Tumor mutations predict HER2-targeted therapy resistance in primary HER2-positive breast cancer

Abstract

The presented study investigated the relevance of mutations in 17 cancer genes and response to neoadjuvant chemotherapy in two clinical cohorts of HER2+ breast cancer. 364 samples from HER2+ tumors of the neoadjuvant studies GeparTrio (no anti-HER2 treatment, n = 71) and GeparSepto (dual HER2 blockade and randomization for paclitaxel vs. nab-paclitaxel, n = 293) were analyzed by targeted next generation sequencing of hot spot regions of 17 genes. Mutations in TP53 (47.3%) and PIK3CA (23.9%) were most prevalent. EGFR, KRAS, NRAS, HRAS were combined to the MAPK module with 2.5% harboring mutations. In GeparSepto, the pCR rate was significantly lower in PIK3CA-mutant vs wild-type (wt) tumors (47.7% vs. 66.7%; p = 0.009). In patients treated with nab-paclitaxel, pCR rates were significantly lower in PIK3CA-mutated tumors compared to wt-tumors (38.7% vs. 72.0%; p = 0.001). In the GeparTrio cohort without neoadjuvant anti-HER2 therapy the pCR rate was 27.3% in the mutant cohort compared to 16.3% in the PIK3CA-wt cohort (p = 0.339). In HER2+ breast cancer, PIK3CA mutations were significantly associated with reduced response to dual HER2 blockade with pertuzumab+trastuzumab as well as reduced response to nab-paclitaxel. This reduction was not observed in GeparTrio without anti-HER2 therapy.

Introduction

Investigating the subtype-specific mutational landscape in breast cancer is of interest and becomes relevant for breast cancer treatment¹. Due to considerable advances in the implementation of new technologies, it is possible to perform targeted next-generation sequencing (NGS) in formalin-fixed paraffin embedded (FFPE) core biopsy samples from clinical studies². This method allows for parallel interrogation of a pre-selected set of genes to obtain a more comprehensive picture of clinically relevant genetic alterations in breast cancer. Mutations in TP53, encoding a non-druggable tumour suppressor-gene and in PIK3CA, encoding a targetable oncogenic driver, are most prevalent in breast cancer³. PIK3CA mutations are more frequently detected in hormone receptor (HR)-positive (+) and HER2-positive (+) breast cancer compared to other subtypes⁴. Previous reports demonstrated that PIK3CA mutations predict lower pathological complete response (pCR) to dual blockade with trastuzumab and lapatinib as well as trastuzumab plus afatinib in HER2+ primary breast cancer^5–9, whereas no correlation of pCR and TP53 mutations was observed¹⁰. In vitro studies have already demonstrated that inhibition of the PI3K pathway in PIK3CA mutated HER2+ breast cancer cell lines can overcome resistance to anti-HER2 therapy¹¹. Ongoing trials investigate promising combinations of anti-HER2 therapy with PI3K inhibitors and endocrine therapy omitting chemotherapy in the metastatic setting¹². The neoadjuvant GeparPiPPa trial investigates the combination of trastuzumab, pertuzumab and endocrine therapy +/- the PI3K inhibitor inavolisib in hormone receptor and HER2+ breast cancer harbouring a PIK3CA mutation.

With respect to the association of PIK3CA mutations and pCR it is still unknown whether PIK3CA mutations are specifically affecting the efficacy of anti-HER2 therapy or whether it reflects a lower sensitivity to neoadjuvant chemotherapy. To address this specific question and to analyse the predictive power of mutations of other breast cancer-related genes and pathways, we employed targeted NGS of 17 breast-cancer related genes in pre-therapeutic biopsies of two prospective neoadjuvant trials: 1) a clinical cohort GeparTrio (G3) that was treated with vinorelbine-capecitabine versus docetaxel-doxorubicin-cyclophosphamide in an early nonresponsive breast cancer setting before the era of trastuzumab, and 2) the HER2+ cohort from the GeparSepto (G7) study with pertuzumab/trastuzumab in addition to an anthracycline/taxane based therapy with randomization for paclitaxel vs. nab-paclitaxel^13–15.

Results

Baseline factors

A total of 448 tumour samples were available in the biobank, and NGS analysis was successful in 364 (81%) of these samples. In the G7 study, 293 of 396 patients with HER2 + breast cancer that had started treatment could be successfully sequenced. Median age in the analysed cohort was 50 years (range 22-75); most tumours were cT1-2 (89.9%); cN0 (54.4%); ductal invasive (88.7%), grade 3 (53.9%), HR+ (69.6%), Ki67 > 20% (69.3%), LPBC-negative (83.2%) (Supplementary Fig. 1). The analysed cohort contained more patients with ER/PgR negative tumours (30.4% vs 17.5%; p = 0.014) and more patients with grade 3 tumours (53.9% vs 39.8%; p = 0.016) than the non-analysed cohort. The median Ki67 was higher (30% vs. 25%; p = 0.027) as was the median stromal TILs (20% vs 10%; p = 0.006). The pCR rate was higher in the analysed compared to the not analysed cohort (62.5% vs. 44.7%; p = 0.002).

The G3 study cohort consists of 586 known HER2+ patients; of those 94 provided a sample and sequencing data from 71 patients could be obtained (Supplementary Table 1).

Frequency of mutations and association with clinical-pathological factors

A total of 290 non-synonymously mutated genes were detected in the 364 tumour samples, the most commonly mutated genes were TP53 (172/364; 47.3%) and PIK3CA (87/364; 23.9%), respectively. EGFR, KRAS, NRAS, HRAS were combined to the MAPK-signalling module with 9/364 (2.5%) mutated tumour samples. 151 tumours had no, 148 one, 58 two, and 7 three or more mutated genes. Overall, 141/252 (56.0%) HR + /HER2+ and 72/112 (64.3%) HR-/HER2+ tumours had a mutation. HR + /HER2+ tumours had TP53 mutations in 112/252 (44.4%); PIK3CA mutations in 53/252 (21.0%) and MAPK-module mutations in 5/252 (2.0%). In HR-/HER2+ tumours 60/112 (53.6%) had a TP53 mutation, 34/112 (30.4%) a PIK3CA mutation and 4/112 (3.6%) MAPK-module mutations (Fig. 1). In the PIK3CA mutant tumours 31 tumours had a mutation in exon 9 and 44 tumours in exon 20, two had a mutation in both exons and 14 tumours neither had an exon 9 nor an exon 20 mutation but another one. PIK3CA mutations correlated with TP53 and MAPK-module mutations: 54/87 (62.1%) of PIK3CA mutant tumours had TP53 mutations compared to 118/277 (42.6%) TP53 mutations in PIK3CA wt tumours (p = 0.002). 5/87 (5.7%) tumours with PIK3CA mutations also had a mutation in the MAPK-signalling pathway compared to 4/277 (1.4%) MAPK-signalling pathway mutations in the PIK3CA non-mutant cohort (p = 0.039).

Fig. 1. Mutation landscape in HER2+ early breast cancer (G3 and G7 trials, n = 364).

A Non-silent mutations were detected in 14 of the 17 investigated genes. For each of the genes, the percentage of the mutated tumors is displayed. B Comparison of the HR+ (n = 252) and the HR- (n = 112) breast cancer subtype in pooled trials.

In the G7 cohort, 45.4% had a TP53 mutation (133/293), and grade 3 tumours had a higher TP53 mutation rate (88/158 (55.7%) for grade 3; 45/135 (33.3%) for grade 1-2; p < 0.001). There was no correlation with the hormone-receptor status (88/204 (43.1%) in HR + ; 45/89 (50.6%) in HR-; p = 0.253). In 65/293 tumours (22.2%) a PIK3CA mutation was found. The PIK3CA mutation rate was 20.1% in HR+ (41/204) and 27.0% (24/89) in HR- cohort (p = 0.222). 56.9% (37/65) of the tumours with a PIK3CA mutation also had a TP53 mutation vs. 42.1% (96/228) of the PIK3CA wild-type cohort (p = 0.048).

In G3 39/71 (54.9%) had a TP53 mutation; 22/71 tumours had a PIK3CA mutation (31.0%) and 5/71 (7.0%) had MAPK-signalling pathway mutations. 24/48 (50.0%) HR+ and 15/23 (65.2%) HR- tumours had a TP53 mutation (p = 0.309). The PIK3CA mutation rate was 25.0% (12/48) in HR+ and 43.5% (10/23) in HR- cohort (p = 0.170). 2/48 HR+ (4.2%) and 3/23 (13.0%) HR- tumours had a MAPK-signalling pathway mutation (P = 0.320). 77.3% (17/22) of the tumours with a PIK3CA mutation also had a TP53 mutation vs. 44.9% (22/49) of the PIK3CA wild-type cohort (p = 0.019). 18.2% (4/22) had a mutation in the MAPK-signalling pathway in PIK3CA mutant cohort vs 2.0% (1/49) in PIK3CA wild-type cohort (p = 0.030).

Correlation of mutations with response (pCR)

In G7, 104/168 (61.9%; 95% confidence interval (CI) 54.1–69.3%) patients with any tumour mutation showed a pCR compared to 79/125 (63.2%; 95%CI 54.1–71.6%) without a mutation (p = 0.903). In G7 no differences in pCR rates were detected based on the TP53 mutation status: the pCR rate was 64.7% (95%CI 55.9–72.7%) in the group with a TP53 mutation (86/133) vs 60.6% (95%CI 52.6–68.2%, 97/160) in the group without (p = 0.545). In the HR+ cohort 53/88 (60.2%; 95%CI 49.2%-70.5%) with TP53 mutation had a pCR compared to 65/116 (56.0%; 95%CI 46.5–65.2%) without mutation (p = 0.570). In the HR- cohort 33/45 (73.3%; 95%CI 58.1–85.4%) with TP53 mutation and 32/44 (72.7%; 95%CI 57.2–85.0%) without TP53 mutation (p = 1.000) had a pCR.

Overall in the G7 cohort, the pCR rate was significantly lower in the PIK3CA mutant tumours compared to the wild-type group (47.7%; 95%CI 35.1%-60.5%, 31/65) vs. 66.7% (95%CI 60.1–72.8%, 152/228); Odds Ratio (OR) 0.46 (95%CI 0.261–0.797), p = 0.009); this was confirmed in multivariate analyses (OR 0.41 (95%CI 0.224–0.742) p = 0.003) (Supplementary Table 2). The observed effect in G7 was particularly strong in the HR- population ((13/24) 54.2%; 95%CI 32.8–74.4%) vs. (52/65) 80.0% (95%CI 68.2%-88.9%); p = 0.029), but was also observed as a trend in the HR+ population ((18/41) 43.9% (95%CI 28.5–60.3%) vs. (100/163) 61.3% (95%CI 53.4–68.9%); p = 0.052) (Fig. 2). There was no significant difference in the pCR rate between PIK3CA mutant and wild-type tumours in the G3 cohort, that was treated without neoadjuvant anti-HER2 therapy ((6/22) 27.3% (95%CI 10.7–50.2%) vs (8/49) 16.3% (95%CI 7.3–29.7%); univariate OR 1.92 [95%CI 0.575–6.419] p = 0.339; multivariate OR 0.92 [95%CI 0.132–6.441] p = 0.935) Fig. 2 (Supplementary Table 2).

Fig. 2.

Pathological complete response rates according to PIK3CA mutation status in HER2+ breast cancer from G3 without anti-HER2 treatment and G7 with Trastuzumab and Pertuzumab overall and by hormone receptor status.

The role for PIK3CA mutations on response to the two different taxanes in G7 was investigated. In the nab-paclitaxel group, pCR rates were significantly lower in patients with PIK3CA mutations compared to those without ((12/31) 38.7% (95%CI 21.8%-57.8%) vs. (85/118) 72.0% (95%CI 63.0%-79.9%); p = 0.001), whereas in the paclitaxel group, there was no significant difference between patients with and without a PIK3CA mutation ((19/34) 55.9% (95%CI 37.9%-72.8%) vs. (67/110) 60.9% (95%CI 51.1–70.1%); p = 0.690) (Fig. 3A) (Table 1; Supplementary Table 2). The respective interaction could be demonstrated in univariate (p_interaction = 0.039) as well as multivariate regression analysis (p_interaction = 0.010) after adjusting for known baseline factors. This effect was seen in HR+ and HR- tumours (Fig. 3B) (Supplementary Table 3). Overall, neither TP53 nor MAPK-module mutations showed a significant association with therapy response. In the wild-type TP53 cohort patients treated with nab-paclitaxel HR- tumours achieved a significantly higher pCR rate than HR+ tumours ((18/21) 85.7% (95%CI 63.7–97.0%) vs. (36/61) 59.0% (95%CI 45.7–71.4%); p = 0.033). This result could not be observed in any other subgroup.

Fig. 3. pCR rates according to PIK3CA mutation status from G7.

A overall and by taxane treatment B by hormone receptor status and taxane therapy.

Table 1.

pCR rates and logistic regression models for pCR in G7 by PIK3CA mutation status in different subcohorts

pCR rate			Univariable analysis for prediction of pCR		Multivariable analysis for prediction of pCR
PIK3CA wild type	PIK3CA mutated	p-value (Fishers exact test)	OR mt vs wt (95% CI)	p-value	OR mt vs wt (95% CI)	p-value
All patients G7 – dual anti-HER2 blockade
66.7% (60.1–72.8%)	47.7% (35.1–60.5%)	0.009	0.456 (0.261–0.797)	0.006	0.407 (0.224–0.742)	0.003
All patients G3 – no anti HER2 blockade
16.3% (7.3–29.7%)	27.3% (10.7–50.2%)	0.339	1.922 (0.575-6.419)	0.288	0.922 (0.132–6.441)	0.935
Subgroup analysis for G7
G7 – all HR+ tumors
61.3% (53.4–68.9%)	43.9% (28.5–60.3%)	0.052	0.493 (0.247–0.986)	0.045	0.497 (0.240–1.029)	0.060
G7 – all HR- tumors
80.0% (68.2–88.9%)	54.2% (32.8–74.4%)	0.029	0.295 (0.108–0.809)	0.018	0.237 (0.074–0.755)	0.015
G7 – all tumors nab paclitaxel therapy
72.0% (63.0–79.9%)	38.7% (21.8–57.8%)	0.001	0.245 (0.107–0.561)	0.001	0.133 (0.048–0.364)	<0.001
G7 – all tumors paclitaxel therapy
60.9% (51.1–70.1%)	55.9% (37.9–72.8%)	0.690	0.813 (0.373–1.769)	0.602	0.913 (0.398–2.092)	0.830
G7 – HR+ nab paclitaxel therapy
65.9% (54.8–75.8%)	31.3% (11.0–58.7%)	0.013	0.235 (0.075–0.742)	0.014	0.193 (0.053–0.703)	0.013
G7 – HR+ paclitaxel therapy
56.4% (44.7–67.6%)	52.0% (31.3–72.2%)	0.818	0.837 (0.339–2.066)	0.700	0.866 (0.334–2.250)	0.768

Correlation of PIK3CA mutations with outcome

Patients enrolled in the G3 trial who had tumours with PIK3CA mutations showed a trend towards worse invasive disease-free survival (iDFS) compared to those with PIK3CA wild-type tumours, although this difference did not reach statistical significance (HR 2.1 (95%CI 0.95–4.78); p = 0.066) (Fig. 4). Interestingly, this observation was not mirrored in the G7 cohort (HR 1.1 (95%CI 0.56–2.01); p = 0.869). When analyzing HR subgroups, no significant differences in iDFS were observed and this was also the case when stratified by taxane treatment (Supplementary Fig. 2).

Fig. 4.

iDFS according to PIK3CA mutation status in HER2+ breast cancer from G3 without anti-HER2 treatment and G7 with Trastuzumab and Pertuzumab.

Regarding overall survival (OS) in the G3 cohort, there was a slight but nonsignificant trend towards poorer survival in patients with PIK3CA mutations (HR 2.2 (95%CI 0.84–5.87); p = 0.109) (Fig. 5). However, no OS difference was seen in the G7 cohort, where patients were treated with pertuzumab and trastuzumab (HR 1.1 (95%CI 0.45–2.81); p = 0.805). Subgroup analyses based on HR status and taxane treatment also showed no significant effect of PIK3CA mutations on OS (Supplementary Fig. 3).

Fig. 5.

OS according to PIK3CA mutation status in HER2+ breast cancer from G3 without anti-HER2 treatment and G7 with Trastuzumab and Pertuzumab.

Discussion

In this analysis a total of 364 formalin-fixed paraffin embedded core biopsy samples from HER2+ tumors were analyzed by targeted NGS, the analysis could be conducted in 81% of samples that were available in the clinical study biobank. We interrogated hot spot regions of 17 genes, confirmed previous results, and found TP53 and PIK3CA mutations to be the most common alterations in HER2+ breast cancer¹⁶. Interestingly, PIK3CA mutation frequencies in G3 reflect other cohorts whereas the mutation rates in the G7 subset (22.2%) are slightly lower¹⁷.

Regarding pCR rates the data are to some extend comparable to the TCGA HER2+ subset (55% pCR) and the analysis of the CALGB40601 (56%) and the GeparSixto set (52%), although the selection of hotspots, the methods used and the cohort characteristics varied^10,18. In contrast to the CALGB analysis we, as well as the Neo-Altto study group, did not find a correlation of pCR with TP53 mutations but with PIK3CA mutation status¹⁹. However, there are remarkable differences between our analysis and the CALGB40601: the anti-HER2 treatment consists of different anti-HER2 agents and the dual blockade of trastuzumab and lapatinib given over 16 weeks with paclitaxel alone resulted in a lower overall pCR rate (ypT0/is ypN0 (52%), whereas in G7 all patients received the dual blockade with pertuzumab and trastuzumab over 24 weeks in addition to a taxane and anthracycline based chemotherapy resulting in a higher pCR rate using the same definition (71.3%).

It has been an open question whether the effect of PIK3CA mutations is specific for the anti-HER2 therapy or if it reflects an altered sensitivity to any type of neoadjuvant treatment, mainly chemotherapy. To answer this question, we have studied a clinical cohort that was treated before the era of trastuzumab with neoadjuvant chemotherapy only. Interestingly, in this cohort PIK3CA mutations were not significantly associated with a lower pCR rate although pCR rates were low in both subgroups. This supports the theory that PIK3CA mutation status predicts resistance specifically to anti-HER2 treatment. In contrast to previous results using trastuzumab and lapatinib as anti-HER2 treatment, there was no interaction between PIK3CA and hormone-receptor status in the cohort with the dual blockade with trastuzumab and pertuzumab^5,7,20.

In the NeoSPHERE study, investigating trastuzumab/pertuzumab in addition to neoadjuvant taxanes, a similar trend for a lower pCR rate in the group with PIK3CA mutation was shown^17,21. The Tryphaena study investigating several chemotherapy regimens with trastuzumab and pertuzumab demonstrated a numerically lower pCR rate in the group with a PIK3CA mutation compared to the wild-type PIK3CA status but the power was too low to show a significant effect (48.7% vs 64.3%; p = 0.172)^22,23. In the Cleopatra study, patients with a PIK3CA mutant tumour had a significantly worse PFS but the mutation status did not predict benefit from dual blockade^24,25. Previous studies discussed included mainly chemotherapy backbones containing paclitaxel or docetaxel. A recent Chinese study compared all three taxanes combined with carboplatin, trastuzumab and pertuzumab and reported the highest pCR rates with nab-paclitaxel. Unfortunately, associations of PIK3CA mutations with pCR were not reported²⁶. In the G7 study, a significant interaction of mutant PIK3CA with the more potent chemotherapy regimen nab-paclitaxel followed by epirubicin and cyclophosphamide could be demonstrated irrespective of the hormone-receptor status. While we do not have experimental data that explain our statistical results on PIK3CA-mutations and worse response towards nab-paclitaxel, several studies reported on mutant PIK3CA conferring resistance towards chemotherapy regimens via modulating and binding to tubulin isoforms that play a major role in mitosis^27,28. Hence, accumulation of clones with mutant PIK3CA could be a resistance mechanism driving cell survival that is actually selected by administration of mitosis inhibitors with a particularly high concentration in target cells as is the case with nab-paclitaxel (compared to standard paclitaxel). In contrast, patients with PIK3CA mutant tumors treated with trastuzumab-emtansine (T-DM1) in metastatic and early breast cancer did not experience worse outcome^29,30.

The NeoPHOEBE trial already started investigating the combination of a PI3K inhibitors with anti-HER2 therapy and paclitaxel³¹, but was suspended due to hepatotoxicity. In the currently recruiting GeparPiPPa trial only patients with hormone receptor positive tumors harboring a PIK3CA mutation can be included and will be treated with the combination of trastuzumab, pertuzumab and endocrine therapy +/- the PI3K inhibitor inavolisib. Future escalation and de-escalation strategies for HER2+ breast cancer should be tailored to taking the PIK3CA mutation status in consideration as PIK3CA mutations lead to a ligand-independent activation of the PI3K–AKT–mTOR axis downstream of HER2. For this reason, dual blockade strategies of HER2 and PI3K appear most promising for these breast cancer cases.

This is the first study investigating the mutational landscape in HER2+ breast cancer that did not receive an anti-HER2 targeted therapy in the neoadjuvant setting. Highlighting that PIK3CA mutations become relevant when the HER2 pathway is inhibited as otherwise the HER2 pathway is the dominant growth promoting factor in these tumors. Therefore, PI3K and HER2 targeting combinations are of key interest in conferring reduced response to HER2 targeted therapies. Moreover, this study shows for the first time limited therapeutic potential of nab-paclitaxel in HER2+ and PIK3CA mutated breast cancer. However, there exist limitations of this analyses as the investigated sample size was small, especially for the G3 cohort, and did not entirely reflect the original G7 cohort so results regarding nab-paclitaxel efficacy and subgroups have to be interpreted with caution.

Targeted NGS on FFPE core biopsies reliably identified in this study the most common point mutations in HER2+ breast cancer. PIK3CA mutations in HER2+ breast cancer predict resistance to anti-HER2 therapy and neither TP53 nor MAPK-module mutations showed a significant effect on response and treatment effect.

Methods

Material, patients and treatment

Pretherapeutic FFPE core biopsies prospectively collected in the German Breast Group tumour biobank from HER2+ breast cancer patients treated within the prospectively randomised GeparSepto (AGO-B-032, NCT01583426) (G7)¹⁵ and GeparTrio (NCT 00544765) (G3)^13,14 trials were used. In the G3 study, the cut-off for the local hormone-receptor status was 10%; the central HER2-status was determined retrospectively, and only HER2+ tumours were included in the mutation analysis. In the G7 study, HER2 status (positive if IHC3+ or ISH ratio >2.0) and hormone receptor status (ER/PR positive if >1% stained cells) was centrally assessed for all patients prior to randomization.

In the G3 study³² women with previously untreated locally advanced or operable (T2-4, M0) primary breast cancer and written informed consent received two cycles of TAC (T, docetaxel 75 mg/m² i.v. on day 1 every 3 weeks; A, doxorubicin 50 mg/m² i.v. bolus; C, cyclophosphamide 500 mg/m2 i.v. bolus on day 1 every 3 weeks). Early responders were randomized to four or six further TAC cycles. Patients without an early response were randomized to four further cycles of TAC or four cycles of NX (N, vinorelbine 25 mg/m² i.v. on days 1 and 8 every 3 weeks; X, capecitabine 2000 mg/m² orally twice daily on days 1–14 every 3 weeks). Patients in this trial did not receive any neoadjuvant anti-HER2-therapy.

The G7 study¹⁵ enrolled women with previously untreated, unilateral or bilateral primary invasive breast carcinoma after written informed consent for study participation. Inclusion criteria were tumour size of ≥2 cm (cT2 - cT4a-d) without additional risk factors or ≥1 cm with either clinical or pathological (positive sentinel node biopsy in case of cN0 or core biopsy in cN + ) nodal involvement, or hormone-receptor-negative, or HER2+ on IHC staining, or a Ki67 > 20%. Patients were randomized to either weekly nab-paclitaxel or solvent-based paclitaxel for 12 weeks followed by 4 cycles of conventionally dosed epirubicin and cyclophosphamide. Nab-paclitaxel was given initially at a dose of 150 mg/m² i.v. weekly but was reduced to 125 mg/m² weekly based on a recommendation of the independent data monitoring committee due to an excess of grade ≥3 peripheral neuropathy after recruitment of 400 patients. Paclitaxel was given at 80 mg/m² i.v. weekly. All patients were scheduled to receive epirubicin 90 mg/m² i.v. and cyclophosphamide 600 mg/m² i.v. each every 3 weeks for 4 cycles. Patients with HER2+ tumors received trastuzumab 8 mg/kg loading dose i.v. followed by 6 mg/kg i.v. and pertuzumab 840 mg loading dose i.v. followed by 420 mg i.v. each every 3 weeks simultaneously to all chemotherapy cycles. After surgery trastuzumab was continued for a total duration of 1 year.

Eligibility criteria were otherwise comparable between the two studies. All patients provided written informed consent for study participation and biomaterial collection. The relevant authorities and ethics committees approved the studies (Bundesinstitut für Arzneimittel und Medizinprodukte (BfArM) — for both GeparTrio and GeparSepto trials, GeparTrio NCT00544765, GeparSepto NCT01583426, Registration dates not applicable) and the study was performed in accordance with the Declaration of Helsinki. The REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) criteria were followed³³.

Mutation assessment

A breast cancer hotspot panel was designed to cover the most frequent mutated hotspots in breast cancer with a special focus on the PI3Kinase pathway³⁴. We included mutation hotspots in known oncogenes and tumour suppressor genes that were detected in at least 5 breast carcinomas in the COSMIC (catalogue of somatic mutations in cancer) database (http://cancer.sanger.ac.uk/cosmic, 6.11.2014). Additionally, we included hotspots of the genes of PIK3 pathway that were detected in at least 3 breast carcinomas in COSMIC. The resulting DNA sequencing panel included the following 17 genes: TP53 (6 amplicons), PIK3CA (7 amplicons), CDH1 (2 amplicons), FBXW7 (2 amplicons), PTEN (2 amplicons) as well as AKT1, ATM, BRAF, ERBB2, EGFR, ESR1, FGFR2, HNF1A, HRAS, KRAS, NRAS, SF3B1 (one amplicon each). Samples having a minimum coverage of 500 at the two most important mutation hotspot in PIK3CA (p.542/p.545 and p.1047) were included in the study. Mean coverage was 6520 (G3) and 6346 (G7) per amplicon, respectively. Only non-synonymous mutations with allele frequencies ≥10% were taken into consideration.

DNA preparation

Two consecutive 5 µm thick sections of FFPE core biopsies were prepared. The first section was stained with hematoxylin/eosin and the tumor containing area was assessed by a pathologist. Sections with tumor cell content ≥20% were subjected to fully automated DNA extraction using VERSANT kPCR Sample Prep (Siemens Healthcare GmbH, Erlangen, Germany) following the manufacturer’s instructions.

Sequencing analysis

Semiconductor sequencing was executed on an Ion Torrent Personal Genome Machine (PGM) using the customer designed gene panel IAD68218_166.

Total nucleic acid concentrations were measured with a Qubit fluorometer HS DNA Assay (Life Technologies, Carlsbad, CA, USA) and a TaqMan RNase P Detection Reagents Kit (Life Technologies). 5 µL gDNA were subjected to library preparation with customer made IAD68218_166 panel and Ion AmpliSeq™ Sample ID Panel (Life Technologies) to prevent sample swap. The final library was quantified using an Ion AmpliSeq Library Kit 2.0 (Life Technologies). Samples were 16-fold multiplexed and amplified on Ion Spheres Particles using the Ion OneTouch™ 200 Template Kit v2 DL (Life Technologies). After library enrichment and quality control on a Qubit instrument (Ion Sphere Quality Control Kit, Life Technologies), samples were sequenced using the Ion 318 chip v2 with an adapted standard protocol using 330 flows³⁵. Base calling and alignment to the human genome (hg19) were executed with the Torrent Suite Software 4.0.3.

NGS data processing

After gene selection we employed the dbSNP database and the Exome Variant Server (http://evs.gs.washington.edu/EVS/) to identify and exclude common germline variants from the analysis^36,37. Raw data processing and variant calling were executed using the recent Ion Torrent standard protocol (somatic variant calling: low stringency). VCF and BAM files were obtained from the Ion Torrent server. VCF files were imported into Ion Reporter and annotated using the workflow “Annotate variants single sample, version 4.4”.

The mean coverage of each amplicon in each sample was obtained from the BAM files. Cases with coverage ≥500 of both of the amplicons PIK3CA_exon10_p.545 and PIK3CA_exon21_p.1047 were included. Only variants with a frequency of ≥10% were used for further analysis.

Statistical analysis

The Clopper-Pearson method was used for confidence intervals for pCR rates. Associations between the genes status (wild-type vs mutant), clinico-pathological characteristics and pathological complete response rate were investigated with Fisher´s exact tests for binary and χ2 tests for categorial variables. In addition, univariate analyses using a binary logistic regression model were performed to estimate the magnitude of the effect. OR with 95%CIs and 2-sided p-values are presented. In addition, multivariable logistic regression models were constructed to adjust for clinical variables such as age (continuous), tumour stage (T1-2 vs. T3-4), nodal stage (N0 vs. N + ), grading (G1-2 vs. G3), Ki67 (continuous), HR (pos vs. neg), lymphocyte predominant breast cancer (>= 60% tumour infiltrating lymphocytes) (LPBC yes vs no), arm (G7Pac vs. G7nab-Pac vs. G3), HER2 treatment (yes vs no)³⁸. Interaction tests based on logistic regression were performed. Subgroups by cohorts (G3, G7), treatment and HR within G7 were analysed. All p-values are two-sided, with a p-value ≤ 5% considered to be statistically significant. No adjustment for multiple testing was performed. Statistical analysis was performed using SAS version 9.2 (SAS Institute Inc, Cary, NC) and R version 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Author contributions

MvM., K.W., V.N., S.L. wrote the main manuscript, interpreted the results and prepared the figures.MU, Christine Solbach, A.S., J.B., S.S., C.H., T.L., V.B.R., J.H., P.A.F., B.A., K.R., M.R. provided patient data and samples and interpreted the results. N.P., P.J., D.T., A.L., Christine Sers, M.H., W.S., C.D. processed the samples, carried out the analyses and interpreted the results. All authors reviewed, edited and approved the manuscript.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

Marion T. van Mackelenbergh reports personal fees, honoraria or travel grants from Amgen, AstraZeneca, Daiichi Sankyo, Genomic Health, GSK, Lilly, Molecular Health, MSD, Mylan, Novartis, Pfizer, Pierre Fabre, Roche, Seagen. Karsten Weber and Valentina Nekljudova report to be GBG Forschungs GmbH, employees. GBG Forschungs GmbH received funding for research grants from Abbvie, AstraZeneca, Celgene, Novartis, Pfizer, Daiichi Sankyo, Gilead, MolecularHealth, Menarini Group, Greenwich LifeSciences (paid to the institution); Honoraria for Consulting or Advisory Role from Pfizer, Roche, Novartis, Seattle Genetics, Lilly, AstraZeneca/MedImmune, Bristol-Myers Squibb, Abbvie, Amgen, Diiachi Sankyo, GlaxoSmithKline, Eisai Europe, Relay Therapeutics, Sanofi, Olema Pharmaceuticals, Menarini Group, MSD Oncology, BeiGene, Bicycle Therapeutics, BMS, Jazz Pharmaceuticals (paid to the institution); honoraria as invited speaker from AstraZeneca, Daiichi Sankyo Europe GmbH, Novartis, Pfizer, Roche, Gilead Sciences, Seattle Genetics, Menarini Group, Agendia, Bayer, Medscape, MSD, Stemline Therapeutics, Streamedup! (paid to institution). GBG Forschungs GmbH has the following royalties/patents: EP14153692.0, EP21152186.9, EP15702464.7, EP19808852.8 and VM Scope GmbH. Michael Untch reports personal fees for lectures and/or consultancy from Abbvie, Amgen, AstraZeneca; Daiichi Sankyo, Lilly, MSD Merck, EurBio Scientific, Myriad Genetics, Novartis, Pierre Fabre, Pfizer, Gilead, Roche; Sanofi Aventis, Seagen, Menarini Stemline; BeOne, Sandoz Hexal. Andreas Schneeweiss reports Research Grant from Celgene, Roche, AbbVie. Personal Fees (Travel expenses): Celgene, Roche, Pfizer, AstraZeneca. Personal Fees (Honoraria): Roche, Celgene, Pfizer, AstraZeneca, Novartis, MSD, Tesaro, Lilly, Seagen, Gilead, GSK, Bayer, Amgen, Pierre Fabre.Paul Jank reports grants and travel expenses from Gilead Sciences GmbH. Jens Blohmer reports Honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events: Payment for presentations at satellite symposia: AstraZeneca, Daiichi-Sankyo, Eisai, Gilead, MSD, Lilly, Novartis, Pfizer, Roche, Seagen. Support for attending meetings and/or travel: AstraZeneca, Daiichi-Sankyo, Gilead, Pfizer, Roche. Participation on a Data Safety Monitoring Board or Advisory Board: AstraZeneca, Daiichi-Sankyo, Eisai, Gilead, MSD, Lilly, Novartis, Pfizer, Roche, Seagen. Claus Hanusch reports advisory role and speakers bureau for AstraZeneca, Roche, Novartis and Aristo Pharma. Theresa Link reports receiving honoraria for lectures or presentations from Amgen, Roche, MSD, Novartis, Pfizer, Lilly, GSK, Gilead, AstraZeneca, Daiichi Sankyo, Stemline, Seagen; is on advisory board/plays an advisory role for MSD, Roche, Pfizer, Lilly, Myriad, Esai, GSK, Gilead, Daiichi Sankyo, Roche, Astra Zeneca; and has received support for attending meetings and/or travel from Pfizer, Astra Zeneca, Gilead, Daiichi Sankyo, Stemline. Vesna Bjelic-Radisic reports honoraria for lectures: Amgen, Roche, Astra Zeneca, Novartis, Pfizer, Comessa, Lilly, Gilead, Mentor, pfm, Mammotome, AB: Roche, Astra Zeneca, Novartis, Pfizer, Lilly; clinical trials: Roche, Astra Zeneca, Novartis, Pfizer, Lilly, pfm; research support: Roche, Astra Zeneca, Ratiopharm. Jens Huober received honoraria from Lilly, Novartis, Roche, Pfizer, AstraZeneca, Gilead, Daiichi, Stemline, MSD, and Abbvie, reports consulting or advisory relationships with Lilly, Novartis, Roche, Pfizer, AstraZeneca, Gilead, Daiichi, Abbvie, and received travel expenses from Roche, Daiichi, Gilead, Astra Zeneca. Peter A. Fasching: Advisory Board, Invited Speaker (Personal Fees): Novartis, Daiichi-Sankyo, AstraZeneca, Eisai, Merck Sharp and Dohme, Lilly, SeaGen, RocheGilead, Mylan. Advisory Board, Invited Speaker, Research Grant (Grant, Personal Fees): Pfizer. Institutional Funding (Grant): Biontech, Cepheid. Advisory Board (Personal fees): Pierre Fabre, Agendia, Sanofi Aventis, Menarini, Medac. Invited Speaker (Personal Fees): GuardantHealth. personal fees: Veracyte. Mattea Reinisch reports honoraria and/or consultancy fee from AstraZeneca, Daiichi Sankyo, Gilead, Lilly, MSD, Novartis, Pfizer, Roche, OnkowissenTV, Seagen, Streamed Up and Somatex. M. Reinisch also reports travel support from AstraZeneca, Daiichi Sankyo, Gilead, Lilly, Novartis, Pfizer, Roche. Carsten Denkert reports other support from MSD Oncology, personal fees from Daiichi Sankyo and AstraZeneca, and grants from Myriad Genetics and German Breast Group outside the submitted work; C. Denkert also reports a patent for VMscope digital pathology software with royalties paid, a patent for WO2020109570A1 issued, a patent for WO2015114146A1 issued, and a patent for WO2010076322A1 issued. Sibylle Loibl declares to be GBG Forschungs GmbH employee (CEO); The institution receives grants from AbbVie, AstraZeneca, Celgene, Daiichi-Sankyo, Immunomedics/Gilead, Molecular Health, Novartis, Pfizer and Roche; honoraria for Advisory board from Abbvie, Amgen, AstraZeneca, BMS, Celgene, DSI, EirGenix, Gilead, GSK, Lilly, Merck, Novartis, Olema, Pfizer, Pierre Fabre, Relay Therapeutics, Roche, Sanofi and Seagen; honoraria as invited speaker from AstraZeneca, DSI, Gilead, Novartis, Pfizer, Roche, Seage and Medscape. S. Loibl reports non-financial interest as advisory role in AGO Kommission Mamma, other non-financial interest from AstraZeneca, Daiichi-Sankyo, Gilead, Novartis, Pfizer, Roche and Seagen. GBG Forschungs GmbH has following /patents pending: EP14153692.0, EP21152186.9, EP19808852.8 and receives licensing fees from VM Scope GmbH.

Supplementary information

The online version contains supplementary material available at 10.1038/s41523-026-00948-7.