We found reduced genetic variability in the sulfur creek (El Azufre) population compared to non-sulfidic surface creeks, and a further decrease in the sulfidic cave (Cueva del Azufre). This pattern was observed before as well as after the flood (Fig. ). Before the flood, genetic variability also varied within the Cueva del Azufre, with roughly equal values for observed heterozygosity (HO), expected heterozygosity (HE), and allelic richness (A) in chambers V and X, but distinctly lower genetic variability in the small rearmost chamber XIII. After the flood, however, values for HO, HE, and A in chamber XIII were similar to those of the other cave chambers (Fig. ). Indeed, pair-wise comparisons (paired t-tests) of pre- and post-flood data in chamber XIII indicated a statistically significant increase in HE and A after the flood (Fig. ), and a tendency toward significance (P = 0.06) in the case of HO (Fig. ).
Figure 4 Genetic diversity in eight populations of P. mexicana before (light gray) and after (dark gray) the flood (for populations code see legend to Fig. 1). Depicted are mean (± S.E.) values across loci for observed (Ho, above) and expected heterozygosity (more ...)
Before the flood, individuals from chamber XIII were fixed at three of the nine microsatellite loci examined, but in all cases shared alleles from the other cave chambers after the flood. For example, at locus GAI29B, all fish from chamber XIII invariably were homozygous for an allele of 217 bp length before the flood, but after the flood also exhibited two additional alleles (225 and 227 bp), both of which had previously been detected only in chamber X. Likewise, all chamber XIII fish before the flood were homozygous for one allele (175 bp) at locus GTII33, while after the flood an additional allele (173 bp) was detected that had previously been found only in cave chamber V. Across loci, we found 12 different novel alleles to occur in chamber XIII after the flood that had previously been recorded exclusively in one or both of the other cave chambers, which was also reflected by large and significant pair-wise Fst-values for the comparison of pre- and post-flood data (e.g., 0.173 in chamber XIII; Table ). Altogether, this suggests extensive flood-induced population mixing among the different cave chambers within the Cueva del Azufre.
Pair-wise genetic divergence (Fst-values).
Genetic diversity was also affected by the flood in the central river of our study system (Río Oxolotán), where water currents were the strongest during the flood. Here, an impoverished genetic diversity was detected after the flood, and pair-wise comparisons (paired t-tests) for HO, HE, and A were highly significant in all cases (Fig. ).
The AMOVA on pre-flood data assigned 21.48% of the total variation to variability among populations (FST = 0.215, P < 0.0001). This can be divided further into (high) variability among habitats (18.64%) and (low) variation among populations within these habitats (2.83%). After the flood, a slightly lower overall FST-value was detected (FST = 0.196, P < 0.0001). The proportion of the total variation among habitats was 17.86% and that among populations within habitats made up 1.74% of the total variation.
The Mantel test on pair-wise FST-values explained 86.6% of the variation in post-flood FST-values. Pre- and post-flood FST-values were highly correlated (r = 0.931, P < 0.001), suggesting that general patterns of genetic differentiation were largely unaffected by the flood. Notably, there were significant effects of perturbation within the cave system, as significant genetic differentiation among the three sampled cave chambers (V, × and XIII) was detected before the flood (pair-wise FST-values between 0.066 and 0.134), but not after the flood (FST-values close to zero; Table ). Moreover, significant FST-values were detected for the comparisons of pre- and post-flood samples from the same locality in the case of chambers V and XIII (Table ). This further substantiates the idea that the flood brought about increased population mixing (immigration and emigration) between different cave chambers.
The partial Mantel tests indicated that the effects of habitat type and distance, overall, did not differ prior and after the flood (Table ). In both cases, pair-wise FST-values were significantly lower between sites of the same habitat type than between sites of a different habitat type (pointing towards a pattern of 'isolation-by-adaptation'), while distance between sites did not have a significant influence on genetic differentiation.
Partial Mantel tests on Fst-values and number of migrants prior and after the flood.
Results from STRUCTURE for both data sets found the best statistical support for clustering all individuals into k
= 3 groups according to habitat type (Fig. ). While the first cluster (white) was composed of fish from the Arroyo Bonita, Arroyo Tres, and Río Oxolotán, the second cluster (gray) was made up mostly of El Azufre I and El Azufre II, and the third (black) of individuals from the cave. The occurrence of several cluster 3 (black) specimens also in El Azufre I (and to a lesser degree in El Azufre II) is indicative of some degree of unidirectional migration out of the cave into the adjacent sulfur creek (i.e
., El Azufre I). This pattern was detected already before the flood, but appeared more prominent afterwards (Fig. ). If we consider the highest probability of assignment (to the gray vs. the black cluster; see [97
] for method) of any specimen in the sulfur creek, the percentage of specimens assigned to the black cluster had a tendency to increase from 15% (6/40) before the flood to 32% (18/57) after the flood (two-tailed Fisher's exact test: P
Figure 5 Population assignment using STRUCTURE 2.3.2 Beta . For both data-sets [before (top) and after the flood (bottom)], k = 3 was recovered as the most likely number of genetic clusters.
Genetically detected dispersers
The STRUCTURE analysis already pointed towards a tendency for increased migration out of the cave into the sulfur creek in the course of the flood (see above and Fig. ). According to our GENECLASS analysis, relative numbers of first-generation migrants were highest within the same habitat type and distinctly lower among different habitat types (Fig. ). No migrants were detected between the clear water surface sites and the sulfidic surface or cave sites. Overall, migration patterns pre- and post-flood were strikingly similar in the GENECLASS analysis; with the exception of migration among the three cave chambers, which was increased after the flood (Fig. ).
Figure 6 Numbers of first generation migrants as calculated using GENECLASS 2.0  before (top) and three months after the millennium flood in fall 2007 (bottom). The width of the arrows is proportional to the fraction of migrants in the 'recipient' population. (more ...)
The Mantel test explained 71.1% of variance in numbers of migrants between sites, and pre- and post-flood numbers of migrants were highly correlated (r = 0.843, P < 0.001). The partial Mantel tests indicated that migration patterns were affected similarly by the predictor variables before and after the flood (Table ). The only significant predictor of the number of migrants between sites was whether or not they belong to the same habitat type. There was no significant effect of distance, and migration was not more common from sulfidic to non-sulfidic (Δ H2S in Table ) or from cave to surface habitats (Δ light).
Population assignment using life history trait characterization
After the flood, females of two populations [sulfur creek (EA I and EA II) and non-sulfidic surface habitats (RA and AB)] suffered a strong decrease in female condition (i.e., female fat content), while embryo condition (i.e., embryo fat content) was decreased in all ecotypes. However, regardless of the flood event, the Cueva del Azufre and El Azufre were clearly distinct from the benign surface habitats (RA and AB) in that they produced few but large offspring before and after the flood (Table ). Furthermore, there was a pronounced difference in offspring size and fecundity in the benign surface habitats. After the flood, AB females produced more but also larger offspring than did RA females before the flood (Table ).
Female life history traits for Poecilia mexicana from three different habitat types, sampled before and after the 2007 flood (Means ± S.E.).
Classification success for a separation by habitat type did not vary greatly before the flood (89.8%) and after the flood (87.0%), while the cross-validation DFA success was 66.2% (Fig. ). Cross-validation success was highest in the El Azufre (100.0%), still good for the Cueva del Azufre (81.0%), but low in the non-toxic surface habitats (40.5%). In all cases, the most important life history traits for successful separation were fecundity and embryo lean weight (Table ).
Figure 7 Group centroids (± S.D.) of discriminant function analyses (DFAs) for separation of populations from different habitats based on female life history traits while controlling for female size (SL) and embryo stage. (A) before and (B) after the flood (more ...)