Ideas For Refining Causative Mutation By Gwas

Question

[Pcrer]

We ran a GWAS on farm animals using thousands of individuals and a QTL was found for a specific trait (complex trait).

But I know it is not so easy fine mapping the causative mutation(variation) due to the density of chips(only 50K, for the animals not so dense like human chips).

And next, we resequenced several genes in this QTL-genomic region(3 Mb) just using a DNA panel (16 individuals) and hundreds of SNPs were detected.

Finally, only a few missense SNPs were selected and typed but non-significant.

Now, I am concentrating on one gene which the two most significant markers are within one of introns, we genotyped the unique missense mutation (non-significant) and 1 promoter SNP which highly linked with the second top marker and of course it is significant but have't reach the higher level compared top marker.

So we propose this promoter SNP is not the causative one and we have to continue seeking.

I knew it is hard to find a true causative mutation but we might quite close. Thus the next step we may select some of putative SNPs for genotyping and I am still confusing about it.

Someone who works for GWAS especially animals can give some hint or better ideas to do it well.

Thanks a million.

Answer 1

[Larry_Parnell]

One need not work with GWAS or animals to assist in this question. This is a common problem in genetics: what genetic difference(s) is the source of the observed difference in phenotypes.

With recent work by Folkersen et al (2010, Circ Cardiovasc Genet), it is apparent that many such genetic differences associating with traits also affect gene expression - and at a considerable distance. Also see the expression QTL (eQTL) work by Stranger et al, among many others.

What I suggest is to ask which genes within a zone defined by the GWAS show significant or interesting or relevant expression in pertinent tissues, cell types or under some relevant condition. For example, if the QTL is for meat quality or fat marbling in beef or pork, then I would look for genes involved in muscle development, expressed in muscle, responsive to fat in the diet and I would look for such genes in a variety of model organisms. As we know, many fat metabolism genes in the mouse or pertinent to our work on fat metabolism and diet choices in humans, for example. Thus, a gene involved in the relevant biology of your phenotype is likely to carry variants that affect the phenotype you're studying in detail.

The key here is gene expression and the SNP may reside in the 3'-UTR (mRNA stability or microRNA interaction), in an intron (mRNA splicing or transcription factor binding), in a promoter, or even far away (in an enhancer).

[Pcrer]

Thanks for your generous reply. There are remind me should think about all the situation may happen. and consider the function part.

[Pcrer]

And the most important, if u wanna investigate the expression of genes, you need decide which gene is the true causative gene and putative SNPs, analyze whether there are expression differences between different genotypes for this marker.

Answer 2

[Khader Shameer]

Here is my two cents. Have you explored the possibility to perform imputation of your GWAS data to obtain significant hits by extrapolating the experimental data with a reference dataset. I am not sure if there is an extensive project like HapMap for animals that you are working on. This review article will be a good start for the concepts of imputation of GWAS. You may also go through this application note from Illumina that describes the concepts and the technical details of imputation. Different tools are available for imputation of GWAS data (for example IMPUTE or ProBABEL)

[Larry_Parnell]

Imputation can be both easy and difficult in animals. The easy part are teh super long stretches of LD in, for example, dogs (a rather in-bred group). The hard part is the obvious - lack of data on variants. My feeling is the best examples in this regard are to be found in dog and cow; horse trait mapping is a bit behind but may also offer ideas. Mouse is an obvious option but the data and resources are too dense to compare to the 50K set described in the question.

[Khader Shameer]

Thanks for this Larry !