Thread by @jakphd

Our #COVID19 pan-ancestry exome-wide meta-analysis across 586,157 individuals (20,952 SARS-CoV-2 positive cases, 4928 hospitalized, 1304 severe) is now out in @AJHGNews.

What did we find? To be frank, not much, but here's a 🧵.

cell.com/ajhg/fulltext/…

Within 3 cohorts with #COVID19 phenotypes (@uk_biobank, @GeisingerHealth, @PennMedicine), we analyzed rare variants (MAF<0.5%; ~7million) and burden tests within each ancestry separately and meta-analyzed our results. Incorporating non-EUR data increased our case N by ~10%.

For any given burden test or rare variant to be significant, we set a conservative Bonferroni significance threshold at 9.6e-10 given the number of traits & variants tested.

Here's Manhattan plots for 20,952 SARS-CoV-2 cases vs 565,205 controls with nothing remotely significant.

Focusing on 4928 hospitalized COVID19 cases sadly didn't yield anything Bonferroni significant here either.
In case you are curious, the top gene was MARK1 (P=2.9e-8) and the top individual RV was a splice region variant in WDR78 (P=2.8e-9).

Lastly, 1304 severe COVID19 cases (those on a ventilator or who had passed away) compared to 528,758 controls was similarly flat.

The sub-significant tower in the burden tests is TLR7 (P=4.3e-8), a gene first implicated by @ahoischen and co. in ja.ma/34OEDJT.

A full list of sub-significant variants and burden tests that would pass a standard genome-wide significance threshold of 5e-8 (which is really inappropriate for rare variants) can be found in Table 1. But please, treat these like you would any other non-significant variant!

Why do I say this?

Because even our replication attempts of the most biologically interesting rare variants failed - huge thanks to @GBL_Unit, @BrentRichards19, @GPovysil, @kirylukk, @kennethbaillie, @AMeynert and co for helping with these, sadly failed, replications!

Even had we focused on say 1) 281 genes with GWAS loci from @covid19_hgi, or 2) 32 COVID19 therapeutic-target genes or genes involved in SARS-CoV-2 (e.g., ACE2, TMPRSS2) we still wouldn't have found anything statistically convincing (Tables S6 and S8).

Lastly, we tried to replicate the other COVID19 exome studies such as the type 1 interferon signaling @ScienceMagazine paper from Zhang et al. 2020 that found a small increase of rare pLoFs in 13 interferon genes (P=0.01).

science.sciencemag.org/content/370/65…

Despite having 8x cases and 1000x controls as the original paper, we couldn't replicate this association regardless of using 1) PTVs or PTVs+missense, 2) restricting to singletons, 3) only using severe cases (Table 2 and Table S7).

Our study is now the 2nd paper reporting an inability to replicate the biologically-attractive work of Zhang et al., @ScienceMagazine 2020. @GPovysil, @kirylukk, and co. also couldn't replicate it as seen in their recent paper jci.org/articles/view/…

A huge amount of work across all of @Regeneron went into this paper and couldn't have been done with the help from my fellow co-first author, Julie Horowitz, and leadership from Manuel Ferreira, @aris_baras, @marchini, @jgreid and so many others.

However, the real heroes of any exome paper are those working tirelessly on QC!

Huge props to @jdbackman and @marchini for building a SVM to perform some fantastic exome QC (in the same manner as used by @gnomad_project and TOPmed). You can trust exome data without it!

Lastly, a sensible meta-analysis of rare variants and burden tests wouldn't have been possible without REGENIE developed by Joelle Mbatchou and @marchini.

You can find REGINI on github here: github.com/rgcgithub/rege…