In this study, to our knowledge we present the largest analysis of PIK3CA oncogenic mutations to date, using an unselected routine cohort of early stage breast cancer patients (n = 1123). The principal aim of our study was to evaluate the prevalence of PIK3CA-mutations and their associations with clinical and histopathological parameters and secondary the relation of a PIK3CA-mutation to clinical outcome. Since the landmark study of Samuels et al. [20], it has been known that presence of somatic PIK3CA-mutations promotes cancer progression also in breast cancer. Most previous publications have reported on heterogeneous sample sets including mixed sets of early stage and metastatic breast cancer patients from retrospective studies. Instead, our findings of a prospective well-defined homogenous cohort of early-stage breast cancer patients provide new insights to the realistic frequency of PIK3CA-mutations overall and in subgroups, as well as their association with recurrence-free interval and overall survival.

We detected an overall somatic mutation rate of 26.7% (300 of 1123 samples) when testing the three most common hot spots C775 (H1047R), C763 (E545K), and C760 (E542K) (https://www.mycancergenome.org/), which represent 87% of all mutations currently known in the PIK3CA gene [11]. Of interest, the highest frequencies (> 30%) of PIK3CA gene mutations were detected in tumours with more favourable characteristics (G1, G2, ER-positive, PgR-positive, luminal-like, HER2-negative), which is in line with most other studies and available data in the COSMIC database http://www.sanger.ac.uk/cosmic/http://www.sanger.ac.uk/cosmic/ [21]. In tumours with high risk biology (HER2-positive, TNBC) we found the lowest rate of PIK3CA-mutations (15.2% and 11.4), which is also consistent with published data [11].

These findings provoke the question why PIK3CA-mutations are more frequently detected in ER-positive disease. The current state of research postulates that PIK3CA-mutation-dependent activation of AKT phosphorylates and activates ER leading to transcriptional activity of ER in an oestradiol-independent manner and consecutively to preferential growth of ER-positive cancer [22, 23]. Thus, mutated PI3K likely promote ER-positive cancer growth and may explain the overrepresentation of PIK3CA-mutated tumours in luminal and well-differentiated breast cancer. In addition, PIK3CA-mutations are considered an early event in breast cancer development since they were detected even in small tumours as well as in non-invasive precursor lesions, like DCIS [24]. In contrast, fast growing ER-negative and undifferentiated tumours, however, may be derived from different precursor cells and independent of activating PIK3CA-mutations.

The second objective of our study was the prognostic impact of PIK3CA-mutations, and we did not find any association with recurrence free interval (RFI) or overall survival (OS) within the entire cohort of 1123 patients. We choose RFI as endpoint since we wanted to analyse the clean disease-related impact of PIK3CA-mutations. We tested an unselected and rather large cohort of early breast cancer patients, thus we assume that our data provide a realistic view, demonstrating a lack of a general impact of PIK3CA-mutations on the course of disease in breast cancer. This is in contrast to the published meta-analyses and single studies on PIK3CA-mutations that present conflicting results on its association to prognosis; studies found an association to better survival (e.g. Dumont et al., Pang et al. [25, 26]) as well as to inferior survival (Sobhani et al., Fan et al.) [12, 27]. These divergent results might presumably result from the heterogeneity of the populations that were studied with regard to sample size, subgroups, and type of treatment, so, selection bias cannot be excluded in these analyses (for review see [28]).

The published results are also inconclusive with regard to a potential predictive impact of PIK3CA-mutations. However, in our study we found a significant predictive value of PIK3CA-mutations in luminal breast cancer by observing more disease-related events in patients with PIK3CA-mutations. Most importantly, there was an association to the type of endocrine therapy: We found a significant impact of PIK3CA-mutations on the effect of adjuvant aromatase inhibitors, but no impact on the effect of adjuvant tamoxifen. This observation may be explained by PI3K-triggered estradiol-independent activation of the ER that can be observed in estradiol-deprived situations created by aromatase inhibition but might be blocked by ER-modulation through tamoxifen as postulated by Campbell and colleagues [22]. This differential therapy response has also been described for advanced BC by Ramirez-Ardila et al. [29].

It has to be acknowledged that these relations are complex and other mechanisms are involved. For example, recent findings suggest that PI3K pathway alterations might be associated with the composition of the tumour microenvironment in luminal breast cancer, including the attraction of CD8-positive T-cells [30]. Our observations are fully in line with the data of Stemke-Hale and colleagues who also did not find an association between PIK3CA-mutations and the effect of adjuvant tamoxifen [31]. However, data are again heterogeneous; some authors described resistance to tamoxifen [32, 33], whilst others found significantly improved endocrine sensitivity to tamoxifen if PIK3CA-mutations were detected [34].

The results from our observational cohort study support the finding that PIK3CA-mutations may indicate resistance to aromatase inhibitor therapy; however, prospective studies are lacking.

In patients with HER2-positive breast cancer we found no impact of PIK3CA-mutations on RFI. Similarly, a well described pooled analysis of five prospective clinical trials showed no significant impact of PIK3CA-mutation on the course of disease in patients with adequately treated HER2-positive BC, although the PIK3CA-mutated group had a significantly lower pCR rate [35]. In an uniformly treated early-stage HER2-positive Danish cohort, the PIK3CA-alterations predicted a significantly worse OS (adjusted HR 2.14), but had no significant impact on invasive disease free survival (iDFS) presumably due to the small sample size [36].

An exploratory analysis of the CLEOPATRA trial identified a subgroup of HER2-positive PIK3CA-mutated patients who were resistant to anti-HER2-therapy with trastuzumab and pertuzumab (worse OS if mutated, adjusted HR 1.48, p = 0.0025) [37]. Contradictory results may be explained by the continued activation of PI3K and an inhibitory effect on HER2 signaling [38]. Thus, in patients with HER2-positive tumours the impact of PIK3CA-mutations is not clear; at least the effects are small and not significant.

The presence of PIK3CA-mutations may have a favourable impact in early TNBC, suggested by a 2.3-times improved RFI and a 3-times improved OS, which is in line with Mosele and Takeshita, even though they worked with samples from advanced BC [39] analysing cell-free DNA [40].

To the best of our knowledge, this is the first study using CART in order to find out if combinations of variables could predict the risk of an RFI event. For nearly one fifth of the patients the gene modifications seem to have a relevant prognostic impact depending on the SHR status of the tumour. Overall, using the CART algorithms (see Fig. 3) for 121 of 1123 patients, the presence of gene alterations predict a worse prognosis in defined subgroups. Patients with PIK3CA-mutated tumours (n = 20) had lower pCR rates than wildtype tumours. This observation is similar to other studies [35].

Our real world data from a multicentre cohort adds exploratory, but valuable information, as our patients were consecutively enrolled in the daily clinical routine.