On 2015 Sep 26, Eric Fauman commented:

I applaud the authors for identifying novel genetic associations with metabolites but I disagree with their interpretations and conclusions in several regards.

As tempting as it is to use eQTL data to assign causal genes to SNPs it is frequently seen that SNPs tag expression of unrelated genes as often as they tag the true causal gene for the given trait.

In this study the most obvious example is at rs2066938 where the authors report eQTL associations with 5 egenes (RNF10, MLEC, UNC1198B, CAMKK and COQ5), but not ACADS which is almost certainly the true causal gene, as the authors acknowledge in the text.

At the ARG1 and CRAT loci, other genes have stronger eQTL signals so here too the eQTL data is incomplete.

The ALMS1/NAT8 locus is less clear, but previous authors have assigned this locus to NAT8 given the association with N-acetylornithine and NAT8's presumed acetylation function. The biochemical linkage of N-acetylornithine and arginine in the urea cycle suggests that NAT8 is also the causal gene for this paper. If NAT8 is truly the causal gene, the eQTL data missed it at this locus.

A striking example of the over-reliance on eQTL data in this paper is at the "PPP1R16A" locus which associates with the ratio of aspartic acid to alanine. This SNP is in fact just upstream of GPT which encodes glutamic-pyruvic transaminase, also known as alanine transaminase. GPT is a far more plausible causal gene even though it is not one of the 10 egenes listed for this SNP. Interestingly, there is a coding variant in GPT in reasonable LD with the lead SNP (rs1063739, r2=0.77).

In fact 6 of the loci are linked to coding variants in the most probable causal gene (NAT8, GPT, ACADS, SLC22A16, MCCC1 and CPS1).

Again, it's great to see new SNP-metabolite associations still emerging. However any GWAS interpretation must make use of all biological lines of evidence and not rely only on one or two types of data or analysis.

On 2015 Sep 26, Eric Fauman commented:

I applaud the authors for identifying novel genetic associations with metabolites but I disagree with their interpretations and conclusions in several regards.

As tempting as it is to use eQTL data to assign causal genes to SNPs it is frequently seen that SNPs tag expression of unrelated genes as often as they tag the true causal gene for the given trait.

In this study the most obvious example is at rs2066938 where the authors report eQTL associations with 5 egenes (RNF10, MLEC, UNC1198B, CAMKK and COQ5), but not ACADS which is almost certainly the true causal gene, as the authors acknowledge in the text.

At the ARG1 and CRAT loci, other genes have stronger eQTL signals so here too the eQTL data is incomplete.

The ALMS1/NAT8 locus is less clear, but previous authors have assigned this locus to NAT8 given the association with N-acetylornithine and NAT8's presumed acetylation function. The biochemical linkage of N-acetylornithine and arginine in the urea cycle suggests that NAT8 is also the causal gene for this paper. If NAT8 is truly the causal gene, the eQTL data missed it at this locus.

A striking example of the over-reliance on eQTL data in this paper is at the "PPP1R16A" locus which associates with the ratio of aspartic acid to alanine. This SNP is in fact just upstream of GPT which encodes glutamic-pyruvic transaminase, also known as alanine transaminase. GPT is a far more plausible causal gene even though it is not one of the 10 egenes listed for this SNP. Interestingly, there is a coding variant in GPT in reasonable LD with the lead SNP (rs1063739, r2=0.77).

In fact 6 of the loci are linked to coding variants in the most probable causal gene (NAT8, GPT, ACADS, SLC22A16, MCCC1 and CPS1).

Again, it's great to see new SNP-metabolite associations still emerging. However any GWAS interpretation must make use of all biological lines of evidence and not rely only on one or two types of data or analysis.