PIK3CA Mutations in Advanced Cancers: Characteristics and Outcomes

PIK3CA mutations are frequently diagnosed in diverse cancers and may predict response to PI3K/AKT/mTOR inhibitors. It remains unclear whether they are associated with other characteristics. We analyzed characteristics and outcome of 90 consecutive patients with diverse advanced tumors and PIK3CA mutations and 180 wild-type PIK3CA controls matched by tumor type, gender, and age referred to the Clinical Center for Targeted Therapy. PIK3CA and MAPK mutations (KRAS, NRAS, and BRAF) were analyzed using polymerase chain reaction-based DNA sequencing. The most frequent PIK3CA mutations were E545K (31/90, 34%), E542K (16/90, 18%) in exon 9, and H1047R (20/90, 22%) in exon 20. PIK3CA mutations compared to wild-type PIK3CA were associated with simultaneous KRAS (p=0.047) and MAPK mutations (p=0.03), but only MAPK mutations were confirmed as having an independent association in multivariate analysis. Rates of lung, bone, liver and brain metastases were similar in PIK3CA-mutant and wild-type patients. Patients with PIK3CA mutations treated on trials with PI3K/AKT/mTOR inhibitors had a higher partial/complete response (PR/CR) rate than wild-type PIK3CA patients treated with their best phase I therapy (10/56, 18% vs. 12/152, 8%; p=0.045), but not a prolonged progression-free survival. Patients with H1047R PIK3CA mutations had a higher PR/CR rate with PI3K/AKT/mTOR inhibitors compared to wild-type PIK3CA patients treated with their best phase I therapy (6/16, 38% vs. 12/152, 8%; p=0.003). In conclusion, PIK3CA mutations in diverse cancers were not associated with clinical characteristics, but were correlated with MAPK mutations. PIK3CA mutations, especially, H1047R, were associated with attaining a PR/CR to PI3K/AKT/mTOR pathway inhibitors.


INTRODUCTION
PIK3CA mutations frequently occur in diverse cancers and are associated with constitutive activation of the PI3K/AKT/mTOR pathway. [1][2][3][4][5] In addition, PIK3CA mutations predicted sensitivity to PI3K/AKT/ mTOR inhibitors in multiple tumor types in preclinical and early clinical experiments. [1,2,[5][6][7][8][9][10][11][12] A seminal question is whether PIK3CA mutations are associated with a distinct phenotypic taxonomy. Retrospective studies in colorectal cancer demonstrated that PIK3CA mutations in exon 20 encoding for the kinase domain, but not in exon 9 encoding for the helical domain, are associated with resistance to EGFR-targeting monoclonal antibodies. [13] In addition, our group reported that, regardless of histology, PIK3CA mutations often coexist with mitogenactivated protein kinase (MAPK) mutations, such as mutated KRAS, NRAS, and BRAF. [14] A partial answer to the question posed about the relationship between PIK3CA mutations and specific subtypes of cancer is generally that different cancers seem to have different types of PIK3CA mutations and associations with still other mutations. [15] For example, in colorectal cancer PIK3CA mutations in exon 9, but not exon 20, trended toward an association with KRAS mutations, whereas only PIK3CA exon 20 mutations were associated with KRAS mutations in ovarian cancer. [13,16] Other oncogenic mutations have also been correlated with clinical characteristics and outcome. For example patients with advanced cancers and BRAF mutations have less soft tissue, retroperitoneal, lung metastases and more brain metastases. [17] In colorectal cancer, BRAF mutations predicted poor outcome and KRAS mutations were associated with lung metastases. [13,18] We investigated characteristics and outcomes of patients with advanced cancers with and without PIK3CA mutations.

Patients
We retrospectively reviewed clinical and pathological characteristics and treatment outcomes of 90 consecutive patients with advanced tumors harboring PIK3CA mutations who had been referred to the Clinical Center for Targeted Therapy at The University of Texas MD Anderson Cancer Center (MD Anderson) starting in October 2008. To define distinguishing features of advanced cancers with PIK3CA mutations, we selected a control group of 180 patients with wild-type (wt) PIK3CA advanced cancers matched in a 2:1 ratio by tumor type, gender, and age (+/-5 years) to patients with PIK3CA mutations referred to the MD Anderson Clinical Center for Targeted Therapy (CCTT) during the same period of time.
Data were collected from transcribed notes and radiology reports in the electronic medical record and other source documentation. Registering patients in the database, pathology assessment, and mutation analysis were performed at MD Anderson. The study and all treatments were conducted in accordance with the guidelines of the MD Anderson Institutional Review Board.

Tissue samples and mutation analyses
PIK3CA, KRAS, NRAS, and BRAF mutations were investigated in archival formalin-fixed, paraffin-embedded tissue blocks or material from primary or metastatic lesions obtained from diagnostic and/or therapeutic procedures. All histologies were centrally reviewed at MD Anderson. Mutation testing was performed in the Clinical Laboratory Improvement Amendmentcertified Molecular Diagnostic Laboratory within the Division of Pathology and Laboratory Medicine at MD Anderson. DNA was extracted from microdissected paraffin-embedded tumor sections and analyzed using a polymerase chain reaction-based DNA sequencing method for PIK3CA mutations in codons 532-554 of exon 9 (helical domain) and codons 1011-1062 of exon 20 (kinase domain). Analysis included the mutation hot spot region of the PIK3CA proto-oncogene denoted by Sanger sequencing, following amplification of 276 bp and 198 bp amplicons, respectively, utilizing primers designed by the MD Anderson Molecular Diagnostic Laboratory. Whenever possible, in addition to PIK3CA, mutation analysis was done for KRAS and NRAS codons 12, 13, and 61 mutations, and BRAF codons 595-600 mutations of exon 15 by pyrosequencing, as previously described. [19] Treatment Prior to being treated with Phase I agents, patients typically received the US Food and Drug Administration (FDA)-approved therapy. If available, we collected data from the last FDA-approved therapy in addition to the Phase I therapy given in the CCTT. For patients with PIK3CA mutations, data were recorded for treatment received that included PI3K/AKT/mTOR inhibitors. For wt PIK3CA patients, data from the best phase I therapy were recorded. The response outcome was measured by partial [PR] or complete response [CR] or the absence of PR/CR as well as the duration of progression-free survival [PFS] and overall survival (OS). [20] Statistical analysis Patient characteristics were summarized using descriptive statistics. Response to treatment (PR or CR) was evaluated using Response Evaluation Criteria in Solid Tumors (RECIST 1.0). [20] PFS was defined as the interval from initiation of the selected phase I treatment to disease progression or death. Patients who were alive and not progressing were censored at the date of the last follow-up. Overall survival (OS) was defined as the interval from diagnosis to death and OS on phase I therapy (OS-Ph1) was defined as the interval between recorded initiation of the systemic phase I treatment to death. www.impactjournals.com/oncotarget Patients who were alive were censored at the date of the last follow-up. Distant metastasis-free survival (DMFS) was defined as the interval from diagnosis to development of metastatic disease. The probabilities of PFS, OS, OS-Ph1, and DMFS were estimated using the method of Kaplan and Meier and the time-to-event endpoints were compared among subgroups using the log-rank test. [21,22] Associations between PIK3CA mutation status and categorical variables (ethnicity, biopsy/tissue site, metastatic site, history of deep vein thrombosis [DVT], history of smoking, KRAS mutation, NRAS mutation, BRAF mutation, PR/CR status after last FDA-approved therapy, PR/CR status after PI3K/AKT/mTOR targeted phase I therapy in PIK3CA-mutant patients, or status after the best phase I therapy in wt PIK3CA patients) were assessed using Fisher's exact test. PR/CR rate to prior vs. current therapies in matched paired subjects was assessed using McNemar's test. In addition, univariate and multivariate logistic regression models were fit to assess the associations between PIK3CA mutations and other categorical variables. A Cox regression model was applied to assess the effect of covariates on time-to-event endpoints. All tests were two-sided, and a P value less than 0.05 was considered statistically significant. All statistical analyses were carried out using SPSS 17 computer software (SPSS Chicago, IL) and R version 2.15.0 (R Foundation for Statistical Computing).

Patient characteristics
A total of 270 patients with diverse advanced cancers consisting of 90 patients with PIK3CA mutations and 180 controls with wt PIK3CA (matched by tumor type, gender, and age) were analyzed and their clinical and pathologic characteristics are listed in Table 1. Most patients (79%) were white and women (62%). The median age was 56 years (range,  and patients received the median number of 3 prior therapies (range, 0-12). The most prevalent tumor types were colorectal cancer (27%), breast cancer (17%), ovarian cancer (12%) and endometrial cancer (10%). Lung and/or liver metastases were found in 62% of patients. Brain metastases were found in 13% of patients. Of the 208 patients tested for KRAS, 53 (25%) had a mutation; of the 95 tested for NRAS, 3 (3%) had a mutation; and of the 179 tested for BRAF, 15 (8%) had a mutation. When analyzing tested MAPK mutations, of the 132 patients tested for KRAS, NRAS, and BRAF mutation status (patients were selected for analyses if they had a mutation in KRAS, or NRAS, or BRAF since they are known to be mutually exclusive or if they were tested negative for all 3 oncogenes), 71 (54%) had KRAS, NRAS or BRAF mutations.
For KRAS, NRAS, and BRAF mutations, no patient had more than one type of mutation within each gene, which is not surprising since these mutations are known to be mutually exclusive. [13,23]

Clinical and molecular features associated with PIK3CA mutations
Patients with PIK3CA mutations (n=90) compared to patients with wt PIK3CA (n=180) had a trend to a higher incidence of DVT ( 1). There was no difference with respect to ethnicity, smoking history, number of prior therapies, biopsy/tissue site, and the occurrence of lung, liver, bone, and brain metastases between patients with and without PIK3CA mutations (Table 1). The multivariate regression model, which included MAPK mutation status and history of DVT, confirmed that patients with PIK3CA mutations had a higher prevalence of MAPK mutations (odds ratio There were no other associations with any other assessed characteristics (ethnicity, smoking history, number of prior therapies, biopsy/tissue site, and the occurrence of lung, liver, bone, and brain metastases).

Treatment outcomes with respect to PIK3CA mutation status
We analyzed PR/CR rates from the last FDAapproved treatment in 197 patients with available data and found no statistically significant differences between patients with PIK3CA mutations and wt PIK3CA patients (5/59 [8%] vs. 6/138 [4%], p=0.31) ( Table 3). In contrast, in 208 patients who received phase I systemic therapy, those with PIK3CA mutations treated with a phase I therapy targeting the PI3K/AKT/mTOR pathway had a higher PR/CR rate than wt PIK3CA patients treated with their best phase I therapy (10/56 [18%] vs. 12/152 [8%], p=0.045). We also analyzed PR/CR rate following phase I therapy separately for patients with exon 9 and those with exon 20 PIK3CA mutations. Patients with PIK3CA exon 9 mutations showed a trend toward a higher PR/ CR rate to the phase I therapy with a PI3K/AKT/mTOR inhibitor than patients with wt PIK3CA treated with their best phase I therapy ( We next analyzed PFS after the last FDA-approved therapy (n=197) and phase I systemic therapy (n=208) ( Finally we analyzed OS (measured from diagnosis), OS-Ph1 (measured from recorded phase I therapy), and DMFS (measured from diagnosis to metastatic disease) for patients with PIK3CA mutations and wt PIK3CA patients (Table 3). Patients with PIK3CA mutations had a similar median OS (50.4 months, 95% CI 36.2-64.6) as patients with wt PIK3CA (55.2 months, 95% CI 46.7-63.7, p=0.70). In 142 patients who were initially diagnosed with localized disease, those with PIK3CA mutations had a numerically shorter DMFS (12.3 months, 95% CI 7.5-17.1) than patients with wt PIK3CA (18.8 months, 95% CI 14.5-23.1, p=0.6). In 208 patients who received phase I therapy, those with PIK3CA mutations had a shorter median Ph1-OS (6.6 months, 95% CI 3.9-9.3) than patients with wt PIK3CA (8.6 months, 95% CI 7.1-10.1, p=0.03).
Colorectal cancer was the largest tumor-specific subgroup, consisting of 24 patients with PIK3CA mutations and 48 matched wt PIK3CA controls. We therefore performed subanalysis on this histology. PIK3CA mutations were most frequent in exon 9 ( A paired analysis of colorectal cancer patients for whom we had data on the last FDA-approved therapy and phase I therapy (PIK3CA mutations, n=10; wt PIK3CA, n=25), there was no response noted ( Finally, there was no significant difference in OS, OS-Ph1, and DMFS between colorectal cancer patients with PIK3CA mutations and wt PIK3CA (Table 3).

DISCUSSION
In this study of 90 patients with PIK3CA mutations and 180 wt PIK3CA controls (matched by tumor type, gender, and age) we identified having a history of DVT as the only clinical characteristic potentially associated with PIK3CA mutations. However, this association was not confirmed by multivariate analysis. None of the clinical characteristics including ethnicity, the site of metastases, smoking history, number of prior therapies, OS, OS-Ph1, DMFS was associated with PIK3CA status. In agreement with previous reports, when assessing individuals for MAPK mutations we noted the association between PIK3CA and MAPK mutations (66% vs. 46%, p=0.03) and between PIK3CA and KRAS mutations (34% vs. 21%, p=0.047). [13,14] In disease-specific subanalyses, also in agreement with previous reports, there was a trend toward an association between PIK3CA and KRAS mutations in colorectal cancer (71% vs. 48%, p=0.08); associations between PIK3CA and KRAS (35% vs. 7%, p=0.04), and PIK3CA and MAPK mutations (62% vs. 13%, p=0.02) in ovarian and endometrial cancers. [13,14] We also looked for associations linked to mutations in exon 9 and exon 20, which account for more than 80% of PIK3CA mutations. [24] Overall, PIK3CA exon 9 mutations were associated with simultaneous MAPK mutations (70% vs. 48%; p=0.04), showed a trend toward association with simultaneous KRAS mutations (37% vs. 22%; p=0.05), and albeit a weaker trend toward association with BRAF mutations (14% vs. 7%; p=0.12), whereas PIK3CA exon 20 mutations were not associated with any mutations in the MAPK pathway. This finding is consistent with previous reports from our group and others in colorectal cancer and other tumor histologies. [13,14] Oncogenic mutations often point to the presence of a therapeutic target that might be amenable to directed therapeutic intervention. For example, KIT mutations render patients with gastrointestinal stromal tumor sensitive to KIT tyrosine kinase inhibitors (TKIs), EGFR mutations render patients with NSCLC sensitive to EGFR TKIs, a EML4-ALK fusion renders patients with NSCLC sensitive to ALK inhibitors, and BRAF mutations increase the sensitivity of melanoma patients to BRAF inhibitors. [25][26][27][28] We reported increased PR/CR rates in response to PI3K/AKT/mTOR inhibitors in patients with PIK3CA mutations compared to wt PIK3CA treated in early phase clinical trials. [8] In the current study, we retrospectively evaluated treatment outcomes (PR/CR rate, PFS) on the last FDA-approved therapy and phase I therapy (therapy targeting PI3K/AKT/mTOR pathway in patients with PIK3CA mutations or best phase I therapy [defined by CR/PR or longest PFS] in patients with wt PIK3CA). Overall, there was no difference in PR/CR rate (8% vs. 4%; p=0.31) and PFS (3.0 months vs. 3.2 months, p=0.55) to the last FDA-approved therapy between patients with PIK3CA mutations and wt PIK3CA, but patients with PIK3CA mutations had a higher PR/CR rate to phase I therapy with PI3K/AKT/mTOR inhibitors (18% vs. 8%; p=0.045), which did not, however, translate to a longer PFS (2.0 months vs. 3.7 months; p=0.59) compared to wt PIK3CA patients treated with their best phase I therapy.
In the paired analysis, which included only patients who had available data from treatment with both the last FDA-approved therapy and phase I therapy, we found no difference in PR/CR rates and PFS in patients with PIK3CA mutations and wt PIK3CA. The explanation for this might be because PI3K/AKT/mTOR targeting therapies are effective in only a subset of patients. In NSCLC, specific EGFR mutations such as exon 19 deletions or exon 21 mutations (L858R), render tumors sensitive to EGFR TKIs, whereas an exon 20 mutation (T790M) is associated with therapeutic resistance. [29] A similar scenario is possible for other oncogenic mutations. Indeed, in our current study we noticed that patients with H1047R PIK3CA mutations compared to patients with wt PIK3CA had an increased PR/CR rate (38% vs. 8%; p=0.003) to phase I therapy compared to wt PIK3CA patients. Another explanation might relate to the histological milieu of the PIK3CA mutations. Although, these mutations were found in a variety of cancers, the largest subgroup of patients had colorectal cancer. These patients showed a poor outcome on PI3K/AKT/mTOR, perhaps because of the frequent coexistence of KRAS mutations in this histology. [13,14,30] Finally, we investigated whether PIK3CA mutations have prognostic significance. For instance, EGFR mutations compared to wt EGFR in NSCLC are usually associated with improved treatment outcomes and a longer OS. [31] KRAS mutations in NSCLC are associated with poor outcomes in response to EGFR therapies. [31] In colorectal cancer, BRAF mutations are associated with a shorter survival. [13] Consensus regarding the impact of PIK3CA mutations is contradictory. Some investigators reported better prognosis in certain cancers such as breast cancer with PIK3CA mutations, whereas others suggested that PIK3CA mutations indicate a worse prognosis in colorectal cancer, endometrial cancer and NSCLC. [13,29,[32][33][34][35][36] There was no significant difference in patients with PIK3CA mutations and wt PIK3CA in OS (from the time of first diagnosis; 50.4 months vs. 55.2 months; p=0.70), and DMFS (from time of first diagnosis to development of metastatic disease; 12.3 months vs. 18.8 months; p=0.64); however, patients with PIK3CA mutations had a shorter OS-Ph1 (from initiation of phase I therapy; 6.6 months vs. 8.6 months; p=0.03). In disease-specific sub-analyses, we found a lower survival in patients with colorectal cancer and PIK3CA mutations who, compared to wt PIK3CA patients, experienced a shorter OS-Ph1 (3.6 months vs. 10.3 months; p=0.001). This finding provides another piece of evidence suggesting that colorectal cancer patients with PIK3CA mutations do not do well on a PI3K/ AKT/mTOR targeted therapy.
In conclusion, in the current study of 90 patients with PIK3CA mutations and 180 matched controls with wt PIK3CA we found that there was no PIK3CA phenotype. Overall, PIK3CA mutations were associated with KRAS and MAPK (KRAS, NRAS, BRAF) mutations. Patients with PIK3CA mutations treated with PI3K/AKT/mTOR axis inhibitors had a PR/CR rate of 18%. This PR/CR rate is lower than the rate we previously reported for gynecologic and breast malignancies, in which the PR/