Multiple-testing corrections in case-control studies using identity-by-descent segments

Abstract

Identity-by-descent (IBD) mapping provides complementary signals to genome-wide association studies (GWAS) when multiple causal haplotypes or variants are present, but not directly tested. However, failing to correct for multiple testing in case-control studies using IBD segments can lead to false discoveries. We propose the difference between case-case and control-control IBD rates as an IBD mapping statistic. For our hypothesis test, we use a computationally efficient approach from the stochastic processes literature to derive genome-wide significance levels that control the family-wise error rate (FWER). Whole genome simulations indicate that our method conservatively controls the FWER. Because positive selection can lead to false discoveries, we pair our IBD mapping approach with a selection scan so that one can contrast results for evidence of confounding due to recent sweeps or other mechanisms, like population structure, that increase IBD sharing. We developed automated and reproducible workflows to phase haplotypes, call local ancestry probabilities, and perform the IBD mapping scan, the former two tasks being important preprocessing steps for haplotype analyses. We applied our methods to search for Alzheimer's disease (AD) risk loci in the Alzheimer's Disease Sequencing Project (ADSP) genome data. We identified six genome-wide significant signals of AD risk among samples genetically similar to African and European reference populations and self-identified Amish samples. Variants in the six potential risk loci we detected have previously been associated with AD, dementia, and memory decline. Three genes at two potential risk loci have already been nominated as therapeutic targets. Overall, our scalable approach makes further use of large consortia resources, which are expensive to collect but provide insights into disease mechanisms.

Keywords: Alzheimer’s disease; binary traits; haplotypes; identity by descent; mean-reverting processes; multiple testing.

Figure 1:

Proportion of false positives when strong positive selection is confounding. Bar plots show the proportion of times that we reject the null hypothesis of the IBD rate difference scan in terms of the selection coefficient (x-axis) and the sweeping allele frequency (colors in legend). The demographic scenarios are A) population bottleneck and B) three phases of exponential growth. Each parameter combination is simulated 200 times. The significance threshold is based on the average threshold over all null simulations. The IBD segment detection threshold is ≥ 2.0 cM.

Figure 2:

Genome-wide IBD rate difference scans for sample sets in the Alzheimer’s Disease Sequencing Project. Line plots show standardized IBD rate differences every 0.05 cM (y-axis) for base pair positions along twenty-two human autosomes. The data for each subplot is based on A) AFR ancestry, B) EUR ancestry, and C) Amish sample sets. Horizontal dashed lines show (blue) the autosome-wide median standardized IBD rate difference, (orange) the heuristic threshold of four standard deviations above the median, (green) the discrete-spacing analytical threshold, and (red) the simulation-based threshold. The IBD segment detection threshold is ≥ 2.0 cM.

Figure 3:

Positive selection confounds the case-control scan in European ancestry samples. The base pair position (y-axis) by chromosome number (x-axis) of genome-wide significant loci in the A) case-control and B) selection scans are shown for the European ancestry samples. Genes in the loci appearing in both A) and B) are annotated. ** This locus is identified in the TOPMed European ancestry selection scan of Temple et al. [40].

Figure 4:

Alzheimer’s disease risk loci that are genome-wide significant in whole genome case-control scan. The scatter plot shows the standardized test statistics (y-axis) by autosomal base pair position (x-axis) for four genome-wide significant loci. The test statistics are the IBD rate difference (purple), the IBD rate in cases (red), and the IBD rate in controls (blue). The horizontal purple lines are the genome-wide significance threshold in the case-control scans. The DCDC2C, COLEC11, ADAMTS6, ICE1, NBAS, ASTN1, and BRINP2 genes are shown in shades of gray. The subplot titles give the ancestry cohort.