Identifying and reducing AFLP genotyping error: an example of tradeoffs when comparing population structure in broadcast spawning versus brooding oysters

Abstract

Phylogeographic inferences about gene flow are strengthened through comparison of co-distributed taxa, but also depend on adequate genomic sampling. Amplified fragment length polymorphisms (AFLPs) provide a rapid and inexpensive source of multilocus allele frequency data for making genomically robust inferences. Every AFLP study initially generates markers with a range of locus-specific genotyping error rates and applies criteria to select a subset for analysis. However, there has been very little empirical evaluation of the best tradeoff between culling all but the lowest-error loci to minimize overall genotyping error versus the potential for increasing population genetic signal by retaining more loci. Here, we used AFLPs to compare population structure in co-distributed broadcast spawning (Crassostrea virginica) and brooding (Ostrea equestris) oyster species. Using existing methods for almost entirely automated marker selection and scoring, genotyping error tradeoffs were evaluated by comparing results across a nested series of data sets with mean mismatch errors of 0, 1, 2, 3, 4 and 44%. Artifactual population structure was diagnosed in high-error data sets and we assessed the low-error point at which expected population substructure signal was lost. In both species, we identified substructure patterns deemed to be inaccurate at average mismatch error rates p2 and 44%. In the species comparison, the optimum data sets showed higher gene flow for the brooding oyster with more oceanic salinity tolerances. AFLP tradeoffs may differ among studies, but our results suggest that important signal may be lost in the pursuit of 'acceptable' error levels and our procedures provide a general method for empirically exploring these tradeoffs.

Simple regression models as a threshold for selecting AFLP loci with reduced error rates

“…By using theses loci in further analyses much of the genomic diversity that exists within the dataset is lost. Previous studies have demonstrated that larger datasets having higher error rate can yield greater information than a small number of loci with little error [7]. This reduction in information content may be related to the reduction of loci with moderate allele frequencies, as a result of using fixed error rate thresholds.…”

“…Previous studies have demonstrated inaccurate population substructure patterns in both datasets with high genotyping error rates and datasets using selected loci with very low error rates [7]. These results suggest that a tradeoff exists between reducing error rate and maintaining loci with high information content.…”

“…While next-generation sequencing has created a wealth of genetic information, the AFLP technique continues to provide useful information for numerous experimental questions. This continued use of the AFLP technique is largely due to its ability to screen a large number of genomically representative markers at a substantially lower cost compared to other techniques within non-model species with few genomic resources [6,7]. …”

A call for more transparent reporting of error rates: the quality of AFLP data in ecological and evolutionary research

“…; Holland et al . ; Zhang & Hare ). Detailed reviews of how genotyping error can be generated are provided by Bonin et al .…”

“…). For example, Zhang & Hare () compared the results obtained from AFLP data sets varying in error rates (0, 1, 2, 3, 4 and >4%) and found that inaccurate inferences of a previously determined phylogeographic pattern were made based on data sets with >4% error. While it is possible that genotyping errors may not significantly bias overall conclusions (e.g.…”