We investigated mechanisms of reproductive isolation in livebearing fishes (genus Poecilia) inhabiting sulfidic and nonsulfidic habitats in three replicate river drainages. Although sulfide spring fish convergently evolved divergent phenotypes, it was unclear if mechanisms of reproductive isolation also evolved convergently. Using microsatellites, we found strongly reduced gene flow between adjacent populations from different habitat types, suggesting that local adaptation to sulfidic habitats repeatedly caused the emergence of reproductive isolation. Reciprocal translocation experiments indicate strong selection against immigrants into sulfidic waters, but also variation among drainages in the strength of selection against immigrants into nonsulfidic waters. Mate choice experiments revealed the evolution of assortative mating preferences in females from nonsulfidic but not from sulfidic habitats. The inferred strength of sexual selection against immigrants (RIs) was negatively correlated with the strength of natural selection (RIm), a pattern that could be attributed to reinforcement, whereby natural selection strengthens behavioral isolation due to reduced hybrid fitness. Overall, reproductive isolation and genetic differentiation appear to be replicated and direct consequences of local adaptation to sulfide spring environments, but the relative contributions of different mechanisms of reproductive isolation vary across these evolutionarily independent replicates, highlighting both convergent and nonconvergent evolutionary trajectories of populations in each drainage.

The question of how reproductive isolation between diverging populations evolves remains essential in evolutionary biology. Theory distinguishes between pre- and postzygotic isolating mechanisms, which can act independently or in concert to create varying degrees of reproductive isolation (Coyne and Orr 2004). Ecological gradients are particularly well suited to study the potential role of local adaptation in facilitating reproductive isolation, because they can influence population genetic structure via divergent natural selection. If divergent selection across such gradients is sufficiently strong, it may result in ecologically based reproductive isolation through reduced fitness of dispersers between selective environments (Hendry 2004; Nosil 2004; Nosil et al. 2005) and/or reduced fitness of hybrids between individuals adapted to alternate ecological conditions (Hatfield and Schluter 1999; Via et al. 2000; Rundle 2002). In addition, local adaptation is often intertwined with the processes of reinforcement during which selection promotes assortative mating between individuals adapted to different environmental conditions (Boughman et al. 2005; Rundle et al. 2005; Snowberg and Benkman 2009; Lenormand 2012). Together, these mechanisms are among the central components of ecological speciation (Schluter 2000, 2001; Rundle and Nosil 2005; Nosil 2012; but see Langerhans and Riesch 2013).

Convergent differentiation along replicated selective gradients provides strong evidence for a central role of natural selection in driving adaptive trait divergence (Clarke 1975; Endler 1986; Wood et al. 2005; Elmer and Meyer 2011). In addition, systems in which populations of the same species repeatedly evolved reproductively isolated ecotypes with similar phenotypes are valuable to study the evolution of reproductive isolation (Schluter and Nagel 1995; Johannesson 2001). Examples of convergent ecological speciation have been documented in organisms inhabiting different habitat types, exploiting different resources, and experiencing different predator regimes (reviewed by Langerhans and Riesch 2013). Nevertheless, it is yet unclear (1) whether reproductive isolation barriers evolve convergently across replicated selective gradients in the same way as adaptive phenotypic traits, and (2)—as gene flow between populations can constrain divergence (Slatkin 1987; Hendry et al. 2001)—how convergent or divergent isolation barriers affect gene flow patterns (Räsänen and Hendry 2008). This is of special interest because speciation can be thought of as a continuum, and pairs of populations facing the same divergent ecological selection may be at different stages along the continuum between panmixis and complete reproductive isolation (Hendry 2009; Nosil et al. 2009a; Langerhans and Riesch 2013).

One ecological gradient that fundamentally alters the evolutionary trajectories of populations arises due to naturally occurring, toxic hydrogen sulfide (H2S). H2S-toxicity results predominantly from its interference with mitochondrial bioenergetics and blood oxygen transport, which inhibits aerobic respiration (Bagarinao 1992; Grieshaber and Völkel 1998). Suppression of aerobic respiration is aggravated by the reactivity of H2S leading to extreme hypoxia in aquatic environments. Adverse effects notwithstanding a number of livebearing fishes (family Poeciliidae) thrive in sulfide springs exhibiting continuously high concentrations of H2S (Tobler and Hastings 2011; Tobler and Plath 2011). Adaptation to the toxic conditions in these springs is mediated by complex phenotypic changes, including physiological, morphological, and life-history traits (Riesch et al. 2010, 2011a, 2011c; Tobler et al. 2011). In particular, sulfide spring fish—as compared to close relatives in nonsulfidic habitats—are characterized by increased head size and a correlated increase in gill surface area (Tobler et al. 2008a, 2011; Riesch et al. 2011c; Tobler and Hastings 2011), which facilitates efficient oxygen acquisition in the hypoxic environment and directly affects survival (Plath et al. 2007a, 2010a). Intriguingly, adaptation to H2S gave rise to striking patterns of phenotypic convergence among evolutionarily independent lineages both within and among species (Riesch et al. 2010; Tobler et al. 2011; Tobler and Hastings 2011).

Adaptation to H2S is best studied in fish of the genus Poecilia in southern Mexico, where three evolutionarily independent lineages—two within Poecilia mexicana and a more divergent one described as Poecilia sulphuraria—have colonized sulfidic springs in three river drainages. Phylogenetic analyses suggest that colonization of H2S-springs occurred first in the Pichucalco drainage by P. sulphuraria, which show strong phylogenetic affinity to present Northern Mexican Poecilia mexicana limantouri (Tobler et al. 2011). Invasion of H2S-springs in two other drainages was more recent, and those sulfide spring ecotypes each are closely related to P. mexicana in adjacent nonsulfidic habitats in their respective drainage (Tobler et al. 2011). Still, introgression of nuclear genes cannot be ruled out, as previous phylogenetic analyses were solely based on mitochondrial markers (Tobler et al. 2011).

Potential consequences of local adaptation on gene flow and reproductive isolation between populations from different habitat types have only been investigated in depth in the Tacotalpa river drainage. There, P. mexicana inhabits both nonsulfidic and H2S-containing surface habitats and subterranean ecosystems, in which they are under additional selection from permanent darkness (Tobler et al. 2008a; Riesch et al. 2011a, 2011b). Population genetic analyses indicated reduced gene flow and strong genetic differentiation among populations residing in ecologically different habitat types (Plath et al. 2007a, 2010b), and reproductive isolation is at least in part mediated by a combination of natural and sexual selection against immigrants from other habitat types (Tobler 2009; Tobler et al. 2009).

To date, the degree of reproductive isolation between H2S-adapted lineages of P. mexicana-like fish and adjacent populations from nonsulfidic habitats in other river drainages remains to be investigated. It is yet unclear whether the consequences of local adaptation for gene flow are similar across replicated selective gradients and if similar patterns of natural and sexual selection promoting reproductive isolation are found. Specifically, the present study poses the following questions: (1) Are there any reductions in gene flow between population pairs inhabiting nonsulfidic and H2S-containing habitats within different drainages? To address this, we used 17 unlinked nuclear microsatellites to quantify patterns of gene flow between ecotypes within drainages. Based on previous findings from the Tacotalpa river drainage (Plath et al. 2007a, 2010b), we predicted reduced gene flow between populations inhabiting divergent habitat types over short geographical distances and in the absence of physical barriers; a pattern that would be indicative of “isolation-by-adaptation” (Nosil et al. 2009b). (2) How strong is the contribution of natural selection against migrants (RIm) in reducing gene flow among ecotypes? No barriers other than water chemistry prevent fish movement across habitat types in our study system; hence, we used reciprocal translocation experiments to quantify mortality of migrant individuals. Previous results from the Tacotalpa river drainage indicate that viability of fish is low when transferred between nonsulfidic and H2S-containing habitats (Tobler et al. 2009). (3) What is the role of sexual selection against immigrants (RIs) in reducing gene flow among ecotypes? Both male and female mate choice experiments were conducted, for which focal fish were presented with a potential resident and immigrant mate. Morphological differentiation among ecotypes (Tobler et al. 2011) provides potential cues for assortative mating. (4) Does total prezygotic isolation (T), calculated from empirically established values for RIs and RIm, predict the degree of population genetic differentiation in the three drainages? If selection against immigrants—as determined in this study—accounts for the majority of reproductive isolation, total prezygotic isolation should correlate with the extent of gene flow within each of the three individual drainages. Alternatively, other reproductive isolation barriers not addressed here will need to be considered in future studies.

Material and Methods


Poecilia spp. were collected in the vicinity of the city of Teapa. Here, the mountains of the Sierra Madre de Chiapas meet the wide floodplains of northern Tabasco. Sulfide spring complexes inhabited by Poecilia spp. are located in the foothills of the Sierra Madre and are distributed across three major tributaries of the Río Grijalva (from east to west: Ríos Tacotalpa, Puyacatengo, and Pichucalco). Nonsulfidic and H2S-containing habitats within each drainage are interconnected and not separated by physical barriers that would prevent fish migration. All three rivers eventually join the Río Grijalva and are widely interconnected in the lowlands during the wet season; however, the sulfide spring complexes are permanently separated by mountains (Miller 1966). H2S in this region is likely associated with volcanic activity of the Chichón Volcano (Rosales Lagarde et al. 2006), and average sulfide concentrations in the spring systems range between 23 and 190 μM (Tobler et al. 2006, 2011). In the Río Pichucalco drainage, the sulfide ecotype has been described as a distinct species, P. sulphuraria (Alvarez 1948), which is endemic to sulfide spring complexes at the Baños del Azufre and Rancho La Gloria.

Fish were caught with a seine (4 m long, 4 mm mesh-width). Different sites were sampled for each experiment; an overview is provided in Fig. 1. To ensure that only resident ecotypes were sampled (and not migrating nonresidents), fish were not collected in the mixing zone of sulfidic and nonsulfidic waters. All specimens were individually screened to match their respective ecotype using external morphological features directly upon capture, and all fish collected exhibited the predicted phenotype (see Tobler et al. 2011 for ecotype-specific morphological differences), but it is possible that hybrids might go undetected when using this method. Several other studies that tried to detect ecological speciation used a design in which the divergence between populations from comparable environments was compared to the divergence between populations inhabiting dissimilar environments (e.g., Lu and Bernatchez 1999; Ogden and Thorpe 2002), and ecological speciation was inferred when divergence between populations from dissimilar environments was greater than between populations from similar environments. Because phylogenetically independent invasion of sulfide springs and convergent phenotypic divergence between sulfidic and nonsulfidic ecotypes was already established in this system (Riesch et al. 2010; Tobler et al. 2011) and previous population genetic investigations in the Tacotalpa system included repeated samples within the different habitat types (Plath et al. 2007a, 2010b), our present study concentrated on genetic differentiation and potential gene flow at the interface between populations from sulfidic and nonsulfidic habitats.

Figure 1.

Overview of the study area in Mexico with reference cities in gray. Drainages are underlined, numbers indicate sample sites, black arrows indicate sulfidic sites, and white arrows nonsulfidic sites. 1, Arroyo Caraco; 2, La Gloria; 3, Puente El Azufre II; 4, Baños del Azufre; 5, Puyacatengo road crossing; 6, La Lluvia; 7, Arroyo Cristal; 8, El Azufre I; 9, Arroyo Bonita; 10, El Azufre II. Sulfidic waters usually extend from the point of the black arrows downstream to the nearest confluence with nonsulfidic waters; the sole exception being La Gloria, which is sulfidic for only a few hundred meters.


We used 17 nuclear microsatellite loci to genotype N = 180 fish from six sites (Fig. 1). In the Tacotalpa drainage, samples were obtained from El Azufre I (sulfidic, N = 25) and Arroyo Bonita (N = 25); in the Puyacatengo drainage from La Lluvia (sulfidic, N = 25) and Puyacatengo road crossing (N = 25); and in the Pichucalco drainage from the Baños del Azufre (sulfidic, N = 25) and the western branch of the Río El Azufre at “Puente El Azufre II” (N = 55).

We amplified 12 tri-, tetra-, and penta-nucleotide microsatellites specifically designed for P. mexicana and P. sulphuraria (Slattery et al. 2012), as well as five di-nucleotide microsatellites previously designed for the closely related Amazon molly (P. formosa; Tiedemann et al. 2005). DNA was extracted from ethanol-preserved tissue samples (fin clips) using the DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer's recommendations. Primer pairs for the 17 loci were arranged in three separate multiplex reactions (primer mix 1: EAI_2566, EAI_343, EAI_916, EAI_1039, EAI_808, EAI_568, EAI_1426; primer mix 2: GAI29A, GAV18, GTII33, GAI29B, GTI13B; primer mix 3: AB_195, AB_231, EAI_475, EAI_999, EAI_96) and amplified using the Type-it Microsatellite polymerase chain reaction (PCR) kit (Qiagen). PCR included an initial denaturation step for 5:00 min at 95°C, 30 cycles of 1:30 min at 60°C, and 0:30 min at 72°C, followed by a final extension step for 30:00 min at 60°C. The 5 μL reaction mix included 2.5 μL Type-it master mix, 0.4 μL primer mix, 0.4 μL Q-solution, 0.9 μL RNase-free water, and 0.8 μL template DNA.

ARLEQUIN version 3.5 (Excoffier and Lischer 2010) was used to calculate expected (HE) and observed heterozygosity (HO), to test for deviations from Hardy–Weinberg equilibrium and to calculate pairwise FST-values between populations within each drainage. Allelic richness (A) was calculated using FSTAT version ( All descriptive statistics are presented in Table S1. We tested for null alleles at each locus using Micro-checker version 2.2.3 (van Oosterhout et al. 2004) while pooling all P. mexicana from nonsulfidic sites. STRUCTURE version 2.3.3 (Pritchard et al. 2000) was employed to identify the number of genetically distinct clusters (K) in each drainage with the method presented by Evanno et al. (2005) using the web-based tool STRUCTURE HARVESTER version 0.6.8 (Earl and von Holdt 2012). For each value of K = 1–4, 10 iterations were run using the admixture model with a burn-in period of 100,000 generations, followed by 1,000,000 iterations. Each simulation was performed using an ancestry model incorporating admixture, a model of correlated allele frequencies, and no prior information on locations.


Natural selection against migrants

We conducted reciprocal translocation experiments between representative pairs of sulfidic and nonsulfidic habitats in all three drainages using 20-L plastic buckets as experimental containers. Two holes (18 × 32 cm) were cut on opposite sides of the buckets and then sealed with 1.5-mm plastic mesh to maintain constant exchange of water with the environment. Bucket lids were perforated with ∼50 small holes to facilitate air exchange. Experimental containers were placed directly into a shallow area of the natural habitats and equipped with a 3.5-cm layer of natural substrate. Previous studies have shown that water conditions within these mesocosms closely match the conditions in the adjacent environment (Tobler et al. 2009).

Upon collection, fish were kept in insulated and aerated coolers. Six haphazardly chosen individuals from a given site were then introduced into an experimental bucket. Half of the buckets at each site were set up with resident fish, half with fish from the other habitat type. Transportation and handling times on average were 66 min (range 45–95 min) and were balanced for resident and translocated fish, so differential stress is unlikely to have affected the results, particularly within replicates. Fish were sexed and measured for standard length (SL) prior to introduction. Experiments were terminated after 24 h to quantify mortality, and surviving individuals were released at their original collection site. A subset of data was reanalyzed from Tobler et al. (2009; one set of N = 27 runs including the sulfidic El Azufre I and the nonsulfidic Arroyo Cristal in the Río Tacotalpa drainage) and Plath et al. (2010b; one set of N = 40 runs including the sulfidic Baños del Azufre and an adjacent nonsulfidic tributary in the Río Pichucalco drainage). We completed additional replicates with different sites for the Río Tacotalpa (N = 60 runs between the sulfidic El Azufre II and the nonsulfidic Arroyo Bonita), the Río Pichucalco (N = 60 runs between the sulfidic La Gloria springs and the adjacent nonsulfidic Arroyo Caracol), and the Río Puyacatengo drainage (N = 60 runs between the sulfidic La Lluvia springs and an adjacent nonsulfidic tributary). Essentially, this provided us with two independent sets of replicates for evaluating the performance of fish translocated between sulfidic and nonsulfidic sites for the Tacotalpa and the Pichucalco drainages.

To analyze overall differences in survivability of potential migrants across drainages and sites, we used a generalized linear mixed model (GLMM) with a binomial error distribution and a logit link function; survival (binary data: 1 = survived; 0 = not survived) was used as the dependent variable. We included “population of origin” (sulfide spring or nonsulfidic water), “testing environment” (sulfidic or nonsulfidic), “sex” of the test fish, as well as “drainage” as fixed factors. All possible second- and third-order interactions of the fixed factors were included in the model. Body size of the test fish could have affected survivability (Tobler et al. 2011), so we included “standard length” as a covariate. In addition, “replicate” (i.e., site combination nested within drainage) as well as “Bucket ID” (nested within drainage × replicate × origin × environment) were included as random factors. Removal of nonsignificant interaction terms did not decrease Akaike's information criterion (AIC) for the model; thus, the full model was analyzed. All survival rates are presented as estimated marginal means.

Reproductive isolation through mate choice

Recent studies acknowledge both the role played by female and male mate choice in facilitating reproductive isolation among diverging lineages (e.g., Boul et al. 2007; Gregorio et al. 2012), even though mostly monogamous species were considered so far (Pierotti et al. 2009; Puebla et al. 2011). Thus, we examined both female and male mating preferences for resident versus immigrant phenotypes using simultaneous, binary association preference tests. Female association preferences in another poeciliid fish, the green swordtail (Xiphophorus hellerii), were good predictors of male reproductive success when females are allowed to mate after the association tests (Walling et al. 2010). Furthermore, male association preferences in P. mexicana predict the number of copulations with the preferred female (Plath et al. 2006).

Tests were conducted using wild-caught fish at the aquaculture facilities of the Academic Division for Biological Sciences, Universidad Juárez Autónoma de Tabasco (DACBIOL-UJAT), Villahermosa. Fish were caught from proximate sulfidic and nonsulfidic sites in each of the three drainages (Table S2; Fig. 1), transferred into closed and aerated Sterilite® containers, and immediately brought to the laboratory, where they were kept separated by sex in aerated 70-L tanks for at least 24 h to allow acclimation to laboratory conditions. Fish were kept under a natural, approximately 12 h:12 h light:dark cycle at ambient temperatures and were fed ad libitum with commercially available flake food daily. All tests were conducted in nonsulfidic water, as fish from sulfide-free sites cannot acclimate to sulfide conditions.

Tests were conducted in three identical portable test tanks (42.6 × 30 × 16.5 cm) built with UV-transparent Plexiglas. Each tank was visually divided into three equally sized zones by black marks on the outside. The central zone was designated as a neutral zone, the two lateral zones as preference zones. Two stimulus fish were presented in two smaller auxiliary tanks (19.5 × 30 × 14.5 cm) on either side of the test tank. To avoid disturbance from the outside, we set up all test tanks in large oval tubs that were filled with water to the level inside the test tanks (see Fig. 1 in Bierbach et al. 2011). The entire set-up was placed on a shelf of about 1 m height, and the observer was standing approximately 1.5–2 m away from the test apparatus and observed the fish from diagonally above.

Focal fish were tested with stimulus fish of the opposite sex, one from the same population and one from the respective opposite ecotype within a given drainage. Before each trial, the two stimulus fish were each placed into the auxiliary tanks (see Table S2). Once the stimulus fish were swimming freely, a focal fish was introduced into the center of the test tank. Test fish typically froze on the bottom of the test tank up to a few minutes after they were introduced, so a trial began only after the focal individual resumed swimming freely in the water column. We subsequently measured the time the focal individual spent in each preference zone during a 5-min observation period. To detect side biases, the stimuli were switched between sides immediately after the first 5-min observation period and measurement was repeated. Upon termination of the trials, SL was measured in all fish. Focal and stimulus individuals were used only once.

We summed association times near each stimulus from the two trials (before and after switching of side assignments) and tested for preferences within each test situation (i.e., for each population and sex separately) by comparing individual strength of preference (SOP)-values against a “no choice” expectation of SOP = 0 using one-sample t-tests. The SOP was calculated as:

display math

Hence, SOP-values could range from +1 (complete preference for the resident ecotype) to −1 (complete avoidance of the resident ecotype). SOP-values were used as the dependent variable in a fully factorial general linear model (GLM) to test for differences among drainages, ecotypes, and sexes. Even though an attempt was made to use size-matched stimulus pairs, it was not always possible to assign equal-sized stimulus pairs from our catches, and we included difference in body size (SL) between the two stimulus fish (resident–immigrant) as a covariate in the analysis. Poecilia mexicana females (Tobler et al. 2008b; Bierbach et al. 2011) and males (Plath et al. 2006) prefer larger mating partners irrespective of their own body size.

RIm, RIs, and total isolation (T)

The individual contribution of selection against migrants (RIm) was calculated as 1 − (migrant survival/resident survival) and the individual contribution of sexual isolation (RIs) as 1 − (immigrant mate choice/resident mate choice). This formula was originally developed to contrast heterotypic and homotypic mating (Nosil 2004), and we adjusted it for the binary choice tests employed in this study by using association times as a proxy of individuals’ likelihood to mate. To uncover a potential signature of reinforcement, we tested for a negative correlation between RIs and RIm, which is predicted if low-migrant mortality increases the likelihood of heterotypic mating allowing for direct selection on RIs (see Howard 1993; Higgie et al. 2000; Höbel and Gerhardt 2003; Nosil et al. 2007). We used RIs-values as dependent variable in a GLM, while including RIm as a covariate and “sex” as a factor.

Following Nosil (2004), we estimated the relative contributions of reduced migrant survival and sexual isolation to total premating isolation (T; ranging from 0 to 1; see Ramsey et al. 2003). In these sulfidic/nonsulfidic systems, natural selection against migrants has an immediate impact on fitness (i.e., survival) prior to sexual selection. Thus, the absolute contribution of selection against migrants is RIm, whereas the absolute contribution of sexual isolation is RIs(1 − RIm), and total prezygotic isolation is T = RIm + RIs(1 − RIm). We tested for a correlation between total isolation (average T for males and females of each population) and mean genetic assignment to the nonresident genetic cluster (obtained from a population genetic assignment test and averaged for the ten runs for K = 2) using a nonparametric Spearman rank correlation.



The population assignment test identified K = 2 as the most likely number of genetically distinct clusters in all three drainages with all runs separating the sulfide-spring populations from the populations from surrounding nonsulfidic waters (Fig. 2). Although this separation is relatively clear in the Tacotalpa and Pichucalco drainages, it is considerably weaker in the Puyacatengo drainage, where some degree of gene flow across habitats was uncovered, even though first-generation migrants were not detected in any drainage (Fig. 2). The strong differentiation between ecotypes was corroborated by high pairwise FST-values of 0.296 (Pichucalco) and 0.203 (Tacotalpa drainage). Differentiation in the Puyacatengo drainage was lower, but still significant (FST = 0.103).

Figure 2.

Population assignment using STRUCTURE version 2.3.2. K = 2 was recovered as the most likely number of genetic clusters in every drainage separately; given are individual relative assignment values in black (sulfide springs; left side) and white (sulfide-free waters; right side), divided by a black line and sorted by the relative assignment score for each population separately.


The interaction term between “origin of fish” (sulfidic vs. nonsulfidic habitat) and “test environment” (sulfidic vs. nonsulfidic habitat) had a highly significant effect explaining survival of fish, indicating habitat-specific performance-differences between ecotypes (Table 1). Resident fish consistently had higher survival rates than immigrant fish, and mortality was highest for fish from nonsulfidic waters being transferred into sulfidic water (Fig. 3). Nevertheless, a significant three-way interaction between “origin × test environment × drainage” indicates that performance differences were not uniform across drainages. Survivability of H2S ecotypes in nonsulfidic sites varied across drainages (Fig. 3A), whereas ecotypes from nonsulfidic waters consistently had low survival rates in sulfidic environments (Fig. 3B). In the Tacotalpa drainage, survivability of H2S ecotypes in nonsulfidic environments was almost as low as survivability of the ecotype from nonsulfidic water in the sulfidic environment. In contrast, survivability of H2S ecotypes in nonsulfidic habitats of the Puyacatengo and the Pichucalco drainages was comparatively high. No differences were uncovered among replicate site pairs within drainages, suggesting that survivability estimates for each drainage were not caused by the specific sites we chose for this experiment. Our model also revealed a strong overall effect of the factor “sex” (Table 1), and males generally exhibited lower survival rates than females (Fig. 3C). This pattern varied to some extent across drainages and between sulfidic and nonsulfidic locations (significant three-way interaction “drainage × environment × sex”; Table 1). Our model also identified a significant effect of body size (SL) on survivability, and a post hoc Pearson correlation with standardized residuals for survivability detected a significant negative correlation with SL (rp = −0.220, P < 0.001, N = 1,482; Fig. 3D).

Table 1. Results from a generalized linear mixed model (with binomial error distribution and logit link function) on survival rates during the reciprocal translocation experiment. As random factors we included “site nested within drainage” as well as “bucket ID nested within site and drainage.” The model included “population of origin,” “testing environment,” “drainage,” and “sex” as fixed factors, and “SL” as a covariate. Significant P-values are in bold face. AIC = 7,969.859
Random effectsEstimateSEzP
Replicate (drainage)0.5510.6210.8870.375
Bucket ID (drainage × replicate × origin × environment)0.5630.2132.6240.008
Fixed effectsFdf1df2P
Origin × Environment196.56411459<0.001
Origin × Drainage8.98221459<0.001
Origin × Sex2.558114590.110
Environment × Drainage5.028214590.007
Environment × Sex1.158114590.282
Drainage × Sex6.202214590.002
Origin × Environment × Drainage11.67221459<0.001
Origin × Environment × Sex4.048114590.044
Location × Environment × Sex3.851214590.021
Origin × Drainage × Sex0.256214590.774
Figure 3.

Survival in reciprocal translocation experiments. Within each drainage, fish were either transferred from (A) sulfidic to nonsulfidic sites or (B) from nonsulfidic to sulfidic sites with H2S-ecotypes in black and ecotypes inhabiting nonsulfidic waters in white. R, resident fish; T, transplanted fish. (C) Visualization of the sex-effect from our generalized linear mixed model (GLMM) with females in light gray and males in dark gray. (A–C) Shown are estimated marginal means (+ SEM; SL was fixed at 34.41 mm) from a GLMM. (D) Regression of residual survival versus SL.


In the GLM comparing the SOP of individual focal fish for mating partners from the same population of origin, a significant main effect of “ecotype” and a significant interaction term of “sex by ecotype” were uncovered (Table 2). Closer inspection of SOP values indicates that only fish from nonsulfidic, but not from sulfidic habitats, showed a preference for resident over immigrant mating partners, and females generally expressed stronger preferences than males (Fig. 4). Qualitatively, fish from the Tacotalpa drainage showed weaker association preferences than fish from the other two drainages (Fig. 4A,B); however, neither the main effect nor any interaction involving “drainage” were statistically significant (Table 2).

Table 2. Comparison of the strength of preference (SOP, see main text) for mating partners from the same population in males and females across drainages (three levels) and ecotypes (two per drainage). The full factorial general linear model included the difference in body size between the two stimulus individuals (SL own–SL alien ecotype) as a covariate. The interaction term “drainage × sex × ecotype” was not significant (mean square = 0.56, F1, 354 = 0.61, P = 0.54, partial η2 = 0.003) and thus, removed from the final model. Significant effects are highlighted in bold typeface
SourcedfMean squareFPPartial η2
Sex × Drainage20.130.990.370.006
Drainage × Ecotype20.
Sex × Ecotype10.574.210.0410.012
Stimulus size difference12.5818.91<0.0010.050
Figure 4.

Mate discrimination in female (A) and male (B) Poecilia spp. Within each drainage focal fish could choose between two stimulus fish of the opposite sex, the respective ecotypes from sulfidic and nonsulfidic waters. Positive strength of preference (SOP) values indicate preference for mating partners from the own population. Focal fish stemmed from sulfidic (●) and nonsulfidic sites (○). Results from two-sided one-sample t-tests are shown, testing against the null assumption of SOP = 0, where ** signifies P < 0.01 (from left to right: t30 = −0.12, P = 0.91; t29 = 3.31, P = 0.002; t29 = −1.63, P = 0.12; t29 = 3.14, P = 0.004; t29 = −1.01, P = 0.32; t34 = 1.79, P = 0.083; t29 = −0.86, P = 0.40; t29 = 2.85, P = 0.008; t29 = −0.93, P = 0.36; t29 = 0.73, P = 0.47; t29 = 1.52, P = 0.14; t30 = −0.70, P = 0.49). Shown are means ± SEM.


A GLM using RIs as the dependent variable uncovered a significant effect of the covariate RIm (F1,8 = 10.00, P = 0.013, math formula = 0.555). Closer investigation of the relationship between both variables revealed a negative correlation (Fig. 5). There was also a significant difference between sexes (F1,8 = 5.46, P = 0.048, math formula = 0.406) owing to males showing higher values for RIs than females. The interaction term was not significant (F1,8 = 2.79, P = 0.133, math formula = 0.259).

Figure 5.

Reproductive isolation due to sexual selection (RIs) versus reproductive isolation due to selection against migrants (RIm) for males (triangles) and females (diamonds) from three different drainages. Closed symbols refer to RI-values estimated for fish from sulfidic sites in sulfide-free habitats, open symbols refer to RI-values estimated for fish from sulfide-free sites in sulfidic habitats.


Total prezygotic isolation (T) differed significantly between sexes (t5 = 3.30, P = 0.022) and was higher for males (mean ± SD: 0.842 ± 0.129) than females (0.564 ± 0.230). Mean values of total prezygotic isolation did not correlate with population-wise means of assignment to the nonresident genetic cluster, our estimate of gene flow (Spearman rank correlation: rs = −0.49, P = 0.33, N = 6…; Fig. 6). The same pattern was uncovered when only male (rs = −0.77, P = 0.07, N = 6) or only female T-values were used instead (rs = −0.14, P = 0.66, N = 6).

Figure 6.

Proportion of immigrant assignment based on STRUCTURE version 2.3.2 versus total prezygotic isolation. Closed symbols refer to immigration of fish from sulfidic sites into nonsulfidic habitats, open symbols refer to immigration of fish from nonsulfidic sites into sulfidic habitats. Pi, Pichucalco drainage; Pu, Puyacatengo drainage; T, Tacotalpa drainage.


We investigated genetic differentiation and patterns of gene flow in livebearing fishes of the genus Poecilia, inhabiting nonsulfidic and H2S-containing habitats in three distinct river drainages. We found reduced yet variable gene flow resulting in strong genetic differentiation between ecotypes in all river drainages. The varying degree of genetic differentiation across drainages is indicative of population pairs in contrasting environments being at different stages along the speciation continuum. Nevertheless, high pair-wise FST-values in all cases suggest that adaptation to H2S-containing habitats repeatedly caused the emergence of reproductive isolation across evolutionarily independent population pairs despite the lack of physical barriers that would prevent fish migration. Translocation and mate choice experiments indicated that both natural and sexual selection against immigrants contribute to reproductive isolation between ecotypes, but their relative contributions vary among drainages, reflecting unique evolutionary trajectories in each replicate drainage. Because the inferred amount of gene flow did not correlate with total prezygotic isolation estimated by our experiments, additional reproductive isolation barriers need to be explored to better understand all factors contributing to speciation in this system.


Divergent selection can reduce gene flow in natural populations, but genetic structuring in relation to ecological variables is not ubiquitous among systems in which phenotypically divergent populations experience different selection pressures (reviewed in Thibert-Plante and Hendry 2010). Potentially, fitness of dispersers and their hybrids may not be sufficiently reduced among selective environments to detect a reduction in gene flow in microsatellites or other neutral loci (see Räsänen and Hendry 2008). Our population genetic analyses revealed that fish from nonsulfidic and H2S-containing habitats in all three drainages were genetically divergent and gene flow (estimated from STRUCTURE assignment scores) was virtually absent between population pairs from the Tacotalpa and Pichucalco drainages, corroborating previous studies on the Tacotalpa drainage (Plath et al. 2007a, 2010b). Even though genetic drift can significantly affect the frequency of alleles at neutral loci and its effects cannot be conclusively separated from those of divergent selection especially in cases of very low gene flow (Thibert-Plante and Hendry 2010), our results match previous findings of convergent phenotypic divergence between different ecotypes across all three drainages (Riesch et al. 2010; Tobler et al. 2011), suggesting “isolation-by-adaptation” (Nosil et al. 2009a). In contrast to the Tacotalpa and Pichucalco drainages, a low level of gene flow was detected in the Puyacatengo drainage, suggesting that this population pair falls onto a different point along the speciation continuum than the other two population pairs (Hendry 2009; Nosil et al. 2009a; Langerhans and Riesch 2013). More specifically, there are two drainages (Pichucalco and Tacotalpa) in which the speciation process is well advanced, whereas the population pair from the Puyacatengo drainage is still in the process of speciation. This may be due to a stronger spatial and temporal heterogeneity in the Puyacatengo drainage compared to the other two drainages (authors, unpubl. data), or it could be due to a more recent colonization of sulfidic springs.


Translocation experiments revealed that resident ecotypes in each habitat had higher probabilities of survival compared to immigrant fish, which is indicative of local adaptation (Kawecki and Ebert 2004) and highlights the costs of local adaptation in terms of reduced individual performance in nonnative environments (Nosil 2012). This was particularly evident when fish were moved from nonsulfidic to H2S-containing environments, and high mortalities are readily explained by the lack of adaptation to cope with the extreme environmental conditions in sulfidic habitats, including H2S-toxicity and hypoxia. H2S is a potent respiratory toxicant that is usually lethal to nonadapted organisms in micromolar amounts (Bagarinao 1992; Grieshaber and Völkel 1998). Although the exact physiological mechanisms allowing fish from H2S-containing habitats to withstand continuous exposure to H2S remain unclear, eco-toxicological experiments revealed that they exhibit significantly higher sulfide tolerance than fish from nonsulfidic environments (Tobler et al. 2011). Furthermore, fish from nonsulfidic habitats lack the increased respiratory efficiency of fish from H2S-containing habitats, which—due to their increase in gill size—can acquire oxygen more efficiently at low ambient concentrations (Tobler et al. 2011).

We also found high mortality rates in translocations from H2S-containing into nonsulfidic environments, but the magnitude of this effect varied across drainages. Survivability was comparatively high in the Puyacatengo and the Pichucalco drainages, but in the Tacotalpa drainage it was almost as low as survivability of fish transferred from nonsulfidic into H2S-containing habitats. Mortality of fish from H2S-containing habitats in nonsulfidic environments could be caused by oxidative stress, possibly in combination with poor body condition and energy limitation of H2S-ecotypes (Tobler 2008; Riesch et al. 2010, 2011a). Oxygen can have adverse physiological effects due to its biotransformation into reactive oxygen species, and organisms have evolved biochemical pathways with antioxidant activity to mitigate such effects (Halliwell and Gutteridge 1999). In aquatic organisms, the expression of antioxidant enzymes is often downregulated during periods of hypoxia (Hermes-Lima and Zenteno-Savín 2002; Olsvik et al. 2006), and subsequent exposure to normoxic conditions can cause substantial oxidative stress with profound fitness consequences (Sies 1986).

Both laboratory and field experiments further indicate that survival in adults is dependent on body size (with small individuals having a higher probability of survival than larger ones). We hypothesize that H2S-detoxification becomes inefficient in larger individuals, possibly due to their higher absolute demand for readily available oxygen (Hildebrandt and Grieshaber 2008). In addition, standard length in poeciliid fishes is correlated with age (e.g., Snelson 1989), so the positive correlation between mortality and body size could also reflect a decreased physiological capacity as a result of senescence (e.g., Kirkwood and Austad 2000; but see Reznick et al. 2002). Finally, females consistently outperformed males, which could be caused by different life-history strategies. Males in these species have consistently lower amounts of stored body fat than females and a larger proportional investment into reproduction (measured as gonadosomatic index) in sulfidic compared to nonsulfidic habitats (Riesch et al. 2010, 2011a). In other words, male poeciliids seem to have a higher total investment into reproduction (including costly mate searching and mating behavior) than females even at the risk of lowering survival capability. This interpretation is consistent with a previous study that found male P. mexicana to perform more aquatic surface respiration than females in H2S-containing water (Plath et al. 2007b). Clearly, additional studies are warranted to understand size- and sex specific patterns of mortality mechanistically.


Our mate choice experiment revealed the evolution of assortative mating preferences in at least some of the populations investigated. Particularly, fish from nonsulfidic (but not from sulfidic) habitats showed a preference for resident over immigrant mating partners. The inferred strength of sexual selection against immigrants (RIs) was negatively correlated with the strength of natural selection (RIm). We propose to attribute this pattern to reinforcement, a process by which natural selection strengthens behavioral isolation due to reduced hybrid fitness (Kirkpatrick 2001; Servedio and Noor 2003). Direct selection on mating preferences is more likely to occur when individuals have an increased likelihood of encountering a nonnative mating partner (Coyne and Orr 1997); that is, when RIm is relatively low, like in migrants from sulfidic to nonsulfidic environments. It is important to point out that occasional mating (and thus a role for reinforcement) might still be possible even for migrants from nonsulfidic to sulfidic environments, because our experiments indicate that H2S-related death is not instantaneous upon exposure to sulfidic waters. Furthermore, reinforcement can shape male and female mate preferences even when reductions in hybrid fitness are small (Kirkpatrick 2001; Svensson et al. 2007; Servedio 2007). Future experiments will need to establish (potentially drainage-specific) differences in hybrid offspring fitness both under sulfidic and nonsulfidic conditions (Servedio 2007).

Females consistently expressed stronger preferences than males. Females generally have a lower reproductive potential (Andersson 1994), and their reproductive fitness may thus decrease disproportionally if unfit hybrid offspring were produced (Servedio and Noor 2003). Hence, females are expected to be choosier than males and contribute more to premating isolation between populations (Servedio 2007).

To date, it is unclear on what cues individuals base their preferences. Our experimental design only allowed for the transmission of visual cues, and general phenotypic differences between ecotypes—for example, in body shape (Tobler et al. 2011) and coloration (authors, pers. obs.)—may provide sufficient information. If mating decisions are indeed based on such broad phenotypic differences shaped by natural selection, reproductive isolation through mate choice could evolve rapidly because any detectable modification of a phenotype essentially could serve as a “magic trait” (a character shaped by divergent natural selection that pleiotropically affects mate choice decisions; Servedio et al. 2011). Alternatively, assortative mating may be based on condition-dependent signals. Female preferences for condition-dependent male traits signaling local adaptation through enhanced performance may promote assortative mating even when female preferences are unimodal (van Doorn et al. 2009). Males from H2S-containing habitats are known to have poor overall body condition even in their native environment (Plath et al. 2005; Tobler et al. 2006; Riesch et al. 2011a), which may provide females from nonsulfidic environments with cues to discriminate against alien males.

Our results provide fruitful ground for future studies illuminating mechanisms of assortative mating in this system. If reproductive isolation evolved as a byproduct of adaptation (as predicted by ecological speciation theory; Langerhans et al. 2007; Nosil 2012), across drainage comparisons should uncover signals of reproductive isolation between different ecotypes, but a lack thereof between individuals of the same ecotype. Further cross-drainage experiments are clearly warranted both to make inferences about convergent and nonconvergent mechanisms underlying assortative mating and the potential role of reinforcement (Räsänen and Hendry 2008; Schwartz et al. 2010). Future studies will also need to elucidate the role of nonvisual cues, such as pheromones (Stacey 2003; Stacey and Sorensen 2005; Rosenthal et al. 2011), in mate choice decisions. It also remains unclear whether preferences are innate (reflecting genetically based divergence) or shaped by sexual imprinting (Verzijden et al. 2012), highlighting the need for future experimentation using common-garden reared fish (Jennions and Petrie 1997).


Total reproductive isolation was calculated based on our empirical estimates of natural and sexual selection against immigrants and did not correlate significantly with the degree of gene flow. The virtual absence of gene flow in the Tacotalpa and Pichucalco drainages points towards nearly complete reproductive isolation between ecotypes, which in theory should be reflected by T values around 1 (Ramsey et al. 2003). However, we evaluated T ranging between 0.64 and 0.85 for these drainages, suggesting that additional reproductive isolation barriers, not investigated in this study, are present. Considering the toxic effects of H2S, we suspect that gene flow in this system is constrained by local adaptation. However, we cannot unequivocally establish a causal relationship, because the reverse—gene flow constraining adaptation—also receives support from empirical studies (Räsänen and Hendry 2008). An assessment of multiple ecological and evolutionary factors, including dispersal rates and effective population sizes, is requisite to determine cause and effect (Räsänen and Hendry 2008).


Overall, reproductive isolation and genetic differentiation appear to be consistent consequences of local adaptation to sulfide spring environments. Nonetheless, genetic divergence as well as the relative contributions of different reproductive isolation barriers vary across the evolutionarily independent replicates investigated here, supporting the view that both convergent and nonconvergent aspects of evolutionary differentiation are common during transitions along the speciation continuum even in response to replicated environmental gradients. This could be due to a variety of reasons, among them as yet undetected (i.e., cryptic) differences among seemingly similar environments, genetic covariances of traits, differences in the amount of gene flow between populations (i.e., replicated population pairs falling onto different points along the speciation continuum), or mutation-order speciation (e.g., Kaeuffer et al. 2012; McGee and Wainwright 2013).

Prezygotic isolation through selection against immigrants (either through natural selection or sexual selection) is a key in driving genetic differentiation between sulfidic and nonsulfidic populations in all three river systems. Observed gene flow patterns, however, are lower than the total reproductive isolation calculated based on the empirical patterns of natural and sexual selection against immigrants would predict, which points toward the existence of additional isolation barriers that need further evaluation. Thus, current efforts in identifying genes underlying adaptation to sulfidic spring environments (see Kelley et al. 2012) in conjunction with experiments on hybrid performance, both in the laboratory (to test for intrinsic postzygotic isolation) and under natural conditions (to test for environmentally based postzygotic isolation), are the logical next steps to further evaluate whether and how ecological speciation is operating in our system.

Associate Editor: G. Mayer


The authors thank members of the Tobler lab, G. Mayer, A. P. Hendry, and an anonymous reviewer for their helpful suggestions. Financial support came from the German Science Foundation (DFG, PL 470/1-2, 3-1), the German Academic Exchange Service (DAAD), the initiative Nach-wuchswissenschaftler im Fokus by the presidential office of the University of Frankfurt, and the Herrmann Willkomm-Foundation (to M. Pl.), the research funding program LOEWE – Landesoffensive zur Entwicklung wissenschaftlich-ökonomischer Exzellenz of Hesse's Ministry of Higher Education, Research, and the Arts (to M. Pl. and M. Pf.), the Stiftung Polytechnische Gesellschaft Frankfurt am Main (to HL), the Erwin Riesch-Stiftung (to RR), and National Science Foundation (IOS-1121832, to MT). Collection of fish and experimental work in Mexico were conducted under authorization by the Municipal de Tacotalpa, Tabasco, and the federal agency SEMARNAT.