Gwas Peak In Only One Population, Why?

Question

[Sally]

Hi,

This is a pretty general question I know, but I'm wondering why I might be seeing a statistically significant peak from a GWAS in one population and nothing in another. Statistical power to detect an association is roughly the same in both populations.

Hope somebody can give me some insight.

Sally.

Answer 1

[David Quigley]

• The locus is real, but only has an effect in one population. You could hypothesize that this is due to interaction with other alleles associated with that population, or due to an environmental difference between the populations, or due to differences in the condition. This has been seen several times (though no one has adequate explanation that I'm aware of) in common diseases such as various cancers. Alternately, the effect is real in both populations but the effect is so modest that other environmental or genetic factors hide the signal in one population. That's a different situation from true interaction.

• The locus, despite apparent statistical significance, is in fact only appearing by chance. That's why you need replication for these studies.

• There is a systematic difference in genotyping technique between populations, e.g. different platforms, different DNA extraction methods, hybridization at different genome cores, etc. This difference is affecting this locus, either creating a new signal or removing one selectively.

• The significant probe is flawed somehow: it cross-hybridizes, it has weak signal that is noisy, etc. This gives a false positive result in one population.

Have a look at the raw genotype data to see if the calls are credible in both populations. Look for biases associated with population. Look for systematic clinical/macroscopic differences in your phenotype between populations For example, if this were breast cancer, is one population African-derived and another European-derived? If so, not only do you have different allele frequencies as Jorge points out, you also have different predispositions to ER-positive vs ER-negative cancer (probably due in part to the genetic architecture, but also due to environmental differences). If both populations were Caucasian, is there an ascertainment bias due to different medical systems diagnosing patients differently? Etc...

Answer 2

[Jorge Amigo]

if the statistical power of detection is based on sample number only it may not be enough. allele frequencies vary among different populations, and maybe the alleles that could be associated with the biological trait of interest may not be present at the appropriate frequencies on both populations.

this answer may be obvious, so please forgive me if I didn't get your question right. maybe a deeper description of your experiment would allow a better explanation.

Answer 3

[Larry_Parnell]

Thanks, David, for providing a great answer. I have less to type!

We have often seen examples of David's first point and in many cases environment is the answer. We know this because we look for it. One of our populations has higher fish consumption and lower fat consumption than the US average and so an association appearing there may be real or may also be influenced by those dietary parameters. Thus, replication in a second population may only happen when we split according to dietary fat intake. These often become gene-environment interactions.

Another possibility is the probe in one population is linked to a copy number variant that is causal or more so than the SNP itself and where that CNV is quite rare in the second population.

You could also look for an association in both populations for a relatively similar phenotype. We look at blood lipids a lot and often see correlations between HDL-cholesterol and triglyceride measures. One SNP may associate with both in one population but only one HDL-C in the second.