Alpelisib Monotherapy for PI3K-Altered, Pretreated Advanced Breast Cancer: A Phase II Study

The PI3K pathway is an ancient signaling pathway present in all eukaryotic multicellular organisms (1). Beginning with discoveries that growth factor signaling and oncoproteins modulate the phosphorylation of the plasma membrane lipid inositol (2–4), the multifarious functions of PI3K signaling downstream of receptor-coupled tyrosine kinases, Ras family GTPases, and G protein–coupled receptors have been elucidated (5). Class I PI3K signaling has broad effects on cell growth, metabolism, and survival, as well as more specialized roles in immune cell development (5). It is among the most commonly disrupted pathways in early-stage and advanced cancer, particularly through gain-of-function mutations in the PIK3CA gene, which encodes the class IA p110α catalytic subunit, and loss-of-function alterations of PTEN (6, 7). In breast cancer, alterations in the PI3K pathway occur in approximately 40% of hormone receptor–positive HER2-negative (ER⁺HER2⁻) tumors and 20% of triple-negative cancers (TNBC) regardless of the stage (7–11).

In solid tumors, targeting the PI3K pathway has recently seen success in the clinic with α-subunit–specific inhibitors such as alpelisib (12). The randomized phase III SOLAR-1 trial compared alpelisib in combination with fulvestrant with fulvestrant alone in patients with advanced hormone receptor–positive breast cancer following progression on prior endocrine therapy (13). Alpelisib improved progression-free survival (PFS) in patients harboring PIK3CA mutations determined via tumor sequencing or circulating tumor DNA (ctDNA), leading to FDA approval in this population (14). Conversely, there was no benefit in tumors lacking PIK3CA mutations. Detection of mutations with ctDNA was found to be an effective screening strategy (15).

Following this encouraging evidence that targeting a commonly altered signaling pathway is efficacious, several questions remain. First, the efficacy of alpelisib in later lines of therapy is poorly characterized, particularly in the setting of clinical endocrine therapy resistance or acquired ESR1 mutations (16, 17). This is particularly pertinent as alpelisib therapy may entail challenging toxicity (13, 18). In populations of patients with a wider range of prior treatment exposure than were enrolled in SOLAR-1, smaller early phase studies have tested alpelisib in combination with endocrine therapies with variable results (19, 20). Second, concerns that alpelisib-induced glucose dysregulation and hyperinsulinemia may restore PI3K activity via insulin/IGF signaling have not been investigated beyond preclinical studies (21, 22). Finally, although genomic aberrations in the PI3K pathway are less common in TNBC, the pathway is highly activated at the transcriptional and proteomic levels, providing a rationale for PI3K inhibition in this subtype (8, 23). To explore these questions, we conducted a phase II multicohort study of alpeli­sib monotherapy in advanced ER⁺HER2⁻ breast cancer and TNBC, with rich correlative studies including tumor genomic and ctDNA profiles together with metabolic biomarkers and molecular imaging with F-18 fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT).

From August 2015 to October 2019, 43 patients were enrolled, 33 in the ER⁺HER2⁻ cohort and 10 in the TNBC cohort. Estrogen receptor and HER2 status were determined according to the American Society of Clinical Oncology/College of American Pathologists guidelines (24). The study schema is shown in Fig. 1A, and the disposition of trial participants is shown in Fig. 1B. Among the 33 patients in the ER⁺HER2⁻ cohort, 28 patients were evaluable, with 4 withdrawing for adverse events and 1 for progressive disease prior to response assessment. In the TNBC cohort of 10 patients, 2 ceased trial therapy before the first tumor assessment due to clinical progression, leaving 8 patients evaluable.

Baseline characteristics for the intention-to-treat cohorts are shown in Table 1 and the evaluable cohort in Supplementary Table S1. The median age was 60 years, and all patients were postmenopausal due to age or surgical oophorectomy. The majority of patients had visceral metastatic disease (87.9% ER⁺ and 80% TNBC) and at least three metastatic sites (60.6% ER⁺ and 70% TNBC). In the ER⁺HER2⁻ cohort, all patients had received and developed resistance to multiple lines of prior endocrine therapy. Prior use of CDK4/6 inhibitors was low (24.2%), as these agents were not approved during the trial period. The majority of patients had also received prior chemotherapy (75.8% ER⁺HER2⁻ and 100% TNBC). Patients had received a median of three lines of prior therapy in the ER⁺HER2⁻ cohort (range, 1–9) and two in the TNBC cohort (range, 1–5).

Baseline PI3K pathway alterations that determined eligibility in the ER⁺HER2⁻ cohort were ascertained from tumor sequencing on archival specimens in 30 patients and from ctDNA analysis in 3 patients (Supplementary Table S2). Thirty patients had mutations in PIK3CA, two patients had mutations in PTEN, and one patient had a mutation in PTEN and PIK3R1. Two patients each had two distinct mutations in PIK3CA detected via tumor sequencing. All PTEN mutations were frameshift indels (n = 2) or premature stop codons (n = 1).

In the whole study population, 62.7% of patients (n = 27/43) experienced at least one grade 3 adverse event related to alpelisib, and 4.7% experienced at least one grade 4 event (n = 2/43). The most common grade 3 and 4 adverse events were hyperglycemia in 32.6% of patients (n = 14/43), maculopapular rash in 25.6% (n = 11/43), colitis in 7.0% (n = 3/43), and diarrhea in 7.0% (n = 3/43; Supplementary Table S3). The use of prophylactic antihistamine therapy for rash was not mandated in the protocol. Two patients received prophylactic antihistamines at the start of alpelisib therapy, one of whom still developed a grade 3 rash. Serious adverse events (SAE) considered definitely or potentially related to alpelisib occurred in 28% of patients (n = 12/43; Supplementary Table S4). There were two deaths during study therapy: one due to malignant bowel obstruction from a second malignancy and one due to sepsis that was not considered related to alpelisib. Permanent discontinuation of alpelisib due to an adverse event occurred in 15.2% (5/33) of patients in the ER⁺ cohort. Of the patients who completed at least 28 days of therapy and excluding one patient who took the wrong dose in error, 75% had at least one dose delay (n = 30/40), 27.5% (n = 11/40) had one dose reduction, and 20% (n = 8/40) had two dose reductions. The median relative dose intensity was 86% (range, 21%–100%).

Results are summarized in Table 2 for the intention-to-treat and evaluable cohorts. The median follow-up was 44.6 months at the time of data cutoff. In the TNBC cohort, the overall response rate (ORR) was 0% (n = 0/10) and clinical benefit rate (CBR) was 0% (n = 0/10). Recruitment to this cohort was terminated due to a lack of efficacy in line with the study protocol.

In the ER⁺ cohort, the centrally reviewed ORR by RECIST 1.1 criteria was 30% (n = 10/33, 95% confidence lower bound 22.6%) and the CBR was 36% (n = 12/33, 95% confidence lower bound 26.9%). The median duration of response was 5.6 months [n = 10, 95% confidence interval (CI), 4.5–upper bound not calculable (NA)]. The median PFS was 4.6 months (n = 33; 95% CI, 3.7–7.3) and median OS was 16.1 months (n = 33, 95% CI, 8.5–34.5), reflecting the advanced state of the disease included in this study. Nine percent of patients (3/33) continued study therapy beyond 12 months without progression. For subsequent analyses, we focus on the evaluable cohort, which has the most complete correlative data.

PET scans at 8 weeks typically showed evidence of stable metabolic disease (SMD) or partial metabolic response (PMR). PMR according to European Organization for Research and Treatment of Cancer (EORTC) criteria (25) was significantly associated with ORR by RECIST (P < 0.01, Fisher exact test) but no clinical benefit in the ER⁺ cohort. However, 44% of patients with a PMR failed to achieve a partial response (PR) by RECIST, indicating that PET response assessment had low specificity for predicting RECIST response. A metabolic response on PET appeared associated with a longer time on study, as previously reported with the PIK3CA inhibitor buparlisib (BKM120; ref. 26), although this association was not statistically significant (P > 0.1, log-rank test, Supplementary Fig. S1). Given the central role of the PI3K pathway in insulin receptor signaling, pharmacologic inhibition of the pathway has the potential to reduce cellular glucose uptake and consequently the transit of the PET tracer FDG into cells, thus confounding response assessment due to pharmacodynamic effects. Insulin resistance is a well-known on-target effect of PI3K inhibitors such as alpelisib (13).

To investigate this further, the metabolic effects of alpelisib were evaluated in an unplanned exploratory analysis of serial HbA1C, C-peptide, and fasting glucose. The Homeostasis Model Assessment of insulin resistance (HOMA2-IR) calculator was used to estimate changes in insulin sensitivity from C-peptide and glucose levels to provide a normalized index for comparison between patients (27). Comparing baseline to the first time point where blood was collected for metabolic markers, alpelisib induced elevation of HbA1C (n = 37, median increase 1%, 95% CI, 0.80–1.35, P < 0.001 Wilcoxon signed-rank test; Supplementary Fig. S2A); insulin levels were measured via C-peptide (n = 38, median increase 1 nmol/L, 95% CI, 0.75–1.4, P < 0.001; Supplementary Fig. S2B) and fasting glucose (n = 41, median increase 0.9 mmol/L, 95% CI, 0.3–2.2, P < 0.01). The HOMA2-IR also increased in line with insulin and glucose levels (n = 37, median increase 2.36, 95% CI, 1.72–2.55, P < 0.00; Supplementary Fig. S2C). Compared with baseline, fasting C-peptide levels increased in every patient assessed except for two patients who were not taking alpelisib at the time of blood collection due to dose interruptions for toxicity.

Preclinical models have found that hyperinsulinemia induced by PI3K inhibition may restore PI3K signaling in cancer cells despite therapeutic inhibition (21, 22). In an exploratory analysis, patients with PR or clinical benefit showed similar increases in HOMA2-IR (P = 0.059, Wilcoxon rank-sum test on per-patient increase between response groups) and C-peptide levels (P = 0.11) at the first available time point after treatment initiation compared with baseline. Visualizing the change in C-peptide and HOMA2-IR over time in relation to response and progression events did not reveal a clear relationship (Supplementary Fig. S3). Interestingly, patients with a PR by RECIST had higher baseline levels of insulin resistance (median difference 0.45; 95% CI, 0.11 = 1.33, P < 0.05, Wilcoxon rank-sum test) and C-peptide levels (median difference 0.2, 95% CI, 0.05–0.6, P < 0.05) than patients with stable disease (SD)/progressive disease (PD). HOMA2-IR and C-peptide levels were higher in patients with PR versus SD versus PD (HOMA2-IR P < 0.01 and C-peptide P < 0.05 for trend; Supplementary Fig. S4). This suggests the possibility of “insulin addiction” in parallel with “oncogene addiction” in which tumor cells in a high-insulin environment are reliant on it to sustain growth. Baseline body mass index (BMI) did not differ between patients with PR versus not (P = 0.26) or clinical benefit versus not (P = 0.42).

The relationship of genomic profile with clinical outcome was explored by merging data obtained from tumor sequencing (Supplementary Table S5) and ctDNA digital droplet PCR (ddPCR) and targeted sequencing (Supplementary Tables S6 and S7). Figure 2A shows the waterfall plot for objective response in the ER⁺ cohort with key PI3K pathway alterations from tumor sequencing and ctDNA assays. There was no relationship between the type of PIK3CA mutation and outcomes. Figure 2B summarizes time on study, RECIST response, time to objective response, PET response, and mutations detected with tumor sequencing and/or ctDNA assays for the ER⁺ cohort. ctDNA assays expanded the number of detected PI3K aberrations. Tumor sequencing detected 29 PI3K aberrations, and ctDNA assays detected 45 aberrations. Twenty-five aberrations were detected by both modalities, leaving 4 tumor-only aberrations and 20 ctDNA-only aberrations. With tumor sequencing alone, 2 of 28 patients had multiple aberrations, whereas with combined tumor and ctDNA assays, 12 patients had multiple aberrations detected (Supplementary Table S6). A similar pattern was seen across other frequently altered genes. This was particularly relevant for ESR1, in which ctDNA-detected mutations at baseline were identified in an additional six patients over and above the five patients in whom mutations were detected with tumor sequencing. Aberrations in ESR1 showed an association with clinical benefit (Fisher exact test P < 0.05, FDR = 0.43). The presence of aberrations in TBX3 was also associated with ORR (Fisher exact test P < 0.05, FDR = 0.24). No (0/4) patient with PTEN alterations derived clinical benefit from alpelisib monotherapy in the ER⁺ cohort.

Due to this greater ascertainment of genomic aberrations with ctDNA, and the fact that ctDNA assays were performed contemporaneously with the initiation of trial therapy, we explored the ctDNA correlates of clinical outcome in more detail. Thirty-two of the 33 (97%) patients in the ER⁺ cohort had their qualifying PI3K pathway somatic mutation detected in plasma using ddPCR. In the TNBC cohort, which was unselected for PI3K pathway alterations, 8 of 10 (80%) patients nevertheless had a detectable alteration in the pathway. Thirty of the 32 (94%) confirmed plasma PI3K pathway somatic mutations by ddPCR in the ER⁺ cohort were detected at baseline, with 1 case detected only from week 8 (ID 13), and the other only at the end of treatment (EOT; ID 40; Fig. 3A). Across the entire cohort, 42% of the qualifying PI3K pathway mutations detected by ddPCR in plasma were PIK3CA H1047R, followed by PIK3CA E545K (21%), PIK3CA E542K (7%), and PTEN mutations (7%; Supplementary Fig. S5). In the ER⁺ cohort, there was no association between baseline PI3K pathway mutant ctDNA levels (copies/mL) and the clinical endpoints of ORR or CBR.

In the ER⁺ cohort, 27 of 28 patients with clinically evaluable disease for ORR and clinical benefit had plasma available for serial ddPCR testing (Fig. 3A). ctDNA levels of the PI3K pathway mutations generally tracked with disease response with a reduction in levels in patients with PR and rising ctDNA levels prior to the detection of PD (Fig. 3B). The median lead time between the first rise in ctDNA levels from nadir and clinical or radiologic progression was 54 days (range, 1–247 days) in 15 of the 27 patients in whom lead time was evaluable.

We next assessed the change in ctDNA levels between baseline and week 8 of alpelisib therapy. PIK3CA-mutant ctDNA levels declined across the ER⁺ and TNBC cohorts (n = 28) from baseline to week 8, with a median decrease of 155.6 copies/mL (P = 0.013, Wilcoxon matched-pairs signed-rank test; Supplementary Fig. S6A). In contrast, a decrease in ctDNA levels at week 8 was not observed in patients (n = 5) with non-PIK3CA mutations (PIK3R1, PTEN, and ERBB2; Supplementary Fig. S6B). The decline in ctDNA levels at week 8 was greatest in patients (n = 8) who experienced a PR (P = 0.008) compared with patients who experienced SD (n = 13) or PD (n = 2; Fig. 3C).

In keeping with these findings, there was an association between ctDNA decline at week 8 and clinical benefit (n = 11, P = 0.0049; Fig. 3D). Importantly, ctDNA decline at week 8 was also associated with improved PFS (median PFS 170 vs. 109 days, HR 0.24; 95% CI, 0.083–0.67, P = 0.0065; Fig. 3E). Complete ctDNA clearance at week 8 was additionally associated with prolonged OS (median OS 964 days vs. 297 days, HR 0.32; 95% CI, 0.13–0.79, P = 0.044; Supplementary Fig. S7).

In addition to the assessment of PI3K pathway mutations by ddPCR, ctDNA was also subjected to targeted sequencing at baseline and at EOT to identify co-occurring somatic mutations (Supplementary Tables S7 and S8). Figure 4A shows the baseline ctDNA mutational profile detected by sequencing for the evaluable ER⁺ cohort (n = 28). Ninety-three percent (26/28) of the ER⁺ cohort had at least one PIK3CA mutation except for ID 39 and ID 13, in whom PTEN and PIK3R1 mutations were detected as the only PI3K pathway alterations, respectively. The commonest mutations detected to co-occur with PIK3CA were in MAP3K1 (n = 12, 46%), ESR1 (n = 11, 42%), TP53 (n = 10, 38%), RB1 (n = 6, 23%), ATM (n = 5, 19%), CDH1 (n = 5, 19%), and KMT2C (n = 5, 19%).

Mutations in PIK3CA were the dominant mutation [i.e., showing the highest variant allele fraction (VAF)] in 12 of 26 (46%) of ER⁺ patients at baseline; however, this did not influence CBRs. Patients with dominant PIK3CA mutations showed similar rates of clinical benefit to those with subclonal mutations [5/12 (42%) versus 7/14 (50%), respectively, P = 0.71]. Multiple PIK3CA mutations were detected in 8 of 26 (31%) patients of the ER⁺ cohort but were not significantly associated with objective response (P = 0.66) or CBR (P = 0.68; Supplementary Table S6).

A total of 11 of 28 (39%) patients in the ER⁺ cohort had an ESR1 mutation detected from ctDNA sequencing at baseline. Mutations in ESR1 at baseline were significantly associated with clinical benefit from alpelisib monotherapy (P = 0.019; Fig. 4A and B). As shown in Fig. 4C, they were also significantly associated with improved PFS (median 250 vs. 138 days; HR 0.22, 95% CI, 0.078–0.60, P = 0.0032). In order to explore the ctDNA dynamics of coexisting ESR1 mutations, serial monitoring of ESR1-mutant ctDNA levels was performed by ddPCR in 8 of 11 patients with paired baseline and week 8 plasma available (Supplementary Table S9). In concordance with the PIK3CA-mutant ctDNA dynamics, a decline in ESR1-mutant ctDNA levels between baseline and week 8 was greatest in patients who experienced a PR and/or clinical benefit, although this was not statistically significant (Supplementary Fig. S8A and S8B). A decline in ESR1-mutant ctDNA levels by 8 weeks was associated with a trend toward improved median PFS, and complete clearance of ESR1-mutant ctDNA by 8 weeks was significantly associated with improved PFS (Supplementary Fig. S8C and S8D).

PTEN mutations were identified in four individuals at baseline, all of whom showed no clinical benefit from alpel­isib monotherapy. Across the cohort, although several other co-occurring mutations were identified at baseline, no other mutations were associated with treatment response (Fig. 4A).

At the EOT time point, the PIK3CA and ESR1 mutations detected in ctDNA at baseline were maintained in the majority of ER⁺ cohort study patients [24/26 (92%) and 8/11 (73%), respectively, Fig. 5]. However, significant genomic evolution following therapy was observed. The most common newly identified mutations at EOT not present at baseline were in MAP3K1 [n = 11/28 (39%)], TP53 [n = 9/28 (32%)], PTEN [n = 7/28 (25%)], GATA3 [n = 6/28 (21%)], CDH1 [n = 5/28 (18%)], and EGFR [n = 5/28 (18%)]. Importantly, there was no significant difference in the representation of MAP3K1, TP53, and PTEN mutations at EOT in patients who had achieved clinical benefit versus not, implying that primary and acquired resistance to alpelisib may differ in tempo but not in a mechanism. Acquired ESR1 mutations following alpelisib monotherapy were rarely observed, and there was evidence of deselection of ESR1 mutations in 3 of 11 (27%) patients at EOT when compared with baseline.

This study contributes to the evidence on the efficacy of alpelisib monotherapy in heavily pretreated patients with advanced breast cancer. We found that this agent as monotherapy had no clinical activity in triple-negative breast cancers, regardless of the presence of PI3K pathway aberrations. Targeting the PI3K pathway in combination with chemotherapy has shown efficacy in advanced TNBC (28, 29). Our findings of limited efficacy with alpelisib monotherapy may in part reflect the difficulties in treating advanced TNBC beyond first line and suggest that combinations with chemotherapy or other antiproliferative agents (30) are an advisable strategy. In ER⁺ disease, however, clinical activity was observed despite a heavily pretreated population, with some patients achieving long-term benefit (36% ≥6 months, 9% ≥12 months). These data support PIK3CA mutations as targetable oncogenic drivers in selected patients with advanced, heavily pretreated disease. Toxicities were similar to those reported in other studies (13). PET response at 8 weeks was poorly predictive of ORR or clinical benefit, but lack of PET response showed a trend toward predicting inferior outcomes.

Elevated insulin secretion and insulin resistance following alpelisib occurred in all patients, as has been found in prior studies (19, 31). We did not find evidence that the development of treatment-induced hyperinsulinemia resulted in treatment resistance and progression. In fact, we found that patients with higher baseline insulin levels and insulin resistance may benefit more from alpelisib in terms of response rate. This finding is exploratory and requires confirmation in other cohorts, but several lines of evidence point to the importance of insulin homeostasis and breast cancer risk. Goodwin and colleagues found that in nondiabetic women with early-stage breast cancer, higher fasting insulin was associated with a higher risk of recurrence and death (32). Higher blood glucose levels are also associated with inferior OS outcomes in advanced TNBC (33). A recent study of the Women's Health Initiative cohort determined that higher insulin resistance as determined with the HOMA2-IR model was a risk factor for postmenopausal breast cancer (34).

Additional preclinical and clinical evidence supports our findings that disruption of insulin signaling may be therapeutically relevant depending on the prior insulin exposure of the tumor. Gallagher and colleagues studied breast tumor development in a mouse model with constitutive insulin resistance and hyperinsulinemia (35). They found that in hyperinsulinemic mice, the growth of orthotopic tumors was accelerated compared with normoinsulinemic mice. Furthermore, treatment with the pan-class I PI3K inhibitor buparlisib showed a larger benefit in the hyperinsulinemic versus normoinsulinemic mice, particularly in the Met-1 tumor model, which harbors a PIK3CA mutation (36). Considering that higher BMI is an imperfect surrogate marker of insulin resistance and hyperinsulinemia, Kim and colleagues conducted a retrospective study of patients with breast and gynecologic malignancies harboring PIK3CA mutations and receiving PI3K-targeted therapies stratified according to BMI (37). They found that patients with a higher BMI had a better response rate and longer OS compared with patients with lower BMI. These preclinical and human data support the hypothesis that “insulin addiction” may occur in the setting of hyperinsulinemia and somatic PIK3CA mutation prior to any PI3K-directed treatment.

Mounting evidence for the importance of hyperinsulinemia-induced aberrant PI3K signaling suggests one strategy to improve the efficacy of PIK3CA inhibition could involve concomitant reduction of hyperinsulinemia through other means including dietary manipulation or the use of antidiabetic drugs such as SGLT2 inhibitors. Given the broad effects of PI3K signaling in promoting and maintaining oncogenesis, such strategies may enhance the efficacy of any therapy and are under investigation in clinical and preclinical contexts (38–40). This preliminary finding of insulin addiction in PI3K-mutant tumors also suggests that patients with insulin resistance or frank diabetes may benefit from alpelisib and should not be automatically excluded from therapy, assuming the attendant disruption in glycemic stability can be managed.

Our study demonstrates the utility of early PIK3CA-mutant ctDNA dynamics to predict objective tumor response, clinical benefit, and PFS in patients with ER⁺ metastatic breast cancer on alpelisib monotherapy as late-line treatment. ctDNA decline at 8 weeks was shown to be associated with objective responses, presence of clinical benefit, and improved PFS. Furthermore, complete clearance of PIK3CA-mutant ctDNA was associated with prolonged OS. These findings are consistent with several recent studies that have demonstrated the role of early ctDNA dynamics in predicting treatment response in metastatic breast cancer (41–43). In addition, we identified a median ctDNA lead time of 54 days between the first point of ctDNA rise (from nadir) and clinical progression in 56% of ER⁺ patients. These results demonstrate the ability of rising ctDNA levels to reflect treatment resistance prior to evidence of clinical progression, as has been shown in previous studies (41). Monitoring ctDNA dynamics during PI3K inhibitor therapy could pinpoint patient populations that require add-on combination therapies or an early switch in therapy.

Treatment response with alpelisib monotherapy in ER⁺ patients was independent of PIK3CA-mutant clonal dominance in ctDNA, as patients with subclonal PIK3CA mutations showed similar rates of clinical benefit. Our study included a high number of patients with multiple-PIK3CA mutations (31%), but this was not associated with treatment response. Other recent studies have shown variable results with respect to the role of multiple PIK3CA mutations and treatment response to PI3K inhibitors, and additional larger studies will be needed to confirm the clinical implications of multiple PIK3CA mutations in this setting (43, 44).

A large proportion (39%) of the ER⁺ cohort had an ESR1 mutation detected at baseline, which was in keeping with the significant prior endocrine therapy exposure of the study population (45–49). Acquired ESR1 mutations, not surprisingly, were rarely observed following alpelisib monotherapy. Importantly, co-occurring baseline ESR1 mutations identified through ctDNA analysis were associated with clinical benefit and improved PFS following alpelisib monotherapy. This finding was in contrast to a recent phase I/II study of combination therapy with alpelisib and an aromatase inhibitor (either letrozole or exemestane), which reported ESR1 mutations to be a mechanism of resistance to the combination strategy (16). The results highlight potential differences in the role of ESR1 mutations for predicting response to alpelisib monotherapy versus combination therapy with aromatase inhibition and suggest that successful PIK3CA inhibition and sustained treatment effect using alpelisib monotherapy is possible in ER⁺ patients harboring combination PIK3CA and ESR1 mutations following multiple prior lines of therapy. It is plausible that discontinuation of endocrine therapy in these patients may relieve the selective pressure that resulted in the ESR1 alteration arising.

PTEN mutations at baseline were observed in patients with primary resistance to treatment, and acquired PTEN mutations were also observed at the EOT time point. Despite the small sample size, these findings support those of other studies and highlight the potential role of PTEN loss as a mechanism of both primary and acquired resistance to alpelisib therapy through bypass PI3Kb activation (50). ctDNA sequencing at the EOT time point also demonstrated further evidence of genomic evolution with the frequent acquisition of MAP3K1 and TP53 mutations. The predominance of acquired MAP3K1 mutations was consistent with findings of a recent study showing that activation of the MAPK pathway mediates resistance to the PI3Kδ inhibitor idelalisib (51).

In addition to ESR1, aberrations of TBX3 were tentatively associated with responses. TBX3 is a transcription factor, and loss of function has been noted as a potential driver alteration enriched in lobular breast carcinomas (9, 52). TBX3 has context-dependent effects on proliferation in breast cancer cell lines and may function to repress PTEN (53, 54). Due to the small number of patients with TBX3 alterations, only limited conclusions can be drawn and require confirmation in larger cohorts.

The efficacy reported here is substantially inferior to the outcomes seen in the SOLAR-1 phase III trial of alpelisib and fulvestrant versus fulvestrant alone, in which the median PFS was 11.0 months and the CBR was 61.5% (13). The overall response in SOLAR-1 was 26.6% compared with 30% here. The SOLAR-1 trial population had better performance status, less visceral disease, less metastatic disease burden, and fewer lines of prior therapy when compared with the patients enrolled in our trial. Furthermore, in our trial, 100% of patients in the ER⁺ cohort were refractory to endocrine therapy. In the phase Ib trial reported by Juric and colleagues, a more advanced, chemotherapy-experienced, and endocrine-resistant population treated with fulvestrant and alpelisib displayed a median PFS of 9.1 months in tumors with PIK3CA mutations, along with an ORR of 29% (19). Combining alpelisib with fulvestrant has a strong preclinical rationale, as PI3K inhibition alone promotes expression of the estrogen receptor, and combining fulvestrant and PI3K inhibition enhances efficacy (17). These effects may explain why we observed a less durable benefit with alpelisib monotherapy despite a response rate of 30%, suggesting that PI3K inhibitors should in general be combined with fulvestrant or other selective estrogen receptor–degrading compounds (SERD) even in endocrine-refractory populations. Correlative data from SOLAR-1 and other trials of PI3K inhibition plus SERD therapy (55) would help to clarify the significance of PIK3CA and ESR1 comutation.

We acknowledge a number of limitations of our study, including the small patient population, the exploratory nature of some analyses, and the focus of our ctDNA genomic analysis on the detection of somatic mutations without encompassing additional assessment of copy-number alterations. However, our study has provided unique insights into the genomic landscape predicting response and resistance to alpelisib monotherapy in breast cancer patients with PI3K alterations without the influence of endocrine therapy. We conclude that alpelisib monotherapy may be a relevant treatment alternative in patients with PIK3CA and ESR1 alterations and confirm that ctDNA dynamics can be useful for identifying patients with favorable responses to PI3K monotherapy.

A phase II, single-center, open-label trial of alpelisib monotherapy was conducted with two cohorts. The ER⁺ cohort enrolled patients with ER⁺HER2⁻ advanced breast cancer harboring a somatic genomic aberration in the PI3K pathway. The triple-negative cohort enrolled patients with ER-, progesterone receptor–negative (PR-), and HER- advanced breast cancer, without the requirement for a genomic aberration in the PI3K pathway. All patients in the ER⁺ cohort were required to have had prior treatment in the metastatic setting. All patients in the TNBC cohort were required to have developed metastatic disease within 12 months of adjuvant systemic therapy or received at least one line of therapy in the metastatic setting. All patients were required to have progressed on prior therapy. Key inclusion criteria were as follows: male or female patients ages 18 years or older with histologically confirmed breast cancer; adequate organ function; and measurable disease by RECIST 1.1. Key exclusion criteria included untreated central nervous system involvement (treated disease was allowed if clinically stable at least 4 weeks from treatment) and known type I diabetes mellitus or uncontrolled type II diabetes mellitus. The hormone receptor (via IHC) and HER2 status (via IHC and in situ hybridization assays) of each patient's tumor was ascertained from archival specimens that had been tested in approved laboratories prior to study enrolment. For the ER⁺ cohort, tumors were required to show estrogen receptor staining in 1% or more of tumor cells and be negative for HER2 as per American Society of Clinical Oncology/College of American Pathologists guidelines (24). In the TNBC cohort, tumors were required to show less than 1% staining for the estrogen receptor and be negative for HER2.

For the TNBC cohort, a Simon's two-stage optimal design was followed with an unacceptable response rate of 5% and an acceptable response rate of 25%. With a significance level of 5% and power of 80%, planned accrual was nine patients in stage 1. An additional eight patients would be recruited in stage 2 if at least one objective response was observed in stage 1. For the ER⁺ cohort in which there was preliminary evidence for efficacy in this population at the time of study design, planned accrual was 30 patients at a significance level of 5% and power of 83% to exclude an unremarkable ORR of 5% in favor of an ORR of 25% of more and provide additional power for correlative studies.

All patients provided written informed consent in alignment with the Declaration of Helsinki recommendations. The study protocol was approved by the Peter MacCallum Cancer Centre Human Research Ethics committee (HREC #14/176) and registered with the Australian New Zealand Clinical Trials Registry (ACTRN12615000850572) and ClinicalTrials.gov (NCT02506556). All patients were recruited and treated at the Peter MacCallum Centre (Melbourne, Australia) from August 26, 2015, to October 21, 2019. The final trial participant ceased therapy on April 3, 2020. Alpelisib was supplied by the manufacturer (Novartis).

Alpelisib was administered at a dose of 350 mg daily in 28-day cycles in both cohorts. For all patients, treatment continued until disease progression, intolerance, or withdrawal of trial participation. Tumor response was assessed every 2 cycles (8 weeks) with CT for the first 24 weeks and every 12 weeks thereafter until progression or study cessation. A whole-body bone scan was performed if clinically indicated at these time points. FDG-PET was performed at baseline and after 2 cycles, and the response was determined according to the EORTC criteria (25). Response assessment was conducted by the study investigators and was also centrally reviewed. Patients were followed for OS after ceasing trial therapy.

The primary objective of the study was to evaluate the efficacy of alpelisib using the objective response rate according to RECIST 1.1, defined as the percentage of patients achieving a complete or partial objective response (CR or PR) by central review. Responses did not require confirmation. Secondary objectives were determined by the investigator and included evaluating the efficacy of alpelisib using CBR, defined by the percentage of patients who achieve a CR or PR, or SD, for 6 months or more duration. Other secondary objectives included time to objective response in patients with CR or PR, duration of response in patients with CR or PR, and PFS (as assessed by the investigator). Objectives were determined in the intention-to-treat population and the evaluable cohort that reached the first restaging time point. All time-to-event data used the first day of alpelisib therapy as the baseline. Safety and tolerability were also assessed according to Common Terminology Criteria for Adverse Events (CTCAE) 4.03 criteria, and adverse events grade 3 or more, SAEs, and suspected unexpected serious adverse reactions were reported. Exploratory translational objectives were to investigate the efficacy of alpelisib with regard to the primary and secondary objectives in relation to genomic aberrations of the PI3K pathway and other driver alterations determined via tumor and ctDNA assays. The relationship between response on FDG-PET and the primary and secondary objectives was an additional exploratory objective. Investigation of metabolic parameters in relation to treatment outcomes was not prespecified.

Tumor sequencing was performed using an in-house custom hybrid capture panel, commercial hybrid capture panels, and amplicon assays on formalin-fixed, paraffin-embedded tumor samples with confirmed tumor content reviewed by a pathologist. Details of our in-house hybrid capture approach used for the majority of cases have been previously published (56). A validated pipeline was used to call and curate variants and copy-number alterations. For the commercial assays, alterations reported as significant by the vendor were used for analysis.

Cell-free DNA was extracted from 2 to 4 mL of plasma with ddPCR analysis performed using singleplex assays targeting the qualifying somatic mutations in the PI3K pathway and tracked serially at 8-week intervals for each patient using the Bio-Rad QX-200 system. ctDNA levels were reported in copies per milliliter as the average value between two technical replicates of the same plasma DNA sample. ctDNA was determined to be detected if at least one positive droplet signal was detected in both replicates but undetected if a positive droplet signal was detected in just one of the replicates.

In parallel, targeted sequencing using a custom panel of 394 amplicons across 39 genes frequently mutated in breast cancer was applied to plasma samples collected at baseline and EOT time points. Somatic mutations were called from targeted sequencing if detected in both duplicate reactions at baseline (unless also detected by ddPCR) and meeting all selected curation criteria (VAF ≥1%, read depth of ≥100, and variant depth of ≥10) and not detected in the matched germline blood sample. At the EOT time point, mutations were classified as maintained rather than acquired if these were previously detected in both targeted sequencing duplicate reactions at baseline or confirmed by ddPCR at baseline (57, 58). Patients were considered evaluable for the lead time analysis if ctDNA rise from nadir was detected from serial ddPCR analysis before clinical or radiologic progression was confirmed.

The translational analyses involving tumor and ctDNA sequencing, including both ddPCR and targeted sequencing, were preplanned in the study protocol without specific power calculations. Quoted P values should therefore be considered nominal. All statistical analyses were performed using the GraphPad Prism version 8.1.0 (GraphPad Software) and R statistical software version 4.0.2 (R Core Team; 2021; https://www.R-project.org). The Wilcoxon rank-sum test was used for unpaired continuous data and the Wilcoxon signed-rank test for paired data, both with two-sided alternative hypotheses. One-sided exact 95% CIs were calculated for the primary endpoint (ORR) using the Clopper–Pearson method. Hypothesis testing for increasing metabolic markers across ordered response categories was performed with one-sided Jonckheere–Terpstra tests. Hypothesis testing for proportional data was conducted using the two-sided Fisher exact test. P values less than 0.05 were considered statistically significant. P values were not adjusted for multiple testing unless an FDR was quoted, calculated with the method of Benjamini–Hochberg.

Survival curves were computed with the Kaplan–Meier estimator in GraphPad Prism. Associations between PFS and ctDNA (both ctDNA levels and ctDNA mutational profile) were determined using Cox proportional hazards models implemented in R. The Grambsch–Therneau test was used to check for the proportional hazards assumption, which was satisfied in all the fitted Cox proportional hazards models.

The processed mutation data from tumor sequencing are available in Supplementary Table S5. Processed mutation data from ctDNA ddPCR and targeted sequencing at baseline and EOT are available in Supplementary Tables S7 and S8.

P. Savas reports grants from Cancer Council Victoria and Novartis during the conduct of the study, as well as grants from Roche/Genentech outside the submitted work. R.J. Hicks reports other support from Telix Pharmaceuticals outside the submitted work. P.A. Francis reports other support from Novartis outside the submitted work. C.K. Lee reports personal fees and other support from Novartis, personal fees from GSK, Pfizer, and Takeda, and grants and personal fees from Roche, AstraZeneca, Amgen, and Merck KGA outside the submitted work. S.-J. Dawson reports grants from Roche/Genentech and Cancer Therapeutics CRC, and personal fees from AstraZeneca and Inivata outside the submitted work. S. Loi reports other support from research funding to her institution from Novartis, Bristol Meyers Squibb, Merck, Puma Biotechnology, Eli Lilly, Nektar Therapeutics AstraZeneca, Roche/Genentech, and Seattle Genetics, nonfinancial support as a consultant (not compensated) from Seattle Genetics, Novartis, Bristol Meyers Squibb, Merck, AstraZeneca, and Roche/Genentech, and other support as a consultant (paid to institution) from Aduro Biotech, Novartis, GSK, Roche/Genentech, AstraZeneca, Silverback Therapeutics, G1 Therapeutics, PUMA Biotechnologies, Pfizer, Gilead Therapeutics, Seattle Genetics, and Bristol Meyers Squibb outside the submitted work. No disclosures were reported by the other authors.

P. Savas: Conceptualization, resources, data curation, formal analysis, investigation, visualization, writing–original draft, project administration, writing–review and editing. L.L. Lo: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S.J. Luen: Conceptualization, data curation, formal analysis, visualization, writing–review and editing. E.F. Blackley: Data curation, writing–review and editing. J. Callahan: Resources, data curation, writing–review and editing. K. Moodie: Data curation, writing–review and editing. C.T. van Geelen: Data curation, investigation, writing–review and editing. Y.-A. Ko: Investigation, project administration. C.-F. Weng: Data curation, investigation. L. Wein: Data curation, investigation. M.J. Silva: Data curation, investigation. A. Zivanovic Bujak: Data curation, investigation. M.M. Yeung: Data curation, formal analysis, visualization, writing–review and editing. S. Ftouni: Data curation, investigation, writing–review and editing. R.J. Hicks: Conceptualization, resources, supervision, writing–review and editing. P.A. Francis: Conceptualization, resources, supervision, methodology, writing–review and editing. C.K. Lee: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. S.-J. Dawson: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing. S. Loi: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank the following: Novartis for supplying alpelisib; the Molecular Genomics Core and Pathology Department at the Peter MacCallum Cancer Centre for their assistance; and Dr. Madawa Jayawardana for advice on statistical analysis. The authors also acknowledge the generous support of Jeff Eisman. Financial support (grants with source details for each author) is as follows: P. Savas: fellowship support from Cancer Council Victoria; L.L. Lo: PhD scholarship from the National Health and Medical Research Council, the Royal Australasian College of Physicians, and the National Breast Cancer Foundation; S.J. Luen: fellowship support from the Peter MacCallum Cancer Centre; R.J. Hicks: Australian National Health and Medical Research Council Practitioner Fellowship; S.-J. Dawson: CSL Centenary Fellowship and National Health and Medical Research Council Investigator grant (#1196755); and S. Loi: National Breast Cancer Foundation of Australia Endowed Chair and the Breast Cancer Research Foundation, New York.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.