Flexible oviposition behavior enabled the evolution of terrestrial reproduction

Significance

How did animals evolve terrestrial reproduction? We demonstrate that behavioral flexibility of individuals to lay eggs both in water and on land, coupled with eggs capable of developing in either environment, likely facilitated the evolution of obligate terrestrial reproduction in a lineage of Neotropical treefrogs. While aquatic reproduction is ancestral in the group, reproductive flexibility is common and is associated with the evolution of larger eggs capable of greater resistance to desiccation in air. This research has implications for evolutionary transitions between other such supposedly discrete states and implies that behavioral flexibility is important in facilitating such evolutionary shifts.

Abstract

Among vertebrates, nearly all oviparous animals are considered to have either obligate aquatic or terrestrial oviposition, with eggs that are specialized for developing in those environments. The terrestrial environment has considerably more oxygen but is dry and thus presents both opportunities and challenges for developing embryos, particularly those adapted for aquatic development. Here, we present evidence from field experiments examining egg-laying behavior, egg size, and egg jelly function of 13 species of Central and South American treefrogs in the genus Dendropsophus, which demonstrates that flexible oviposition (individuals laying eggs both in and out of water) and eggs capable of both aquatic and terrestrial development are the likely factors which enable the transition from aquatic to terrestrial reproduction. Nearly half of the species we studied had previously undescribed degrees of flexible oviposition. Species with obligate terrestrial reproduction have larger eggs than species with aquatic reproduction, and species with flexible reproduction have eggs of intermediate sizes. Obligate terrestrial breeding frogs also have egg masses that absorb water more quickly than those with flexible oviposition. We also examined eight populations of a single species, Dendropsophus ebraccatus, and document substantial intraspecific variation in terrestrial oviposition; populations in rainy, stable climates lay fewer eggs in water than those in drier areas. However, no differences in egg size were found, supporting the idea that the behavioral component of oviposition evolves before other adaptations associated with obligate terrestrial reproduction. Collectively, these data demonstrate the key role that behavior can have in facilitating major evolutionary transitions.

One goal of biology has been to order the natural world into discrete categories. These categories are designed to reflect natural variation but are often subject to debate (1) since any attempt to place biological diversity in groups is subject to potentially artificial delineations. Indeed, many traits that we think of as occupying alternative discrete states may in fact be more continuous than currently appreciated. For example, C₃ vs CAM metabolism in plants has long been considered a dichotomous trait but recent analyses have revealed some continuity between states, with certain species capable of engaging in both forms of CO₂ fixation (2, 3).

Another such seemingly dichotomous trait in annelids, arthropods, mollusks, and vertebrates is between breeding in water versus breeding on land. Because of the physical differences between water and land (4), oviparous animals are generally thought to have embryos adapted to survive and develop in just one of these habitats. Aquatic eggs should, in theory, be capable of surviving on land as long as they can avoid desiccation (5). On the other hand, terrestrial eggs that are adapted to the high oxygen environment of air may have difficulty surviving in water (6).

Amphibians provide an excellent group for studying the evolution of reproducing on land. Although aquatic reproduction is ancestral in amphibians, reproducing on land has evolved dozens of times, including to fully terrestrial egg-laying species that lack a larval stage (i.e., direct development) or even live birth (7–11). In frogs specifically, most species actually lay eggs out of water, approximately half with eggs that hatch into an aquatic tadpole stage (i.e., semiterrestrial reproduction) and half with direct development (11). Recent analyses suggest that the most common evolutionary progression is from aquatic egg-laying to semiterrestrial egg-laying to direct development (11). Although we know a considerable amount about the different reproductive modes of amphibians, we know little about the processes that facilitate evolutionary shifts between such modes (11). One possible evolutionary route from aquatic to semiterrestrial oviposition is flexible reproduction, wherein an individual animal can lay eggs both in water and on land and have those eggs successfully develop in either location. Such a reproductive strategy has previously been documented in the Neotropical treefrog Dendropsophus ebraccatus, which can lay eggs in water or on land and will do so in response to the shade above a pond and the presence of aquatic predators (12, 13). An ability to adjust the allocation of eggs laid aquatically or terrestrially in response to environmental variation is termed reproductive mode plasticity.

It seems plausible that the transitions from water to land in amphibians may have been similar to those that occurred in the early amniotes, making amphibians an important group for testing hypotheses about the shift to terrestrial from aquatic reproduction. All hypotheses thus far about the evolution of terrestrial life in tetrapods have failed to consider that flexible reproduction may have played a role, instead positing that embryos somehow went from aquatic to terrestrial development in one discrete step (8, 9, 14). For example, when describing the evolutionary shift of vertebrates to land, Romer (9, 15) proposed that it was “probable that the egg came ashore before the adult was fully ready for land life” but said nothing more about the role of reproduction in the transition to land.

We test the hypothesis that flexibility in reproduction has aided the transition to terrestrial reproduction using frogs of the genus Dendropsophus, which contains approximately 100 species of small South and Central American treefrogs (16). The genus mirrors the pattern seen in tetrapods more generally; aquatic reproduction is the ancestral state but terrestrial oviposition has evolved repeatedly (12). In all species, eggs develop and hatch into tadpoles that develop in water (16). Looking across species of Dendropsophus and across populations of one of its species D. ebraccatus (SI Appendix, Fig. S1), we tested for reproductive mode plasticity and asked the following questions: 1) How widespread is flexible reproduction? 2) What role might flexible reproduction have had on the transition to obligate terrestrial reproduction? 3) What adaptations permit eggs of species with obligate terrestrial reproduction to survive on land? 4) Are there environmental variables associated with variation in the degree of terrestrial reproduction?

Results

Behavioral Assays of Reproductive Mode.

We conducted behavioral assays for reproductive mode plasticity in 13 species of Dendropsophus (Fig. 1, Table 1, and SI Appendix, Fig. S1). Our initial focus was on species in the clade most closely related to D. ebraccatus (the D. leucophyllatus clade), but we opportunistically tested any other species of Dendropsophus that we encountered in the field. We collected 1 to 25 pairs of frogs per species and allowed them to breed overnight in shaded or unshaded experimental habitats, previously shown to influence aquatic versus terrestrial egg allocation (12), allowing the assessment of their oviposition site choices in these two environments.

Fig. 1.

Results of behavioral assays for reproduction and phylogenies of 13 species of Dendropsophus treefrogs and one outgroup species showing ancestral state estimation of reproductive mode. (A) The results of behavioral assays for reproduction in shaded (dark) or unshaded (light) breeding enclosures. Boxplots show the proportion of eggs laid out of water in assays of each species and points represent individual females. The box-and-whisker plot shows the median (thick horizontal line), interquartile range (top and bottom of the box), and the most extreme values (whiskers). (B) Bayesian phylogeny. Numbers at nodes represent posterior probabilities in the consensus tree. (C) A maximum likelihood phylogeny. Numbers at nodes indicate bootstrap support generated from 1,000 replicates. In both trees, the color of the branches in the phylogeny indicates estimated ancestral reproductive mode, from obligate aquatic reproduction shown in blue to obligate terrestrial reproduction shown in green, and degrees of flexible reproduction in shades of orange. Mode of reproduction, shown by branch colors, was estimated by treating reproduction as a continuous variable ranging from fully aquatic (0) to fully terrestrial (1). Pie diagrams at nodes indicate the estimated probability of the ancestral state being either aquatic (blue), flexible (orange), or terrestrial (green) reproduction when considering reproduction of extant taxa as a categorical variable.

Table 1.

Reproduction of 13 species of Dendropsophus treefrogs measured in shaded and unshaded breeding enclosures

Species	Treatment	N	Terrestrial eggs laid	Aquatic eggs laid	Proportion flexible	χ2	P
D. bifurcus	Shaded	10	143 ± 15	54.8 ± 14.3	0.80	4.91	0.027
D. bifurcus	Unshaded	10	101 ± 14.4	107 ± 18	1.0
D. bokermanni	Unshaded	1	66	0	0
D. ebraccatus	Shaded	8	207 ± 25.9	56.8 ± 19.2	0.75	3.97	0.046
D. ebraccatus	Unshaded	8	176 ± 30.1	139 ± 26.3	1
D. frosti	Shaded	1	70	0	0
D. leucophyllatus	Shaded	2	102 ± 52	192 ± 7.5	1
D. leucophyllatus	Unshaded	2	93.5 ± 28.5	214 ± 72.5	1
D. microcephalus	Shaded	3	0	205 ± 19.7	0
D. microcephalus	Unshaded	3	0	206 ± 47.2	0
D. minutus	Shaded	4	12.5 ± 12.5	195 ± 12.5	0.25	0.14	0.712
D. minutus	Unshaded	4	0 ± 0	127 ± 32.5	0
D. parviceps	Shaded	1	0	302	0
D. parviceps	Unshaded	1	0	323	0
D. rhodopeplus	Shaded	3	0	256 ± 58.2	0
D. rhodopeplus	Unshaded	2	0	369 ± 103	0
D. riveroi	Shaded	2	0	146 ± 2.5	0
D. riveroi	Unshaded	2	0	96 ± 7	0
D. sarayacuensis	Shaded	11	85.7 ± 6.48	0	0
D. sarayacuensis	Unshaded	11	92.5 ± 8.38	0	0
D. timbeba	Shaded	3	61.3 ± 30.7	238 ± 46.3	0.33	3.01	0.083
D. timbeba	Unshaded	3	0	262 ± 30	0
D. triangulum	Shaded	10	21.7 ± 10.1	324 ± 27.8	0.50	0.68	0.408
D. triangulum	Unshaded	13	7.0 ± 4.2	259 ± 21.0	0.38

Shown are the number of pairs of tested, the average (± SEM) number of eggs laid terrestrially or aquatically, and the proportion of females with flexible reproduction. For species with at least a sample of size of three pairs tested in each type of enclosure and which did not have obligate aquatic or terrestrial oviposition we conducted a generalized linear mixed model to test for differences in aquatic vs terrestrial egg allocation in shaded and unshaded enclosures, and the χ² statistic and associated P-value are shown.

Females generally allocated eggs into multiple discrete egg masses, laying 3.3 ± 0.1 masses (mean ± SEM), with some masses fully terrestrial, others fully aquatic, and some containing eggs that spanned the air–water interface. One species (D. sarayacuensis) showed obligate terrestrial reproduction and two more appear likely to have obligate terrestrial reproduction (D. bokermanni and D. frosti), although only one pair of each of these species was tested due to their rarity (Fig. 1A and Table 1). Four species showed obligate aquatic reproduction (D. parviceps, D. riveroi, D. rhodopeplus, and D. microcephalus), although the numbers of pairs tested for some species are limited. The remaining six species (D. timbeba, D. minutus, D. ebraccatus, D. triangulum, D. leucophyllatus, and D. bifurcus) showed some degree of flexibility, with the same female laying both aquatic and terrestrial eggs in a single bout of reproduction (Fig. 1A and Table 1). At one extreme were two species (D. minutus and D. timbeba) that both laid 100% of eggs in the water in unshaded cages, but laid a small proportion of eggs out of water in shaded cages where eggs would be more protected from desiccation (Fig. 1A and Table 1). Dendropsophus triangulum laid >90% of its eggs in water in both shaded and unshaded cages (Fig. 1A and Table 1). Dendropsophus leucophyllatus, considered to have obligate terrestrial oviposition [e.g., (16)], laid approximately 1/3 of its eggs terrestrially in both conditions (Fig. 1A and Table 1). At the other end of the aquatic-terrestrial oviposition spectrum were two species with mostly terrestrial oviposition and reproductive mode plasticity. These species, D. bifurcus and D. ebraccatus, laid approximately 16 and 22 % more eggs out of the water in shaded cages, respectively (Fig. 1A and Table 1). In all cases, as many or more eggs were laid in the water in unshaded cages compared to shaded cages.

To estimate the relationships between species of Dendropsophus, we constructed Bayesian and maximum likelihood phylogenies of the 13 species we studied. Our Bayesian tree was well resolved at all nodes (Fig. 1B). The maximum likelihood tree had a nearly identical topology with one exception (the placement of D. timbeba as sister to the rest of Dendropsophus, rather than part of a clade of D. parviceps, D. frosti, and D. bokermanni) and less bootstrap support (i.e., <85%) at two nodes (Fig. 1C). Given that nearly half of the extant species we tested showed some degree of flexibility in aquatic vs terrestrial egg deposition, we chose to estimate the ancestral state of reproductive mode by treating reproductive mode as a continuous trait. Ancestral character estimation clearly suggested that aquatic reproduction is ancestral in the clade and that flexible reproduction was widespread in the evolutionary history of the genus. This view of ancestral states did not change when viewing reproduction as having three discrete states (i.e., aquatic, flexible, and terrestrial). Nearly 40% of nodes in both the Bayesian and ML phylogenies (5 out of 13) are considered most likely to have had flexible reproduction when based on discrete categories (Fig. 1).

The strongest data for the role of flexible reproduction in evolution come from the clade containing the obligate terrestrial breeding species D. sarayacuensis. The ancestral state reconstruction strongly suggests that an ability to lay both aquatic and terrestrial eggs preceded fully terrestrial reproduction (Fig. 1). The nature of the second transition to terrestrial reproduction in D. bokermanni and D. frosti is murkier, as these species and their close relatives are quite rare, forcing small sample sizes. Our behavioral assays do not indicate a polymorphism, with some individuals laying terrestrial eggs and others laying aquatic eggs, but rather that most individuals flexibly allocate their eggs above or below the water during a single bout of reproduction (Fig. 1A and Table 1). Interestingly, we also found that one species (D. triangulum) appears to be reverting to primarily aquatic oviposition from an ancestor with more flexible oviposition.

In addition to looking across species, we also assayed eight populations of D. ebraccatus for flexible reproduction. Six populations laid substantial numbers of eggs in both air and water (Fig. 2A and Table 2). Two populations (Durango, Ecuador, and Piro, Costa Rica) had essentially obligate terrestrial reproduction, with each population laying 98% of eggs out of water in both shaded and unshaded habitats, whereas all other populations had flexible reproduction to some degree (Fig. 2A and Table 2). Frogs in two populations (Gamboa and Rambala, Panama) laid significantly more eggs out of the water in shaded than unshaded enclosures (Fig. 2A and Table 2), indicating plasticity in reproduction, and most of the remaining populations showed a similar pattern, even when there was not a statistical difference in egg deposition between enclosures. The one exception to this was the northernmost population (el Ocote, Mexico) where frogs laid somewhat more eggs in water in shade, but this difference was not significant (Fig. 2A and Table 2).

Fig. 2.

Results of behavioral assays for reproduction and population structure and variation in reproduction of Dendropsophus ebraccatus in Central America and Ecuador. (A) The results of behavioral assays for reproduction in shaded (dark) or unshaded (light) breeding enclosures. Box-and-whisker plots show the proportion of eggs laid out of water in assays of each population and points represent individual females. The box-and-whisker plot shows the median (thick horizontal line), interquartile range (top and bottom of the box), and the most extreme values (whiskers). (B) Median joining haplotype network of eight populations of D. ebraccatus based on 1,442 base pairs of mitochondrial DNA (ND1 and 16S). Each colored circle shows a unique haplotype whereas small black circles represent hypothetical, unsampled haplotypes. Different colors indicate localities, and the diameter of the circle represents the number of individuals with that haplotype. Tick marks or numbers represent the number of substitutions between haplotypes. (C) Estimated population genetic clusters as generated in STRUCTURE from an average of 59,198 variable SNP loci per site obtained through RAD-seq. Each bar represents a different individual and the colors of each bar represent the estimated probability of belonging to one of four genetic clusters. Localities included in each cluster are listed at the top.

Table 2.

Reproduction of Dendropsophus ebraccatus treefrogs at eight populations measured in shaded and unshaded breeding enclosures

Population	Treatment	N	Terrestrial eggs laid	Aquatic eggs laid	Proportion flexible	χ2	P
Bartola	Shaded	10	254 ± 12.1	33 ± 10.2	0.70	0.40	0.53
Bartola	Unshaded	10	244 ± 27.7	53 ± 21.7	0.50
Cockscomb	Shaded	13	193 ± 12.5	54 ± 13.4	0.85	0.38	0.54
Cockscomb	Unshaded	12	167 ± 25.7	70 ± 24.9	0.58
CRARC	Shaded	13	198 ± 10.5	55 ± 9.7	0.92	2.17	0.14
CRARC	Unshaded	14	158 ± 20.8	96 ± 23.3	1
Durango	Shaded	6	84 ± 13.3	0	0	0.58	0.44
Durango	Unshaded	7	98 ± 17.5	5 ± 4.7	0.29
Gamboa	Shaded	8	207 ± 25.9	57 ± 19.2	0.75	3.97	0.046
Gamboa	Unshaded	8	176 ± 30.1	139 ± 26.3	1
el Ocote	Shaded	8	167 ± 17.9	166 ± 19.1	1	2.24	0.13
el Ocote	Unshaded	9	212 ± 31.5	148 ± 36.5	0.78
Piro	Shaded	14	246 ± 10.2	5 ± 2.5	0.43	0.01	0.92
Piro	Unshaded	12	225 ± 14.7	7 ± 3.4	0.33
Rambala	Shaded	10	185 ± 24.2	71 ± 28.2	0.70	5.88	0.015
Rambala	Unshaded	10	156 ± 21.0	142 ± 19.8	1

Shown are the number of pairs tested in shaded and unshaded breeding enclosures, their average (± SEM) number of eggs laid terrestrially or aquatically, and the proportion of females with flexible reproduction. For all populations, we conducted a generalized linear mixed model to test for differences in aquatic vs terrestrial egg allocation in shaded and unshaded enclosures, and the χ² statistic and associated P-value are shown.

To understand the relationships between populations of D. ebraccatus, we constructed a haplotype network based on 1,442 bp of mitochondrial DNA. There is a deep level of divergence between D. ebraccatus in South America (Durango, Ecuador) and populations in Central America, with over 200 substitutions between them (Fig. 2B). The haplotype network implies a complex evolutionary history of dispersal by D. ebraccatus in Central America. The Ecuadorian population is more closely related to those on the Caribbean coast of Costa Rica and Nicaragua than to the most geographically close populations in Panama (Gamboa and Rambala), which are most closely related to a population on the Pacific side of Costa Rica (Piro). No population shared haplotypes with other populations, although some haplotypes were more closely related to those in neighboring populations than to haplotypes in their own population (Fig. 2B).

For a more fine-scale view of population differentiation among populations of D. ebraccatus, we also used RAD-seq to identify an average of 51,918 polymorphic SNP loci per population (SI Appendix, Table S1). Using the program STRUCTURE (17), we identified four main genetic clusters: 1) Mexico and Belize, 2) the two populations from Panama, 3) the Caribbean side of Costa Rica and 4) the Pacific side of Costa Rica (Fig. 2C). Population genetic estimates (Ф_ST and F_ST) showed substantial divergence even among geographically close populations within a single genetic cluster (Table 3). F_ST estimates based on the mitochondrial sequence data were similar (Table 4). The two most northern populations had noticeably fewer variable loci than the four southern populations from Costa Rica and Panama (SI Appendix, Table S1). Thus, when considering both the sequence and RAD-seq data together, it appears that the two populations which have evolved terrestrial reproduction (Durango, Ecuador, and Piro, Costa Rica) lack gene flow from other populations in our study, which maintain greater degrees of aquatic egg-laying.

Table 3.

Mean Ф_ST (upper diagonal) and F_ST (lower diagonal) between six populations of Dendropsophus ebraccatus from across Central America, based on an average of 59,198 SNPs per population identified via RAD-seq (N = 10 individuals per population)

	el Ocote	Cockscomb	Piro	CRARC	Rambala	Gamboa
el Ocote		0.323	0.765	0.688	0.733	0.694
Cockscomb	0.240		0.799	0.727	0.760	0.752
Piro	0.424	0.467		0.356	0.645	0.636
CRARC	0.336	0.371	0.216		0.532	0.509
Rambala	0.430	0.469	0.319	0.240		0.440
Gamboa	0.396	0.436	0.298	0.227	0.196

Table 4.

Mean F_ST (lower diagonal) between eight populations of Dendropsophus ebraccatus from across Central America and Ecuador calculated from a 1,442 bp section of mitochondrial DNA (N = 8 to 30 individuals per population)

	Durango	Gamboa	Piro	Cockscomb	CRARC	Bartola	Rambala
Gamboa	0.274
Piro	0.344	0.094
Cockscomb	0.361	0.105	0.158
CRARC	0.314	0.075	0.125	0.136
Bartola	0.310	0.067	0.117	0.129	0.097
Rambala	0.282	0.049	0.098	0.109	0.078	0.071
el Ocote	0.358	0.098	0.152	0.164	0.130	0.122	0.102

The Relationship Between Egg Diameter and Reproductive Mode.

It has long been assumed that frogs which breed on land generally have larger eggs than those which lay eggs in water (10, 18). Using a database of frog egg diameters from Summers et al. (19), we confirm this assumption (Fig. 3A; F_1,331 = 199.2, P < 0.00001). Terrestrial eggs were on average over twice the diameter of aquatic eggs (3.41 mm vs 1.66 mm) which translates to an 8.7-fold difference in volume (20.76 mm³ vs 2.39 mm³), assuming eggs to be spherical. However, the increase in surface area is only 4.2-fold between terrestrial and aquatic eggs (36.5 mm² vs 8.7 mm²), thus making the surface area to volume ratio for terrestrial eggs roughly half that of aquatic eggs (1.8 vs 3.6). As these differences evolved long ago, we evaluated variation in egg size across the evolutionary transition from aquatic to terrestrial egg-laying in Dendropsophus. We obtained egg diameter data for 10 species of Dendropsophus, and found a sharp and significant transition between species with aquatic reproduction and small eggs to those with terrestrial reproduction and larger eggs, with species having flexible reproduction falling in the middle (Fig. 3B; phylogenetic binomial GLMM: P = 0.007). A similar pattern was not seen among populations of D. ebraccatus; there was no relationship between egg size and the degree of terrestrial reproduction (Fig. 3C; binomial GLMM: χ² = 0.28, P = 0.60).

Fig. 3.

Variation in egg diameter across frogs in general, species of Dendropsophus and populations of D. ebraccatus. (A) A box-and-whisker plot showing egg diameter of frog species with obligate aquatic or terrestrial oviposition. The box-and-whisker plot shows the median (thick horizontal line), interquartile range (top and bottom of the colored box), and either the most extreme values (whiskers without points) or 1.5 times the interquartile range and outliers (ends of the whiskers followed by points). (B) The relationship between egg diameter and the proportion of eggs laid terrestrially within the genus Dendropsophus. Large colored circles represent the average egg diameter and degree of terrestrial reproduction for each species, whereas small colored circles represent oviposition for individual clutches laid in shaded and unshaded behavioral assays. (C) Egg size variation across eight populations of D. ebraccatus in relation to mode of reproduction. Large colored circles represent the average egg diameter and degree of terrestriality for each population of D. ebraccatus, whereas small colored circles represent oviposition for individual clutches laid in shaded and unshaded behavioral assays. Different colors in (B) and (C) represent different species or populations, respectively. The values for egg diameter of individual clutches in (B) and (C) represent the average of five eggs removed from each clutch laid at that site or by that species. Curves were calculated from generalized linear mixed effects models using a binomial error distribution. The gray shaded region is a 95% confidence interval.

Rates of Water Gain in Dendropsophus and D. ebraccatus Eggs Masses.

Terrestrial egg masses need to absorb and retain water from the environment to keep developing embryos suitably hydrated (20), as desiccation can be a major source of mortality (21). The jelly that surrounds frog eggs is a matrix of fibrous glycoproteins which can take up water and slow its loss to evaporation. In addition to its function in preventing desiccation, the jelly of terrestrial egg masses provides structural support to the embryos and can help dissuade predators (20). To assess how well the eggs of Dendropsophus treefrogs are adapted for developing out of water, we created groups of 10 eggs with their surrounding intact egg jelly, submerged them in water and weighed them over time. This provided an estimate of both the rate that the egg mass could absorb water and the maximum water-holding capacity of the egg jelly. We were able to obtain hydration data for four species of Dendropsophus. The mass of eggs was significantly related to treefrog species, duration of time submerged, and the interaction between species and time (Fig. 4A; LMM: species, χ² =54.0, P < 0.000001; time, χ² = 71.1, P < 0.000001; species X time, χ² =19.9, P = 0.0002). Egg masses of D. sarayacuensis acquired water more quickly and were capable of holding more water than egg masses of the other species. Dendropsophus leucophyllatus eggs held the next most water, followed by D. bifurcus and D. ebraccatus, which did not differ from one another (post hoc comparisons: D.s. vs all – P ≤ 0.0001, D.l. vs all – P ≤ 0.002, D.e vs D.b. – P = 0.98). We similarly obtained water absorption data for eight populations of D. ebraccatus, and clutch mass once again differed by population, time submerged, and their interaction (Fig. 4B; LMM: time, χ² = 279.0, P < 0.000001; site, χ² = 136.3, P < 0.000001; time X species, χ² = 46.7, P < 0.000001). In particular, egg masses in the Durango, Ecuador population were capable of absorbing water more rapidly than those of all other populations (post hoc comparisons: Durango vs all others, P ≤ 0.0001). Eggs from the remaining populations fell roughly into two groups: the Gamboa, Rambala, and CRARC populations differed significantly from Piro, Bartola, and Cockscomb (all P ≤ 0.02). Populations within each group were not significantly different (all P ≥ 0.11). Eggs from el Ocote were not different from any other population except Durango (all P ≥ 0.07).

Fig. 4.

Maximum rates of water absorption by frog egg masses after submergence in water. (A) Absorption rates for four species of Dendropsophus. (B) Absorption rates for eight populations of D. ebraccatus. Mass gain via absorption of water was measured by submerging terrestrially laid egg masses in water and recording the increase in mass over time. Curves were calculated from linear mixed effects models; the shaded region is the 95% confidence interval.

Climatic Variation and Reproductive Mode.

To determine what aspects of climate might be related to terrestrial reproduction, we analyzed the degree of terrestrial reproduction in different populations of D. ebraccatus in relation to latitude and eight bioclimatic variables from the WorldClim 2.0 database (22). Increased terrestrial egg-laying was most significantly associated with decreasing latitude, greater precipitation during the warmest quarter of the year, and high isothermality, a measure of climatic stability across the year (Fig. 5; binomial GLMM: latitude, χ² = 7.2, P = 0.007, precipitation, χ² = 6.8, P = 0.0009, isothermality, χ² = 6.1, P = 0.014). Both precipitation and isothermality were inversely correlated with latitude, to different degrees (Pearson’s correlation: precipitation, R = −0.81, P = 0.015, isothermality, R = −0.97, P < 0.00001).

Fig. 5.

Degree of terrestrial reproduction in Dendropsophus ebraccatus in relation to latitude and climate. Data for eight populations are plotted against (A) latitude, (B) precipitation during the wettest quarter of the year, and (C) isothermality (the ratio of the mean daily temperature fluctuation to the annual temperature range). Curves were calculated from generalized linear mixed effects models using a binomial error distribution; the gray shaded region is the 95% confidence interval.

Discussion

The colonization of land from water was a fundamental transition that first occurred in plants, then arthropods, mollusks, and annelids, and most recently in vertebrates (8). Here, we provide empirical evidence that flexible reproductive behavior may help animals evolve from reproducing in water to reproducing on land. We quantified three aspects of amphibian reproduction that are essential to terrestrial reproduction: 1) the oviposition site choices of adults, 2) the size of eggs, and 3) the functional ability of the egg jelly to absorb water. Collectively, these data indicate that the capacity to lay both aquatic and terrestrial eggs has helped some lineages of Dendropsophus treefrogs transition from ancestral obligate aquatic egg-laying to obligate terrestrial egg-laying, and that this behavioral capacity likely evolved before increases in egg diameter and enhanced egg jelly function. Looking across species, we find that species with terrestrial oviposition have eggs that are larger and that their jelly absorbs water more rapidly and can hold more water than the jelly of aquatically laid eggs. However, our work with the widespread species D. ebraccatus demonstrates that changes to the egg mass and associated jelly evolve after the behavioral shift to terrestrial oviposition. We document two genetically distinct populations of D. ebraccatus with apparently obligate terrestrial oviposition. These populations, however, do not have larger eggs than populations that still lay some eggs in water. One of these populations (Durango) did have egg jelly that can hold more water, suggesting that the egg jelly may be more mutable than egg size.

While larger eggs are obviously not necessary for developing on land, looking across frogs in general suggests that they are likely advantageous. Larger eggs provide offspring with more energy to develop more complex structures (14) but are presumably more energetically costly to produce and may not be able to obtain enough oxygen in water to survive (6). Thus, even if larger eggs are advantageous, species that lay eggs in water are constrained to have relatively small eggs because of oxygen limitations. Small eggs have a higher surface area-to-volume ratio, enabling greater absorption of oxygen per unit of mass and therefore allowing development in a more oxygen-limited environment. As long as those eggs have not evolved to be too large for aquatic development, they may still be oviposited in water and are capable of surviving. Red-eyed treefrogs (Agalychnis callidryas), for example, have substantially larger eggs than Dendropsophus, which take longer to develop and will die if submerged before hatching competence (6). Even for a species like D. ebraccatus that will actively choose aquatic oviposition in unshaded conditions, development can be delayed and mortality increased in water (21).

For anamniotic eggs, such as those of fish and amphibians, the biggest risk associated with developing on land is desiccation (5, 23), whereas the primary benefit appears to be a reduction in predation risk (21, 24, 25). The jelly surrounding amphibian eggs is a matrix of fibrous glycoproteins, hypothesized to evolve by adding or subtracting layers, resulting in an enhanced or reduced ability to retain water and greater structural rigidity to prevent collapse of the egg mass in air (20). However, thicker egg jelly also reduces oxygen permeability (20, 25) meaning that jelly that improves survival on land may directly impair embryo survival in water.

In addition to physical changes to the egg jelly, many frog species have evolved behaviors to reduce the risk of desiccation. For example, multiple lineages of frogs in South America, Africa, and India roll a leaf around their terrestrial eggs to protect them from predation and desiccation (26–28). Other species, such as Neotropical glass frogs and southeast Asian shrub frogs, may physically sit on terrestrial eggs in order to prevent water loss (29, 30). For eggs adapted for aquatic development, such risk of desiccation is particularly acute. Thus, one other way to reduce mortality from desiccation on land is to lay only part of the clutch out of the water. This could be done as a bet-hedging strategy, wherein animals split their eggs between water and land in accordance with general levels of risk in each environment or via plasticity based on the risk that ovipositing frogs assess at the moment (12, 13). Our findings show that these behaviors can go in both directions: Some species, such as D. timbeba or D. minutus, have principally aquatic reproduction but lay a small number of eggs out of the water in shaded environments, whereas others like D. ebraccatus and D. bifurcus lay the majority of their eggs out of the water but allocate more to aquatic oviposition sites in unshaded habitats.

Some of the species and populations we tested, such as D. bifurcus or D. ebraccatus, demonstrate reproductive mode plasticity (i.e., the capacity to flexibly alter the allocation of aquatic versus terrestrial eggs in different environmental conditions) whereas others appear to be relatively inflexible across environments but still have an ability to lay eggs both in water and on land (e.g., D. leucophyllatus). It is worth pointing out that the strength of response in D. ebraccatus to shade cloth was not as strong as has been previously found in truly forested sites (12), indicating that the tests presented here likely represent a conservative estimate of each populations’ or species’ capacity for plasticity. Furthermore, decisions about aquatic and terrestrial oviposition are likely influenced by more than just shade, such as the presence of egg predators in the water (13). Since females were only able to be tested once in a single environmental condition, it is unknown how consistent oviposition choices are across environments. Both variation in the behavior of the same individual and variation in the behavior of different individuals would account for the differences observed here, and future research should explore repeatability of oviposition decisions.

Our mitochondrial and genome-wide SNP data show that the two populations of D. ebraccatus with essentially obligate terrestrial oviposition were also genetically distinct from all other populations in our study (Fig. 2). Neither population had larger eggs, but one population (Durango) has also evolved a functionally enhanced egg jelly, similar to the obligate terrestrially breeding D. sarayacuensis, and this population is the most genetically distinct of those studied here. Although we found a sharp increase in egg diameter across Dendropsophus species that span the transition from aquatic to terrestrial reproduction, we found no such variation within D. ebraccatus. Thus, the behavioral aspect of reproduction has evolved in some populations of D. ebraccatus, but egg size has not. This could mean that there has not been sufficient time for selection to act on egg size, or it could be the result of shifting environmental conditions over time.

Our analysis of several bioclimatic variables indicated that, among populations of D. ebraccatus, increasingly wet and stable climates near the equator are associated with an increase in terrestrial oviposition. Greater rainfall is associated with more terrestrial egg-laying in amphibians at a global scale (7) but it is not a universal relationship. For example, a recent examination of African toads has found that terrestrial reproduction is instead associated with steep terrain and a lack of standing water bodies (31). Dendropsophus is a clade of South American origin (32), and we find that as D. ebraccatus has expanded north into generally drier areas, it has actually reverted to more aquatic reproduction (Fig. 5). That said, rainfall alone does not explain the evolution of reproduction. For example, the CRARC population in our study receives more rainfall than the population at Piro, yet frogs at CRARC laid considerably more eggs in the water. This may be due to gene flow from populations that receive less rainfall, which would reduce the effect of selection for terrestrial egg-laying. In this case, the population at Bartola, Nicaragua experiences a much drier climate and is therefore likely under selection for more aquatic oviposition but is genetically closely related to the CRARC population we examined (Fig. 2B).

Overall, the variation present in Dendropsophus is reflective of vertebrates in general: Aquatic reproduction is ancestral and terrestrial reproduction has evolved repeatedly (7, 10, 11). The important difference is that this variation occurs in a relatively small clade (~100 sp. of Dendropsophus) of closely related species (16). Our data allow us to present a hypothesis for the progression of events leading to the evolution of obligate terrestrial reproduction. The necessary environmental constraint is that terrestrial reproduction can only evolve in moist or humid environments which reduce mortality from desiccation. 1) Species begin by laying a small proportion of their eggs out of the water likely in habitats that reduce desiccation risk, such as in D. minutus or D. timbeba. 2) Over evolutionary time, species increase the relative allocation of eggs out of the water until terrestrial oviposition is the predominant mode, as in species like D. ebraccatus. 3) Once species have evolved to lay more eggs on land than in water, selection for thicker egg jelly and increased egg diameter occurs. 4) If selection for terrestrial reproduction is strong enough, eggs evolve to be too large to develop in water, thereby locking species into obligate terrestrial oviposition. The role of behavioral flexibility in such evolutionary transitions may prove to be a common phenomenon in other such evolutionary transitions between seemingly bimodal, discrete states.

Materials and Methods

Field Assays of Reproductive Behavior.

Between December 2011 and December 2013, we conducted assays of oviposition site choice in 13 species of Dendropsophus across 12 sites in Central and South America, including eight populations of D. ebraccatus (SI Appendix, Fig. S1). At each site, we built eight 1 m diameter × 2 m tall round mesh breeding enclosures, each supported by three aluminum poles, anchored by guy lines, and set inside a 30 cm deep × 1 m diameter plastic pool to create a “pond” (SI Appendix, Fig. S2). Each enclosure was supplied a mixture of locally collected aquatic and terrestrial oviposition sites, including sticks, emergent plants (e.g., aquatic grasses or pickerelweed), and floating plants (e.g., water hyacinth). Frogs could also oviposit on the mesh wall of the cage or the aluminum supports (SI Appendix, Fig. S2). We placed all enclosures in relatively open habitats and then covered half of them in an 80% opaque shroud to create artificially shaded habitats.

Up to eight pairs of frogs per night were collected from ponds at each site, either in amplexus or while still searching for a mate, between 2100 and 2400 h. Pairs of frogs were placed into 1 gallon plastic bags and were transported to the breeding enclosures where they were randomly assigned to either a shaded or unshaded enclosure for the night. Two species (D. triangulum and D. leucophyllatus) were unlikely to go into amplexus if only a single male was placed into an enclosure with a female, so we placed two males with each female of these species; the increased male–male competition ensured females always went into amplexus and laid eggs.

Little is known about how oviposition sites are selected in Dendropsophus treefrogs, but females appear to be largely responsible for selecting oviposition sites (18, 33, 34). In our experiments, pairs oviposited overnight in their breeding enclosure, and we recorded the locations of embryos the next morning. Our observations indicate that pairs took considerable time to explore the enclosure before laying eggs. For each pair of frogs tested, we recorded each individual egg as either terrestrial (solely in contact with air) or aquatic (partially or wholly in contact with water). A subset of eggs was set aside for measurements of egg diameter or desiccation rate. All other eggs were allowed to hatch and were returned to the site of collection. Adult frogs were returned to the site of collection after being measured (snout–vent length and mass), photographed, and toe clipped, if appropriate.

Estimating the Evolution of Reproduction in Dendropsophus.

We generated a phylogeny for the 13 species for which we had conducted reproduction trials. We downloaded sequences from GenBank (accessed January 2023) for all genes possible for at least one individual from each nominal species in the genus and one outgroup species (Xenohyla truncata). When possible, we ensured that sequences were obtained from locations as geographically close to our field sites as possible, and that multiple genes for a species were obtained from the same individual. We only included loci that were available across at least half of the species. Our final matrix included five mitochondrial genes: the small (12S) and large (16S) subunits of the mitochondrial ribosome genes, cytochrome b (cytB), cytochrome oxidase 1 (CO1), NADH:Ubiquinone Oxidoreductase Core Subunit 1 (ND1), and one nuclear gene: recombination-activating gene 1 (RAG1), for a total of 5,876 bases (SI Appendix, Table S2). Sequences of each gene were aligned separately using MAFFT (35) and the best substitution model for each gene was estimated using ModelTest-NG (36), executed within raxmlGUI 2.0 (37) (SI Appendix, Table S3). The two ribosomal genes were concatenated and treated as a single gene. A maximum likelihood (ML) tree of concatenated sequences, with the appropriate model applied to each gene partition, was estimated with RAxML-NG v1.0.3 (38) implemented in raxmlGUI v2.0 (37) using 1,000 bootstrap iterations to estimate node support. A Bayesian phylogeny was estimated using Mr. Bayes v.3.2.7 (39) with the same partitioning scheme as in the ML reconstruction. Mr. Bayes was run with a temperature of 0.01 in four chains for 5x10⁵ steps saving every 100, which allowed the average standard deviation of split frequencies to fall below 0.01, and the potential scale reduction factor to be equal to 1.00. Convergence was also determined in multiple runs, which produced the same topology.

All statistical analyses were conducted in R v4.2.2 (40). Data from the Gamboa population of D. ebraccatus were used to represent D. ebraccatus in analyses in which other species of Dendropsophus were included. The phylogeny of Dendropsophus and its outgroup was used to estimate ancestral states of reproduction as both a categorical variable (aquatic, terrestrial, or flexible) using the ace function in the ape package (41) and as a continuous variable ranging from 0 (completely aquatic oviposition) to 1 (completely terrestrial reproduction) using the anc.ML function in the phytools package (42). The average proportion of terrestrial eggs laid in behavioral trials (i.e., in both shaded and unshaded enclosures) was used as the continuous measure of reproduction in extant species.

Population Genetic Structure of D. ebraccatus.

We conducted two analyses to understand the relatedness of populations of one species, D. ebraccatus. We obtained tissue from toe clips from 8 to 30 individuals in each of eight populations (SI Appendix, Fig. S1). For all populations, we extracted DNA from toe clips using Qiagen DNeasy kits and amplified a 1,442 bp section of mitochondrial ND1 and 16s. Primers were standard treefrog primers (43). A median joining haplotype network was constructed in PopART (Population Analysis with Reticulate Trees) (44) with reticulation tolerance set to zero to visualize relationships of D. ebraccatus. We also conducted RAD-seq (Restriction Enzyme associated DNA sequencing) for 10 individuals/population from six D. ebraccatus populations in Central America (SI Appendix, Fig. S1) in order to provide a more fine-scale estimate of population differentiation than might be provided by the mitochondrial data. RAD-seq libraries were sequenced on an Illumina HiSeq at Oregon State University’s sequencing core facility. Sequences were aligned using Stacks v2.41 (45) on the Vassar College computing cluster Hopper. Sequences were first filtered with the process_radtags function, and were aligned using the denovo_map.pl pipeline followed by the population’s function, which was used to calculate population genetics statistics (F_ST and Ф_ST). F_ST is a measure of the genetic distance between two populations as calculated from the frequency of alleles or haplotypes, whereas Ф_ST takes into account the genetic distance among alleles or haplotypes. We also calculated F_ST based on our mitochondrial sequence data in the MMOD package in R (46). After preliminary runs allowing 1 to 8 mismatches per stack (M), Stacks was run finally allowing two mismatches per stack for each individual (M), 1 mismatch between individuals (n), and allowing loci to be called based on their presence within a single individual (r) in a single population (p). Stacks identified 305,705 loci, of which 239,659 were variable. SNP loci identified by Stacks were exported for population genetic analysis in STRUCTURE v2.3.4 (17). STRUCTURE was run for K equal to 1 to 6, and 10 separate runs were conducted for each K value. The most appropriate K was selected using STRUCTURE Harvester (47).

Measures of Egg Diameter.

To assess variation in egg diameter among Dendropsophus frogs, we took high-resolution digital photographs of five freshly laid eggs per female with a Nikon D80 DSLR camera with a Tokina 90 mm macro lens and external flash. Embryos were removed from the egg mass but were maintained inside their egg capsule and were placed onto a piece of 10 mm × 10 mm grid paper for photography. Egg diameter was measured in ImageJ (48) in two perpendicular axes, which were averaged. The measurements of the five embryos were then averaged for each female. In total, we obtained egg size data for 10 species of Dendropsophus and all eight populations of D. ebraccatus. See SI Appendix, Table S4 for sample sizes and measurements for each species and population.

We assessed the relationship between Dendropsophus egg diameter and the degree of terrestrial reproduction with a phylogenetic generalized linear effects mixed model (GLMM) assuming a binomial error distribution using the pglmm function in the phyr package (49, 50). pglmm allows the inclusion of a phylogeny as a random effect and uses the variance–covariance matrix derived from branch lengths of the phylogeny to account for relatedness among species. We used the Bayesian phylogeny for the analyses presented here, although the ML tree yielded essentially identical results. The number of eggs laid terrestrially and aquatically in shaded behavioral assays was coded as a two-column response variable, allowing the model to calculate the proportion of eggs laid terrestrially. Mean egg diameter of the clutch was the predictor. We also included nonphylogenetic random effects for species, site, and species nested within site to account for the lack of independence of individuals of the same species at the site. The analysis of different populations of D. ebraccatus was conducted using the glmmTMB function in the glmmTMB package and included site as the random effect. Since there was no relationship between egg diameter and terrestrial oviposition among populations of D. ebraccatus (see results), there was no need to use a phylogenetic GLMM. Both the phylogenetic and nonphylogenetic GLMMs included an individual level random effect to control overdispersion (51). Significance was assessed using nested likelihood ratio tests, and model fit was always checked by inspecting Q–Q plots.

We assessed egg size variation among presumed aquatically and terrestrially breeding frogs of all kinds using a published database of frog egg diameters from Summers et al. (19). We assessed reproductive mode (aquatic or terrestrial eggs, or eggs laid in a foam nest) by cross-referencing each species in the Summers et al. (19) database with published descriptions of life history in www.amphibiaweb.org and www.iucn.org. The Summers et al. (19) database contained egg diameter for 379 species. We could not find any information about oviposition site for 16 species. We also removed 30 species that lay eggs in foam nests, resulting in a final dataset of 205 species of aquatic breeding frogs and 128 species of terrestrially breeding frogs. We tested for differences in egg diameter between aquatic and terrestrially breeding frogs with a linear model (i.e., identical to a two-sample t-test but with a much larger sample size).

Measures of Egg Hydration Rate.

To estimate how well eggs are adapted to terrestrial development, we calculated the rates of hydration for egg masses across species and populations. We separated clusters of 10 eggs from freshly laid egg masses and placed them on a piece of plastic. Eggs were weighed with a portable digital scale accurate to 0.01 g and then were fully submerged in clean pond water, after which they were reweighed periodically. Eggs were blotted dry with a paper towel prior to each weighing. Hydrating egg masses were weighed 9.3 ± 2.8 times, every 3.3 ± 1.5 h but as often as every 10 min at the start of each trial.

We did not assess the rate of hydration for eggs that had been laid in the water during oviposition trials or that had been rained on overnight because they had already absorbed a substantial amount of water and would not yield accurate data about the rate of water uptake into the egg jelly. Thus, we obtained egg hydration data for four species (D. bifurcus, D. ebraccatus, D. leucophyllatus, and D. sarayacuensis) and for all eight populations of D. ebraccatus. See SI Appendix, Table S5 for sample sizes of each species or population.

We analyzed the accumulation of mass over time for submerged egg masses with linear mixed models (LMMs) in the lme4 package (50). Clutch mass was the response variable, and time submerged, species or site, and the interaction between species or site and time were fixed effects. Time was included as a random slope and frog pair as a random intercept to control for the repeated measurements of each egg mass over time. Time submerged and clutch mass were log-transformed to improve model fit. Post hoc comparisons of species or populations were conducted with the emmeans package (52).

Climate Analysis.

We analyzed variation in reproduction among different populations of D. ebraccatus using latitude and bioclimatic variables from the WorldClim 2 database (22) as predictors. The WorldClim 2 database contains 19 variables that describe bioclimatic variation across global land areas. To avoid possibly detecting spurious relationships, we only assessed variables that might logically affect terrestrial reproduction: annual mean temperature, isothermality (a measure of climatic stability), mean temperature of the wettest quarter, annual precipitation, precipitation of the wettest month, precipitation seasonality, precipitation of the wettest quarter and precipitation of the warmest quarter.

Analyses of the eight chosen bioclimatic variables and latitude on terrestrial reproduction were analyzed with separate binomial GLMMs as above. The average number of terrestrial and aquatic eggs laid at each site was a two-column table for the response variable, latitude, or one of the bioclimatic variables at each site was the predictor, and an observation level random effect was included in each model to control for overdispersion.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Data from behavioral experiments, mtDNA sequence data for D. ebraccatus populations data have been deposited in Dryad (53), GenBank (54). All study data are included in the article and/or SI Appendix.