Excess average estimated identical-by-descent in genotype data

Question

(Cross-post with Biostars: https://www.biostars.org/p/470788/)

We have some genotype data that we are putting through quality control in PLINK 1.9. As part of this QC, we have limited the data to subjects of self-reported and genomic European ancestries, and have run basic QC steps, removing call rates below 95% for individuals and variants, and removing variants excessively out of Hardy-Weinberg equilibrium (< 10^-10).

Pruning for LD and limiting to the autosomes, we have calculated estimated IBD between our participants and the estimated the average pairwise IBD for each participant. We have found that an unusual number of our participants (233/13049) show a high mean pairwise IBD with others (>3SD from the cohort mean, > 0.028 in this instance, maximum 0.075). Examining examples of the individuals in question shows that this is driven by low-level relatedness (~0.03-0.1) with many other individuals, rather than a small subset of relatives in the data. In addition, the relatedness in each case stems entirely from IBD1 (one shared variant) and not at all from IBD2 (two shared variants). Individual average IBD correlates negatively (-0.69) with the genome-wide inbreeding coefficient.

I have one idea about what could be causing this (partial cross-contamination of the samples), but I am interested to know if others have other ideas what would be the cause?

Answer 1

A clear negative association within an inbreeding coefficient suggests convergent evolution driven by shared selective pressure. There is an outside chance of purifying selection, or else balancing selection, but selection is the key word. The phenotype of the locus needs investigating, i.e. outside nearly neutral theory what is constitutes positive selection?

Allopatry caused by genetic isolation and genetic drift is your null hypothesis. However Darwinian positive selection is the alternative hypothesis.

I would be wary against suggesting contamination unless you've a clear suspicion this is occuring. I do accept that in human genetics its often difficult to avoid, particularly in ancient DNA type work. In this case you would need to try and control for it. The argument I would put against contamination is that 'genetic relatedness' is localised within the genome, whereas contamination would be a genome-wide phenomena.

Convergence is common within for example HLA genotypes (T-cell) loci between global populations.

You could perform formal tests for postive selection either within pop gen (the locus will associate with significant homozygosity) or phylogeny (point mutation).