This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Forum: Why isn't GWAS used in oncology?

From what I can gather, GWAS is not used in oncology because ~ cancer is characterized by rare/ random mutations that amount to an overall burden on the genes that they up/down regulate. Is this assumption correct?

gwas association oncology studies somatic cancer
• 4,637 views • link
updated by LChart LChart • written by LayneSadler LayneSadler

this is one of those, just need to read and think type things.

...so, when you do a GWAS, what is enrollment like, and whats the goal?

• 0 views • link
written by LauferVA LauferVA

I'm not sure it's correct to say that GWAS isn't used in oncology.

• 0 views • link
written by i.sudbery i.sudbery

the classic GWAS studies were done with arrays - only several million of pre-defined markers are genotyped across the genome and the results are mapped to actual causing variants using linkage disequilibrium principles. not gonna work for cancer.

don't get me wrong many cancer driver mutations are recurrent, but the analysis of differences in frequencies is just a part of GWAS.

• 0 views • link
written by German.M.Demidov German.M.Demidov

1 answer

Genome-wide association studies are used all the time in oncology. Using WGS to look for recurrent somatic mutations (at a single-variant or gene level) that associate with severity, resistance, prognosis, or other property of the tumor is, by definition, a genome wide statistical scan for association with outcome.

That said, most of the time "GWAS study" implies a large-scale microarray-based study to associate germline inherited mutations with a phenotypic outcome. While germline inherited mutations may have modifier effects on disease progression and/or treatment responses, of more direct interest are the mechanisms of causal somatic mutations in the tumor.

One of the reasons that microarrays work well for germline studies is the presence of LD such that variants on the array will "tag" causal variants missing from the array, and even collections of low-frequency causal variants can be partially tagged. However somatic mutations are rare, and are almost always not present on existing microarrays. Further, even recurrent mutations can be assumed to occur on a random genetic background; so there is no statistical LD enabling these somatic mutations to be 'tagged' by microarrays -- hence why studies of tumor genetics typically use sequencing as opposed to microarray.

• 0 views • link
written by LChart LChart

+1 for mentioning LD for microarray-based studies.

Quick follow-up question: In the context of somatic mutation-based GWASs (e.g. using WGS from tumor), do they ever consider the clonality of individual mutations (i.e. using mutation allele frequency or MAF)? Like OP, I didn't know GWAS was used in oncology "all the time". I'm curious to know if they are specific features of such GWASs. Also, I'd greatly appreciate if you can point us to one or two nice papers that performed GWAS in oncology -- if you know such papers on top of your head.

• 0 views • link
written by sbstevenlee • sbstevenlee

I think its misleading to say that somatic, mutation-based GWASs are routine. Or rather, perhaps its misleading to say that large, WGS-based mutation studies are GWAS. Even if they seem to be a genome-wide statistical scan for association with an outcome, I think the term GWAS suggests a certain set of methodologies, that does not really bare that much resemblance to the methodologies used in WGS studies of tumors.

GWAS tests for an association between _inheriting_ a particular SNP (single nucleotide polymorphism), or haplotype, and probability of expressing a particular trait (e.g. having a disease). They do this by comparing the frequency of having a particular variant at a particular location in a set of individuals with the trait and a set without.

WGS scans of cancer patients don't compare cases and controls. Rather, they compare healthy and tumorous tissue from the same person. They then look across their cohort to identify significantly mutated genes. That is, genes that more are frequently mutated than would be expected by random chance, given the number of mutations, irrespective of the location of those mutations within the gene. Indeed, two individuals carrying a mutation at the same point independently is unlikely.

These studies do also sometimes identify recurrent mutations (generally at the level of amino-acid change, rather than base change), so called hotspots, that could be thought of as somewhat analogous to hits from GWAS, but the statistical reasoning that leads to their identification is different, and also two patients that carry a mutation at the same place are carrying different mutations, even if they result in the same change. The genetics is very different from inheriting a SNP.

GWAS studies, as geneally understood, are however, using in oncology. Generally to identify SNPs that are informative for cancer risk. For example, breast cancer GWAS like this one, are of great use in designing screening programs. In this case the PRS built from the GWAS hits can alter the age at which at risk individuals need to commence screening by decades, massive reducing the risk of unneccessary treatment in the low risk, and decreasing the chance of missed tumors in the high risk.

• 0 views • link
written by i.sudbery i.sudbery

Right, I'm talking about producing an actual case-control association between each variant and the phenotype of interest using a tool like PLINK/SAIGE. Not just some burden of SNP counts.

• 0 views • link
written by LayneSadler LayneSadler

Since LD deals with haplotype-inherited SNPs, I don't understand why LD has anything to do with cancer. Doesn't the presence of a tumor mean any SNPs can arise regardless of what is inherited?

• 0 views • link
written by LayneSadler LayneSadler

Yes, but also .. I guess no.

First, there are familial cancer syndromes. Li Fraumeni, etc. runs in families. Thus, these are germline variants and subject to the same forces as any other germline variant... so LD is relevant there and could be exploited to analyze a GWAS of familial cancer predisposition.

Ok, onto somatic. Here's the thing. All the cancer cells arise from the first cell to transform. This cell will have (presumably) some early and some driver events in its genome already.

Any somatic variant that arises after this (in that cell or any daughter cell) will be found in together with those first variants. Any subsequent mutations that arise will also be in linkage with the primary events, but they will only be present in a subset of the daughter cells ... make sense? you will have a clonal, hierarchical structure that does in fact deviate from the Law of Independent Assortment.

• 0 views • link
written by LauferVA LauferVA

LD is basically what allows "typical" microarray GWAS studies to work.

Somatic mutations are typically not in statistical linkage with anything, so the same technique doesn't work in the setting of associating somatic mutations with tumor phenotypes.

• 0 views • link
written by LChart LChart

Log in to answer this question.

More posts like this