PIK3CA and PTEN Gene and Exon Mutation-Specific Clinicopathologic and Molecular Associations in Colorectal Cancer

This article is featured in Highlights of This Issue, p. 3109

Phosphoinositide 3-kinase (PI3K) and PTEN are key positive and negative regulators of the PI3K pathway, respectively, involved in cell growth, survival, proliferation, motility, and glucose homeostasis (1). Activating mutations in PIK3CA, the gene encoding the catalytic p110α subunit of class IA PI3Ks, occur in approximately 15% of human colorectal cancers. The majority (∼80%) of PIK3CA mutations cluster in 2 “hotspot” regions, the helical domain (exon 9) and the kinase domain (exon 20; refs. 2–5). In addition, inactivating PTEN mutations or loss of protein expression is found in approximately 5% and 30% of sporadic colorectal cancers (6–8).

PIK3CA and PTEN status are implicated as markers of colorectal cancer response to mitogen-activated protein/extracellular signal–regulated kinase (MEK) inhibitors and anti-EGF receptor (EGFR) antibody therapy (cetuximab or panitumumab). Some but not all studies of anti-EGFR antibody treatment in metastatic colorectal cancer have reported resistance for tumors with PIK3CA mutation or PTEN loss (9–11), and similar results have been obtained in preclinical studies of MEK inhibition for KRAS-mutated colorectal cancer cell lines (12). Conversely, KRAS mutation has been associated with resistance to single agent PI3K inhibitors (13), but such cell lines may respond to concomitant inhibition of the PI3K and mitogen-activated protein kinase (MAPK) pathways (14). On the basis of these and other findings, drug combination studies of MAP2K and PI3K or mTOR inhibitors have been proposed tailored to tumor PI3K and MAPK pathway mutation status (15).

Despite their clinical importance, data on the baseline relationships of PIK3CA and PTEN mutations with clinicopathologic characteristics, molecular features, and prognosis remain controversial. For PTEN, inactivating mutations have been shown to be associated with microsatellite instability (MSI) status, occurring in approximately 20% of MSI-high (MSI-H; refs. 16–18) and less than 5% microsatellite stable (MSS) tumors (19, 20). However, associations with advanced tumor stage and poor outcome are less certain (11, 21–24). PIK3CA mutations have been reported to be more frequent in women (8, 25, 26), proximal (26), well-differentiated (5), and mucinous tumors (5), but these findings have not been consistent (4, 8, 27, 28). Some studies have reported significant cooccurrence of PIK3CA and KRAS mutations (5, 26, 28–32), whereas others have failed to show this (8, 25). Other, inconsistent data have been reported for PIK3CA mutation and MSI or CpG island methylator phenotype (CIMP) status (4, 8, 26, 28, 31, 32). Poorer outcomes have been suggested for PIK3CA-mutated tumors for patients with early-stage resectable disease (27, 28), but this was not replicated in a recent large series (32).

Emerging data indicate that the conflicting PIK3CA association data may in part be explained by intrinsic differences between exon 9 and 20 mutated cancers. Two recent studies on stage I–IV colorectal cancers have suggested that exon 20 mutations correlate more strongly with BRAF mutation, CIMP-high/-low (CIMP-H/-L) and MSI-H status (31, 32), and one reported that exon 9 mutations may more closely associate with KRAS mutation (31, 32). The latter has further been observed in a study on metastatic colorectal cancer (30). However, these suggestions have been based on pairwise comparisons between PIK3CA exonic mutant and wild-type tumors, rather than direct comparisons between exon 9 and 20 mutated cases required to substantiate such relationships.

The prognostic value of PIK3CA mutation may also vary by exon mutation status, with one study reporting an association of exon 20 but not exon 9 mutations with worse outcome in stage III tumors (33). However, another large study of patients with stage I–IV colorectal cancer found no effect of either exon 9 or 20 mutations on survival; instead, a small proportion of tumors with concomitant exon 9 and 20 mutations were reported to show poor outcomes (32). In addition, exonic location of PIK3CA mutation may influence response to anti-EGFR antibody therapy, with a recent analysis of multiple cohorts suggesting that mutations in PIK3CA exon 20 but not exon 9 are predictive of tumor resistance (34).

To comprehensively define the characteristics of PIK3CA- and PTEN-mutated colorectal cancer, we have analyzed 1,093 stage I–IV cancers for gene and exon mutation-specific relationships with patient clinicopathologic and tumor molecular features including KRAS, BRAF, MSI, and CIMP status. Associations with prognosis were investigated for persons with resected stage II/III cancer. To refine results on PIK3CA exon 9 and 20 mutation-specific associations, our data were integrated with 17 published cohorts comprising a total of 5,594 patients with colorectal cancer.

A total of 1,093 patients with colorectal cancer were recruited from the Royal Melbourne Hospital (Parkville, VIC, Australia), Western Hospital Footscray (Footscray, VIC, Australia), St Vincent's Hospital Sydney (Darlinghurst, NSW, Australia), and the Royal Adelaide Hospital (Adelaide, SA, Australia). The study was approved by ethics committee, and patients gave informed consent. Fresh-frozen (n = 744) or formalin-fixed paraffin-embedded (n = 349) tumor and matched normal specimens were sourced from hospital tissue banks. Ninety-four cancers were stage I, 278 stage II, 525 stage III, and 196 stage IV. Four hundred and forty-four cancers were from the proximal colon (cecum to transverse colon), 342 from the distal colon (splenic flexure to sigmoid colon), and 306 from the rectum (unavailable for 1 case). None of the patients had clinical features of familial adenomatous polyposis, Lynch, or other familial cancer syndromes. Treatment and follow-up data were prospectively recorded. Patient characteristics are summarized in Table 1.

Hematoxylin and eosin (H&E)–stained tissue sections were reviewed, and for tumor samples areas with more than 60% neoplastic cells were macrodissected. DNA was extracted using standard protocols and PCR-amplified for the Bethesda consensus panel of microsatellite markers (BAT25, BAT26, D2S123, D5S346, and D17S250) using fluorescently labeled primers (35). Reaction products were analyzed on a 3130xl Genetic Analyzer (Applied Biosystems). MSI-H was diagnosed if instability was evident at 2 or more markers.

Mutation screening for PTEN (exons 3–8), PIK3CA (exons 9/20), KRAS (codons 12/13), and BRAF (codon 600) was conducted by Sanger sequencing using the BigDye Terminator v3.1 Ready Reaction Mix (Applied Biosystems; details available from authors). Sequencing reaction products were analyzed on a 3730xl DNA Analyzer (Applied Biosystems), and detected mutations confirmed by resequencing of tumor and matched normal DNA from new PCR product.

Tumor CIMP status was determined using MethyLight real-time PCR for the Weisenberger and colleagues 5 marker panel (IGF2, SOCS1, RUNX3, CACNA1G, and NGN1) and the reference gene ALU (C-4; ref. 36). The percentage of methylated reference (PMR) was calculated for GENE:ALU ratio of template amount in a sample against GENE:ALU ratio of template amount in methylated reference DNA. Tumors with a PMR more than 10 for 3 to 5 CIMP markers were classified as CIMP-H, those with 1 to 2 methylated markers as CIMP-L, and 0 methylated markers as CIMP-0.

LOH at the PTEN locus was determined from single-nucleotide polymorphism (SNP) array data for tumor and matched normal samples (Human610-Quad BeadChip arrays; Illumina) using OncoSNP software (Isis Innovations) as described previously (37). SNP call rates for normal samples were more than 98% and for tumor samples were more than 97%.

Statistical analyses were conducted using the statistical computing software R (R Development Core Team, 2011). For univariate analyses, differences between groups were assessed using Fisher exact test for categorical variables and Kruskal–Wallis test for continuous variables. Multivariate analyses for association between gene mutation and clinicopathologic or molecular features were conducted using logistic regression. Integrated multicohort analysis for PIK3CA exon mutation-specific associations was conducted using the DerSimonian–Laird random effects pooling method (ref. 38; rmeta R package version 2.16) and mixed effects logistic regression with the association of interest treated as fixed effect and the study as random effect. Interstudy heterogeneity was assessed using Woolf test. Changes in proportions of tumor MSI/CIMP/KRAS/BRAF genotypes according to PIK3CA exon 9 or 20 mutation status were estimated using mixed effects logistic regression and a 3-stage hierarchical multinomial-Dirichlet model (Supplementary Methods). Outcome analyses for patients with resected stage II/III colorectal cancer were conducted for 5-year disease-free survival (DFS). DFS was defined as time from surgery to the first confirmed relapse, with censoring done when a patient died or was alive without recurrence at last contact. Survival curves were generated according to the method of Kaplan and Meier. Univariate survival distributions were compared using the log-rank test, and multivariate analyses used Cox proportional hazards models. All statistical analyses were two-sided and considered significant if P < 0.05.

A population-based series of 1,093 patients with stage I–IV colorectal cancers was screened for somatic mutations in PIK3CA (exons 9 and 20), KRAS (codons 12–13), and BRAF (codon 600) as well as MSI status. Seven hundred and forty-four tumor samples were analyzed for mutations in PTEN (exons 3–8), and 489 cases were analyzed for CIMP status. Mutations in PIK3CA were detected in 11.9% (130 of 1,093), PTEN in 5.8% (43 of 744), KRAS in 34.1% (372 of 1,091), and BRAF in 9.4% (103 of 1,093) of cases. MSI-H was exhibited by 14.7% (159 of 1,083) of cancers. CIMP-H and CIMP-L were observed in 17.4% (85 of 489) and 21.9% (107 of 489) of cases, respectively.

Out of 133 PIK3CA missense mutations, 66.2% (n = 88) occurred in exon 9 and 33.8% (n = 45) in exon 20 (Supplementary Table S1). Activating mutations at codons 542 (n = 30), 545 (n = 41), and 1,047 (n = 39), the established main mutation hotspots in p110α, accounted for 82.7% (110 of 133) of all exonic mutations. Among 130 PIK3CA-mutated tumors, 97.7% (n = 127) had 1 somatic mutation and 2.3% (n = 3) had 2 somatic mutations. The latter group comprised 1 case with mutations in both exons 9 and 20 and 1 case each with 2 mutations in exon 9 or 20.

Among 43 PTEN-mutated tumors, 76.7% (n = 33) had 1 and 23.3% (n = 10) had 2 or more detected somatic mutations (Supplementary Table S2). Missense mutations were the most common type of alteration (47.3%, 26 of 55), followed by frameshift (38.2%, 21 of 55), nonsense (9.1%, 5 of 55), and splice-site mutations (5.5%, 3 of 55). Of note, 49.1% (27 of 55) of PTEN mutations mapped to the phosphatase domain and 50.9% (28 of 55) to the C2 tensin-type domain of the protein (Supplementary Fig. S1). For missense mutations in the phosphatase domain (n = 22), only Ala126Ser and Arg130Gln (3 instances) were in the vicinity of the catalytic site, whereas most of the remaining changes were buried in the structure and predicted to destabilize the protein (Supplementary Table S3).

Data for allelic loss at the PTEN locus were available from SNP arrays for 631 colorectal cancers, with 15.1% (n = 95) of cases showing LOH. Presence of LOH was significantly associated with PTEN mutation, with 31.6% (12 of 38) of mutated tumors showing loss as compared with 14.0% (83 of 593) of wild-type tumors (P = 0.001). Overall, 52.6% (20 of 38) of PTEN-mutated tumors showed evidence of biallelic inactivation in the form of mutations (8 cases) or 1 mutation and LOH (12 cases).

Among 744 colorectal cancers with complete mutation data for both PIK3CA and PTEN, 16.9% (n = 126) of cases showed mutations in either PI3K pathway member. A subset of 1.2% (n = 9) of tumors harbored mutations in both genes, perhaps suggesting synergy, although this was only borderline statistically significant (P = 0.092).

Clinicopathologic and molecular correlates of PIK3CA mutation were evaluated overall and according to exonic location. Overall, PIK3CA gene mutation was significantly associated with older age at diagnosis, proximal tumor site, mucinous histology, and presence of KRAS mutation (P ≤ 0.037 for all comparisons), but not with patient gender, tumor stage, differentiation, BRAF mutation, MSI, or CIMP status (Table 1). In analyses stratified by MSI status, the relationships with older age, proximal tumor site, and KRAS mutation remained significant for MSS cancers (P ≤ 0.010 for all comparisons), whereas the relationship with mucinous histology was observed for MSI-H cancers only (P = 0.004; Supplementary Table S4).

Although some of these associations were evident for both exon 9 and 20 mutations separately, others seemed to be exon-specific (Table 2). Compared with PIK3CA wild-type status, mutation of either exon 9 or 20 alone was significantly associated with proximal tumor location (P ≤ 0.002 for both comparisons). In contrast, mucinous histology was specific to exon 20–mutated tumors when compared with both wild-type and exon 9–mutated tumors (P ≤ 0.045 for both comparisons). The significant relationship with older age at diagnosis and KRAS mutation was observed for exon 9 but not exon 20 mutation as compared with wild-type (age: P = 0.023 vs. P = 0.139; KRAS: P < 0.001 vs. P = 0.318), although the direct comparisons between exon 9- and 20-mutated cases did not reach statistical significance (age, P = 0.810; KRAS, P = 0.095). Common and differential relationships by exon mutation status remained apparent in analyses stratified by MSI status. In MSS cancers, exon 9 and 20 mutations were both associated with proximal tumor location (P ≤ 0.024 for both comparisons), whereas associations with older age and KRAS mutation were significant only for exon 9 (age, P = 0.014; KRAS, P < 0.001). In MSI-H cancers, exon 9 mutation occurred with KRAS mutation (P = 0.042), and exon 20 mutation with mucinous histology (P = 0.009; Supplementary Table S5).

In multivariate logistic regression analysis to assess for independent relationships between PIK3CA mutation and clinicopathologic or molecular features, proximal tumor location was significantly associated with overall PIK3CA gene mutation (P = 0.002; Supplementary Table S6). For a model excluding CIMP status, which was only available for a limited subset of patients, KRAS mutation was further independently associated with overall PIK3CA gene mutation (P = 0.034; Supplementary Table S6). In multivariate analyses by exonic site, a significant independent association with proximal location was observed for either exon 9 or 20 mutation compared with PIK3CA wild-type (P ≤ 0.028 for both comparisons), whereas an independent relationship with KRAS mutation was observed only for exon 9–mutated tumors (P = 0.01; Supplementary Tables S7 and S8). Direct comparison of exon 9 and 20 mutations provided suggestive evidence of an overrepresentation of KRAS mutation in exon 9–mutated tumors (P = 0.029), although this was not found in the model excluding CIMP status (P = 0.247; Supplementary Table S9).

In 744 patients with colorectal cancer evaluated for PTEN, presence of mutation was significantly associated with proximal tumor location, mucinous histology, BRAF mutation, MSI-H, and CIMP-H status (P ≤ 0.014 for all comparisons), but unrelated to patient age, gender, tumor stage, differentiation, and KRAS mutation (Table 1). When considering MSS cases only, no significant associations were observed, but for MSI-H cases, the positive association between PTEN and BRAF mutation remained significant (P = 0.019; Supplementary Table S10). Consistent with defective DNA mismatch repair, MSI-H cancers showed an overrepresentation of frameshift mutations in 2 poly-adenine tracts located in PTEN exons 7 and 8 as compared with MSS cancers (MSI-H: 46.2%, 12 of 26; MSS: 3.4%, 1 of 29; P < 0.001).

In multivariate logistic regression analysis including all clinicopathologic and molecular features, MSI-H remained independently associated with PTEN mutation (P < 0.001; Supplementary Table S11). For a model excluding CIMP status, BRAF mutation additionally reached statistical significance for independent association (P = 0.037; Supplementary Table S11).

For patients with resected stage II or III colorectal cancer, the influence of PI3K pathway mutation status on DFS was assessed using Cox proportional hazards analysis. Clinical follow-up information was available for 585 patients analyzed for PIK3CA mutation and 381 patients analyzed for both PIK3CA and PTEN mutation. The median duration of follow-up was 32 months for the former and 33 months for the latter patient group.

In both univariate and multivariate analyses adjusted for age at diagnosis, gender, tumor location, stage, differentiation, MSI status, and adjuvant treatment (Table 3), neither PIK3CA nor PTEN mutation were associated with the risk of recurrence [PIK3CA mutation: multivariate HR, 0.79, 95% confidence interval (CI), 0.50–1.25; PTEN mutation: multivariate HR, 0.78, 95% CI, 0.32–1.86). There was no evidence for differential outcomes when classifying PIK3CA mutations into those affecting exon 9 or 20 (exon 9 vs. wild-type: multivariate HR, 0.87, 95% CI, 0.52–1.46; exon 20 vs. wild-type: multivariate HR, 0.64, 95% CI, 0.28–1.47; Supplementary Table S12). Results were similar when separating patients into groups who did and did not receive adjuvant chemotherapy (Supplementary Tables S13–S15). Accordingly, overall PI3K pathway mutation status, defined as either PIK3CA mutation, PTEN mutation, or both, showed no association with outcome (multivariate HR, 0.85; 95% CI, 0.52–1.40; Supplementary Table S16).

To further refine our analysis of the clinicopathologic and molecular associations of PIK3CA exon 9- or 20–mutated tumors, we identified previous publications reporting such details using the PubMed literature search engine (www.ncbi.nlm.nih.gov/pubmed), the COSMIC database (www.sanger.ac.uk/genetics/CGP/cosmic) and cross-referenced citations. A total of 17 eligible cohorts were identified and data retrieved for patient gender, tumor stage, location, MSI status, CIMP status, BRAF mutation, and KRAS mutation (Supplementary Table S17). Tumors with mutations in both PIK3CA exons 9 and 20 were excluded. To maintain consistency for evaluation of KRAS and BRAF, mutation data were limited to KRAS codons 12 and 13 and the BRAF codon 600 (V600E). ORs for integrated studies were calculated for PIK3CA exon 9 or 20 mutation versus wild-type, and exon 20 versus exon 9 mutation using 2 alternative approaches, the DerSimonian–Laird random effects pooling method and mixed effects logistic regression.

In the integrated multicohort analysis using the DerSimonian–Laird approach (Fig. 1; Supplementary Table S18), either exon 9 or 20 mutation was significantly more common in proximal (exon 9: OR, 1.64, 95% CI, 1.27–2.13; exon 20: OR, 1.92, 95% CI, 1.28–2.88), CIMP-L (exon 9: OR, 1.87, 95% CI, 1.15–3.02; exon 20: OR, 3.11, 95% CI, 1.72–5.65) and KRAS-mutated tumors (exon 9: OR, 2.34, 95% CI, 1.91–2.86; exon 20: OR, 1.55, 95% CI, 1.16–2.07) as compared with PIK3CA wild-type. In addition, exon 20 mutation was associated with MSI-H (OR, 2.61; 95% CI, 1.86–3.66), CIMP-H (OR, 3.37; 95% CI, 1.80–6.28), and BRAF mutation (OR, 2.15; 95% CI, 1.42–3.24). Directly comparing exon 20- and exon 9–mutated cancers, exon 20 mutation was significantly more prevalent in MSI-H (OR, 3.47; 95% CI, 2.22–5.43), CIMP-H (OR, 2.69; 95% CI, 1.23–5.86), BRAF-mutated (OR, 3.15; 95% CI, 1.83–5.44), and KRAS wild-type (OR, 0.69; 95% CI, 0.50–0.96) cases. There was no evidence of significant interstudy heterogeneity for any of these molecular variables (Supplementary Table S18). Integrated multicohort analysis using the mixed effects logistic regression identified the same associations, and additionally provided suggestive evidence for an inverse association between exon 9 mutation and BRAF mutation and an underrepresentation of exon 20 mutations in stage III versus stage I/II tumors as compared with wild-type (Supplementary Fig. S2).

Analyses excluding data from this study reproduced these results, although in the direct comparison between exon 20- and exon 9–mutated cancers, the exon 20 mutation association with KRAS wild-type reached only borderline statistical significance (Supplementary Fig. S3).

To further define the differential associations of PIK3CA exon 9 or 20 mutation with tumor location, MSI, CIMP, KRAS, and BRAF mutation, combined analysis was conducted for cohorts with available complete patient data to test for differences in proportions of tumor compound genotypes [n = 1,318; this study, The Cancer Genome Atlas (TCGA; ref. 39) and Whitehall and colleagues (31)]. The 8 most numerous states plus an omnibus state consisting of the remaining genotypes were modeled using mixed effects logistic regression or, alternatively, a 3-stage hierarchical multinomial-Dirichlet model.

In the compound genotype analysis using mixed effects logistic regression (Fig. 2), exon 9 and 20 mutation were both significantly underrepresented in distal MSS/CIMP-0/KRAS^wt/BRAF^wt tumors as compared with PIK3CA wild-type (exon 9: OR, 0.49, 95% CI, 0.27–0.89; exon 20: OR, 0.18, 95% CI, 0.07–0.45). Relative to wild-type, exon 9 mutation was also significantly more common in proximal MSS/CIMP-0/KRAS^mut/BRAF^wt cancers (OR, 2.70; 95% CI, 1.41–5.18) and proximal MSS/CIMP-L/KRAS^mut/BRAF^wt cancers (OR, 3.81; 95% CI, 1.83–7.95). All of these associations were replicated using the multinomial-Dirichlet model (Supplementary Fig. S4). Direct comparisons between PIK3CA exonic mutants did not reach statistical significance, although power to detect such differences was limited because of small sample sizes.

This study presents the first substantial survey of somatic PTEN mutations in sporadic colorectal cancer and is the most comprehensive analysis to date of PIK3CA exon mutation-specific associations with patient characteristics, tumor molecular features and outcome. Our findings show significant gene and exon mutation-specific differences in clinicopathologic and molecular associations for PIK3CA and PTEN and clarify previous data on the prognostic value of these changes in resected stage II/III colorectal cancer.

Somatic mutations in PTEN exons 3 to 8 were detected in 6% of sporadic colorectal cancers, with approximately half localized in the phosphatase domain and half localized in the C2 tensin-type domain of the protein. Although there was a significant overrepresentation of tumors with “two hits,” consistent with a classical tumor suppressor role, 47% of PTEN mutated tumors had only one detected mutation and lacked LOH. Although some of these tumors may carry mutations in exons not evaluated in our screen or may have lost expression of the wild-type protein, a potential role for PTEN haploinsufficiency has been suggested in previous functional studies (40). Consistent with existing reports, PTEN mutations were significantly more common in MSI-H tumors as compared with MSS tumors (16–20). In keeping with the characteristics of the MSI-H/sessile-serrated pathway of colorectal tumorigenesis (41, 42), PTEN mutation was further associated with proximal tumor location, mucinous histology, BRAF mutation, and CIMP-H status in our cohort. Notably, within MSI-H cancers, the positive association between PTEN and BRAF mutation remained significant, suggesting selection for comutation of these PI3K and MAPK pathway members in this colorectal cancer subtype.

Consistent with previous studies, PIK3CA mutations in exon 9 or 20 were detected in 12% of sporadic colorectal cancers with mutation of the former exon about twice as frequent as the latter (2–5). Approximately, 0.3% of colorectal cancers had 2 detected mutations as reported by others (2, 11, 26, 30, 32). Overall, PIK3CA gene mutation was significantly associated with older age at diagnosis, proximal tumor location, mucinous histology, and presence of KRAS mutation, relationships which have been described in a number of reports (5, 26, 28–32). However, when stratified by exonic location, significant differences were evident between PIK3CA exon 9 and 20 mutation in our cohort. Compared with both wild-type and exon 9–mutated tumors, the relationship with mucinous histology was specific to exon 20 mutation. Furthermore, the associations with older age at diagnosis and KRAS mutation seemed stronger for exon 9–mutated tumors, although the direct comparison with exon 20–mutated tumors did not reach statistical significance. A PIK3CA exon 9 mutation-specific association with KRAS mutation has been suggested by others, but a statistical difference as compared with exon 20 mutation was also not formally shown in these reports (30–32). In an analysis stratified by MSI status, the association between PIK3CA exon 9 mutation and KRAS mutation was observed for both MSS and MSI-H cases, but the association between PIK3CA exon 20 mutation and mucinous histology was evident for MSI-H cases only.

Given the limited power to show PIK3CA exon mutation-specific clinicopathologic and molecular associations in our and previous studies, we integrated our results with findings from 17 published cohorts, together encompassing 5,594 patients. In the combined analysis, PIK3CA exon 9 or 20 mutation were both significantly associated with proximal tumor location, CIMP-L and KRAS mutation, in both instances showing a corresponding underrepresentation in distal tumors with MSS/CIMP-0/KRAS^wt/BRAF^wt genotype. Directly comparing PIK3CA exonic mutants, exon 20 mutation was specifically associated with features of the MSI-H/sessile-serrated pathway including MSI-H, CIMP-H, and BRAF mutation. The association of exon 20 mutation with mucinous histology observed in our cohort and the suggestive decrease in stage III as compared with stage I/II cancers in the combined cohorts are further consistent with the established characteristics of this pathway (41, 42). In contrast, PIK3CA exon 9 mutation was more strongly associated with KRAS mutation, and showed overrepresentation in proximal tumors with MSS/CIMP-0/KRAS^mut/BRAF^wt or MSS/CIMP-L/KRAS^mut/BRAF^wt genotype. The molecular basis for these differential exon-specific associations remains to be elucidated, but may be related to these 2 classes of mutations causing gain of p110α function through different mechanisms. Exon 9–mutated p110α has been shown to induce cell transformation independently of binding to p85 but requires interaction with RAS-GTP, whereas exon 20–mutated p110α is active in the absence of RAS-GTP binding but highly dependent on the interaction with p85 (43).

Preclinical and clinical studies suggest that PIK3CA mutation and PTEN loss of protein are important predictors of resistance to MEK inhibitors and anti-EGFR antibody therapy (12–14), and drug combination studies are proposed on the basis of PI3K and MAPK pathway mutation status. The latter include combinations of MAP2K and PI3K or mTOR inhibitors for KRAS/PIK3CA, KRAS/PTEN, and BRAF/PIK3CA double-mutant tumors, MAP2K or BRAF and PI3K or mTOR inhibitors for BRAF/PTEN double-mutant tumors, and PI3K or mTOR inhibitors for PIK3CA/PTEN double-mutant tumors (15). Optimization of the design of such clinical trials for colorectal cancer requires a detailed knowledge of the prevalence of these respective mutation genotypes. Here, we estimate that approximately 6.0% of sporadic colorectal cancers are KRAS/PIK3CA and 1.5% BRAF/PIK3CA double-mutant (integrated analysis), and 1.5% are KRAS/PTEN, 1.9% BRAF/PTEN, and 1.2% PIK3CA/PTEN double-mutant (this study). The low frequencies of these double-mutant cases underscore the need for collaborative international efforts to undertake such drug combination studies.

In our cohort, PIK3CA, PTEN, or overall PI3K pathway mutation status were not associated with prognosis of resected stage II/III colorectal cancer, and similar results were obtained when analyzing PIK3CA exon 9 or 20 status separately. Findings were similar when separating patients into groups who did and did not receive adjuvant chemotherapy. Our results for overall PIK3CA mutation contrast with some smaller previous studies reporting shorter RFS for PIK3CA-mutated stage II/III colorectal cancer (27) and inferior time to local failure for stage I–III rectal cancer (44). However, in line with our findings a recent analysis of 1,170 patients with stage I–III colorectal cancer did not find an association between PIK3CA mutation status and cancer-specific survival, either overall or for exon 9 or 20 alone, although adjuvant treatment details were not available (32). Taken together, the current evidence does not support a role for PIK3CA and/or PTEN mutation as a prognostic biomarker for colorectal cancer.

Potential caveats to our findings include that mutation screening was limited to PIK3CA mutation hotspots and certain PTEN exons. Although the selected regions have been shown to comprise the majority of mutations in colorectal cancer (www.sanger.ac.uk/genetics/CGP/cosmic), this may have impacted on the power to detect significant clinicopathologic, molecular, and outcome associations. Furthermore, loss of PTEN protein expression, observed in approximately 30% of colorectal cancers (6–8), is a documented alternative to mutation, and our study will therefore have underestimated the overall frequency of PTEN inactivation. Although no significant interstudy heterogeneity was observed in our integrated multicohort analysis for molecular variables, some heterogeneity was evident for patient gender.

In conclusion, our data suggest a model in which PIK3CA exon 20 and PTEN mutation are both associated with the sessile-serrated pathway of tumorigenesis (proximal/MSI-H/CIMP-H/BRAF^mut), whereas exon 9 (and to a lesser extent exon 20) mutation seems to be associated with the so-called traditional serrated pathway (proximal/CIMP-L/KRAS^mut; Fig. 3). These findings highlight the PI3K pathway as a preferential therapeutic target in these distinct colorectal cancer subtypes, a contention supported by drug treatment studies in colorectal cancer cell lines showing increased sensitivity of MSI-H cases to the PI3K pathway inhibitors rapamycin and LY-294002 (45). Our results underscore the importance of considering mutation-associated phenotypic and molecular heterogeneity, both between and within individual cancer genes, in the development of molecularly targeted therapies.

No potential conflicts of interest were disclosed.

Conception and design: F.L. Day, L. Lipton, Z.-Z. Xu, B.A. Leggett, R.L. Strausberg, O.M. Sieber

Development of methodology: F.L. Day, A. Sakthianandeswaren, Z.-Z. Xu, B.A. Leggett

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F.L. Day, M. Christie, C. Tsui, J. Tie, J. Desai, Z.-Z. Xu, P. Molloy, V. Whitehall, B.A. Leggett, I.T. Jones, S. McLaughlin, R.L. Ward, A.R. Ruszkiewicz, J. Moore, D. Busam, R.L. Strausberg, P. Gibbs, O.M. Sieber

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F.L. Day, R.N. Jorissen, D. Mouradov, M. Christie, J. Desai, Z.-Z. Xu, P. Molloy, N.J. Hawkins, A.R. Ruszkiewicz, Q. Zhao, P. Gibbs, O.M. Sieber

Writing, review, and/or revision of the manuscript: F.L. Day, R.N. Jorissen, L. Lipton, A. Sakthianandeswaren, M. Christie, J. Desai, P. Molloy, V. Whitehall, B.A. Leggett, I.T. Jones, R.L. Ward, N.J. Hawkins, A.R. Ruszkiewicz, Q. Zhao, R.L. Strausberg, P. Gibbs, O.M. Sieber

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F.L. Day, A. Sakthianandeswaren, M. Christie, S. Li, C. Tsui, J. Tie, B.A. Leggett, A.R. Ruszkiewicz

Study supervision: A. Sakthianandeswaren, I.T. Jones, O.M. Sieber

The authors thank all individuals who participated in this study and colleagues who undertook sample and clinical data collection. The authors also thank the Victorian Cancer Biobank for the provision of patient specimens and BioGrid Australia for providing de-identified clinical data.

This study was supported by the CSIRO Preventative Health Flagship through an Australian Cancer Grid Project Grant (to L. Lipton, P. Gibbs, and O.M. Sieber), the National Health and Medical Research Council through a Project Grant (Application ID 489418; to L. Lipton, O.M. Sieber, P. Gibbs, and R.L. Ward), the Hilton Ludwig Cancer Metastasis Initiative (to L. Lipton, P. Gibbs, and O.M. Sieber), and the Victorian Government through a Victorian Cancer Agency Translation Cancer Research Grant (to L. Lipton, P. Gibbs, and O.M. Sieber). F.L. Day and M. Christie are supported by the Cancer Council Victoria through a Postgraduate Cancer Research Scholarship and L. Lipton by the CSIRO Preventative Health Flagship through a Clinical Researcher Fellowship. J. Tie is supported by the Victorian Government through a Victorian Cancer Agency Researcher Fellowship and an ASCO Cancer Foundation Young Investigator Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.