Significance of PIK3CA Mutations in Patients with Early Breast Cancer Treated with Adjuvant Chemotherapy: A Hellenic Cooperative Oncology Group (HeCOG) Study

Background The PI3K-AKT pathway is frequently activated in breast cancer. PIK3CA mutations are most frequently found in the helical (exon 9) and kinase (exon 20) domains of this protein. The aim of the present study was to examine the role of different types of PIK3CA mutations in combination with molecular biomarkers related to PI3K-AKT signaling in patients with early breast cancer. Methods Tumor tissue samples from 1008 early breast cancer patients treated with adjuvant chemotherapy in two similar randomized trials of HeCOG were examined. Tumors were subtyped with immunohistochemistry (IHC) and FISH for ER, PgR, Ki67, HER2 and androgen receptor (AR). PIK3CA mutations were analyzed by Sanger sequencing (exon 20) and qPCR (exon 9) (Sanger/qPCR mutations). In 610 cases, next generation sequencing (NGS) PIK3CA mutation data were also available. PIK3CA mutations and PTEN protein expression (IHC) were analyzed in luminal tumors (ER and/or PgR positive), molecular apocrine carcinomas (MAC; ER/PgR negative / AR positive) and hormone receptor (ER/PgR/AR) negative tumors. Results PIK3CA mutations were detected in 235/1008 tumors (23%) with Sanger/qPCR and in 149/610 tumors (24%) with NGS. Concordance between the two methods was good with a Kappa coefficient of 0.76 (95% CI 0.69–0.82). Lobular histology, low tumor grade and luminal A tumors were associated with helical domain mutations (PIK3CAhel), while luminal B with kinase domain mutations (PIK3CAkin). The overall incidence of PIK3CA mutations was higher in luminal as compared to MAC and hormone receptor negative tumors (p = 0.004). Disease-free and overall survival did not significantly differ with respect to PIK3CA mutation presence and type. However, a statistically significant interaction between PIK3CA mutation status and PTEN low protein expression with regard to prognosis was identified. Conclusions The present study did not show any prognostic significance of specific PIK3CA mutations in a large group of predominantly lymph-node positive breast cancer women treated with adjuvant chemotherapy. Further analyses in larger cohorts are warranted to investigate possible differential effect of distinct PIK3CA mutations in small subgroups of patients.


Methods
Tumor tissue samples from 1008 early breast cancer patients treated with adjuvant chemotherapy in two similar randomized trials of HeCOG were examined. Tumors were subtyped with immunohistochemistry (IHC) and FISH for ER, PgR, Ki67, HER2 and androgen receptor (AR). PIK3CA mutations were analyzed by Sanger sequencing (exon 20) and qPCR (exon 9) (Sanger/qPCR mutations). In 610 cases, next generation sequencing (NGS) PIK3CA mutation data were also available. PIK3CA mutations and PTEN protein expression (IHC) were analyzed in luminal tumors (ER and/or PgR positive), molecular apocrine carcinomas (MAC; ER/PgR negative / AR positive) and hormone receptor (ER/PgR/AR) negative tumors.

Results
PIK3CA mutations were detected in 235/1008 tumors (23%) with Sanger/qPCR and in 149/ 610 tumors (24%) with NGS. Concordance between the two methods was good with a Kappa coefficient of 0.76 (95% CI 0.69-0.82). Lobular histology, low tumor grade and luminal A tumors were associated with helical domain mutations (PIK3CAhel), while luminal B with kinase domain mutations (PIK3CAkin). The overall incidence of PIK3CA mutations was higher in luminal as compared to MAC and hormone receptor negative tumors (p = 0.004). Disease-free and overall survival did not significantly differ with respect to PIK3CA mutation presence and type. However, a statistically significant interaction between PIK3CA mutation status and PTEN low protein expression with regard to prognosis was identified.

Introduction
Phosphoinositide 3-kinase-protein kinase B (PI3K-AKT) is an important intracellular pathway regulating numerous cellular activities, mostly survival, proliferation, growth and glycogen metabolism. It is activated in many different tumor types and its therapeutic targeting has attracted significant interest [1]. PI3K-AKT signaling is an important driving pathway in breast cancer, triggered by growth factor receptor activation, such as insulin-like growth factor receptors (IGF-R), epidermal growth factor receptor (EGFR) and HER2, by steroid hormone receptors, estrogen (ER), progesterone receptors (PgR), as well as by genetic aberrations of some of its components, more frequently HER2 amplification, PIK3CA mutations and phosphatase and tensin homolog (PTEN) loss [2]. Also, evidence of cross-talk between androgen receptors (AR) and the PI3K-AKT pathway has been reported [3][4][5]. Pathway activation leads to phosphorylation of AKT at two different amino acid residues, pAKT473 and pAKT308. AKT directly phosphorylates and activates mTOR. PTEN is a lipid phosphatase that inhibits AKT activation [6].
Adding more complexity to the model, different PIK3CA mutations exist, with two common hotspot areas, the helical (exon 9, PIK3CAhel) and the kinase (exon 20, PIK3CAkin). The exact mechanism by which these different types of mutations affect the PI3K pathway is not yet clarified. Both PIK3CAhel and PIK3CAkin were found to exert gain-of-function and transforming activity. However, more recent data revealed that each type of mutation might lead to different interactions with protein partners and different tumorigenic potential in animal models [7].
In the present study, different types of PIK3CA mutations were analyzed with regard to their frequency, their association with basic characteristics and their prognostic significance in patients with early breast cancer. Also, the possible interactions of these mutations with multiple other biomarkers, such as EGFR, AR and members of the IGF and PI3K-AKT pathway, described in our previous reports [8][9][10], were investigated.

Patients and Methods Patients
In this prospective/retrospective translational research study, patients from two randomized adjuvant chemotherapy trials were included. The characteristics of these trials are described in brief in S1 Table and detailed information about them can be found in the respective original papers [11][12][13]. At that time, trastuzumab was not licensed as adjuvant treatment; however it was administered in HER2 positive patients upon relapse [14]. The present translational research protocol was approved by the Bioethics Committee of the Aristotle University of Thessaloniki School of Medicine and patients provided written informed consent for the use of their biological material for future research purposes.

Tumor tissue samples, processing, methods and biomarkers
For all methods, formalin-fixed paraffin-embedded (FFPE) tumor tissue samples were used. Tissue blocks were collected retrospectively in the first trial (HE10/97) and prospectively in the second (HE10/00). Tumors were histologically evaluated on hematoxcylin & eosin (H&E) sections for tumor presence and marked for the most tumor dense areas. Tumor cell content (TCC) was assessed as the ratio of cancer vs. non-cancer cells in these areas, which were manually macrodissected from 10 micron unstained sections for subsequent DNA extraction. Next, a second H&E evaluation round was employed for obtaining cores for tissue microarray (TMA) construction, which were used for immunohistochemistry (IHC) and fluorescent in situ hybridization (FISH). TMA blocks were constructed with a manual tissue microarrayer (Beecher Instruments, Sun Prairie, WI), using 2 cores per tumor, each 1.5 mm in diameter, along with orientation and IHC control sample cores. Cases with low tumor tissue availability were represented in TMAs only.

PIK3CA mutations
DNA was extracted with dual nucleic acid extraction using silica-coated magnetic beads (Versant Tissue Preparation Reagents, Siemens Healthcare Diagnostics, Tarrytown, NY) from 1008 samples according to the manufacturer's instructions.
PIK3CA mutations were investigated with classic methods and with next generation sequencing (NGS) (Fig 1). Classic methods included Taqman-MGB-SNP genotyping assays (Applied Biosystems, Foster City, CA) for E542K and E545K (exon 10, coding exon 9) and subsequent allelic discrimination in a 7900HT (Applied Biosystems) real time PCR (qPCR) system (Razis, Bobos et al. 2011). These methods were applied on 1008 tumor FFPE DNA samples. For mutations in exon 21 (coding exon 20), Sanger sequencing (Sanger) on nested PCR products with M-13 coupled primers was performed in 10 ul reactions with the Big Dye Teminator kit v.1.1 (Applied Biosystems). Sequences were visualized upon capillary electrophoresis in an ABI3130XL genetic analyzer, and were initially called with the Sequencing Analysis software v.5.2. The nested intron-spanning gene-specific primer pair was 5'-TTTTCTCAATGAT GCTTGGCT-3' (forward) and 5'-CCTGCTGAGAGTTATTAACAGT-3' (reverse). This method switch was considered necessary for exon 20 in order to avoid detecting sequences from the highly homologous PIK3CA pseudogene on chromosome 22 (LOC100422375).
NGS was applied on libraries constructed with the Ampliseq Library Kit v.2.0 and Ion Xpress barcodes, according to the manufacturer's instructions (Life Technologies / Ion Torrent, Carlsbad, CA), as previously described [16]. In total, 610 tumor DNA samples were examined with this method. PIK3CA coding regions were targeted with 9 amplicons covering the following genomic areas (positions according to GRCh37 / hg19: exon 2 (coding exon 1) ; and, if they were non-annotated. Thus, variants at a frequency of 40% in low performing amplicons were accepted (worst case variants). Amino acid changing variants in coding regions (missense and nonsense) with minor allele frequency <1% if annotated single nucleotide polymorphisms (SNPs) were considered as mutations.

Statistical analysis
Categorical data are displayed as frequencies and corresponding percentages. Possible associations between PIK3CA and clinical characteristics or other markers were investigated by performing Chi-square tests. The presence of PIK3CA mutations, as obtained with Sanger/qPCR and NGS, was compared through the calculation of a) the concordance rate along with the 95% confidence intervals for the estimation, defined as the proportion of the equal results, and b) the Cohen's Kappa coefficient of agreement.
Disease-free survival (DFS) was measured from the date of diagnosis until verified disease progression, death or last contact, whichever occurred first, while overall survival (OS) from diagnosis until death from any cause or date of last contact. Time-to-event distributions were estimated using Kaplan-Meier curves, while log-rank tests and univariate Cox regression analyses were used for assessing statistically significant differences and reporting hazard ratios, respectively. Univariate Cox regression analyses with interactions were performed in order to investigate for possible differences of PIK3CA effects on DFS and OS among the levels of other variables. The identified interactions in the univariate setting between PIK3CA and other markers were adjusted for demographic, clinical and treatment characteristics. Multivariate Cox regression analysis was performed, with a backward selection procedure with a removal criterion of p >0.10 in order to identify significant factors. The clinical variables that were assessed in multivariate analysis were: adjuvant hormonotherapy (Yes vs. No), adjuvant radiotherapy (Yes vs. No), age (<50 vs. In univariate analysis, significance was determined at the level of 5% and all tests were twosided. The statistical analysis complied with the reporting recommendations for tumor marker prognostic studies [17] and was performed using the SAS software (SAS for Windows, version 9.3, SAS Institute Inc., Cary, NC).

PIK3CA mutation analysis
In total, tumor tissues from 1008 patients were evaluated for PIK3CA mutation detection by Sanger/qPCR (see CONSORT diagram in Fig 1). PIK3CA mutations were found with this   Fig 2). In 610 patients (60.5%), PIK3CA mutations were assessed by both Sanger/qPCR and NGS. PIK3CA mutations with NGS were found in 149 patients (24.4%). Comparisons of mutation types and PIK3CA mutation status by both methods are presented in S2 and S3 Tables, respectively. Overall, the discordance rate of PIK3CA status with the two methods was 8.8%, whereas Cohen's Kappa indicated good agreement between Sanger/qPCR and NGS results, with a coefficient of 0.76 (95% CI 0.69-0.82). In general, NGS was more sensitive than Sanger/qPCR in detecting PIK3CA mutations. In the 7 Sanger/qPCR-mutated NGS-wild-type cases the corresponding variants had been filtered out for inadequate quality metrics. In the 44 Sanger/qPCR wild type, NGS positive cases, all but one discrepant PIK3CA variants concerned the most commonly affected c.1624, c.1633, and c.3140 nucleotides with adequate read quality but with 6-12% variant allele frequency in 38/44 cases that might have been missed by the other two methods in samples with low DNA quality. Patients were classified in 3 groups according to their PIK3CA status, i.e. PIK3CA wild-type (PIK3CAwt), PIK3CA kinase domain mutant (PIK3CAkin) and PIK3CA helical domain mutant (PIK3CAhel). Patients with mutations found in both the helical and kinase domains were randomly allocated in the PIK3CAhel group for the purposes of the current analysis.

Association of PIK3CA mutations with basic characteristics and treatment
Basic patients' characteristics can be found in Table 2. All patients except 5 (0.5%) had infiltrated regional lymph node(s) at surgery and approximately equal numbers were pre-and postmenopausal. Differences in the distribution of tumor characteristics were found between the three PIK3CA groups by Sanger/qPCR ( Table 2). PIK3CAhel tumors were more frequently smaller in size, well-moderately differentiated and of lobular histology. PIK3CAkin tumors were less frequently HER2-enriched. PIK3CAmut were significantly more prevalent in luminal tumors (ER/ PgR positive, 27.1%) than in MAC (AR positive and ER/PgR negative, 16%) or hormone receptornegative (AR/ER/PgR negative, 13%) tumors (p = 0.004). Regarding adjuvant treatment, PIK3CAkin patients were more often treated with taxane-based chemotherapy than the other two groups. This may have been caused by slight differences in the inclusion criteria of the two HeCOG trials.
Similarly, PIK3CAmut as revealed with NGS were significantly more often in luminal tumors (27.8%) as compared to MAC (17.8%) and to hormone receptor negative (13.9%) tumors (p = 0.053). Helical PIK3CA mutations with this method were again associated with more favorable disease characteristics, such as histological grade I-II and Luminal A subtype (S4 Table).
The distribution of other biomarkers in the three PIK3CA groups is shown in S5 Table. Notably, EGFR positivity by IHC was observed in a decreasing order from the PIK3CAwt group to the PIK3CAkin and finally the PIK3CAhel group. Also, the frequency of IGF1R-beta high was differentially distributed between the PIK3CA groups.
The above analysis ( Table 2, S4 and S5 Tables) was performed also excluding the 3 cases with mutations at both helical and kinase domains (data not shown). The figures were found very similar, indicating that the decision to include or not these cases in the analysis did not influence the results.

PIK3CA mutations as prognostic factors
After a median follow-up of 121.9 months (range 0.1-191.9), 289 deaths (28.7%) and 365 relapses (36.2%) were recorded. Median DFS and OS did not differ significantly between PIK3CAwt and mutant tumors, while they were also similar in the 3 different PIK3CA groups, as determined with Sanger/qPCR (Table 3) and NGS (S6 Table). Survival curves of the entire population (Sanger/qPCR data) are shown in Fig 3. Only Sanger/qPCR mutations were included in prognostic subgroup analyses because, as shown in S3 Table, categories with helical PIK3CA mutations in non-luminal subtypes were too small for reliable statistical comparisons. Univariate analysis of the prognostic significance of the PIK3CA mutation status stratified for each level of biomarker expression and breast cancer subtypes is depicted in S7   Luminal-HER2 10 (12) 11 (8) 102 (14) HER2-enriched 6 (7) 5 (4) 86 (12) Triple-negative 7 (9) 11 (8) 97 (14) Breast Multivariate analysis of prognostic factors for DFS and OS is described in Fig 4. Nodal stage, tumor subtype, ER status and the type of surgery were found to be important predictors of DFS, while tumor size, nodal stage and tumor subtype were prognostic for OS. Of note, a statistically significant interaction was observed between PIK3CA mutation status, as determined by Sanger/qPCR, and PTEN protein expression levels. Specifically, PIK3CAhel mutations were an adverse prognostic factor only within the group of patients with PTEN low protein expression and, inversely, PTEN low protein expression was a poor prognostic factor only in the PIK3CAhel group.

Discussion
PIK3CA mutations are one of the most frequent genetic alterations in breast carcinomas [18]. Although their role has been the subject of extensive research, our knowledge is limited regarding their significance in malignant diseases. The present study has several strengths: 1) the population was retrieved from two randomized trials with very similar chemotherapy regimens including E and CMF with or without T; 2) more than half of eligible patients (1008 [62%] of the 1681 total eligible patients enrolled in the two randomized HeCOG trials described in Methods) were included in the study; 3) all except 5 patients (1%) were lymph node positive; and 4) PIK3CA mutation analysis was performed by two methods, Sanger/qPCR and NGS. Notably, Jiang et al [19] described intratumoral heterogeneity regarding PIK3CA mutations in 28.4% of cases. NGS is much more reliable in distinguishing tumor from normal DNA or different tumor clones [20]. However, the disagreement rate between the two methods was low. On the contrary, some caveats could be noticed: 1) NGS was performed in a subgroup of 610 patients, which represents 60.5% of the study population; 2) concordance between Sanger/ qPCR and NGS was high but not perfect; 3) HER2 positive patients did not receive the currently standard regimen, which includes trastuzumab, although its activity did not seem to be influenced by PIK3CA genotype, as shown recently by Pogue-Geile et al [21]; and finally 4) the sample size was large but not big enough to offer adequate statistical power to examine small but important subgroups of patients, such as MAC or hormone receptor negative patients with different types of PIK3CA mutations.
The present study is one of the largest among those examining the role of PIK3CA mutations in breast cancer. Our PIK3CA mutation rate (23.3%) was consistent with the literature, as well as our rate of different hotspot mutations [3,14,19,[22][23][24][25][26][27][28][29][30][31][32][33][34]. Also, we confirmed significant associations of different types of PIK3CA mutations with clinicopathological and molecular characteristics. One important question is the possible effect of PIK3CA status on cancer outcome and treatment efficacy. Notably, many studies examining the prognostic significance of PIK3CA mutations in breast cancer [35] have been published, however very few included a large enough number of patients [3,14,19,[22][23][24][25][26][27][28][29][30][31][32][33][34]. Also, less than half of them [22,24,26,28,30,31,33] used data from clinical trials, thus excluding the effect of treatment heterogeneity when evaluating prognostic factors. Up to now, results are largely inconclusive with some studies showing favorable and other adverse impact of PIK3CA mutations on patient outcome, while a significant number of reports did not show any prognostic significance. Moreover, recent studies incorporating patient data from big randomized studies did not show any significant effect of PIK3CA status on recurrence or survival rate [22,24,26,28], which has also been shown in the present study.
Preclinical models [27,36] have demonstrated that PIK3CA mutations alone, compared to PTEN loss or AKT1 mutations, can cause weaker or more inconsistent activation of PI3K-AKT signaling. However, PTEN knockdown was shown to reinforce the ability of PIK3CA mutations to trigger this pathway. Also, PTEN loss increases tumor cell susceptibility to PI3K inhibition compared to PIK3CA mutations [27]. Therefore, PIK3CA mutations do not seem to have Notably, PTEN loss and PIK3CA mutations are regarded in the literature to be the most frequent downstream molecular aberrations affecting PI3K-AKT signaling, nevertheless only a minority of studies has examined the possible role of such combined effect [26,31,33,34]. Interestingly, each one of these studies provided different results, which may be attributed to the diverse population characteristics and treatments applied, for example only HER2 positive [33], only postmenopausal [26,31], or only hormone receptor positive cases [34]. Also, the differences in the methodologies and/or cutoffs used in assessing PTEN protein expression may have further influenced the results as well.
Although initially there was an impression that PIK3CA mutations are mutually exclusive with PTEN loss [37], this was refuted by subsequent studies [31], giving a percentage of approximately 14% of coexistence of both molecular alterations, whereas in our population the coexistence of PIK3CA mutations and low PTEN protein expression was detected in 94 patients (9%). Notably, both studies [26,34] that tested the prognostic significance of PTEN expression separately from PIK3CA mutations did not find any significant results, while those combining them [31,33] reported their independent prognostic significance. In the present study, a significant interaction was detected between PTEN expression and PIK3CA mutational status, confirmed by multivariate analysis. This may imply that patients with tumors bearing both low PTEN protein expression and PIK3CAhel mutations may be characterized by significantly worse prognosis. However, in subgroup analysis, patients with both low PTEN expression and PIK3CAhel mutations had only marginally lower 4-year DFS and OS compared to the other subgroups (p = 0.078). Therefore, this finding is suggestive of a trend and cannot support definitive conclusions, considering also the small size (only 33 patients) of this subgroup.

Conclusions
The present study showed that the different types of PIK3CA mutations do not seem to have any prognostic role in breast cancer patients treated with adjuvant chemotherapy. However, combining the effect of PIK3CA mutation status with that of PTEN expression levels led to the detection of statistically significant interactions with regard to prognosis. As the sample size was relatively small for subgroup analyses, we cannot exclude the possibility that these results might have occurred by chance. Nonetheless, there is strong preclinical rationale to support them and, therefore, this study offers interesting hypotheses to test in the laboratory or in larger cohorts of patients with the above characteristics.
Supporting Information S1