For context: My colleagues and I have been running some GWAS' on a couple of different cohorts for a meta-analysis over the past year or so. I personally have conducted the GWAS of our phenotype in UK Biobank while one of my colleagues has conducted the analysis in All of Us (AoU).
The problem: One aspect of what we are trying to do is run a total GWAS of our phenotype across our cohorts (and meta-analyse them) and then also conduct sex and age-stratified GWAS of the same-phenotype. I.e. re-run the GWAS but only for individuals of a specific age-band or for males/females specifically. This has worked entirely fine across multiple different cohorts in this phenotype with one single exception.
As mentioned, my colleague has been using REGENIE to conduct the GWAS in AoU and the results for almost every single version of the GWAS look eminently sensible. Below, is the Manhattan plot for the complete GWAS (including every individual).
However, when the cohort is stratified by age the results look like this:
This second plot is the same phenotype as in the first plot but ONLY male individuals (the female-only plot looks pretty much exactly the same). Quite obviously we are seeing some absolutely massive P-value inflation across the entire genome, but particularly in the X-chromosome, which just looks absolutely insane.
We hit upon this issue a few months ago and she has since re-run the analysis/double-checked everything and produced exactly the same results, so we are actually at a complete loss as to why this is happening. So I am hoping that we can have some suggestions here as to why this might be an issue.
- We don't think it's an issue with the phenotype/distribution because then the total phenotype plot wouldn't look so sensible
- We don't think it's an issue with REGENIE, because the pipeline seems to be working correctly across each phenotype
- We also don't think it's an issue with the genotypes, because again this X-chromosome inflation is only seen in the sex-stratified analysis
One question we had was whether the fact that AoU is WGS as compared to array genotypes might have some influence here, but we have another cohort which is WGS and it didn't produce anything like this (although admittedly the GWAS there was run with SAIGE.
Any input/suggestions/thoughts would be super useful to try and sort this out!
Sex chromosomes are always a bit tricky to handle in associations studies.
How many male cases and male controls you have for this phenotype? Here, I would go back and look at the genotype data on chr23: can you check that your male cases and male controls roughly have the same number of variants?
Maybe try another GWAS tool (Saige, plink, PRsice etc...) to confirm that the inflation is not an artifact from REGENIE.
Please, keep us updated if you find the reason for this inflation, I am curious to know the reason behind such a massive inflation on just one chromosome.
And just to get this out of the picture: I guess your phenotype does not involve the duplication of a sex chromosome in males?