Converging or diverging? Shape coevolution between a sperm-dependent asexual and its sexual hosts

Asexual species, despite lacking recombination, can evolve in response to environmental changes and influence the evolutionary trajectory of coexisting sexual species. Gynogenesis, where asexual females rely on sperm from males of a different species, offers a unique perspective on the eco-evolutionary dynamics between asexual females and their sexual hosts. The Amazon molly, Poecilia formosa, is a gynogenetic species that primarily uses sperm from two sympatric sexual species: the Sailfin molly (P. latipinna) and the Atlantic molly (P. mexicana). We analyzed shape variation among wild P. formosa, P. latipinna, and P. mexicana females to understand Amazon molly shape variation relative to their sexual hosts. We tested three hypotheses: (i) Amazon mollies mimic their sexual hosts to enhance mating opportunities (sexual mimicry hypothesis); (ii) ecological interactions or male mate choice drive morphological divergence (character displacement hypothesis); and (iii) Amazon mollies exhibit random shape variation due to their asexual nature (null hypothesis). Our findings revealed significant shape variation in Amazon mollies, which differ from their sexual hosts in a host-specific manner (e.g., Amazon mollies with P. latipinna resemble P. mexicana and vice versa), supporting character displacement at the interspecific level in a sexual-asexual system.

Methods
Sampling

To address these questions, we leveraged two independent and complementary datasets that used slightly different methodological approaches. The first dataset included a balanced sampling of P. formosa (N=212) and its host species from six natural populations in Texas, USA and Tamaulipas, Mexico, and one common-garden-reared population (since 1994) from a site where P. formosa coexists with both host species. In this first dataset (Dataset I), geometric morphometric analyses were based on external landmarks on lateral photographs of live specimens taken in the field (Figure 2a). The second dataset (Dataset II) included a larger sample size in terms of the number of sites (N=18) and specimens (N=678). Despite a larger sample size, Dataset II is often unbalanced in terms of species sampled within collection locations (reflecting the relative frequency of species at the time of collection). For Dataset II, specimens were fixed upon collection, and geometric morphometric analyses were based on internal landmarks on lateral radiographs of the preserved specimens (Figure S2.2). All females included in each of the two datasets were sexually mature (>30mm). Both approaches yielded qualitatively remarkably similar results. Given the more balanced distribution of specimens across sample sites in Dataset I, we present the details of the methods and findings based on Dataset I. The methods and results for the Dataset II are presented in Supplementary Material 2.

Geometric morphometric data processing 

For Dataset I, sampling was conducted in September 2001 throughout Texas, USA and Tamaulipas Mexico, using a 6.4 mm mesh, 7.6 m x 1.2m standard minnow seine. The fish were anesthetized using Tricaine methanesulfonate, (MS-222) and photographed on-site. At some sites, Amazon mollies lived syntopically with P. mexicana; at others, they coexisted with P. latipinna (Figure S1; Table S1).

We acquired images from 212 individuals (120 P. formosa females, 102 from wild populations, and 18 from one stock population (IV/5); 58 wild P. latipinna females, and 34 wild P. mexicana females) from seven localities across their natural range and different combinations between Amazon mollies and its sexual hosts (Figure S1; Table S1). We classified sampling points by the presence of Amazon mollies and their syntopy with its sexual hosts, either P. latipinna mainly in Texas, USA or P. mexicana, from Tamaulipas, Mexico. Digital photographs were taken on-site on a laminated millimeter grid using an Olympus Camedia 2500L digital camera. Camera set up, light condition, and individual positioning (left side of the individual positioned at the center of the laminated grid) were standardized as much as possible to avoid photographic errors across sampling points. To have a shape representation from P. formosa females from sites where Amazon mollies coexist with both P. latipinna and P. mexicana, we added one population from a laboratory stock (Figure S1; Table S1). All photos were taken with a Nikon D5200 camera with a Nikon with a Nikon DX 18-55mm lens but otherwise treated the same way as the field photos.

Twelve landmarks and 22 semi-landmarks (Figure 2a) were digitized using TPSDig v. 2.30 (Rohlf, 2015). The software TPSUtil v.1.82 (Rohlf, 2015) was used to compile and convert image files into a ‘.tps’ file for downstream analysis in MorphoJ v. 1.07a (Klingenberg, 2011). In MorphoJ, we superimposed the landmark configurations using generalized Procrustes analysis (Rohlf & Slice, 1990), generated a covariance matrix, and performed all geometric morphometric analyses. We conducted a multivariate regression of Procrustes coordinates (shape) and the centroid size (size) to account for allometric effects on shape variation (Klingenberg & Marugán-Lobón, 2013). We then used the residuals of this regression as size-corrected shape data for the subsequent analysis using MorphoJ.

Geometric morphometric analysis

All geometric morphometric analysis was performed using MorphoJ v. 1.07a. We first used Principal Component Analysis (PCA) to characterize the overall shape differences between females of P. latipinna (sexual host), P. mexicana (sexual host), and P. formosa (asexual sperm-parasite). For PCA, we selected the minimum number of principal axes that together explained at least 75% of the overall shape variation (Figure S3).

We investigated whether Amazon mollies have a shape similar (sexual mimicry hypothesis), dissimilar (character displacement hypothesis), or had little/random morphological variation (null hypothesis) in relation to the sexual hosts with whom they coexist. We compared shape differences amongst these groups using a Canonical Variate Analysis (CVA) to determine the effectiveness of variation in predicting a priori group assignments using a T2 Hotelling test with 1000 permutations with Bonferroni multiple correction. This approach tests against the null hypothesis that group averages have no shape differences. A Procrustres pairwise distance matrix was used as a shape distance matrix among groups. For individual pair-to-pair comparisons, we used Discriminant Function Analysis (DFA). We performed cross-validation and permutation tests (1000 permutations each) for all comparisons between groups to statistically evaluate the discriminant scores, using Procrustres distances. From those permutation tests, an individual classification/misclassification table was generated to assess the percentage of individuals per group correctly assigned by the DFA after 1000 permutations. We reran a CVA as described above but with a dataset containing only Amazon mollies to check for shape differences amongst Amazon mollies according to their syntopic sexual hosts. A warped outline drawing graphic was chosen to visualize shape variation.