Phosphatidylinositol 3-kinases (PI3Ks) are lipid kinases that regulate signaling pathways important for neoplasia, including cell proliferation, adhesion, survival, and motility (1–3). To determine if PI3Ks are genetically altered in tumorigenesis, we sequenced PI3K genes in human cancers and corresponding normal tissue. Hidden Markov models were used to identify eight PI3K and eight PI3K-like genes, including two uncharacterized genes, in the human genome (3)(table S1). We initially examined the sequences of 117 exons that encode the predicted kinase domains of these genes in

35 colorectal cancers (3). PIK3CA, which encodes the p110 catalytic subunit, was the only gene with somatic (ie, tumor-specific) mutations. Subsequent sequence analysis of all coding exons of PIK3CA in 199 additional colorectal cancers revealed mutations in a total of 74 tumors (32%)(Fig. 1, fig. S1, and table S2). We also evaluated 76 premalignant colorectal tumors; only two mutations were found, both in very advanced tubulovillous adenomas greater than 5 cm in diameter. Thus, PIK3CA mutations generally arise late in tumorigenesis, just before or coincident with invasion. Mutations in PIK3CA were also identified in 4 of 15 glioblastomas (27%), 3 of 12 gastric cancers (25%), 1 of 12 breast cancers (8%), and 1 of 24 lung cancers (4%)(table S2). No mutations were observed in 11 pancreatic cancers or 12 medulloblastomas. In total, 92 mutations were observed, all of which were determined to be somatic in the cancers that could be assessed. The sheer number of mutations observed in this gene strongly suggests that they are functionally important. This hypothesis is buttressed by two lines of evidence. First, analysis of the ratio of nonsynonymous (NS) to synonymous (S) mutations is a good measure of selection during tumor progression, as silent alterations are unlikely to exert a growth advantage. The ratio of NS to S mutations in PIK3CA was 92 to 3, far higher than the 2 to 1 ratio expected by chance (P 0.0001). Second, the prevalence of NS changes located in the PI3K helical and kinase domains was 120 per Mb of tumor DNA, more than 100 times higher than the background mutation frequency of nonfunctional alterations observed in cancer cells (P 0.001)(4).

The positions of the mutations within PIK3CA imply that they are likely to increase kinase activity. No truncating mutations were observed and 75% of alterations occurred in two small clusters in the helical and kinase domains (Fig. 1 and table S2). The affected residues within these clusters are highly conserved evolutionarily. The clustering of somatic missense mutations in specific domains is similar to that observed for activating mutations in other oncogenes, such as RAS, BRAF, CTNNB1, and members of the tyrosine kinome (3). To directly test this hypothesis, we expressed the wild-type p110 or a “hot-spot” mutant (H1047R) in NIH3T3 cells and assessed lipid kinase activities in immunoprecipitated PI3K complexes. Expression of mutant p110 conferred more lipid kinase activity