Detection Signatures of Selection in Genomic Data, Lessons Learned from a Replicated Rat Model

The wide availability of affordable next generation sequencing technology and high density SNP chips has resulted in many attempts to "data mine" for features among the genomic data suggestive that such regions have been under selection and hence a QTL associated with the trait being selected. This increasingly popular approach is intriguing, but not without risks, primarily those associated with false positives due to random genetic drift. Secondary to this concern is the ability to fine map regions under selection. Linkage disequilibrium (LD) often limits the ability fine map. One of the goals of this research was to develop recommendations for design and analysis of resource populations to minimize false positives and maximize the ability to fine map QTL. Methods: Our experiment consisted of 2 replicated, bidirectionally selected lines (HAD/LAD) initiated from the N/NIH HS stock. Selection based on the amount of alcohol consumed when given free access to food and water. Sixty generations were completed at the time of sampling. Samples from 10 rats (5 males and 5 females) of each line and replicate were collected and sequenced to an approximate depth of 5X. Reads were mapped to the reference genome and variants (SNPs) called. Signatures of selection (SS) were detected as excessive differentiation, measured as the intra-class correlation due to between line differentiation (θ). Statistical Significance was determined by genome wide permutations. Results: We found that the average haplotype block (HB) was exceptionally small (1.4kb) allowing for fine resolution mapping, not only to the gene, but often to gene regions. The greatest number of SS occurred in promoters and intronic regions, which were 7 and 15 times more abundant than those in exons, respectively. Among the identified SS, 50% were limited to a single region of a gene, indicating that selection operated primarily on these subcomponents, not the entire gene. Thus, most of the genetic variation in AUD is due to regulatory elements, not genes, consistent with other complex traits. The ability to fine map SS resulted from use of a MAGIC cross in the formation of the N/NIH HS. False positives were largely eliminated by use of replicated lines. Without replication, the majority of allele fixations were due to random genetic drift. We show this problem cannot be fixed statistically by increasing stringency for testing. These results raise concern for analysis of SS in non-replicated experiments/data sets.

 

Source of funding: NIAAA P60 AA07611