4 | DISCUSSION

4.1 | Geographic Variation

We detected very little geographic heterogeneity in the distribution of genetic variation among North American blue crabs despite extensive sampling and considerable statistical power. Our estimate of overallFST was 0.0002; without the sample from JAC the estimate would have been 0.0001 and not significantly above zero. In two previous studies, estimates ofFST were on the order of 0.01, 50 times higher than ours. McMillen-Jackson and co-workers (1994) estimated Nm (population size times migration rate) fromFST to be between 11.1 and 19.0, from which their unreported FST estimates can be inferred to have been between 0.013 and 0.022. From RAD sequence data, Plough (2017) estimated FST between a sample of blue crabs from the North American Atlantic and a sample from the Gulf of Mexico at 0.01. It is possible that this discrepancy in the magnitude ofFST estimates is due to accidents of sampling. McMillen-Jackson and co-workers (1994) used only nine polymorphic allozymes, a small sample of loci. Furthermore, they estimated FST with the software BIOSYS-1 (Swofford & Selander, 1981), which doesn’t use the now widely adopted Weir and Cockerham (1984) correction for sampling variance in allele frequency estimates, and therefore is expected to produce upwardly biased estimates ofFST . Plough (2017) surveyed only two North American locations, one may have been unusually divergent like our sample from JAC. Differences in estimates ofFST can also reflect differences in statistical methodology. A survey of microsatellite variation in North American blue crabs yielded average pairwise FST estimates that ranged between 0.003 and 0.008 depending on methodological details (Macedo et al., 2019). None of the pairwise FSTestimates in this study were significantly different from zero when the full set of putatively neutral loci was analyzed by the permutation test in Arlequin (Schneider et al., 2000).
Our study encountered several potential sources of error that could lead to the false appearance of genetic structure in blue crabs or other species. The first is misidentification of specimens. Megalopae of several congeners of C. sapidus are found in the northern Gulf of Mexico and are difficult to distinguish by morphology. This could explain some of the anomalous findings of an allozyme survey of blue crabs from the coast of Texas, which included large temporal differences in allele frequencies among cohorts of settling blue crab megalopae over a span of several months (Kordos & Burton, 1993). The observed pattern closely matches seasonal changes in abundance of the megalopae of a related species, Callinectes similis , and the megalopae of these species are difficult to distinguish (Sullivan & Neigel, 2017). Sullivan and Neigel (2017) also identified the megalopae of C. rathbunae and C. danaeco‑settling with C. sapidus . In one of our samples, 6 out of 41 megalopae that we had identified as C. sapidus by morphology matched the distinctive pattern of homozygous and unscorable loci seen in a specimen of C. rathbunae . Had we not removed these megalopae from our analysis we would have concluded that the sample was genetically differentiated from others and included closely related individuals. Our laboratory’s previous survey of sequence variation in protein-coding genes of blue crabs found nuclear gene haplotypes in blue crabs from Venezuela that were not found in samples from North America (Yednock & Neigel, 2014). These individuals from Venezuela appear to represent either a divergent population of C. sapidus or a different species. They were automatically excluded from the present study because of their unusually low heterozygosity (which also suggests a different species), but if we had included them our estimate ofFST among locations would have been much higher. Another potential source of error is null alleles. Yednock and Neigel (2014) suspected null alleles at the ant locus in blue crabs because of departures from Hardy-Weinberg proportions, and we confirmed this in the present study. In their preliminary analysis, inclusion of data for this locus created the appearance of significant population differentiation not seen with other loci, even after a correction for null alleles was applied (Yednock and Neigel, unpublished). From the above considerations, it appears that known sources of error in genetic marker data tend to inflate estimates of FST and create false appearances of population structure. Large numbers of marker loci are often viewed as providing “high resolution” analysis of population structure (e.g. Davey & Blaxter, 2010) because they increase the statistical power to detect genetic differentiation, but they may also increase the sensitivity of statistical tests to errors in data (Chapuis & Estoup, 2006).

4.2 | Accuracy of Illumina Genotyping

SNP genotypes determined by the Illumina Infinium assay for our study were highly reproducible and in aggregate, the data were consistent with the low population differentiation expected for a species that experiences very high gene flow. Differences between genotypes determined by the Infinium assay and those determined by sequencing are explained as null alleles in the sequencing data. Individuals with anomalous genotypes can be readily explained as misidentified specimens. Although a direct comparison between the accuracy of our data and those based on other types of genetic markers in blue crabs is not feasible, our comparatively low estimates of FST imply greater accuracy. The biology of blue crabs suggestsFST should be very low, and errors in genetic marker data are expected to inflate estimatesFST . The distinction between no genetic differentiation and slight genetic differentiation is not trivial. As Palumbi (2003) pointed out in discussing the relevance of apparent “slight geographic differentiation” to the design of marine reserves “Interpreting the significance of this slight genetic signal has been difficult because even mild genetic structure implies very limited demographic exchange between populations, but slight differentiation could also be due to sampling error.” We hope that our study will serve to stimulate interest in the use of Infinium genotyping for population genetic surveys, and that the error assessment methods we developed will prove useful.

4.3 | Temporal Variation without Sweepstakes Reproductive Success

Sweepstakes Reproductive Success (SRS) occurs when the progeny of a few lucky spawners comprise the majority of a cohort of settling individuals, drastically reducing effective population size and creating immediate genetic drift (Hedgecock & Pudovkin, 2011; Hedrick, 2005). SRS is often considered the de facto cause of temporal variation in allele frequencies in marine populations, although it can be difficult to rule out other mechanisms (Cornwell et al., 2016). We detected temporal variation in allele frequencies for blue crabs across years: among samples from all locations as well as among samples exclusively from Louisiana estuaries. We tested for two spatiotemporal patterns predicted to result from SRS: 1) genetic differentiation among cohorts (chaotic genetic patchiness), and 2) the occurrence of siblings or half-siblings that are the progeny of the lucky spawners (Hedgecock & Pudovkin, 2011; Selkoe, Gaines, Caselle, & Warner, 2006). We found no evidence of chaotic genetic patchiness among 21 collections of settling megalopae. However, we did not sample every settlement event so it is possible that we missed those coinciding with SRS. We therefore also tested for genetic differences among annual cohorts of juveniles. Juveniles represent all the successful settlement cohorts of the previous year, and so would include any from SRS events. However, because temporal variation in juveniles could also be caused by post-settlement mechanisms unrelated to SRS it is important to also test for the second prediction of SRS: the presence of full or half siblings within the samples exhibiting temporal variation. Although we did detect genetic heterogeneity among juveniles from different years, we did not find any instances of siblings or half-siblings among juveniles or indeed among any of the individuals that we genotyped. This suggests that a mechanism other than SRS caused temporal variation. We also uncovered a potential bias in relatedness estimation that can lead to the false appearance of closely related individuals. When we estimated coancestry with allele frequencies from pooled samples it appeared that pairs of closely related individuals occurred in some samples. However, this was an artifact created by differences in allele frequencies among samples, which causes individuals from the same sample to have more alleles in common than expected for unrelated individuals drawn from pooled samples. In this light, we consider the findings of Iacchei and co-workers (2013) who used microsatellite allele frequencies pooled from 17 locations to estimate kinship between lobsters (Panulirus interruptus ). They reported average kinship varied among locations and that related individuals were more often at the same location than expected by chance. From this, they concluded that “siblings more likely settle together than disperse across sites”. Our analysis suggests that observations like this could be artifacts.