Introduction

Environmental variation and adaptation along climatic gradients

Identifying the ecological factors driving phenotypic diversification along climatic gradients lies at the heart of research in biogeography and evolutionary ecology1,2,3. The multivariate variation of ecological conditions along climatic gradients—especially in mean annual temperature4 and (daily or seasonal) temperature fluctuation5,6, but also in predation7,8,9,10 and other biotic factors11,12—creates divergent selective regimes that affect phenotypic traits directly related to fitness, including physiological, morphological, reproductive, and behavioral traits13,14,15,16. Studies over large geographic scales (in terms of longitudinal, latitudinal, and/or altitudinal variation) are likely to capture systematic variation not only in abiotic, but also in biotic selection factors and provide important insights into the mechanisms underlying the observed phenotypic divergence.

Adaptive phenotypic divergence along extensive climatic gradients has been reported for several taxa, including insects17, birds18 and mammals19. Phenotypic variation along latitudinal gradients received most scientific attention20,21,22. Variation in temperature regimes, precipitation, photo- and vegetation periods, to mention but some important abiotic factors, bring about an array of correlated changes in biotic selection factors (e.g., regarding species richness, primary production, and resource availability)23,24. Latitudinal variation in body size has been examined thoroughly25,26,27. Body size is linked to fitness as it not only influences physiological performance in contrasting thermal environments (with passive heat loss being reduced as the body volume-to-surface ratio increases28), but can also affect traits like anti-predator behavior (e.g., through altered maneuverability)29,30,31. In this context, Bergmann’s rule arguably represents the most widely known ecogeographic rule. It states that within a given taxonomic group of endotherms (populations, species, or higher taxonomic levels), larger body size would be predicted in colder environments, i.e., towards higher latitudes32,33. Bergmann’s rule has received extensive support from studies on different endotherms34,35, while the evidence for ectotherms is controversial36,37,38,39. For instance, revisiting n = 703 angling records from populations of 29 North American freshwater fishes, Rypel40 demonstrated that only 38% of species follow Bergmann’s rule, while 34% showed the reversed pattern, and the remaining 28% showed no intraspecific body size variation related to latitude. Another widely known ecogeographic law is Allen’s rule41, which states that the body extremities of endotherms that live under cold climatic conditions (i.e., at higher latitudes) are smaller than those of related taxa living at lower latitudes. Just like an increased body mass (i.e., Bergmann’s rule), shorter body appendages are thought to help increase the body volume-to-surface ratio, thereby minimizing thermal energy loss in cold environments42,43,44.

Does sexual selection contribute to phenotypic divergence along climatic gradients?

Studies throughout the Animal Kingdom reported that body size is not only under natural but also sexual selection, e.g., via mate competition (intrasexual selection) or female mate choice (intersexual selection)45,46,47,48. More generally speaking, phenotypic traits typically considered to show latitudinal divergence in response to natural selection could also diverge—at least in part—through different forms of sexual selection. Regarding body size, this could be true especially for ectotherms, for which the above-mentioned form of natural selection from climatic variation does not readily apply49. Certain forms of sexual selection could be stronger at lower latitudes, where population densities and mate encounter rates can be higher50,51. However, the role played by sexual selection during phenotypic diversification along latitudinal gradients is generally not well understood.

Our present study provides novel insights into the potential contributions of both natural and sexual selection in driving phenotypic variation in invasive Western mosquitofish (Gambusia affinis) along climatic (latitudinal and longitudinal) gradients in the species’ invasive distribution range in China52. However, our study design was not suitable to tease apart the relative influences of natural and sexual selection on body size, and so investigating this question will be reserved to future studies. Still, we provide indirect evidence that both forms of selection are involved in phenotypic diversification along climatic gradients. Specifically, we demonstrate divergence in various phenotypic traits (male and female body size, body shape, and life histories), while including traits that are known to be under strong sexual selection.

Multivariate phenotypic trait divergence in invasive mosquitofish

In an attempt to control mosquito-borne diseases, Western (G. affinis) and Eastern mosquitofish (G. holbrooki) have been introduced to at least 40 countries worldwide53,54,55, including the introduction of G. affinis to large parts of mainland China52,56. A recent study57 demonstrated that latitudinal body size variation of the closely related G. holbrooki in its native range in the Eastern USA is in support of Bergmann’s rule. Considering various ecological factors covarying with climate along the examined stretch of >14 degrees of latitude (such as the thermal regime, local population densities, and habitat productivity), a model selection approach identified the thermal regime as the main selection force driving the pattern of increasing body size with increasing latitude. Reproductive strategies showed r-selected life-history patterns at high latitudes [with high reproductive allocation (RA) and numerous small offspring], which could be owing to higher extrinsic mortality rates. On the other hand, other traits, like body condition and body shape, appear to diverge as a function of habitat productivity and population density. However, in another study49, G. affinis from 27 populations spanning nine degrees of latitude in North America showed a suggestive trend contradicting Bergmann’s rule. Finally, Stockwell and Vinyard58 studied life-history variation of four newly established (invasive) G. affinis populations and found small body size, early maturity, low fat reserves and small embryos in female G. affinis from thermally unstable environments.

In this study, we collected invasive mosquitofish along 17 degrees of latitude and 16 degrees of longitude in mainland China (Fig. 1a). Based on existing theories and recent studies, we tested the following predictions:

  1. (1)

    Body size: Following a previous study in the congener G. holbrooki, we predicted that invasive G. affinis in China have larger body size at higher latitudes, partly because bigger individuals have an advantage in terms of greater overwintering survival in harsh environments59,60. On the other hand, body size could also show a pattern contradicting Bergmann’s rule49: life-history theory predicts that high adult mortality in fluctuating environments (i.e., higher latitudes, and continental/inland sites61) selects for early maturity and thus, small adult body size62. Moreover, lower resource availability in colder environments impairs growth rates63. We refrain from formulating predictions for body size evolution by sexual selection, but we will tentatively discuss our results in light of the insights into the general involvement of sexual selection in driving trait divergence, as obtained from our analyses of the size and shape of the distal part of the male intromittant organ, called gonopodium64,65.

  2. (2)

    Life-history traits: G. affinis at higher latitudes are likely to experience high overwinter mortality66,67. Other environmental factors, such as fluctuating productivity68, should increase (unpredictable) mortality rates. Based on life-history theory62,69, we predicted G. affinis females to produce more but smaller offspring at higher latitudes, and to have a higher total investment into reproduction. More stable and benign conditions at lower latitudes likely result in higher survival and continuously higher population densities. Increased intraspecific competition should favor the production of fewer but bigger offspring, which are more competitive62,70.

  3. (3)

    Morphology: Since female body shape is tightly linked to life-history traits71,72,73, we predicted that divergence in female body shape largely follows patterns predicted for life-history divergence. Populations at higher latitudes—characterized by higher reproductive effort—should have enlarged abdomens to harbor larger broods, more anteriorly positioned pectoral fins and relatively smaller heads than more southern populations. By contrast, males are unlikely to show a similar degree of morphological divergence mirroring life-history divergence.

  4. (4)

    Gonopodium morphology: Populations from lower latitudes and from more coastal areas (which are preferred by mosquitofish53) likely experience more stable and benign environmental conditions. Low overwinter mortality should result in higher population densities and heightened intrasexual competition amongst males. Also, mosquitofish females prefer males with longer gonopodia64, and females are more likely to exert mate choice when population densities are high, as they have more opportunities to choose (even though this effect may be weakened by coercive mating74,75,76,77). This could result in elongated gonopodia (via female choice) and more rigid distal gonopodium tips (a trait that is beneficial to achieve coercive copulations65) towards lower latitudes and in coastal areas.

Figure 1
figure 1

Sampling sites and morphological landmarks. (a) Ten sampling sites across China from which adult G. affinis were collected; city name codes can be found in Table 3. The map was generated using DAVI-GIS v 7.5.0 (http://www.diva-gis.org/). (b) Male (above) and female G. affinis (below) collected in Hangzhou in April 2016. Dots and numbers indicate the 13 landmarks used for morphometric analyses and two additional landmarks (14 and 15) used for the ‘unbending’ procedure in our Procrustes analyses. (c) Exemplary micrograph of a gonopodium showing the 51 morphometric landmarks. Nomenclature of fin rays (3, 4a, 4p, 5a) follows Rosen and Gordon108. (d–f) Exemplary photos of our sampling sites in Ankang, Chaozhou and Nan**g, respectively.

Full size image

Results

Population genetic analyses

We conducted population genetic analyses based on 15 nuclear microsatellites78,79,80. This part of our study served not only as a validation of species identity56, but also tested for ‘unusual’ patterns of population genetic structure (suggesting recent translocations or multiple introductions)—important background information for the interpretation of our data on phenotypic divergence. Standard population genetic parameters for each population can be found in Table S1. We found varying degrees of genetic differentiation between populations, ranging from virtual panmixis (FST = 0.038, between Nan**g and Hangzhou) to moderate genetic differentiation (FST = 0.268, between **-stone-like fashion, or at least some degree of ongoing gene-flow between populations.

Figure 2
figure 2

Population genetic analyses of invasive G. affinis in mainland China. (a) Genetic structure among populations (see Table 1 for population codes). Individual assignment to two genetically distinct clusters using STRUCTURE168. Likelihood of assignment for each individual is shown as a vertical bar. (b) Bayesian inference of the number of genetically distinct clusters (K) among the 10 sampled populations using ΔK169. (c) Principal coordinate analysis (PCoA) showing genetic differentiation between populations according to the first two axes. Percent variance explained is given in parentheses. Triangles indicate positive values of the third axis (12.22% variance explained), while squares indicate negative values. Colors signify assignment to K = 2 genetically distinct clusters in STRUCTURE. (d) Neighbor-joining tree based on genetic distances (Nei’s DA). Numbers at nodes indicate bootstrap support; only values >70 are presented. (e) Correlation between genetic distance and geographic distance. A Mantel test on log-transformed pairwise genetic distances (FST-values) detected a significant effect of log-transformed geographic distances (Z = −103.73, r = 0.37, one-sided P = 0.014).

Full size image

According to bottleneck analyses under three microsatellite evolution models, most of the populations underwent genetic bottlenecks in the recent past—especially the populations from Ankang, ** life-history trait divergence along climatic gradients. Sperm competition intensifies as a function of lower overwinter mortality under benign (southern) conditions118,119. A higher GSI may allow for increased sperm production under these circumstances120,121. However, further studies are required to fully elucidate the impact of population densities on the observed life-history divergence. Likewise, female poeciliids typically produce fewer and bigger offspring that are more competitive under fierce resource competition70,121,122,123. Decreased fat content towards southern populations in both sexes could thus reflect a trade-off between reproductive investment and investment into somatic maintenance in the face of strong resource competition117. We are lacking a clear explanation with respect to the observed divergence of somatic lean weight, but increased lean weight could reflect an adaptation to enhance growth and reproduction during the shorter growing seasons in northern latitudes57.

Females (from northern populations) and males showing increased fat content in inland populations may be indicative of relaxed resource competition towards inland sites. The pattern observed in females from southern sites, however, matches predation-driven patterns described for other poeciliids, with lower fat content, and more but smaller offspring being produced in inland populations14,124,125. Large body size in poeciliids can be accompanied by an increased risk of falling victim to predation126,127, and avian predation exerts strong selection on body size at least in natural G. affinis populations128. However, we are currently lacking empirical data on potential variation in avian predation along the climatic gradients examined here.

Body shape and size variation

We hypothesized that body shape divergence would primarily follow patterns observed for life-history diversification, with enlarged abdominal cavities, anteriorly positioned pectoral fins and smaller heads at higher latitudes. We found patterns of divergence to be seemingly congruent with our predictions, but notably, the pattern was more clear-cut for males than for females. This sheds doubt on our initial hypothesis that body shape would evolve as an indirect consequence of life-history divergence, in which case females should show the strongest body shape divergence. We further predicted increased body size at higher latitudes and inland sites because large-bodied individuals have a higher survival rate in harsh and fluctuating environments. Our predictions were met along the latitudinal gradient (climatic PC1) but reversed along the longitudinal gradient (PC2), and the pattern was only observed in males.

Previous studies on freshwater fishes identified several agents of natural selection to affect body shape and size, including flow regime129, resource availability130,131, and predation risk14,89,90,91,132. We argue that variation in both body shape and size of male G. affinis in our study system is primarily driven by temperature regimes. Low overwinter temperatures at higher latitudes select for endurable individuals with larger body size133 and larger fat reserves134. Larger fat stores could indeed explain enlarged abdominal cavities of male G. affinis from northern populations. Mosquitofish males are much smaller, on average, than females135, and so males may be under stronger selection for increased body size (and more compact shape) at higher latitudes. Moreover, we suggested increased intrasexual competition in southern populations, but also higher predation pressure in the south could translate into more forced copulations10,127. Enlarged caudal regions, smaller heads, and a more elongated body are traits that improve unsteady swimming87,136, which is also important during coercive mating. Moreover, small-bodied males can approach females in the blind portion of their visual field, thus preventing females from fleeing, and have a better maneuverability than large-bodied ones137,138,139.

Predation selects for early maturity and smaller adult body size in numerous poeciliid species140 like Poecilia reticulata14, P. vivipara141, Brachyrhaphis episcopi142 or Phalloceros harpagos143. The smaller body size of males from inland populations could again point towards a role for increased predation risk along the longitudinal gradient (see above), for which we do not currently have empirical evidence at hand. As poeciliid females tend to show a preference for large-bodied males64,175. Following the methods described in Chapuis and Estoup176, we used FreeNA to calculate unbiased FST-values between populations while accounting for potential null alleles. To estimate the degree of isolation-by-distance among populations, we performed a Mantel test with pairwise FST-values (calculated with FreeNA using the ENA correction) and linear geographic distances (obtained from Google Earth) using IBDWS v 3.23 (http://ibdws.sdsu.edu/ibdws/distances.html). We tested for evidence of genetic bottlenecks in each population separately using Bottleneck v 1.2.02177. We used Wilcoxon signed-rank tests to identify recently bottlenecked populations by comparing observed and expected numbers of loci with heterozygosity excess under three mutation models, the infinite allele model (IAM), stepwise mutation model (SMM), and two-phase model (TPM), respectively, as recommended by Luikart and Cornuet178.

We used STRUCTURE v 2.3.4179 to calculate individual assignment probabilities (Q-values) to varying numbers of genetically distinct clusters (K). For each value of K = 1–10, ten iterations were run using the admixture model with a burn-in period of 250,000 generations, followed by a sampling phase of 750,000 iterations. We detected the uppermost level of population differentiation with the method presented by Evanno et al.180 using the web-based tool STRUCTURE HARVESTER v 0.6.94. Furthermore, we calculated genetic distances181 (Nei’s DA) using Populations v 1.2.32 (http://bioinformatics.org/project/?group_id=84) and visualized a neighbor-joining tree using TreeView v 1.6.6182 (http://taxonomy.zoology.gla.ac.uk/rod/rod.html). The bootstrap** procedure implemented in Phylip v 3.695 (http://evolution.genetics.washington.edu/phylip.html) was used to evaluate the significance of tree nodes (based on allele frequencies, with 1,000 bootstrap replicates). Moreover, we analyzed genetic structure among populations by means of a principal coordinate analysis (PCoA) based on pairwise Nei’s using GenAlEx v 6.503183,184.

Body size and life histories

We included n = 184 males and n = 191 females in the analysis of life-history traits (18 to 52 individuals per population). We measured standard lengths (SL) of each individual using digital calipers (accurate to the closest 0.01 mm). Maturity was assessed by inspecting the opened body cavity for develo** ova (females) or mature testes (males). Afterwards, we removed all reproductive tissues and all develo** embryos. We determined the stage of development and number of embryos (fecundity) for each female185. Somatic tissues, along with gonads or embryos, were then dried at 55 °C for 24 hours. To assess female and embryo body condition, dried samples were washed for at least six hours in petroleum ether to extract non-structural body fat and were then re-dried and re-weighed.

We thus assessed standard length (SL [mm]), somatic dry weight [mg], somatic lean weight [mg], and fat content [%] for both sexes, the GSI [%] for males, and fecundity (number of develo** embryos), RA [%], embryo lean weight [mg], as well as embryo fat content [%] in case of females. Reproductive effort (i.e. GSI for males and RA for females) was calculated by dividing gonad dry weight (plus embryo dry weight in the case of females) by the sum of gonad (plus embryo) and somatic dry weights. We log10-transformed SL, somatic dry weight, and somatic lean weight, arcsine (square root)-transformed somatic fat content, GSI, RA and embryo fat content, and square root-transformed fecundity. Z-transformation was subsequently applied to all data to obtain unit-free data with equal variance.

To assess the extent of divergence along climatic gradients, we used MANCOVA using the two climate-related PCs (see above) as covariates. Throughout this study, we also included the interaction term of both climate-related covariates but removed it from the final models if not significant. In male life-history analyses, we further included SL as a covariate, while SL and the embryos’ developmental stage served as additional covariates in the case of females. We ran post-hoc ANCOVAs of the exact same structure as the final retained MANCOVA model, to identify the source(s) of variation in case of significant model terms. To evaluate the relative importance of each term, we estimated effect sizes by calculating Wilk’s partial eta squared (ηp2)85. Furthermore, we report relative variance explained by model terms as the partial variance explained for a given term divided by the maximum partial variance in that model.

Generally, to visualize significant interaction effects, we split the data into inland (climate-related PC2 ≥ median) and coastal populations (PC2 < median) and depict variation along PC1 (latitudinal variation) for both cohorts. The alternative way of depicting variation along PC2, while splitting the data based on median values of PC1, is shown in Supplementary Figs S1 and S2.

Geometric morphometrics

We included n = 191 males and n = 211 females in the analysis of body shape divergence (17 to 57 individuals per population). We took lateral photographs of alcohol-preserved individuals (left body side) that were placed in a paraffin-coated petri-dish alongside a piece of laminated scale grid paper using a Canon EOS 760D single lens reflex camera (CANON INC., Ota-Ku, Japan). Photos were loaded into tps format using tpsUtil software186, after which we digitized 13 landmarks and measured gonopodium length (in the case of males) using tpsDig2 v 2.26187 (Fig. 1b). Landmarks provided adequate coverage of the lateral body contour of mosquitofish82,188. To correct for bending effects, we applied the ‘Unbend specimens’ function in tpsUtil using landmarks 1 and 6, as well as two additional landmarks (14 and 15) that were removed from the final analysis (Fig. 1b). We then applied a full Procrustes fit procedure using the software MorphoJ188. This procedure superimposes shape coordinates in a linear tangent space and automatically excludes variation that is not caused by true shape-variation (i.e. translation, scaling and rotation effects). After extracting shape information, a factor reduction procedure was performed in MorphoJ to reduce data dimensionality. We retained ten morphology-related PCs for both males and females, which accounted for 88.31% (males) and 88.81% (females) of the total morphological variance, respectively.

Our main analytical MANCOVA used morphology-related PCs as dependent variables and log10-transformed centroid size along with the two climate-related PCs (see above) as covariates. Again, post-hoc ANCOVAs on single PCs were conducted as described above. Significant effects for PCs that explained only a small percentage of shape variation (≤6.42%) can be found in Supplementary Figs S3–5.

Gonopodium morphology and length

We assessed morphological information on gonopodium tip structures of n = 183 males from eight populations (12 to 35 individuals per population) as we missed to assess gonopodium morphology in the **amen and Nan**g populations. Because the male gonopodium is a delicate organ, we cut the entire gonopodium and photographed the distal tip laterally (left side) at 100× magnification using an Optec B 302 microscope equipped with an Optec TP510 CCD camera (both from Optec Instrument Co. ltd., Chongqing, China). We used 51 homologous landmarks described by Heinen-Kay and Langerhans54 to capture morphological variation (Fig. 1c). Using similar Procrustes analyses and PCA procedures as described above, we obtained nine PCs that cumulatively explained 89.93% of the total variance. We conducted MANCOVAs using morphology-related PCs as dependent variables, while including log10-transformed centroid size, total gonopodium length (determined during the assessment of body shape information, see above) and both climate-related PCs as covariates. We performed post-hoc ANCOVAs to determine the source(s) of variation in case of significant model terms. We subjected the data on gonopodium lengths (from all 10 populations) to an ANCOVA, using standard length and the two climatic PCs as covariates.

Data availability

The datasets generated and/or analyzed for the current study are available from the corresponding author on reasonable request.