry mass was determined by drying and then weighing developing offspring and reproductive tissues. Mean offspring mass was calculated as the dry mass of all individual offspring in a litter divided by the number of offspring in the litter. Reproductive allocation (RA) was defined as the percentage of total dry mass of the females that was devoted to reproduction at the time of dissection and was calculated as the ovary dry mass divided by the total dry mass of the female.

We then estimated the degree of matrotrophy using the Matrotrophy Index (MI) for each individual. MI is the average dry mass at birth divided by the average dry mass of eggs with blastodiscs (Reznick, Mateos & Springer 2002) and is similar to other approaches used in other taxa (Wourms, Grove & Lombardi 1988; Stewart & Thompson 2003; Thompson & Speake 2006). If a female fully provisions eggs prior to fertilization, then MI has a value <1, usually in the vicinity of 0·6–0·7, because the embryos lose mass between the time when the egg is fertilized and when the embryo is born. If there is substantial post-fertilization provisioning, MI is instead >1. For example, P. prolifica from the Northern Clade has MI values in the vicinity of eight, which means that there is, on average, an eightfold increase in dry mass between fertilization and birth (Pires, McBride & Reznick 2007).

Laboratory studies

We then examined covariation among matrotrophy and the life-history traits through common garden laboratory experiments, using protocols similar to those as used for the Northern Clade (Pires et al. 2011) to facilitate comparison. The virtue of the laboratory studies is that they enable us to control for environmental effects and to quantify additional life-history variables. Briefly, wild-caught fish were brought to the laboratory and raised to at least the second generation in a common lab environment to reduce variation among species that might arise from maternal and environmental effects. Siblings were placed together in groups of five individuals in 8 L aquaria on the day they were born and were fed a diet of liver paste in the morning and brine shrimp nauplii in the afternoon on a daily basis until they reached a weight of 30 micrograms. As offspring number and size at birth vary between species, the time interval between birth and placement in the experiments differed between species. For the Southern Clade species, P. presidionis and P. turneri reached the designated mass within 5–8 days of age, while it took approximately 25 days for P. gracilis and P. scarlii populations to attain that mass.

Once they reached 30 mg, fish were placed in the experiment under one of two separate experimental designs. In design 1, individual F2 fish were placed in separate 8 L aquaria and positioned in the laboratory based on a randomized block design wherein each block contained four tanks of each species, but the orders of the species or population group on the shelves were randomly assigned across blocks. All species in a block were set up within 1 week of each other to keep setup time relatively constant. Once their male siblings or cohort members began to show signs of reaching maturation, mature stock tank males were added to the female tanks for 1 week, every other week, to serve as mates. We timed the addition of males this way because prior research (Reznick 1982) revealed that males and females mature at approximately the same rate, so the maturation of brothers can serve as an indicator of the approaching maturity of their sisters. Providing mature males shortly before females attain maturity assures that females will mate as soon as they are capable of reproducing.

Some species did not reliably produce offspring under design 1, so we also employed a modification of this design, hereafter design 2. After individuals obtained the appropriate size for placement in the study and before their age at first parturition, they were placed in species-specific 20 L group tanks instead of individual 8 L tanks as done in design 1. Six individuals per species per block were housed in each of these group tanks. Once the experimental fish were large enough to avoid risk of being cannibalized, two mature stock tank males were added to the tanks as mates. Developing males were moved to 8 L aquaria when their gonopodium began to elongate. All females were removed once offspring appeared in the group tank (i.e. when the first female had given birth). Once individuals were removed, they were placed in the randomized block setup described for design 1.

As done by Pires et al. (2011) and in prior common garden studies (e.g. Reznick & Bryga 1996), we reared fish at two different ration levels for design 1. Within each species or population of a block, fish were randomly assigned to either a high or low food ration. High and low food rations were identical across species and populations. In design 2, all fish were fed ad libitum food for the duration of the experiment.

In both designs, tanks were checked daily to determine the sex of the individuals and to check for newly born offspring in female only and group tanks. The maturation status of males was checked by observing the degree of metamorphosis in the anal fin, and males were considered mature when the barbed tip of the gonopodium was no longer covered by protective cells (Turner 1941). Once mature, males were euthanized using an overdose of MS-222 and their standard length and wet mass were measured. When new offspring were observed in female tanks, the offspring were removed, enumerated, euthanized using an overdose of MS-222, and preserved in 5% formalin for later determination of dry mass. Females were euthanized and preserved in 5% formalin 60 days after first parturition. Sixty days is approximately two times the amount of time required for an embryo to develop from fertilization to birth. This time interval was thus sufficient for us to collect four or more litters of young from these species because they all have superfetation. All experiments were conducted in the vivarium at the University of California, Riverside (UCR) under protocols approved by the UCR Institutional Animal Care and Use Committee.

We measured MI and the same life-history traits as in Pires et al. (2011), including age and wet mass at first parturition, number of offspring per litter, degree of superfetation, ovary dry mass, mean offspring mass, reproductive allocation, and interlitter interval. Number of offspring per litter, degree of superfetation, ovary dry mass, mean offspring mass, reproductive allocation and the degree of matrotrophy were measured using the same protocols as described previously for the field collections. We used age and mass at first parturition as a proxy for maturity because females in this genus have no external clues to mark the time of their maturity. Interlitter interval was defined as the duration, in days, between the births of two consecutive litters.

Statistical analyses

We first examined whether there were significant differences in the suite of life-history traits among species and populations in the laboratory stocks using multivariate analyses of variance (manova). We included age and wet mass at first parturition, offspring size, litter size, degree of superfetation, interlitter interval and ovary dry mass as variables that describe the life history. We excluded the MI from this analysis because our ultimate goal was to see whether MI predicts the patterns of life-history variation. We also excluded ova dry mass and RA from the multivariate analyses. Ova dry mass is in the denominator of the ratio used to estimate MI and RA is derived from female wet mass, so these two traits were known a priori to be correlated with MI and female wet mass at first parturition, respectively. Ovary dry mass, which is the mass of all developing embryos and associated reproductive tissue, was used instead of RA in multivariate analyses.

We then employed discriminant function analysis (DFA) to characterize the contributions of individual dependent variables to the differences among species. We evaluated canonical variables from the DFA with eigenvalues greater than one (first three) to determine which of the canonical variables was primarily responsible for the separation among species. We used the probability of correct classification as a measure of the degree of separation among species. Next, we used bivariate correlation analyses to determine whether any of the canonical variables was correlated to MI. Finally, we examined the total canonical structure of any canonical axis that was related to MI to determine which of the life-history traits was mostly associated with that axis and ultimately the degree of matrotrophy. The total canonical structure is equivalent to the bivariate correlation between the score on the canonical variable and the life-history traits used to construct it. Individual level data were used for this analysis.

We utilized a similar approach for the life-history analysis of fish from the wild collections. However, instead of DFA, we performed a principle components analysis (PCA) on the population means of each life-history trait. We took this approach because using individual level data in a DFA would force us to reduce the number of life-history traits we could examine in this collection. So, using the PCA allowed us to examine the multivariate relationship among life-history traits, but did not allow us to explicitly test for differences among populations in the life histories. We used a modified suite of variables for the PCA because the variables estimated from the field samples were not identical to those estimated in the laboratory. The life-history traits for the field analyses included minimum size of reproductive females (a field surrogate for the size at first parturition), the mean level of superfetation, the mean number of offspring per litter, the projected mass of offspring at birth (derived from a regression that describes the relationship between stage of development and offspring mass) and the mean total reproductive mass (the dry mass of all developing embryos and ovarian tissues). We retained the first three principle components and calculated the factor loadings (bivariate correlations between each principle component and the life-history traits). Similar to the DFA for the laboratory populations, we then used bivariate correlations to test for significant correlations between each principle component axis and the degree of matrotrophy. The multivariate analyses of the field and laboratory data were conducted in a conventional, non-phylogenetic fashion (i.e. assuming a star phylogeny). We also tested separately for relationships between matrotrophy and each life-history character in the laboratory and field populations of the Southern Clade using bivariate correlations, but results were qualitatively similar to those from the multivariate analyses (Appendix S2, Supporting information).

Finally, we tested for differences between the two clades in their relationship between matrotrophy and each life-history character. We started by including the same life-history traits in both clades as dependent variables in a multivariate analysis of variance (manova). Independent predictors included clade entered as fixed effect, natural log transformed MI as a covariate and the interaction between clade and MI. A significant interaction between clade and MI would mean that the multivariate relationship between the life-history variables and MI differed among the independent origins of matrotrophy. Individual level data from only the lab study were used in this analysis. Next, we used a regression approach to analysis of covariance (ancova) to evaluate which traits contributed to similarities and differences in the relationship between MI and the life-history traits between the clades. The value of the life-history trait was included as the dependent variable and MI was included as a covariate. Clade was entered as a fixed effect and was dummy coded as 0 or 1. The interaction between clade and MI was included to test for differences in the relationship between the life-history trait and matrotrophy between the two clades, but interactions were subsequently removed if non-significant (> 0·05).

Analyses of the bivariate relationships in the Southern Clade alone and univariate comparisons of these relationships between the two clades were conducted on population means using both raw data (i.e. ‘star phylogeny’) and data corrected for phylogenetic relationships using phylogenetic generalized least-squares (PGLS). These analyses were conducted using the Matlab program REGRESSION.m (Ives, Midford & Garland 2007). We employed the most current evolutionary hypothesis for the phylogenetic relationship among the Poeciliopsis species (Mateos, Sanjur & Vrijenhoek 2002) and used arbitrary branch lengths because the divergence time between species or populations was unknown (Pagel 1992). For both the laboratory and field collections, populations within species were included as soft polytomies with branch lengths set to 0·5. We compared the likelihoods of both the star and PGLS analyses to determine which provided a better fit to the data.

Results

Southern clade

Laboratory-reared populations: The manova showed that a significant amount of the variation in life-history traits were attributable to differences among species (Wilks' λ28,192·5 = 0·003, P < 0·0001, Fig. 1). The subsequent DFA showed that all individuals were correctly classified to species. The only misclassifications were between the two populations of P. scarlii. In total, two of ten individuals (20%) from Rio San Blas were misclassified as being from the Rio Tomatlan population. The first canonical variable accounted for 88% of the variation in the life-history traits among the species and was positively correlated with MI (= 0·99, d.f. = 3, P < 0·001, Table 1, Fig. 2). This significant correlation shows that the evolution of increased MI in this clade is indeed correlated with the evolution of a complex of other life-history traits. Specifically, the evolution of increased MI is correlated with the evolution of delayed age at maturity, a larger body size at maturity, a higher degree of superfetation, reduced litter size, reduced ovary dry mass, but increased offspring size at birth (Table 1). Of all the life-history traits, only interlitter interval showed no relationship with the first canonical variable. The other axes accounted for much less variation and were not significantly related to the degree of matrotrophy (Table 1).

Table 1. Bivariate correlation coefficients between either the canonical variables (laboratory) or the principle components (field) and each the life-history traits. Eigenvalues and summary statistics of the correlation between each canonical variable or principle component and degree of matrotrophy
 Canonical variable
Variable123
(A) Laboratory data
Age at first parturition0·341−0·527−0·005
Female wet mass at first parturition0·447−0·591−0·361
Offspring dry mass0·991−0·0590·012
Litter size−0·6780·0670·006
Superfetation0·356−0·5330·184
Interlitter interval−0·0480·813−0·377
Ovary dry mass−0·867−0·0760·229
Eigenvalues29·32·51·2
Proportion of total variance explained0·8770·0750·037
Correlation with MI0·9990·259−0·114
d.f.333
P <0·0010·6740·855
 Principle component
Variable123
(B) Field data
Smallest size of reproductive females0·584−0·0750·525
Superfetation0·430−0·297−0·812
Litter size0·2840·641−0·061
Estimated offspring size at birth0·412−0·5420·235
Ovary dry mass0·4730·449−0·078
Eigenvalues2·1431·7800·650
Proportion of total variance explained0·4290·3560·130
Correlation with MI0·340−0·672−0·240
d.f.141414
P 0·1980·0040·371
Figure 1.

Ordinations of the five populations used in the laboratory study on canonical variables 1 and 2 (a) and canonical variables 3 and 4 (b). Small symbols represent the scores for the individual fish and the large symbols represent the centroids for each population.

Figure 2.

Relationships between Matrotrophy Index and the first three canonical variables (a–c) for the laboratory data and the first three principle components for the wild collection (d–f). Canonical variable scores are the centroids for each species from the discriminant function analysis.

Field collected populations: The PCA yielded two principle components which accounted for similar proportions of the total variation (PC1 = 42·9% and PC2 = 35·6%; Table 1B). PC1 was not significantly correlated with MI, but PC2 was (Table 1B, Fig. 2d–f). The qualitative weighting of the dependent variables in PC2 was the same as in the analysis of the laboratory data. Increased MI was associated with a decrease in the number of offspring per litter but an increase in the minimum size of reproducing females (a surrogate estimate of the size at maturity), degree of superfetation, offspring size and ovary dry mass.

Comparisons with the northern clade

Laboratory-reared populations: The multivariate analysis (manova) that included age at maturity, mass at maturity, offspring size, offspring number, level of superfetation, interlitter interval and ovary dry mass as dependent variables showed that the relationship between these life-history traits and MI was significantly different among the clades (Clade x MI: Wilks' λ7,125 = 0·115, P < 0·0001). In the univariate analyses, for all life-history traits, analyses based on star-phylogenies yielded higher log-likelihoods compared to the analyses that included phylogenetic corrections (Table 2). The two clades were similar to each other in the relationship between MI and the number of offspring per litter, reproductive allocation and ovary dry mass. In both clades and in both the laboratory and field collections, increased MI was related to giving birth to fewer offspring per litter, smaller ovary dry masses and lower values for reproductive allocation (Table 2, Fig. 3).

Table 2. Log-likelihoods and t statistics from linear model with MI and clade as independent variables. Analyses are those with a star phylogeny (‘Star’) analysed with least-squares and those with phylogenic correction (PGLS). Log-likelihoods from full models (including interaction terms) were compared to determine whether the star phylogeny or the phylogenetic correction best fit the data. In all cases, log-likelihoods for star phylogeny provided a better fit. Non-significant interactions were removed in these models, but the log-likelihoods from the full model are shown for comparison with PGLS analyses. Degrees of freedom for laboratory studies are 6 and for wild-caught populations are 28. All variables were natural log–transformed prior to analysis. Values in bold are significant at the 0·05 level
TraitLab-StarLab-PGLSField-StarField-PGLS
Log-likelihoodCladeMIClade x MILog-likelihoodCladeMIClade x MILog-likelihoodCladeMIClade x MILog-likelihoodCladeMIClade x MI
  1. Dry weight of reproductive tissue = average dry weight of all eggs and embryos found in reproductive females; Reproductive Allocation = [dry mass of embryos/(dry mass of embryos + dry mass of female)], from field collections preserved in formalin or [dry mass of embryos/wet mass of alcohol-preserved females)*3·592, for females preserved in alcohol.; Estimated dry weight of offspring at birth = estimated dry weight at stage 45 based on parameters of regression model between stage of development and embryonic dry weight/litter size = litter size of the average-sized female (the average number of young per litter, estimated from applying the regression of litter size on female size to the average female length); Superfetation = average number of litters per female at the time of dissection; Maximum number of litters per female = maximum number of litters found in a female.

  2. a

    field data.

  3. b

    laboratory data.

Age at first parturition11 2·9 0·7 3·6 7·31·10·4 2·6
Smallest sizea/Size at maturityb1·8 5·9 1·4 2·3 −0·4 2·5 11·810·6 3·9 2·7 2·9 4·30·81·10·9
Average size of pregnant females18·2 6·0 2·4 2·6 13·41·2 2·1 1·7
Superfetation−0·11·71·5 2·8 −4·20·60·92−7·30·5 8·2 5·2 −15·80·1 2·9 1·7
Litter size5·21·6 5·1 0·60·7 3·9 1·5−26·90·7 2·1 −35·80·11·50·7
Offspring dry mass11·2 9·6 15·9 14·1 7·0 3·3 9·8 9·3 −17·4 3·4 1·2 3·2 −18·10·71·12
Ovary dry mass−0·1 3·3 7·1 −1·71·4 4·5 0·1−32 3·2 3·3 2·3 −39·80·6 2·1 1·3
Interlitter interval5·111·4 3·3 1·40·20·82·1
Ova dry mass10 7·5 17·2 13 5·8 2·6 10·3 8·3
Reproductive allocation−3·72·3 5·1 −80·8 3·1 0·8−19·8 2·6 3 −28·20·50·90·3
Figure 3.

Relationships between Matrotrophy Index and each life-history trait for the laboratory-reared populations. ▲/dashed line = Southern Clade; ●/solid line = Northern Clade.

There were significant interactions among clades for the relationship between MI and all other dependent variables. The interaction between MI and offspring size was the most dramatic. In the Northern Clade, matrotrophic species gave birth to smaller babies relative to lecithotrophic species, while in the Southern Clade, matrotrophic species instead gave birth to larger babies relative to more lecithotrophic species (Table 2, Fig. 3). The interactions for the remaining variables also revealed significant, but less marked differences among clades in how life histories change in association with the evolution of placentation. In the Northern Clade, increased matrotrophy was associated with a significant decline in the age and size at maturity. In the Southern Clade, increased MI was instead associated with a trend, not significant, towards larger size and later age at maturation (Table 2, Fig. 3, and Appendix S2, Supporting information). Likewise, the evolution of increased MI was associated with significantly shorter interlitter intervals in the Northern Clade, but with a non-significant trend towards longer intervals in the Southern Clade (Table 2, Fig. 3, and Appendix S2, Supporting information). Increased MI was related to higher degrees of superfetation in both clades. In the Northern Clade, there was a significant increase in superfetation in association with increased MI, while in the Southern Clade there was a non-significant positive correlation between MI and superfetation (Table 2, Fig. 3, and Appendix S2, Supporting information). Finally, both clades showed a negative relationship between MI and ova dry mass, but the relationship was steeper in the Northern Clade (Table 2, Fig. 3). The preponderance of significant interactions between MI and clade for so many dependent variables is a signature of the differences between the clades in the association between the evolution of MI and the evolution of the remainder of the life history.

Field collected populations: Patterns observed for the field collected populations were the same as those described earlier for the laboratory-reared populations. Again, analyses based on star-phylogenies yielded higher log-likelihoods compared to the analyses that included phylogenetic corrections (Table 2). Litter size decreased with increasing MI in both clades and did not differ between clades (Table 2, Fig. 4). Reproductive allocation also decreased significantly with increase in MI in both clades, but the Northern Clade had significantly higher RA for a given level of matrotrophy (Table 2, Fig. 4). There was a significant interaction between MI and clade for the remaining life-history traits. First, increase in matrotrophy was associated with smaller offspring in the Northern Clade but larger offspring in the Southern Clade (Table 2, Fig. 4). Secondly, increased matrotrophy was associated with a smaller minimum and mean size at maturation in the Northern Clade, whereas in the Southern Clade, increased matrotrophy was associated with a larger minimum and mean size at maturation (Table 2, Fig. 4). Thirdly, increasing matrotrophy was associated with a higher degree of superfetation in both clades, but the rate of increase was higher for the Northern Clade (Table 2, Fig. 4). Finally, ovary dry mass significantly decreased with increase in MI in both clades, but did so at a higher rate in the Northern Clade (Table 2, Fig. 4).

Figure 4.

Relationships between Matrotrophy Index and each life-history trait for the field populations. ▲/dashed line = Southern Clade; ●/solid line = Northern Clade.

Discussion

There was a significant association between the evolution of MI and the evolution of the remainder of the life history in the Southern Clade. This pattern was much more evident in the laboratory than the field data, we presume because in the laboratory we are able to control for environmental effects and hence reduce residual variance. We are also able to estimate a wider spectrum of life-history variables in the laboratory, including age and size at maturity and frequency of reproduction. These additional variables contributed to our ability to discriminate among species that differ in maternal provisioning. The virtue of reporting laboratory and field results is that they enable us to generalize our observations to all species in the clade and a larger number of populations. The correlation we observed between the evolution of the maternal provisioning and the remainder of the life history suggests a causal relationship, as postulated by the various life-history facilitation hypotheses. The catch is that the nature of these correlations is quite different from those in the Northern Clade.

How are the two clades different?

In the laboratory, the association between the degree of matrotrophy and a multivariate measure of the life history was different between the two clades. In one dependent variable (offspring size), the relationship between the life history and MI was the opposite in the two clades and significant in each of them. In four dependent variables (age at maturity, mass at maturity, superfetation and interlitter interval), the relationship between the dependent variable and MI was the opposite in the two clades, but the within clade relationship was significant in only one of the clades. In one dependent variable (ova dry mass), there was a significant interaction between the life history and MI, but the slopes among the clades had the same sign. Finally, three of the nine traits (litter size, reproductive allocation and ovary dry mass) showed no significant interaction between the life history and MI. The different associations between MI and age at maturity, size at maturity, offspring size and interlitter interval are enough to invalidate the life-history facilitation hypotheses for Poeciliopsis. Given such differences among such closely related clades, we feel safe in saying that these hypotheses have no general explanatory value and can be rejected. This also means that there is no simple bridge to be found between the evolution of matrotrophy and demographic life-history theory.

We note that the same divide was revealed in a comparison of the life histories of fish in the family Zenarcopteridae [fresh water half beaks; (Reznick, Meredith & Collette 2007)]. The two genera in this study, Dermogenys and Nomorhamphus, included species that varied in the presence and absence of matrotrophy and superfetation; there was at least one independent origin of matrotrophy in each of these genera. In Dermogenys, the evolution of increased matrotrophy was associated with the production of fewer, larger offspring per litter. In Nomorhamphus, the evolution of increased matrotrophy was instead correlated with the production of more, smaller offspring per litter.

It might be tempting to postulate that the placenta could be adaptive in a conditional or context-specific way, such as to enhance the evolution of offspring size in one context but facilitate the evolution of earlier maturity and an increased rate of offspring production early in life in another context. However, invoking such an alternative negates our goal of seeking a general explanation for the evolution of the placenta. More to the point, our prior experience in the study of life-history evolution in guppies (Reznick, Bryga & Endler 1990; Reznick & Bryga 1996; Reznick, Rodd & Cardenas 1996; Reznick et al. 1997) shows that all of these life-history traits can be substantially different among populations within a species and that populations have sufficient genetic variation to sustain rapid evolution of all of these traits, without any recourse to their evolution being driven by the correlated evolution of some other complex trait. Said differently, there is no need for anything like a facilitation hypothesis to explain the scope of life-history evolution that is displayed by this family of fishes.

How are these clades the same?

There was, however, one feature of the life history that evolved in a consistent fashion in concert with the evolution of increased matrotrophy. In both the Northern and Southern Clades, we found that the evolution of increased MI was correlated with the evolution of a smaller ovary dry mass and a lower value for reproductive allocation. Both of these variables characterize the mass of reproductive tissues relative to somatic tissues, and presumably the volume of developing young. The reason we see this common feature in both clades is that the evolution of increased post-fertilization maternal provisioning is attained primarily by reducing the size of the egg at fertilization, rather than increasing the size of the offspring at birth. When this change is combined with superfetation, it means that early stage embryos have a much smaller mass and volume than late stage embryos. Their combined volume, and relative dry mass, can be consistently smaller than in non-placental species, even if there is no reduction in the rate of offspring production. Reznick, Meredith & Collette (2007) obtained the same result for the two independent origins of matrotrophy in the two genera of freshwater halfbeaks. In spite of the differences between the genera in the way offspring size and offspring number change with increased MI, both lineages display a consistent and significant decline in reproductive allocation in correlation with the evolution of increased matrotrophy.

This similar trend in four out of four lineages suggests that there could be a common adaptive explanation for the evolution of matrotrophy in all four of them. Trexler & DeAngelis (2003), Thibault & Schultz (1978) and Miller (1975) proposed that placentation could be an adaptation that reduces the profile of pregnant females and hence reduces the cost of locomotion. It could thus serve as an adaptation to life in streams with high flow rates.

Plaut (2002) and Ghalambor, Reznick & Walker (2004) have shown that a locomotor cost of reproduction exists in related lecithotrophic species of Poeciliidae, Gambusia and Poecilia, respectively. Both of these species lack superfetation. In their case, the cost of locomotion increased as the single litter of young progressed through development because, even though they were declining in dry mass, they increased in wet mass and volume by a factor of three to four between when the egg was fertilized and when the embryo was fully developed and ready to be born. It was this increase in wet mass and volume that was associated with a decline in acceleration and maximum swimming speed. Unfortunately, there currently is no empirical data on how matrotrophic reproduction alleviates these locomotor costs. One possible catch in applying this logic for how the evolution of increased MI might affect ovary volume and locomotion is that our conclusion is based on dry mass, which implicitly assumes that it would carry over to wet mass and volume. If species with high matrotrophy for some reason have more moisture associated with reproductive tissues than do lecithotrophic species, then this hypothesis might not be viable.

Conflict versus adaptation?

Here and in earlier publications, we have addressed most of the proposals for the adaptive significance of placentation. All features of adaptive facilitation hypotheses, save the possible link to ovary volume and locomotion, have been shown to lack generality. We have also addressed Trexler & DeAngelis' (2003) general model for the evolution of matrotrophy and can at least show that the conditions that favour the evolution of matrotrophy in the context of their model are limited. We did so by testing their assumption that placental species have the capacity to fertilize a large number of small eggs with a minimum of resources and then adjust brood size to food availability by aborting some embryos. If they can do so, then the conditions that favour the evolution of matrotrophy are far easier to satisfy than if they cannot. We have shown in species that represent four independent origins of placentation that placental species cannot abort embryos in response to low food availability (Heterandria formosa - Reznick, Callahan & Llauredo 1996; Poeciliopsis prolifica - Banet & Reznick 2008; Poeciliopsis turneri - Banet, Au & Reznick 2010; Phalloptychus januarius - Pollux & Reznick 2011). In all four species, we found that females are unable to abort developing offspring. This result does not disprove the Trexler-DeAngelis hypothesis, but does narrow the range of circumstances in which the hypothesis could apply.

In these same experiments, we also found that the females of placental species responded to reduced food rations by producing smaller babies, rather than aborting some of them. In contrast, guppies, which are lecithotrophic, respond to a reduction in food availability by producing larger offspring (Reznick & Yang 1993); these offspring have a strong selective advantage to smaller offspring when food is scarce, but not when it is abundant (Bashey 2006) which suggests that the production of larger offspring in response to low food represents adaptive phenotypic plasticity. Prior research on a diversity of organisms has yielded similar results (reviewed by Reznick & Yang 1993). If this association between offspring size and resource availability is general, then the production of smaller young in response to reduced rations by all four placental species would be maladaptive.

An alternative proposal for the evolution of placentation is that it evolves as a by-product of intergenomic conflict (Haig 1993; Crespi & Semeniuk 2004). The argument is that the prior evolution of livebearing creates an enlarged forum for parent–offspring conflict, which is driven by the differences in what defines the fitness of parents versus offspring (Trivers 1974). A consequence of this difference is that the quantity of resources that is optimal for the embryo to get from its mother is greater than is in the best interest of the mother to give to the embryo. The placenta evolves as the battle front between mother and embryo, or as the locus where there is selection on a mother's ability to regulate the embryo's access to resources and on the embryo to acquire more resources from its mother. The possible failure of all adaptive hypotheses for the evolution of the placenta makes this alternative explanation more attractive to us, but what is required is positive evidence for conflict, not a failure of all adaptive alternatives. While there is abundant evidence for conflict in mammalian placental reproduction (e.g. Haig 1993), we have only begun to generate evidence that addresses the conflict hypothesis in the Poeciliidae (e.g. O'Neill et al. 2007; Schrader & Travis 2008, 2009).

Acknowledgements

We wish to thank Yuridia Reynoso, who performed or oversaw all of the dissections used to characterize the life histories. We would also like to thank Alex Mamaril and Samantha Natividad for their help in the care and maintenance of the laboratory fish populations. Doug Nelson, from the University of Michigan Museum of Zoology, Lynn Parenti from the U. S. National Museum, John Lundberg from the Academy of Natural Sciences in Philadelphia and Bob Vrijenhoek generously gave us access to their collections of wild-caught fishes for use in dissection and life-history characterization. Mariana Mateos arranged for permits to work and collect in Mexico. This work was supported by a grant from the US National Science Foundation (DEB-0416085).

Ancillary