Conclusions and remaining questions
We show here that allozyme supernatant and formalin-fixed samples can be used for both phylogenomic and many population genomic applications. In particular, allozyme supernatant replicates performed similarly to the frozen tissue pellet replicates in all analyses. On the other hand, only six of ten formalin-fixed samples recovered >25% of SNPs and were useful for analyses. The four formalin-fixed samples that failed had lower extraction yields and high levels of exogenous DNA, potentially corresponding to the amount of time larval samples were left in formalin (10+ years). We recommend that libraries derived from formalin-fixed DNA should be sequenced at greater depths, and multiple samples could be included for lineages or populations of interest in the event that some samples fail. We also document potential biases associated with combining RADseq and capture datasets in shared analyses, including biases in heterozygous SNP calls, clustering by replicate type rather than by specimen, and systematic differences in estimates of genetic diversity. We found that mapping reads to longer reference sequences derived from assembled contigs rather than mapping directly to RAD loci addressed some of these discrepancies. However, systematic biases between RADseq and capture replicates remained in our dataset and we caution that researchers should be aware of these issues especially for studies in which such a bias could impact the interpretation of results (e.g., inferring changes in heterozygosity through time).