Connect Eqtl And Gwas Study

Question

[Bioscientist]

eQTL tries to regress each gene expression against each SNP, in order to find those regulatory elements. And eQTL uses "normal" samples, right? (by normal I mean "no disease" like those in 1000genome project)

GWAS compares SNPs between normal(control) and disease(test) samples, trying to find out those higher-frequency variants enriched for diseases.

Suppose now I've GWAS results at hands, say, top-ranking SNPs associated with schizophrenia, and I would like to find GWAS SNPs residing within eQTL.

So my questions are:

1. Can I use already available eQTL database, or should I establish such database myself? My concern is, each eQTL study from other groups has its own bias (sample size, population structure, ethnics...); so as far as I know, different eQTL database can vary dramatically, correct?

2. If to choose available eQTL database, what criterion should I take into accounts? For example, if the schizophrenia samples in GWAS are neuronal tissues, then should I refer to neuron-specific database? Or if it's blood, then blood-specific database? Also need to guarantee the ethnics consistent? Say both GWAS and eQTL database should be derived from Caucasians?

Thanks

Answer 1

[Larry_Parnell]

With regard to question 1, you can use available eQTL data but keep in mind that the typical tissue or cell type tested is not so relevant to schizophrenia (brain). Most eQTL data for human gene expression is taken from PBMCs and lymphoblastoid cells. Yes, different eQTL databases can show substantial variation. So, look across several and integrate results. In short, an eQTL in PBMCs may not be relevant to brain expression, but it might and so it is worth considering its GWAS overlap as you select candidate loci for validation/replication.

Q2: I would try to keep the ethnicity as consistent as possible. Linkage disequilibrium becomes an issue when you change ethnic groups. The eQTL SNP and the GWAS SNP may not be the same SNP, but in high LD, for example. However, knowing the LD constraints when you have GWAS in one group (CEU or CEU-like) and eQTL info from a very different population can be difficult to fully incorporate into the analysis. Still, one could use such data from a markedly different population and simply reduce its value in making the GWAS-eQTL link.

[Tommy Carstensen]

@Larry_Parnell What is the best eQTL database out there?