Fishing for Selection, but Only Catching Bias: Examining Library Effects in Double-Digest RAD Data in a Non-Model Marine Species

Double-digest, restriction-site associated DNA sequencing (ddRAD) has become a powerful and useful approach for population genomics, especially for non-model organisms. However, once population-genomics studies extend beyond a single library preparation, imprecision of agarose gel-size selection during library preparation has the potential to introduce bias if the same set of homologous genomic fragments across all samples is not sequenced or if SNP allelic identity is linked to RAD genomic fragment length. To test this, we examined SNP data from a project on red snapper (Lutjanus campechanus) involving four different libraries, two prepared with agarose gel-based size selection, and two prepared with automated size selection via a Pippin Prep. Individuals from two geographic localities were sequenced from multiple libraries across different size-selection techniques. Each locality is a single population so SNPs affected by library bias would have high within-locality F_ST values across different libraries. We used F_ST based outlier detection to identify 3,710 SNPs with potential library bias from a total of 38,298 raw SNPs. Stringent bioinformatic filtering of the raw SNPs failed to remove all of the potentially biased SNPs, with over 500 remaining in the filtered set of ~5,000 SNPs. Potentially biased SNPs also included 71 of 111 SNPs identified as outlier loci, and could have be mistaken for candidate loci under selection. Our results indicate that there is significant library bias across multiple ddRAD libraries, especially when using gel-size selection, but that the bias can be detected when individuals from the same locality are included in multiple library preparations.