Prognostic role of PIK3CA mutations and their association with hormone receptor expression in breast cancer: a meta-analysis

The phosphatidylinositol-4, 5-bisphosphate 3-kinase, catalytic subunit alpha (PIK3CA) gene is frequently mutated in breast cancer (BCa). Sex hormone receptors (HRs), including estrogen receptor (ER) and progesterone receptor (PR) play pivotal roles in BCa. In this study, we evaluated the association between PIK3CA mutations and ER/PR expression and the prognostic role of PIK3CA mutations in BCa patients, and in particular, HR-positive BCa. Thirty-two studies involving 5719 cases of BCa obtained from database searches were examined. PIK3CA gene mutations correlated significantly with ER/PR expression (p < 0.00001) and relapse-free survival (RFS) (hazard ratio [HR] 0.76, 95% confidence interval [CI] 0.59–0.98, p = 0.03) but not overall survival (OS) (HR 1.14, 95%CI 0.72–1.82, p = 0.57) in unsorted BCa patients. PIK3CA mutations were not associated with OS (HR 1.06, 95%CI 0.67–1.67, p = 0.81) or RFS (HR 0.86, 95%CI 0.53–1.40, p = 0.55) in HR-positive BCa patients. In conclusion, PIK3CA mutations were significantly related to ER/PR expression and RFS in unsorted BCa patients. However, the clinical implications of PIK3CA mutations may vary according to different mutant exons. And PIK3CA mutations alone may have limited prognostic value for HR-positive BCa patients.

B reast cancer (BCa) is one of the most common cancers among women, with more than 1,300,000 new cases and about 450,000 deaths reported each year worldwide 1 . This highly heterogeneous disease is divided into subgroups on the basis of molecular signatures, clinicopathologic features, and responses to therapy. Hormone receptors (HRs), including estrogen receptors (ERs) and progesterone receptors (PRs) are the most important markers of BCa. Most BCa cases are HR-positive (HR1), and ER-positive (ER1) BCa accounts for up to 80% of BCa cases among women 45 years and older 2,3 . Endocrine therapy is regarded as the cornerstone of ER1 BCa treatment. However, because of de novo or acquired resistance to endocrine therapy, prognosis is still poor for many ER1 BCa patients. Therefore, finding new effective treatment methods for ER1 BCa patients resistant to endocrine therapy is imperative.
After the TP53 gene, the phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha (PIK3CA) gene is the most frequently mutated gene in BCa. Phosphatidylinositol 3-kinase (PI3K) is composed of an 85-kD (p85) and a 110-kD (p110) subunit. When coupled to activated tyrosine kinases via p85 (the adaptor subunit), p110 (the catalytic subunit) phosphorylates the 3-hydroxy group of inositol phospholipids. Gain-of-function mutations in PIK3CA have been found in different types of cancers including BCa. The mutations result in PI3K activation independent of upstream signaling and constitutive activation of the downstream AKT pathway and may contribute to oncogenesis 4 . The frequency of PIK3CA mutations in BCa cases ranges from 16.4 to 45% 5 . There are 3 mutation ''hotspots'' in the PIK3CA gene: E542K, E545K at exon 9 (helix domain) and H1047R at exon 20 (kinase domain). The 3 hotspots represent almost 80% of PIK3CA mutations and lead to constitutive PI3K activity by different mechanisms 6 .
Aberrant activation of the PI3K pathway is thought to contribute significantly to endocrine therapy resistance in patients with ER1 BCa 7 . There is evidence showing that endocrine therapy combined with p110 inhibitors is an effective treatment for ER1 BCa cases, including those with PIK3CA mutations 8 . The synthetic lethal interaction is a promising approach that needs further studies. Testing of several p110 inhibitors is underway in phase II clinical trials. Therefore, evaluation of the relationship between HRs and PIK3CA mutations in BCa is neces-sary. It is also of great clinical interest to determine whether PIK3CA mutations are prognostic factors in HR1 BCa patients.

Results
Search results and description of eligible studies. A total of 1903 potentially relevant citations were retrieved. After exclusion of nonhuman studies, reviews, and duplicates, two authors independently perused the titles and abstracts of the articles. After screenings, 68 articles were chosen for further full-text review. Ultimately, 32 eligible studies were included in our meta-analysis 5, (Figure 1).
The 32 eligible studies were published from 2004 to 2014 and involved 5719 cases. Data from the studies were grouped as follows: group A evaluated the relationship between PIK3CA mutations and ER (26 studies) or PR (20 studies) expression in BCa patients, group B (12 studies) and group C (8 studies) evaluated the relationship between PIK3CA mutations and the outcomes of all BCa patients and HR1 BCa patients, respectively. In the 32 selected studies, the percentage of patients with PIK3CA mutations ranged from 7.1% to 44.6%, and the percentage of ER1 patients ranged from 48.1% to 84.0%. For PR, the percentage ranged from 41.4% to 64.8%. In the B and C groups, the median follow-up time ranged from 50 to 153.6 months.
ER and PR expression and PIK3CA gene mutations in BCa patients. The relationship between PIK3CA gene mutations and ER expression was investigated in 4754 patients from 26 selected studies (Group A, the ER arm) using a fixed-effect model (Table 1). There was a significant association between PIK3CA gene mutations and ER expression in the patients in this group (odds ratio [OR] 1.92, 95%CI 1.65-2.23; P , 0.00001; Figure 2). Then we performed a separate analysis for PR expression in 3507 patients from 20 studies (Group A, the PR arm) using a fixed-effect model (Table 1), and found that PR expression was also significantly associated with PIK3CA mutations (OR 1.88, 95% CI 1.61-2.20; P , 0.00001) ( Figure 3). Direct sequencing was the most frequently used method for detecting mutations in the selected studies. We introduced subgroups and found that direct sequencing and the other mutation detection methods produced similar results (p 5 0.13).
PIK3CA gene mutations and prognosis in all BCa patients. Analyses were conducted to evaluate the relationship between PIK3CA gene mutations and prognosis as defined by overall survival (OS) and relapse-free survival (RFS) in all BCa patients (group B) ( Table 2). Because of significant heterogeneity among the group B studies for OS (P 5 0.008; I 2 5 66%), a random-effect model was used to assess OS correlations. However, because there      was no inter-study heterogeneity among the group B studies for RFS (P 5 0.93; I 2 5 0%), a fixed-effect model was used to assess RFS correlations. For OS, 7 studies involving 2105 patients were analyzed and no significant association between PIK3CA mutations and OS was found (HR 1.14, 95% CI 0.72-1.82; P 5 0.57) (Figure 4). We also performed analysis for different exons. For exon 9 mutations, a significant worse OS was found (HR 1.42, 95% CI 1.02-1.99; P 5 0.04). In addition, for exon 20, the results of OS did not reach a significant level (HR 1.63, 95% CI 0.93-2.85; P 5 0.09) (Figure 4). For RFS, 5 studies involving 1913 patients were analyzed, and a significant relationship between PIK3CA gene mutations and prolonged RFS was observed (hazard ratio 0.76, 95% CI 0.59-0.98; P 5 0.03) (Fig. 5).
PIK3CA gene mutations and prognosis in HR1 BCa patients. Publication bias. Publication bias was not investigated when the number of studies was less than 10 because of the low sensitivity of qualitative and quantitative tests 40 . When the number of studies was more than 10, bias was assessed by Begg's funnel plots. No evidence of obvious asymmetry was found in this analysis by visual evaluation (data not shown).

Discussion
Recently, several studies evaluating the prognosis of BCa patients suggest that PIK3CA mutations are ''good mutations''. Our metaanalysis shows that PIK3CA gene mutations are significantly associated with both ER and PR expression, which are believed to be favorable clinicopathologic features of BCa. Furthermore, in unsorted BCa patients with PIK3CA mutations, RFS was significantly improved. There are some possible explanations for the puzzling favorable effects of PIK3CA mutations. First, signaling pathways downstream of PI3K may not be active in some BCa patients with PIK3CA mutations. Loi et al. found that PIK3CA mutations were associated with relatively low mTORC1 signaling and that some AKT-regulated genes were repressed in BCa patients with PIK3CA mutations 31 . Second, dysregulated gene expression resulting from PIK3CA mutations may be advantageous. Cizkova showed that the Wnt pathway was dysregulated and WNT5A was overexpressed in ER1 BCa patients with PIK3CA mutations 41 . Interestingly, WNT5A expression has been associated with favorable outcomes in patients with invasive breast tumors 42 . Third, PIK3CA, like many other oncogenes, may induce senescence, resulting in a less aggressive phenotype after cell transformation 43,44 .
Despite of this, there was only an insignificant connection between PIK3CA mutations and OS. The improvement in RFS but not OS may suggest a BCa specific effect of PIK3CA mutations. However, considering specific exons, the effects seemed weak or even contradictory. In the future, more studies focusing on specific exons mutations, including the non-hotspot mutations of PIK3CA, are warranted.
Whether PIK3CA mutations contribute to endocrine therapy resistance remains unclear and intriguing. Another important finding of this study was that PIK3CA mutations did not affect either OS or RFS in HR1 BCa patients. In most of the studies selected for our Table 2   It also should be noted that there is some dissociation between PIK3CA mutations and activation of signaling pathways downstream of PI3K. In some phase I clinical trials, PIK3CA mutations were not strongly related to responses produced by PI3K inhibitors 17 . In our study, PIK3CA mutations were associated with favorable pro-  gnostic factors such as ER and PR expression, but are unlikely to be the single pivotal determinant of favorable responses to endocrine treatment. The gene signature associated with PIK3CA mutations was indicative of better clinical outcomes in ER1/HER22 BCa patients 45 . Perhaps its gene signature is more important than the PIK3CA mutation itself in respect to prognosis. Studies determining whether PIK3CA mutations are beneficial to tamoxifen-treated HR1 BCa patients with other molecular features such as PTEN loss or AKT1 mutations are warranted.
There were some limitations to our study. First, we only analyzed available data in the literature. Second, because of significant heterogeneity, we used the random effect model, which is not as reliable as the fixed-effects model, in some analyses. Third, we only included articles that were published in English, and language bias might exist. Fourth, data extracted from the literature may not be as reliable as data generated directly. Fifth, several related studies of high quality were not included in our analysis because ideal unified prognosis parameters were lacking. Finally, the inclusion criteria and treatment procedures were not strictly unified in the studies used for our analysis. These differences are also a potential source of heterogeneity. Therefore, a cautious interpretation of our findings is warranted given possible bias in our meta-analysis.
In summary, our results show that PIK3CA mutations are significantly related to the ER and PR expression status of BCa patients. They also correlated with improved RFS in unsorted BCa patients, but not with OS or RFS in HR1 BCa patients. As a potential biomarker, PIK3CA mutations were not prognostic for HR1 BCa patients or, most notably, ER1 BCa patients. Future studies are needed that collectively explore the possible roles of PIK3CA mutations, the activation of signaling pathways downstream of PI3K, and other important biomarkers such as the genes encoding the components of the PI3K/AKT/mTOR pathway.

Methods
Literature search and eligibility criteria. We searched PubMed and Embase databases up to April 2014 for English-language titles or abstracts that included the words ''phosphoinositide-3-kinase'', ''PIK3CA'', ''mutation'', ''breast cancer'', or ''breast neoplasms''. We also screened the references of the retrieved articles and relevant reviews for additional articles. A published article was included if it (1) evaluated the association between PIK3CA mutations and ER or PR expression in BCa patients or the association between PIK3CA mutations and BCa prognosis; (2) had sufficient data for estimating an OR with a 95% CI or a HR with a 95% CI; and (3) evaluated OS, RFS, or other survival index. The exclusion criteria were as follows: (1) letters, reviews, conference abstracts, and case reports; and (2) articles that did not provide sufficient information such as a HR for OS or had data that could not be extracted.
Data extraction and quality assessment. Two authors independently screened all publications by title or abstract for inclusion in our study. Discrepancies were resolved by group discussion, and data were extracted from eligible publications. The following information was collected: name of the first author, year of publication, source of patients, study design, mean age of the patients, percentage of ER1 and PR1 patients, percentage of patients with PIK3CA mutations, the region of the sequenced PIK3CA mutations, mutation analysis methods, outcome of BCa patients, and median follow-up time (months, range). The studies were assessed for quality according to the Newcastle-Ottawa quality assessment scale, and articles with 5 stars or more qualified for our study 46 .
Statistical analysis. An OR with a 95% CI was used to assess the strength of the association between PIK3CA mutations and ER or PR expression status. The primary end points were RFS and OS. A HR and a 95% CI were used to estimate the impact of PIK3CA mutations on RFS and OS. When a HR and a 95% CI were not given in the article, estimated values were derived indirectly from Kaplan-Meier curves using the methods described by Tierney et al. 47 . Kaplan-Meier curves were read by an Engauge Digitizer, version 4.1 (http://digitizer.sourceforge.net/), and the data from the curves were entered in the spreadsheet appended to Tierney's report 47 . A combined HR . 1 implied a worse survival for groups of patients with PIK3CA mutations. Cochran Q and I 2 statistic values were used to assess heterogeneity among the studies. For the Q statistic, a P value , 0.10 was considered statistically significant for heterogeneity 48 , and the random effects model was calculated according to the DerSimonian-Laird method 49 .Otherwise, the fixed-effects model (Mantel-Haenszel method) was used. I 2 , 50% was considered acceptable. If significant heterogeneity was found, a random-effects model was used for meta-analysis. Statistical analyses were performed using Table 3 | Main characteristics of studies that evaluated the relationships of PIK3CA mutations and the OS/RFS in HR1 breast cancer patients