if these findings are correct, what will it mean for GWAS?

Well, it means it may be damn difficult to unravel the molecular mechanisms underlying many GWAS association signals. People are currently operating under the broad assumption that causative genes are likely to be fairly close to association signals, but this finding suggests that many disease-associated variants operate through effects on distant genes. I guess it will be especially difficult to track these genes down for diseases that mainly affect inaccessible tissues (e.g. brain, heart), from which it's very hard to collect gene expression data from large, well-controlled cohorts.

However, the effects may be most dire for the post-sequencing personal genomics industry. Cheap whole-genome-sequencing will throw up any number of rare variants in every customer, and assigning even probabilistic function to them will be incredibly difficult if the majority of expression-altering variants act in trans - so this is potentially a pretty depressing result for companies like Knome and 23andMe looking to convert genomic data into medical predictions.

There are some serious caveats with this study - e.g. it's unclear to me whether the method used here to study between-population variance will necessarily predict the architecture of within-population variance (I'd be interested in hearing p-ter's take on this), and it will be important to look at effects in multiple tissues - but overall it strikes me as rather worrying...

Hi Dan,
Good point. The authors try to address this point by looking at whether known cis and trans eQTLs differ in their fst distribution. For example,. if trans eQTLs tended to have higher fst this would lead to a more variation being explained by trans ancestry. They find no strong difference. Though this does not rule out the possibility that there are some highly differentiated pleiotropic trans eQTLs causing this result.

G

G, FYI, moderation is on, so your comment won't show up immediately. just thought you should know in case you didn't since you had 3 duplicate comments in a row submitted.

if a trait is primarily additive (e.g. IQ), will it be spared the slow-down in understanding its genetic architecture from these gene-expression differences?

has anyone seen a strong attempt to use gene expression variation to predict phenotype variation outside of tumors? like for example diabetes? not with beta cell RNA extracts, but with say white blood cells as the tissue?

for example, has wtccc started testing gene expression from their subjects?

thanks Razib. I didn't realize that moderation was on.

lol,

I haven't read the paper in detail yet, but this deCODE study analysing gene expression networks and their relationship with obesity seems pretty well-regarded in the field.

As for the WTCCC: most of the samples didn't have cell lines collected, only DNA (since the cost and ethical hassles associated with establishing cell lines are non-trivial). But there are several manuscripts in the works using gene expression data from other samples to inform analysis of the WTCCC (and other GWAS) results; you can get an idea of what to expect from this recent review.

This is consistent with the idea that trans variation segregates within populations, while cis variation differentiates species. I think G pointed this out in his post on the topic.

Trans variation should be common due to copy number variation. Multiple copies of the DNA that codes the trans regulatory elements should change the expression levels of the regulatory elements. Extra copies would also be free to mutate and take on new functional roles. Fast track adaptation.