BKM120

Buparlisib plus fulvestrant in postmenopausal women with hormone-receptor-positive, HER2-negative, advanced breast cancer progressing on or after mTOR inhibition (BELLE-3):a randomised, double-blind, placebo-controlled, phase 3 trial

Summary
Background Activation of the PI3K/AKT/mTOR pathway occurs frequently in breast cancer that is resistant to endocrine therapy. Approved mTOR inhibitors effectively inhibit cell growth and proliferation but elicit AKT phosphorylation via a feedback activation pathway, potentially leading to resistance to mTOR inhibitors. We evaluated the efficacy and safety of buparlisib plus fulvestrant in patients with advanced breast cancer who were pretreated with endocrine therapy and mTOR inhibitors.Methods BELLE-3 was a randomised, double-blind, placebo-controlled, multicentre, phase 3 study. Postmenopausal women aged 18 years or older with histologically or cytologically confirmed hormone-receptor-positive, HER2-negative, locally advanced or metastatic breast cancer, who had relapsed on or after endocrine therapy and mTOR inhibitors, were recruited from 200 trial centres in 22 countries. Eligible patients were randomly assigned (2:1) via interactive response technology (block size of six) to receive oral buparlisib (100 mg per day) or matching placebo starting on day 1 of cycle 1, plus intramuscular fulvestrant (500 mg) on days 1 and 15 of cycle 1 and on day 1 of subsequent 28-day cycles. Randomisation was stratified by visceral disease status. The primary endpoint was progression-free survival by local investigator assessment as per the Response Evaluation Criteria In Solid Tumors (RECIST) version 1.1 in the full analysis population (all randomised patients, by intention-to-treat). Safety was analysed in all patients who received at least one dose of treatment and at least one post-baseline safety assessment.

This study is registered with ClinicalTrials.gov, number NCT01633060, and is ongoing but no longer enrolling patients.Findings Between Jan 15, 2013, and March 31, 2016, 432 patients were randomly assigned to the buparlisib (n=289) or placebo (n=143) groups. Median progression-free survival was significantly longer in the buparlisib versus placebo group (3·9 months [95% CI 2·8–4·2] vs 1·8 months [1·5–2·8]; hazard ratio [HR] 0·67, 95% CI 0·53–0·84, one-sided p=0·00030). The most frequent grade 3–4 adverse events in the buparlisib versus placebo group were elevated alanine aminotransferase (63 [22%] of 288 patients vs four [3%] of 140), elevated aspartate aminotransferase (51 [18%] vs
four [3%]), hyperglycaemia (35 [12%] vs none), hypertension (16 [6%] vs six [4%]), and fatigue (ten [3%] vs two [1%]). Serious adverse events were reported in 64 (22%) of 288 patients in the buparlisib group versus 23 (16%) of 140 in the placebo group; the most frequent serious adverse events (affecting ≥2% of patients) were elevated aspartate aminotransferase (six [2%] vs none), dyspnoea (six [2%] vs one [1%]), and pleural effusion (six [2%] vs none). On-treatment deaths occurred in ten (3%) of 288 patients in the buparlisib group and in six (4%) of 140 in the placebo group; most deaths were due to metastatic breast cancer, and two were considered treatment-related (cardiac failure [n=1] in the buparlisib group and unknown reason [n=1] in the placebo group).Interpretation The safety profile of buparlisib plus fulvestrant does not support its further development in this setting. Nonetheless, the efficacy of buparlisib supports the rationale for the use of PI3K inhibitors plus endocrine therapy in patients with PIK3CA mutations.

Introduction
Hormone-receptor-positive breast cancer accounts for approximately 75% of breast cancer diagnoses worldwide.1 Endocrine therapy is the mainstay of treatment for hormone-receptor-positive advanced breast cancer, but resistance eventually develops and disease progression occurs.2 Various different mechanisms of resistance toendocrine therapy have been identified, including oestrogen-independent activation of the oestrogen receptor, increased signalling through the RAS/MAPK, NFκB, or PI3K/AKT/mTOR pathways, and maintenance of cyclin D1 expression.3 PI3K/AKT/mTOR activation is frequently observed in hormone-receptor-positive breast cancer that is resistant to endocrine therapy.4 The BOLERO-2 phase 3 study showed that combining endocrine therapy with the mTOR inhibitor everolimus prolongs progression-free survival when compared with endocrine therapy alone in postmenopausal patients with hormone-receptor-positive, HER2-negative, advanced breast cancer.5 In the same population, combining everolimus with either tamoxifen or fulvestrant also provided efficacy benefits when compared with endocrine therapy alone.6,7 Despite an expectation that tumours dependent on a hyperactive PI3K/AKT/mTOR pathway might show a greater benefit from everolimus than from endocrine therapy alone, data from this study suggested that patients with wild-type PIK3CA achieved a longer median progression-free survival than did those with PIK3CA mutations.8 Therefore, it might be expected that treatment with a direct PI3K inhibitor could yield greater benefit in patients with PIK3CA mutations than in those with wild-type PIK3CA. Preliminary data from in-vitro and in-vivo studies of the PI3K/AKT/mTOR pathways have shown that mTOR inhibition elicits AKT phosphorylation via several feedback activation pathways that could result in resistance to mTOR inhibitors.9–12 As shown in preclinical studies, PI3K inhibitors can attenuate or abrogate AKT phosphorylation following mTOR inhibition.9–12

Accordingly, inhibition of PI3K might be clinically important in patients progressing on mTOR inhibitor treatment. The BELLE-3 study was therefore designed to test this hypothesis, by comparing the efficacy and safety of buparlisib or placebo when combined with fulvestrant in postmenopausal women with hormone-receptor-positive, HER2-negative, advanced breast cancer who had been treated with an aromatase inhibitor, progressing on or after endocrine therapy plus mTOR inhibitor therapy. Buparlisib is an oral pan-PI3K inhibitor that targets all four isoforms of class I PI3K (α, β, δ, and γ).13 In the phase 3, randomised, placebo-controlled BELLE-2 trial,14 the combination of buparlisib plus fulvestrant was shown to have clinical activity and manageable safety in postmenopausal women with hormone-receptor-positive, HER2-negative, advanced breast cancer progressing on aromatase inhibitor therapy but naive to mTOR inhibitor therapy. Prespecified exploratory analyses in BELLE-2 showed that patients with the PIK3CA mutation in circulating tumour DNA (ctDNA) at study entry had meaningful clinical benefit from the combination regimen.14 On the basis of these findings, a secondary objective was added to BELLE-3 to assess the efficacy of buparlisib or placebo in combination with fulvestrant in patients with PIK3CA-mutant and wild-type status detected in ctDNA.BELLE-3 was a randomised, double-blind, placebo-controlled, phase 3 trial done at 200 trial centres in 22 countries (appendix pp 2–10).

Eligible patients were postmenopausal women aged 18 years or older with histologically or cytologically confirmed hormone- receptor-positive, HER2-negative, locally advanced or metastatic breast cancer pretreated with aromatase inhibitors and resistant to endocrine and mTOR inhibitor combination therapy (defined as progression during or within 30 days of the end of combination therapy for advanced breast cancer, or during or within30 days of the end of either endocrine or mTOR inhibitor therapy if the patient had to stop a study drug because of toxicity). At least one measurable or non-measurable lesion as per the Response Evaluation Criteria In Solid Tumors (RECIST), version 1.1, was required, as well as Eastern Cooperative OncologyGroup (ECOG) performance status score of 2 or lower and adequate archival or new tumour tissue for PI3K-related biomarker analysis. Eligible patients needed adequate bone marrow and organ function as assessed by laboratory tests. Patients previously treated with PI3K or AKT inhibitors, fulvestrant, more than one line of chemotherapy for metastatic disease, with a score of 12 or more on the 9-item Patient Health Questionnaire (PHQ-9) and 15 or more on the 7-item Generalized Anxiety Disorder (GAD-7) questionnaires, or with a medically documented history of active mood disorder or grade 3 anxiety (as per the Common Terminology Criteria for Adverse Events [CTCAE], version 4.03) were excluded.All participating patients provided written informed consent. The study was done in accordance with the Declaration of Helsinki and guidelines for good clinical practice, and the protocol and any amendments were approved at each site by local independent ethics committees and institutional review boards. A steering committee oversaw the conduct of the trial in accordance with the protocol.

An independent data monitoring committee was responsible for monitoring safety and efficacy (assessing criteria for early stopping because of futility based on progression-free survival) in the trial participants. The data monitoring committee met after the first 30 patients had been treated for more than 2 months and subsequently met every 6 months.Interactive response technology was used to randomly assign eligible patients (in a 2:1 ratio) to receive either buparlisib plus fulvestrant or placebo plus fulvestrant on the basis of stratification by presence or absence of visceral disease. Absence of visceral disease was defined as lesions present only in bone, skin, nodes, or soft tissue, or as locally advanced disease without organ infiltration. Randomisation was done with a block size of six within each stratum. At enrolment, investigators registered patients into the interactive response technology system by use of identifying information; patients were assigned seven-digit numbers that were retained throughout their participation in the study. A patient randomisation list was produced via a validated system to automate the random assignment of patient numbers to randomisation numbers linked to a medication number and treatment group, ensuring that treatment assignment was unbiased and concealed from patients and investigators (randomisation numbers were not communicated to investigators). Novartis Drug Supply Management (BSP Pharmaceuticals, Latina Scalo, Italy) produced a separate medication randomisation list via a validated system that automated random assignment of medication numbers to study treatment packs. Patients and investigators (including local radiologists) remained masked to treatment assignment from the time of randomisation to the time of final overall survival analysis. Premature unblinding of study drug assignment was only permitted for patient emergencies. Identification of study treatments was concealed by identical appearance, packaging, labelling, and schedule of administration.Patients received oral buparlisib or placebo (100 mg once daily starting on day 1 of cycle 1) plus intramuscular injections of open-label fulvestrant (500 mg on days 1 and 15 of cycle 1, and on day 1 of subsequent cycles) in 28-day treatment cycles.

Buparlisib and matching placebo as 10 mg and 50 mg hard gelatine capsules packaged in bottles were provided by Novartis Drug Supply Management. Fulvestrant was prescribed by investigators and supplied in accordance with local practice and regulation. Treatment continued until disease progression (as per RECIST version 1.1), unacceptable toxicity, death, or discontinuation for any other reason. Prespecified reasons for treatment discontinuations included adverse events, loss to follow-up, study treatment non-adherence, patient or physician decision, progressive disease, protocol deviation, study termination, technical issues, or death. Crossover from placebo to buparlisib was not permitted. Patients unable to tolerate the protocol-specified dose could have dose adjustments to continue study treatment. Initially, up to three levels of dose reduction of buparlisib and placebo were allowed: 80 mg per day continuously; 100 mg per day on five days of seven; and80 mg per day on five days of seven, with no dose re-escalations during subsequent cycles. After protocol amendment (Aug 14, 2015), only dose reductions of 100 mg per day or 80 mg per day (both on five days of seven) were permitted; patients receiving 80 mg per day continuously were switched to 100 mg per day on five days of seven. No dose modification was permitted for fulvestrant. Study medications could be resumed at the same dose if interrupted for any reason other than toxicity (eg, physician decision). If buparlisib was withheld because of toxicity for 28 or more days or fulvestrant withheld for 35 or more days, treatment was permanently discontinued.Tumour assessments (by CT or MRI scans) were done at screening and every 6 weeks until disease progression; imaging data were also collected for central radiological assessment done by the blinded independent review committee.

Clinical and laboratory assessments, including physical examination, vital signs, performance status, echocardiogram, and biochemistry, were done at baseline and on day 1 of every cycle until the end of treatment. Cardiac imaging was assessed at screening, day 1 of cycle 4, and day 1 of every fourth cycle thereafter. Haematology was assessed weekly at baseline until day 22 of cycle 2, and on day 1 of subsequent cycles until the end of treatment. Liver parameters (transaminases and bilirubin) were monitored weekly during the first two cycles. Patients completed health-related quality of life questionnaires (European Organisation for Research and Treatment of Cancer quality-of-life questionnaires EORTC QLQ-C30, and QLQ-BR23) at screening, day 1 of cycle 1, day 15 of cycle 2, day 1 of cycle 4, and day 1 of subsequent cycles until end of treatment. Adverse events were assessed continuously as per CTCAE version 4.03. All patients were followed up for survival status every 3 months regardless of the reason for treatment discontinuation, except for those who withdrew consent for survival follow-up or were lost to follow-up.For the secondary biomarker analyses, PIK3CA mutation status detected on ctDNA was assessed in hotspots from exons 9 and 20 on plasma samples from blood collected during screening or at day 1 of cycle 1.15 The Inostics BEAMing (Sysmex, Hamburg, Germany) assay was used for ctDNA analysis. The mutations assessed were Glu542Lys, Glu545Lys, Glu545Gly, and Gln546Lys for exon 9 and Met1043Ile, His1047Tyr, His1047Arg, and His1047Leu for exon 20. PIK3CA mutation status was also assessed in new or archival primary or metastatic tumour tissue by PCR by use of the Roche cobas PIK3CA assay (Roche, Pleasanton, CA, USA) covering exons 7, 9, and 20.The primary endpoint was local investigator-assessed progression-free survival as per RECIST version 1.1 in the full analysis population (all randomised patients), defined as the time from randomisation to either first documented disease progression or death from any cause. Progression-free survival was also analysed by central radiology review.

The key secondary endpoint was overall survival, defined as the time from randomisation to death from any cause in the full analysis population. Other secondary endpoints were safety (based on frequency of adverse events, physical and radiological examinations, and laboratory evaluations); local investigator-assessed objective response (best overall response of either complete or partial response according to RECIST version 1.1); local investigator-assessed clinical benefit (best overall response of complete response, partial response, or stable disease at ≥14 weeks according to RECIST version 1.1); progression-free survival by ctDNA PIK3CA status (this endpoint was updated in the protocol from an exploratory to a secondary endpoint on Aug 14, 2015, because a meaningful clinical benefit was observed in the BELLE-2 study for patients with the PIK3CA mutation in ctDNA at study entry who received buparlisib plus fulvestrant14); overall survival by ctDNA PIK3CA status; and quality-of-life outcomes according to the patient-reported EORTC QLQ-C30 and EORTC QLQ-BR23. Time to deterioration of ECOG performance status, PHQ-9, GAD-7, and pharmacokinetics of buparlisib plus fulvestrant were also assessed. Exploratory endpoints included progression-free survival by PIK3CA status in tumour tissue. The results of pharmacokinetics and quality-of-life outcomes are not reported here.The primary efficacy endpoint was analysed with a stratified log-rank test (according to visceral disease status) at a one-sided 2·5% level of significance in the full analysis population. We estimated that 313 progression-free survival events would provide 90% power to detect a hazard ratio (HR) of 0·67 (corresponding to a 50% improvement in median progression-free survival from 3 months in the placebo group to 4·5 months in the buparlisib group) at a one-sided 0·025 level of significance. Approximately 420 patients (280 in the buparlisib group and 140 in the placebo group) were required, assuming a recruitment period of approximately 29 months. One prespecified interim futility analysis was done for progression-free survival with gamma spending function (γ=0) when 136 (43·5%) of the targeted 313 events were documented. Sample size calculations were done with EAST 5·4. Following recommendations of the data monitoring committee, the trial continued after the interim futility analysis.

If progression-free survival was statistically significant, overall survival was tested with a stratified log-rank test (stratified according to visceraldisease status) at a one-sided 0·025 level of significance. We estimated that 319 deaths would provide 80% power to detect an HR of 0·71 (corresponding to a 40% improvement in median overall survival from 15 months in the placebo group to 21 months in the buparlisib group) at a one-sided 0·025 level of significance.Efficacy analyses were done on the full analysis population according to the intention-to-treat principle. Progression-free survival was censored at the last adequate tumour assessment date before the analysis cutoff date if no progression-free survival event was observed before the cutoff date. If the patient started a new antineoplastic therapy, the censoring date was the date of the last adequate tumour assessment before initiation of the therapy or before the cutoff date, whichever came first. If an event was observed after two or more missing or non-adequate tumour assessments, it was censored at the last adequate tumour assessment date. Median progression-free survival (with two-sided 95% CIs) was estimated with the Kaplan-Meier method. Stratified Cox regression was used to estimate HRs along with two-sided 95% CIs. Proportional hazards assumption was verified with graphical plots. Central radiology assessment of progression-free survival by the blinded independent review committee was done with the same analytical methodology as that of the primary analysis.

A prespecified sensitivity analysis for progression-free survival was based on the per-protocol population (a subset of randomised patients excluding those with protocol deviations that could have affected the primary endpoint). A stratified multivariate Cox proportional hazards model was fitted to assess the robustness of the primary endpoint of progression-free survival by considering the prognostic baseline demographic and disease characteristics, also used as a basis for a stratified Cox model for the prespecified subgroup analyses of progression-free survival. A maximum of three analyses were planned for overall survival by use of a 3-look Lan-DeMets (O’Brien-Fleming) efficacy stopping boundary. The first overall survival interim analysis was planned at the final analysis for progression-free survival. Overall survival was censored at the last known date the patient was alive. The distribution of overall survival was estimated with the Kaplan-Meier method. The proportion of patients with an overall response and clinical benefit was summarised with 95% CIs based on the Clopper-Pearson method. Analyses of progression-free survival, overall survival, and the proportion of patients achieving an overall response were repeated in patients with mutant and wild-type PIK3CA detected on ctDNA. All safety analyses were done on the safety population (all patients who received at least one dose of study treatment and had at least one post- baseline safety assessment) according to treatment received on day 1. Exposure was summarised with summary statistics, including median and IQR. All statistical analyses were done and figures generated with SAS, version 9.4.

This trial is registered with ClinicalTrials.gov, number NCT01633060.This study was designed by the trial sponsor, Novartis Pharmaceuticals Corporation, in conjunction with the authors and study steering committee members. The sponsor provided buparlisib and placebo, and representatives of the trial sponsor collected and analysed the data. All authors had full access to study data for interpretation and analysis, were involved in the development and approval of the manuscript, and had final responsibility for the decision to submit the manuscript for publication. The manuscript was prepared by the authors with assistance from a medical writer funded by the sponsor. All authors assume responsibility for the accuracy and completeness of the data and vouch for the fidelity of the trial to the protocol.ResultsBetween Jan 15, 2013, and March 31, 2016, 647 patients were screened and 432 were randomly assigned to receive either buparlisib plus fulvestrant (n=289) or placebo plus fulvestrant (n=143), forming the full study population assessed for the primary endpoint (figure 1). Patients were stratified by visceral disease status as defined at randomisation according to interactive response technology (289 [67%] of 432 present and143 [33%] of 432 absent). 428 patients were assessed for safety: 288 in the buparlisib group and 140 in the placebo group. The four remaining patients were not treated because of protocol deviations (n=3) or technical problems (n=1; figure 1) and were excluded from the safety analysis. Treatment was ongoing in 37 (13%) of 289 patients in the buparlisib group and in 12 (8%) of 143 in the placebo group at the time of analysis.

The most frequent reasons for treatment discontinuation in the buparlisib and placebo groups were disease progression, adverse events, and patient or guardian decision (figure 1). The per-protocol analysis comprised 271 patients in the buparlisib group and 133 patients in the placebo group, with 18 patients in the buparlisib group and 10 in the placebo group excluded because of protocol deviations.Baseline characteristics were balanced between the two treatment groups (table 1). All patients had previously received mTOR inhibitor therapy; 286 (99%) of 289 patients in the buparlisib group and 142 (99%) of 143 in the placebo group received everolimus, and three (1%) of 289 and one (1%) of 143 had received ridaforolimus. 263 (91%) patients in the buparlisib group and 120 (84%) in the placebo group progressed during previous mTOR inhibitor treatment or within 30 days of completion, while 24 (8%) and 23 (16%) discontinued mTOR inhibitor treatment because of safety reasons and had progressed during single-agent endocrine treatment (data were missing for two [1%] patients from the buparlisib group). Median duration of treatment with an mTOR inhibitor was 8·0 months (IQR 4·3–12·7) in the buparlisib group and 8·6 months (4·5–12·8) in the placebo group. 103 (36%) of 289 patients in the buparlisib group and 49 (34%) of 143 in the placebo group had received chemotherapy for metastatic disease. 202 (70%) patients in the buparlisib group and 95 (66%) in the placebo group had received at least two previous lines of endocrine therapy for metastatic disease.At data cutoff (May 23, 2016), 319 progression-free survival events were reported in the buparlisib (n=202) and placebo (n=117) groups.

Median progression-free survival follow-up was 8·3 months (IQR 4·2–20·7) in the buparlisib group and 12·0 months (IQR 4·7–13·7) in the placebo group. Local investigator-assessed progression- free survival was significantly longer in the buparlisib group than in the placebo group (HR 0·67, 95% CI 0·53–0·84, one-sided p=0·00030; figure 2). Median progression-free survival was 3·9 months (95% CI 2·8–4·2) in the buparlisib group and 1·8 months (1·5–2·8) in the placebo group, while the 6-month progression-free survival was 31% (95% CI 25–37) in the buparlisib group and 20% (13–28) in the placebo group. Central radiology assessment of progression-free survival was consistent with these findings (appendix p 16). Progression-free survival in the per-protocol population was consistent with the primary analysis results (median (figure 3). Analysis of progression-free survival by visceral disease status defined by metastatic sites at baseline showed that in the buparlisib group versus placebo group the HR was 0·56 (95% CI 0·43–0·74; p<0·0001) in patients with visceral disease at baseline (median progression-free survival 3·1 months [95% CI 2·8–4·2] vs 1·5 months [1·4–1·8]), whereas the HR was 0·96 (0·61–1·50; p=0·43) in patients with non-visceral disease at baseline (median progression-free survival 4·2 months [3·0–5·6] vs 4·1 months [2·6–5·7]; appendix p 17).91 (31%) of 289 deaths in the buparlisib group and46 (32%) of 143 deaths in the placebo group were observed among the prespecified target of 319 for the interim overall survival analysis. The study did not cross the efficacy stopping boundary for overall survival at the first interim analysis. Investigator-assessed overall response is shown in table 2. Information on PIK3CA mutation status was available from ctDNA analysis for 232 patients in the buparlisib group and 116 in the placebo group, and from PCRanalysis of tumour tissue for 216 patients in the buparlisib group and 104 in the placebo group. 135 (39%) of 348 patients had the PIK3CA mutation and 213 (61%) of 348 had wild-type PIK3CA according to ctDNA analysis, whereas 109 (34%) of 320 had the PIK3CAmutation and 204 (64%) of 320 had wild-type PIK3CA according to PCR analysis of tumour tissue (table 3). Among the 135 patients who were classified as having the PIK3CA mutation by ctDNA, 100 [43%] of 232 patients in the buparlisib group and 35 [30%] of 116 in the placebo group had PIK3CA mutations identified on exons 9 or 20; three patients in the buparlisib group had PIK3CA mutations on both exons 9 and 20. The following PIK3CA mutations for exons 9 and 20 were most frequently detected at baseline in the buparlisib and placebo groups: Glu545Lys (35 [35%] of 100 patients and 15 [43%] of35 patients), Glu542Lys (22 [22%] and three [9%]), His1047Arg (42 [42%] and 12 [34%]), and His1047Leu (three [3%] and four [11%]). 75 (26%) of 289 patients in the buparlisib group and 34 (24%) of 143 in the placebo group were classified as having the PIK3CA mutation detected by PCR. Six (2%) of 289 patients in the buparlisib group and one (1%) of 143 in the placebo group were classified as having unknown PIK3CA mutation status detected by PCR. PIK3CA mutation status among420 patients with archival tumour tissue was predominantly based on primary tumour samples (in 305 [73%] patients). In patients with PIK3CA mutations detected by ctDNA and PCR, buparlisib plus fulvestrant significantly improved progression-free survival compared with that for placebo plus fulvestrant (figures 4 and 5). However, the interaction test showed that this was not significant for PIK3CA status assessed by ctDNA (treatment–effect interaction p=0·113) but it was significant for PIK3CA status assessed by PCR (treatment–effect interaction p=0·0151).Concordance in PIK3CA status between tumour samples (assessed with PCR) and matched ctDNA isolated at study entry (assessed with the BEAMing assay) was analysed on the basis of 256 matched patients (174 in the buparlisib group and 82 in the placebo group; 59% of the study population; table 4). Baseline disease characteristics of the 256 matched patients (174 in the buparlisib group and 82 in the placebo group) were similar to those of the overall patient population (data not shown). For the buparlisib group, overall concordance was observed in 141 (81%) of 174 patients. In the 33 (19%) cases of discordance, a PIK3CA wild-type tumour sample and PIK3CA-mutated ctDNA were reported in 18 (10%) patients and PIK3CA mutations detected in the tumour sample but not in ctDNA in ten (6%) patients. In the remaining five cases (3%), patients had unknown PIK3CA status in tumour tissue and wild-type ctDNA. For the placebo group, overall concordance was observed in 71 (87%) of 82 patients. In the 11 (13%) cases of discordance, a PIK3CA wild-type tumour sample and PIK3CA-mutated ctDNA were reported in three (4%) patients and PIK3CA mutations detected in the tumour sample but not in ctDNA in seven (8%) patients. In the remaining one case (1%), the patient had unknown PIK3CA status in tumour tissue and wild-type ctDNA. Distribution of visceral disease was similar in patients with PIK3CA-mutant and wild-type disease. Among patients with ctDNA-assessed PIK3CA mutation, 73 (73%) of 100 patients in the buparlisib group and 27 (77%) of 35 in the placebo group had visceral disease, while 96 (73%) of 132 in the buparlisib group and 55 (68%) of 81 in the placebo group with ctDNA-assessed wild-type PIK3CA had visceral disease. A similar result was observed in archival tumour samples for patients with the PIK3CA mutation detected. Among patients with archival tissue-assessed PIK3CA mutation, 57 (76%) of 75 patients in the buparlisib group and 25 (74%) of 34 in the placebo group had visceral disease, while 96 (71%) of 135 in the buparlisib group and 48 (70%) of 69 in the placebo group with archival tissue-assessed wild-type PIK3CA had visceral disease.428 patients were included in the safety population (288 in the buparlisib group; 140 in the placebo group). Median duration of exposure to study treatment was longer in the buparlisib group than in the placebo group (3·3 months [IQR 1·8–5·7] vs 2·3 months [1·8–4·6]); median relative dose intensity was 99·4% (IQR 82·0–100·0) versus 100·0% (100·0–100·0). Median duration of follow-up (from randomisation to data cutoff) was 16·3 months (IQR 8·9–24·0). In the safety population, 282 (98%) patients in the buparlisib group and 128 (91%) in the placebo group had at least one adverse event. The most frequent adverse events (affecting ≥10% of patients, all grades and grade 3–4) including those not related to study treatment in the buparlisib group were elevated alanine aminotransferase (ALT), elevated aspartate aminotransferase (AST), hyperglycaemia, nausea, diarrhoea, and fatigue (table 5). The most frequent adverse events in the placebo group were fatigue, nausea, and elevated AST. Grade 3 or 4 adverse events were reported by 177 (61%) of 288 patients in the buparlisib group and 47 (34%) of 140 in the placebo group; the most frequent grade 3–4 adverse events in the buparlisib versus placebo group were elevated alanine aminotransferase (63 [22%] of 288 patients vs four [3%]of 140), elevated aspartate aminotransferase (51 [18%]vs four [3%]), hyperglycaemia (35 [12%] vs none),hypertension (16 [6%] vs six [4%]), and fatigue (ten [3%] vs two [1%]). Serious adverse events were reported by 64 (22%) of 288 patients in the buparlisib group and 23 (16%) of 140 in the placebo group; the most frequent serious adverse events (affecting ≥2% of patients) were elevated aspartate aminotransferase (six [2%] vs none), dyspnoea (six [2%] vs one [1%]), and pleural effusion (six [2%] vs none; appendix pp 11–15). Dose interruptions due to adverse events were more frequent in the buparlisib group than in the placebo group (103 [36%] of 288 patients vs 12 [9%] of 140), as were dose reductions (90 [31%] vs 11 [8%]) and dose discontinuations with orwithout fulvestrant (61 [21%] vs seven [5%]). The most common adverse events leading to permanent discontinuation of the study drug (>2% in either group, all grades; grade 3–4), including those unrelated to study treatment, in the buparlisib group were elevated ALT (16 [6%] of 288 patients; 11 [4%]), elevated AST (12 [4%];seven [2%]), and depression (six [2%]; two [1%]). The most common adverse events leading to permanent discontinuation of study drug in the placebo group were elevated ALT (one [1%] of 140 patients; one [1%]), elevated AST (two [1%]; one [1%]), and depression (one [1%]; none). Medical review confirmed that one patient (in the buparlisib group) in the study met the criteria for Hy’s law. Suicidal ideation was reported in the buparlisib (five [2%]) and placebo (one [1%]) groups; three suicide attempts were reported in the buparlisib group. Ten (3%) improved progression-free survival compared with that for placebo plus fulvestrant in postmenopausal women with aromatase inhibitor-resistant, hormone-receptor- positive, HER2-negative, advanced breast cancer.The BELLE-3 trial met its primary endpoint. Patients predominantly received study treatment as third-line therapy for advanced disease, and progression occurred during mTOR inhibitor treatment or within 30 days from the last dose in almost 90% of patients. Although approximately 50% of patients progressed before the first check-up, the rationale for studying the combination in this particular population was to overcome resistance to mTOR inhibitors by targeting the PI3K pathway upstream and to potentially address the unmet medical need in this population. As the majority of patients were on their third line of treatment, progression-free survival was inevitably low. Progression-free survival was in favour of combination therapy versus fulvestrant alone in patients with PIK3CA mutations (detected from tumour tissue or ctDNA isolated at study entry). This observation in patients positive for the PIK3CA mutation is consistent with the exploratory ctDNA results of the BELLE-2 trial,14 where the same combination was compared with fulvestrant alone in mTOR inhibitor- naive patients with ctDNA PIK3CA mutations (n=200; median progression-free survival 7·0 vs 3·2 months, HR 0·56; 95% CI 0·39–0·80; nominal one-sided p<0·001). Together, these data indicate that PIK3CA status is a potential predictive biomarker for a benefit with buparlisib treatment in such patients. However, in BELLE-2, no statistically significant progression-free survival benefit with buparlisib plus fulvestrant was recorded in patients with PI3K pathway activation in tumour tissue. The sensitivity of the test used to assess PIK3CA status in tumour tissue in BELLE-2 (Sanger sequencing) compared with PCR in BELLE-3 might have contributed to the differences observed in the two studies. Moreover, in contrast to the BELLE-3 study, patients in BELLE-2 were not permitted to receive mTOR inhibitors before study entry.We observed a similar result with the buparlisib combination in patients with wild-type PIK3CA (with ctDNA samples only) and mutated PIK3CA (with tumour tissue or ctDNA samples), raising the possibility that buparlisib might have some activity in patients with tumours harbouring wild-type PIK3CA pretreated with an mTOR inhibitor. This observation could be related to the inhibitory effect of buparlisib on the different PI3K isoforms or to activation of the PI3K pathway through other mechanisms (eg, copy number amplification, deletions, mRNA expression, or protein phosphorylation).16 The findings of another report indicate that the mTOR pathway is activated in breast cancer liver metastases;17 since almost 50% of patients in the BELLE-3 study had liver metastases, this finding might provide further explanation for the observed benefit of buparlisib treatment regardless of PIK3CA mutation status. This finding, however, was inconsistent with the efficacy of buparlisib observed in patients with wild-type PIK3CA tumours in BELLE-2. This discrepancy between the BELLE-2 and BELLE-3 studies could be explained by the inherent population and tumour sample heterogeneities associated with subgroup analyses. A related limitation of BELLE-3 is that progression-free survival by ctDNA-assessed PIK3CA status was not initially planned but added subsequently via a protocol amendment as a secondary endpoint. Notably, the presence of PIK3CA mutations in ctDNA was higher in the buparlisib group (100 [43%] of 232 patients) than in the placebo group (35 [30%] of 116 patients). This difference was not reflected in the PCR analysis of archival tissue, where the PIK3CA mutation was reported in 75 (26%) of289 patients in the buparlisib group compared with 34 (24%) of 143 in the placebo group. The imbalance could be attributed to an absence of stratification based on ctDNA-assessed PIK3CA status at randomisation.The BELLE-3 study represents a large cohort (n=256) for biomarker analysis of tumour tissue and matched ctDNA. The 17% discordance in PIK3CA status between tumour tissue and ctDNA was consistent with that of previous reports based on a smaller sample size.15 Since PIK3CA mutation status according to tumour tissue was mainly based on primary tumour samples (73%), such discordance could be because of molecular changes occurring between primary tumours and metastases, as reported in various other cancers.15,18,19 Despite this discordance, to our knowledge, BELLE-3 is the first study to suggest that PIK3CA mutations detected at the primary tumour level might predict the activity of PI3K inhibitors in advanced settings. This activity could be due to pretreatment of all patients with an mTOR inhibitor, which might upregulate the PI3K pathway via feedback activation, thus increasing the degree of concordance between PIK3CA mutations detected in tumour tissue and PI3K pathway activation status.20 In addition to the predictive value of PIK3CA mutations on the benefit of buparlisib in this pretreated population, an interaction between buparlisib treatment activity and baseline visceral disease status was observed, suggesting that patients deriving the largest benefit from combination therapy included those with visceral disease, independent of PIK3CA status. This result was mainly the consequence of the limited efficacy of fulvestrant alone in patients with visceral disease, and points to an interaction between fulvestrant activity and disease sites, as previously reported.21 Given the multiple subgroups analysed in this study, the possibility that the difference between progression-free survival in patients with visceral and non-visceral metastases might have arisen by chance cannot be discounted. In this study, ctDNA was used to ascertain the PIK3CA mutation status of patients’ tumours. However, further value might be gained from this assessment method by monitoring ctDNA over the course of treatment to give an early indication of the development of acquired resistance and guide treatment decisions.22 Over the past 5 years, the number of targeted therapies for hormone-receptor-positive advanced breast cancer has increased with the approval of the mTOR inhibitor everolimus and the CDK4/6 inhibitors palbociclib and ribociclib, while promising results have been shown for abemaciclib.23,24 Advances continue to be made with novel agents targeting key activation pathways in varying stages of development.25 The future role of PI3K inhibitors should be considered alongside these treatment options, with attention paid to individual genomic profiles. Precision medicine, where patients’ genomic profiles are assessed to direct treatment, is of increasing importance in tailoring these treatments to individual patients. Results of preclinical studies have suggested that CDK4/6 inhibitors might sensitise PIK3CA-mutant breast cancer to agents that target the PI3K/mTOR pathway, leading to the investigational use of CDK4/6 and PI3K inhibitor combinations in this setting.26 Other genomic alterations in the PI3K/AKT/mTOR pathway that are linked to breast cancer include PTEN loss and AKT mutations. PTEN influences multiple crucial cellular processes, and loss of PTEN activity has been linked to many primary and metastatic malignancies, including breast cancer. AKT kinases mediate signalling pathways downstream of PI3K and regulate diverse cellular processes; mutations of AKT result in aberrant signalling and have therefore been associated with various cancers, including breast cancer.28The safety profile of buparlisib was broadly consistent with previous reports of its safety when given as a single agent29 or in combination with paclitaxel, fulvestrant, or letrozole,14,30,31 with hyperglycaemia, rash, gastrointestinal disorders (nausea and diarrhoea), and psychiatric disorders being commonly reported adverse events. Notably, elevated liver aminotransferases were among the most frequently reported adverse events in previous studies of the combination of buparlisib and fulvestrant,14,32 consistent with the findings of the BELLE-3 study. In BELLE-2, among patients receiving buparlisib plus fulvestrant, grade 3–4 elevations in ALT were observed in 25% of patients and elevations in AST in 18%,14 which is consistent with the grade 3–4 incidence reported here (22% for ALT and 18% for AST). However, the most noteworthy difference between the safety profile seen in this study and that of BELLE-2 was the incidence of suicidal ideation. This event was reported in five (2%) of 288 patients in BELLE-3 and led to three suicide attempts, compared with three (1%) of 573 patients in BELLE-2 and no suicide attempts.14 Suicidal ideation was not reported in the phase 1 study.32Upon recommendation of the data monitoring committee, the trial continued after the interim futility analysis, which was done when 136 (43%) of the targeted313 progression-free survival events had been documented. Over the course of the study, various protocol amendments were introduced to mitigate toxicity issues, including modifications to and clarifications of guidelines for management of toxicities, additional exclusion criteria, and changes in the dose reduction schedule. However, despite an adequate benefit–risk ratio, it was decided by the trial sponsor that development of buparlisib should be discontinued in the breast cancer setting, on the basis of the results of buparlisib studies in different indications and in phase 3 studies in breast cancer (BELLE-2 and BELLE-3), and that alternative options should be pursued. The mood disorders described with buparlisib have been linked to its ability to penetrate the blood–brain barrier.33 However, the concept of combination treatment with endocrine therapy and PI3K inhibitors remains attractive and is supported by preclinical and clinical data.14,34 Pan-PI3K inhibitors such as buparlisib might be limited by adverse events arising from a broad spectrum of off-target effects.14 Preclinical findings have indicated that HER2-driven tumours rely on p110α;35 α-selective PI3K inhibitors might therefore be more effective and better tolerated than buparlisib.14 PI3Kα-selective inhibitors are currently under clinical evaluation in combination with endocrine therapy in patients with hormone-receptor-positive, HER2-negative breast cancer. Phase 3 trials investigating an α-selective PI3K inhibitor combined with fulvestrant in patients with hormone-receptor-positive or oestrogen-receptor-positive, HER2-negative, advanced breast cancer progressing on or after receiving aromatase inhibitor therapy include SOLAR-1 (alpelisib; NCT02437318) and SANDPIPER(taselisib; NCT02340221). Prospectively designed validation studies are ongoing to test the interaction between the PIK3CA mutation and the activity of α-selective PI3K inhibitors, and to ascertain the most appropriate timing and sequence (with respect to other treatments) for PI3K inhibition treatment. Phase 1/2 trials are also being done to investigate the combination of PI3K inhibition plus cyclin-dependent kinase inhibition with ribociclib plus letrozole (NCT01872260) or fulvestrant (NCT02088684) in hormone-receptor-positive, HER2-negative, advanced breast cancer. Results of the BELLE-3 study highlight a strong unmet medical need in patients pretreated with everolimus who develop metastatic breast cancer in later lines. Treatment guidelines recommend sequential endocrine therapy for hormone-receptor-positive metastatic breast cancer, but they do not specify a preferred sequence after first-line treatment, and the clinical decision will depend on previous therapy, patient history, and disease status.23,36,37 The absence of benefit seen in the placebo group in the refractory patient populations in the BELLE-3, BELLE-2, and PALOMA-3 studies brings into question whether further studies in this population should randomly assign patients into a single-agent endocrine therapy group.14,3 In conclusion, this trial shows that a pan-PI3K inhibitor such as buparlisib, combined with endocrine therapy, is superior to endocrine therapy alone in patients with hormone-receptor-positive, HER2-negative, advanced breast cancer who were pretreated with an mTOR inhibitor as last line of therapy before study entry. On the basis of the safety profile of buparlisib, we do not recommend its further exploration in this setting. However, the present results set the stage for ongoing and future trials testing α-selective PI3K inhibitors plus endocrine therapy in patients with PIK3CA BKM120 mutations.