Johns Hopkins University, Baltimore, Maryland
A geneticist wonders what it takes to prove causality.
In the post-genomic era, we are increasingly confronted by a torrent of variation data, originating from gene sequence, copy number and methylation patterns. To complicate matters further, I anticipate that a notable fraction of variation among individuals will be found to be relatively rare events. This would severely hamper our ability to implement statistical methods to associate variants with disease pathogenesis.
A recent paper by Carpten et al. (Nature 448, 439–445; 2007) highlights just how difficult solving this problem can be. The authors found a somatic missense mutation in AKT1 in a small number (2–6%) of breast, colon and ovarian cancers, and expended considerable effort establishing its link to tumour development. Experiments included solving the AKT1 protein’s crystal structure; calculating the predicted effect of the missense change on the protein’s conformation and binding abilities; gauging phosphorylation rates of the protein; identifying cellular localization; measuring transformational competency of the mutant versus wild-type allele; and checking the mutant protein’s ability to induce cancer in a mouse model.
In light of recent efforts to understand the total mutational load in cancer (for some examples see F. Dahl et al. Proc. Natl Acad. Sci. USA 104, 9387–9392; 2007; C. Greenman et al. Nature 446, 153–158, 2007; T. Sjöblom et al. Science 314, 268–274; 2006), these data are both exciting and sobering, because the idea of performing such an exhaustive analysis on a large allelic series is not tenable. The challenge, therefore, is to solve this problem by developing functional assays that are physiologically relevant; amenable to at least medium throughput; and applicable to a range of mechanistic questions (not just neoplasia, for example).