Tests of association for rare variants: case control mutation screening

Two recent reviews, one in Nature Reviews Genetics from Bansal et al. (Statistical analysis strategies for association studies involving rare variants Nature Reviews Genetics 11, 773-785 (2010))¹ and one elsewhere², examined the emerging area of rare variant association studies. These reviews nicely describe the progression from association studies for common SNPs towards those for rare variants. We would like to add to these discussions a strategy that has been used by several groups for rare variant case-control association studies, developed independently of genome wide SNP association studies (GWAS) and largely confined to cancer genetics. We refer to this strategy as case-control mutation screening (CCMS).

Ideas contributing to CCMS are as follows. First, linkage analysis shows that evidence from many, individually very rare, sequence variants at the same locus can be combined³. Second, clinical testing of susceptibility genes such as BRCA1 and BRCA2 has shown that testing can be based on sequencing rather than genotyping. Third, the integrated evaluation of unclassified variants in BRCA1 and BRCA2 has shown that in silico assessment of rare variants — currently, rare missense substitutions (rMSs) — can be used to grade variants on the basis of predicted severity without attempting to dichotomize as deleterious versus neutral⁴. Finally, lessons from GWAS tell us that well-powered CCMS studies will be large, usually multi-center and often multi-ethnic, and therefore must be analyzed by statistical methods that allow for covariates.

The development of CCMS can be traced through the efforts of the genetics community to understand the contribution of heterozygous sequence variation in ATM to risk of breast cancer (Table 1). Analysis of ATM CCMS data started with a case-control study that employed a cohort allelic sums test limited to protein truncating variants plus variants that clearly damage splice junctions (T+SJVs)⁵, Analyses progressed to a two-pronged strategy of analyzing the pool of ATM T+SJVs in one logistic regression and the pool of rMSs in a second logistic regression⁶. The subtlety in this latter approach lies in combining all of the rMSs into a single categorical variable that incorporates prior information, such as sequence conservation, and grades the severity of rMSs from likely harmful to likely benign^4,6. This variable is easily assessed in a logistic regression test for trend, thus minimizing the multiple testing problem while accommodating epidemiologic covariates. We believe that this form of CCMS, augmented by steadily improving statistical methods^7,8, will be useful for identifying genes that harbor variants conferring intermediate risk, especially those in which most pathogenic variants are rare and either reduce or ablate function.

Table 1.

Key CCMS and related analyses of association of rare ATM variants with breast cancer

Study	Year of publication	Key elements
FitzGeral et al.5	1997	CCMS study, CAST test of T+SJVs.
Gatti et al.9	1999	Theoretical study predicting relative importance of ATM missense substitutions.
Sommer et al.10	2003	CCMS study incorporating prediction of missense substitution severity.
Renwick et al.11	2006	Well-powered CCMS study, combined CAST of T+SJVs and pathogenic rMSs.
Tavtigian et al.6	2009	CCMS meta-analysis, combined CAST of T+SJVs and graded rMS trend test.
Bernstein et al.12	2010	Case-case analysis, supports importance of “predicted damaging” rMSs.

CAST, cohort allelic sums test. T+SJVs, protein truncating plus splice junction variants. rMSs, rare missense substitutions.

Going forward, improving the accuracy and scope of methods of predicting sequence variant severity is a key goal. To this end, the Critical Assessment of Genomic Interpretation community exercise (http://genomeinterpretation.org/) will illuminate the capabilities of current approaches and inform their further development. An important additional issue is that methods for predicting gene dysfunction must be sufficiently transparent that other researchers can readily replicate predictions and judge the effects of hidden multiple testing, introduced by prediction of sequence variant severity, on CCMS data analysis.