Purpose:

The mTOR complex C1 (mTORC1) inhibitor everolimus in combination with the aromatase inhibitor exemestane is an effective treatment for patients with hormone receptor—positive (HR+), HER2-negative (HER2), advanced breast cancer (HR+/HER2 aBC). However, everolimus can cause hyperglycemia and hyperinsulinemia, which could reactivate the PI3K/protein kinase B (AKT)/mTORC1 pathway and induce tumor resistance to everolimus.

Experimental Design:

We conducted a multicenter, retrospective, Italian study to investigate the impact of baseline and on-treatment (i.e., during first 3 months of therapy) blood glucose levels on progression-free survival (PFS) in patients with HR+/HER2 aBC treated with everolimus-exemestane.

Results:

We evaluated 809 patients with HR+/HER2 aBC treated with everolimus-exemestane as any line of therapy for advanced disease. When evaluated as dichotomous variables, baseline and on-treatment glycemia were not significantly associated with PFS. However, when blood glucose concentration was evaluated as a continuous variable, a multivariable model accounting for clinically relevant patient- and tumor-related variables revealed that both baseline and on-treatment glycemia are associated with PFS, and this association is largely attributable to their interaction. In particular, patients who are normoglycemic at baseline and experience on-treatment diabetes have lower PFS compared with patients who are already hyperglycemic at baseline and experience diabetes during everolimus-exemestane therapy (median PFS, 6.34 vs. 10.32 months; HR, 1.76; 95% confidence interval, 1.15–2.69; P = 0.008).

Conclusions:

The impact of on-treatment glycemia on the efficacy of everolimus-exemestane therapy in patients with HR+/HER2 aBC depends on baseline glycemia. This study lays the foundations for investigating novel therapeutic approaches to target the glucose/insulin axis in combination with PI3K/AKT/mTORC1 inhibitors in patients with HR+/HER2 aBC.

Translational Relevance

Everolimus and other PI3K/protein kinase B (AKT)/mTOR complex 1 (mTORC1) pathway inhibitors are associated with metabolic adverse events, including hyperglycemia/diabetes and hyperinsulinemia. The impact of baseline and on-treatment blood glucose levels on the clinical efficacy of everolimus-based combinations remains poorly defined. Here we performed a large observational study, showing an interaction between baseline and on-treatment glycemia in affecting the risk of disease progression in patients with advanced breast cancer treated with everolimus-exemestane combination. In particular, patients with normal baseline glycemia have significantly worse clinical outcomes if they experience on-treatment hyperglycemia. This study supports the use of early alterations in blood glucose concentration as a biomarker of everolimus-exemestane efficacy and provides the rationale for exploiting novel metabolic interventions as anticancer strategies in advanced breast cancer.

The PI3K/protein kinase B (AKT)/mTOR complex 1 (mTORC1) pathway is the most commonly dysregulated oncogenic axis in hormone receptor–positive (HR+), HER2-negative breast cancer (HER2; refs. 1–4). In both preclinical and clinical studies, the PI3K/AKT/mTORC1 pathway has been crucially implicated in stimulating HR+/HER2 breast cancer cell growth, proliferation, and survival, as well as in causing primary or acquired tumor resistance to endocrine therapies (ET; refs. 5–7). In line with this preclinical evidence, randomized phase III trials showed that inhibiting different nodes of the PI3K/AKT/mTORC1 axis in combination with standard ETs results in a significant prolongation of progression-free survival (PFS) when compared with ET alone in patients with HR+/HER2 advanced breast cancer (aBC; refs. 8, 9). In particular, the BOLERO-2 trial demonstrated that the mTORC1 inhibitor everolimus in combination with the steroidal aromatase inhibitor exemestane improves PFS when compared with exemestane alone in patients with postmenopausal HR+/HER2 aBC progressing after/on prior nonsteroidal aromatase inhibitor (NSAI) therapy (8). More recently, the PI3K inhibitor alpelisib in combination with the antiestrogen fulvestrant significantly prolonged PFS when compared with fulvestrant alone in patients with PIK3CA-mutated HR+/HER2 aBC progressing on previous aromatase inhibitor (AI) therapy (9).

Metabolic adverse events (AE), including hyperglycemia, hypercholesterolemia, and hypertriglyceridemia, are common in patients treated with PI3K/AKT/mTORC1 inhibitors (8–11), and are considered a class effect of these drugs. In particular, hyperglycemia occurs in up to 17% of patients with HR+/HER2 aBC treated with everolimus (8, 12), and results from a combination of impaired pancreatic β-cell function, enhanced glycogen breakdown in the liver, and insulin resistance, which impairs glucose uptake in the skeletal muscle and adipose tissue (13–16). In turn, everolimus-induced hyperglycemia can cause compensatory hyperinsulinemia, which could reactivate the insulin receptor (IR)/PI3K/AKT/mTORC1 pathway and make cancer cells resistant to everolimus-exemestane (17). In line with this hypothesis, a small retrospective Italian study showed that higher blood glucose levels during everolimus-exemestane therapy correlate with worse PFS in patients with HR+/HER2 aBC (18). Moreover, one recent preclinical study indicated that PI3K inhibitor-induced increase of serum insulin concentration in patients with cancer might be sufficient to reactivate the PI3K/AKT/mTORC1 pathway, thus resulting in resistance to PI3K inhibition in HR+/HER2 breast cancer cell lines and murine models (19).

Here, we performed a large, multicenter, retrospective study to investigate the impact of blood glucose levels on the efficacy of everolimus-exemestane treatment in patients with HR+/HER2 aBC. We provide first evidence that both baseline and on-treatment glycemia are associated with everolimus-exemestane efficacy, and this effect is largely attributable to the interaction between these two variables.

Patient population and enrollment criteria

This was an observational, retrospective, multicenter study conducted in 20 Italian Cancer Centers [Fondazione IRCCS Istituto Nazionale dei Tumori di Milano (coordinating center); Istituto Oncologico Veneto di Padova; Policlinico Umberto I di Roma; Azienda Ospedaliero Universitaria Pisana; Azienda Ospedaliera Policlinico di Modena; Ospedale Policlinico San Martino di Genova; Ospedale Belcolle di Viterbo; Istituto Europeo di Oncologia, IRCCS - IEO di Milano; FPO-IRCCS Candiolo Cancer Institute; Humanitas Clinical and Research Center - IRCCS di Milano; ASST di Cremona; Istituto Nazionale Tumori Regina Elena - IFO di Roma; Spedali Civili di Brescia; Ospedale “Vito Fazzi” di Lecce; Istituto Scientifico Romagnolo per lo Studio e la Cura dei Tumori (IRST) IRCCS di Meldola; Università Federico II di Napoli; ASST Santi Paolo e Carlo di Milano; ASST Fatebenefratelli Sacco di Milano; IRCCS Centro di Riferimento Oncologico di Aviano; Azienda Sanitaria Universitaria Friuli Centrale, Udine]. Data were collected through an electronic database.

The main enrollment criteria consisted in: (i) age ≥18 years; (ii) histologically/cytologically confirmed diagnosis of HR+/HER2 advanced (inoperable locally advanced or metastatic) breast cancer; (iii) postmenopausal status, as defined as: (a) patients of age equal to or higher than 60 years, (b) patients of age lower than 60 years but with amenorrhea from at least 12 months that was not related to the administration of chemotherapy or luteinizing hormone releasing hormone (LHRH) analogs, (c) premenopausal/perimenopausal patients receiving LHRH analogs in combination with everolimus-exemestane, or (d) patients with ovarian ablation either through radiotherapy or bilateral ovariectomy; (iv) treatment for at least 1 month with daily everolimus (initial dosage of 10 mg/day) plus exemestane (25 mg/day) between October 2012 and July 2019 outside clinical trials sponsored by pharmaceutical companies; (v) disease recurrence or progression after/on prior therapy with NSAIs plus/minus cyclin-dependent kinase 4/6 (CDK 4/6) inhibitors; (vi) availability of at least one measurement of plasma glucose concentration at the initiation of everolimus-exemestane therapy or at 1, 2, or 3 months after treatment initiation); (vii) any number of previous lines of treatment for advanced disease; and (viii) any prior therapy for localized disease, including (neo)adjuvant chemotherapy, surgery, and ETs. Patients with de novo metastatic disease at diagnosis were included as well. Prior NSAI therapy should not necessarily be the last treatment before everolimus-exemestane therapy. All patients were followed up until death, loss of contact, or time of data lock (July 31, 2019). Written informed consent was obtained from all patients who were alive at the time of study conduction. The study was carried out in accordance with the Good Clinical Practice guidelines and the Declaration of Helsinki. The study protocol was first approved by the Ethics Committee of the coordinating center (internal registration number of the study: INT 30/18), and then approved by ethics committees and/or institutional review boards at each participating site.

Study objectives and statistical plan

The primary objective of the study was to investigate the association between the onset of early hyperglycemia and the PFS of patients with HR+/HER2 aBC treated with everolimus-exemestane. Early hyperglycemia was defined as equal or higher than 126 mg/dL average fasting plasma glucose concentration during the first 3 months of everolimus-exemestane treatment (i.e., excluding baseline evaluation). PFS was defined as the time between everolimus-exemestane initiation and the detection of clinical/radiological disease progression (according to RECIST v1.1) or patient death from any cause, whichever occurred first. For sample size calculation, we assumed that 80% of patients had an average glycemia below 126 mg/dL during the first 3 months of everolimus-exemestane therapy and that normoglycemic patients had median PFS of 7 months (8). With these assumptions, to detect a HR of progressive disease of 1.43 in hyperglycemic versus normoglycemic patients with 90% statistical power and two-sided α error of 0.05, an accrual of approximately 800 patients was estimated. The HR threshold of 1.43 was chosen on the basis of a preliminary, monocentric evaluation performed in the first 110 patients treated with everolimus-exemestane at the coordinating center.

Secondary objectives of the study were: (i) to investigate the association between baseline hyperglycemia (defined as fasting blood glycemia ≥126 mg/dL measured within 28 days before the initiation of everolimus-exemestane) and patient PFS; (ii) to evaluate the association between the onset of precocious hypercholesterolemia and hypertriglyceridemia, as defined as average fasting plasma cholesterol and triglycerides ≥200 mg/dL and ≥170 mg/dL, respectively, during the first 3 months of everolimus-exemestane treatment, and patient PFS; (iii) to investigate the association between baseline hypercholesterolemia (≥ 200 mg/dL) or baseline hypertriglyceridemia (≥170 mg/dL) and PFS; and (iv) to assess the impact of baseline and on-treatment glycemia, cholesterolemia, and triglyceridemia, as evaluated as continuous variables, on PFS. Patients who had not experienced disease progression or death at data cutoff and analysis were censored at the time of last disease evaluation or last follow-up.

Glucose, cholesterol, and triglyceride evaluation

Measurement of fasting (at least 8 hours after the last meal) plasma glucose, cholesterol, and triglyceride concentration was performed at baseline and before initiating a new treatment cycle as per clinical practice; data regarding metabolite measurements at baseline and at 1, 2, and 3 months were collected whenever available. For the purpose of the study, metabolite measurements obtained during the first 3 months of everolimus-exemestane treatment (i.e., excluding baseline evaluations) were summarized as average, maximum, and absolute differences with respect to baseline levels (delta). The average was defined as the arithmetic mean of metabolite concentrations during the study treatment (baseline excluded). The maximum was defined as the highest value of metabolite measurement during the first 3 months of everolimus-exemestane therapy (baseline excluded). The delta was defined as the absolute difference between maximum and baseline values for each metabolic variable. Baseline, average, and maximum values were analyzed both as dichotomous variables, with a cutoff of 126, 200, and 170 mg/dL for plasma glucose, cholesterol, and triglycerides, respectively, and as continuous variables. On-treatment changes of each metabolic parameter were evaluated by comparing baseline measurements with the average value of the same parameter during the first 3 months of treatment.

Statistical methods

Standard descriptive statistics were used to summarize clinical and biological patients' characteristics. Both paired and unpaired t tests were used to compare baseline and on-treatment concentration of metabolic parameters, adjusting P values for multiple comparisons through the Benjamini–Hochberg procedure. Median patient follow-up was quantified with the reverse Kaplan–Meier estimator (20). Survival analysis methods were used to analyze PFS. Survival curves and related descriptive statistics were obtained with the Kaplan–Meier method and comparisons between curves were performed with the log-rank test. Multivariable analyses were performed according to a two-step strategy. In the first step, we modeled covariates by resorting to a random forest method (21). This approach was used to guide and benchmark the subsequent use of more conventional modeling methods according to the following endpoints: detection and exclusion of prognostically irrelevant covariates (based on minimal depth statistic); guidance on the presence of nonlinear effects of continuous predictors or interactions among covariates; and joint predictive performance. The second step relied on the use of Cox regression modeling, with the proportional hazard assumption checked by testing and plotting Schoenfeld residuals. For all continuous variables, nonlinear effects were handled by means of restricted cubic splines. Cox model results were summarized using HRs, together with the corresponding 95% confidence intervals (CI) and Wald P values, while overall model performance was assessed in terms of discrimination with the bootstrap-adjusted Harrell c index. In Cox models, the HR for continuous variables was reported as the HR related to the interquartile range (IQR; interval between the 75th and 25th quantiles). Given the presence of missing data, we performed Cox model analyses both on complete datasets and after 10-fold multiple imputation (22). In addition, a landmark analysis was conducted to explore a possible bias introduced by the time-dependent assessment of metabolic parameters during the first 3 months of treatment; in this landmark analysis, we investigated the impact of baseline and on-treatment glycemia on patient PFS after excluding patients undergoing disease progression during the first 3 months of therapy.

Statistical analyses were carried out with SAS (version 9.4, SAS Institute) and R software (version 3.6.1, R Foundation for Statistical Computing). Statistical significance was set at the conventional 5% two-sided threshold.

Patient population

We evaluated a total number of 848 patients. Of these, 35 patients were excluded because of the lack of at least one blood glucose measurement at baseline or during the first 3 months of treatment, while 4 patients were excluded because of the unavailability of the date of last follow-up. The study CONSORT diagram is shown in Supplementary Fig. S1. Finally, 809 patients fulfilling all the enrollment criteria and treated with the everolimus-exemestane combination between October 2012 and July 2019 were included. Baseline patient and disease characteristics are displayed in Table 1. All patients had received prior therapy with NSAIs in the adjuvant or advanced treatment setting, while 54% of them received antiestrogens (i.e., fulvestrant and/or tamoxifen) for the treatment of advanced disease. At data cutoff and analysis, 775 patients had experienced disease progression during everolimus-exemestane treatment, and 435 patients had died. Median follow-up time was 37.4 months (IQR: 22.8–56.4), with median PFS of 7.13 months (IQR: 3.8–12.9) and median OS of 32.1 months (IQR: 15.9–54.8).

Table 1.

Patient and tumor characteristics.

CharacteristicTotal patients, N = 809
 N (%) 
ECOG PS 
 0 567 (70.2) 
 1 227 (28.1) 
 2 14 (1.7) 
 NA 
Use of metformin 
 Started before EVE-EXE 62 (7.8) 
 Started during EVE-EXE 31 (3.9) 
 NA 15 
Sites of metastatic disease 
 Lymph nodes 307 (38.0) 
  NA 
 Bones 590 (73.2) 
  NA 
 Liver 258 (32.0) 
  NA 
 Lungs 229 (28.4) 
  NA 
 CNS 21 (2.6) 
  NA 
 Soft tissues 86 (10.7) 
  NA 
 Others 68 (8.4) 
  NA 
Visceral disease 450 (55.8) 
 NA 
Prior antineoplastic therapies 
 Prior adjuvant ET 569 (70.9) 
  NA 
 Prior adjuvant ChT 453 (56.6) 
  NA 
 Prior anthracycline treatment 499 (61.9) 
  NA 
 Prior taxane treatment 418 (51.9) 
  NA 
 Prior antiestrogens 434 (53.6) 
  NA — 
Everolimus dose variations  
 Full dose 428 (52.9) 
 Reduction (5 mg) 325 (40.2) 
 Interruptiona 56 (6.9) 
  NA — 
 Median (IQR) 
Age, years 63 (56–70) 
 NA 
BMI 24.7 (22.2–27.8) 
 NA 36 
Disease-free interval, monthsb 54 (19–106) 
 NA 23 
Line of everolimus treatment  
 ET + ChTc 3 (2–4) 
 ET onlyc 2 (2–4) 
 NA 
CharacteristicTotal patients, N = 809
 N (%) 
ECOG PS 
 0 567 (70.2) 
 1 227 (28.1) 
 2 14 (1.7) 
 NA 
Use of metformin 
 Started before EVE-EXE 62 (7.8) 
 Started during EVE-EXE 31 (3.9) 
 NA 15 
Sites of metastatic disease 
 Lymph nodes 307 (38.0) 
  NA 
 Bones 590 (73.2) 
  NA 
 Liver 258 (32.0) 
  NA 
 Lungs 229 (28.4) 
  NA 
 CNS 21 (2.6) 
  NA 
 Soft tissues 86 (10.7) 
  NA 
 Others 68 (8.4) 
  NA 
Visceral disease 450 (55.8) 
 NA 
Prior antineoplastic therapies 
 Prior adjuvant ET 569 (70.9) 
  NA 
 Prior adjuvant ChT 453 (56.6) 
  NA 
 Prior anthracycline treatment 499 (61.9) 
  NA 
 Prior taxane treatment 418 (51.9) 
  NA 
 Prior antiestrogens 434 (53.6) 
  NA — 
Everolimus dose variations  
 Full dose 428 (52.9) 
 Reduction (5 mg) 325 (40.2) 
 Interruptiona 56 (6.9) 
  NA — 
 Median (IQR) 
Age, years 63 (56–70) 
 NA 
BMI 24.7 (22.2–27.8) 
 NA 36 
Disease-free interval, monthsb 54 (19–106) 
 NA 23 
Line of everolimus treatment  
 ET + ChTc 3 (2–4) 
 ET onlyc 2 (2–4) 
 NA 

Note: Data are presented as N (%) unless otherwise specified.

Abbreviations: BMI, body mass index; ChT, chemotherapy; CNS, central nervous system; ECOG PS, Eastern Cooperative Oncology Group performance status; ET, endocrine treatment; EVE-EXE, everolimus-exemestane; IQR, interquartile range; NA, not available.

aEverolimus precocious interruption was defined as treatment suspension at least 3 months before disease progression.

bDefined as the time between surgery for the primary tumor and diagnosis of distant relapse.

cDefined as the everolimus-exemestane treatment line for advanced disease considering both previous ET and ChT, and ET only, respectively.

Effect of everolimus-exemestane on blood metabolic parameters

Details about baseline and on-treatment metabolic biomarkers are described in Supplementary Table S1. At baseline, fasting plasma glucose measurements were available for 722 (89.2%) patients; of these, 79 (10.9%) patients were hyperglycemic according to the prespecified threshold (i.e., ≥ 126 mg/dL). At 1, 2, and 3 months after everolimus-exemestane initiation, plasma glucose measurements were available for 692 (85.5%), 643 (79.5%), and 537 (66.4%) patients, respectively. Consistent with the study assumptions, 186 (24.1%) of 772 patients with at least one available on-treatment plasma glucose measurement were found to be hyperglycemic (i.e., average plasma glucose concentration ≥ 126 mg/dL).

Blood glucose, cholesterol, and triglyceride concentration significantly increased after the first month on therapy (when compared with baseline values), and remained stable between the first and second month, with an initial reduction of blood glucose and cholesterol levels after 3 months (Supplementary Table S1; Supplementary Fig. S2). Overall, average blood glucose, cholesterol, and triglyceride concentration during the first 3 months of treatment was significantly higher when compared with baseline measurements (Supplementary Fig. S2).

Patient and treatment characteristics according to on-treatment glycemic status are summarized in Supplementary Table S2. Overall, normoglycemic and hyperglycemic patients were well balanced with respect to these factors, with the exception that hyperglycemic patients were significantly older and had higher body mass index (BMI). In addition, hyperglycemic patients were more likely to receive metformin as an antidiabetic medication, started either before or during everolimus-exemestane treatment.

Regarding plasma cholesterol and triglyceride concentration, baseline measurements of these parameters were available for 536 (66.3%) and 500 (61.8%) patients, respectively, with a total number of 340 (63.4%) hypercholesterolemic (≥200 mg/dL) and 93 (18.6%) hypertriglyceridemic (≥170 mg/dL) patients. At 1, 2, and 3 months after everolimus-exemestane initiation, blood cholesterol measurements were available for 477 (59.0%), 421 (52.0%), and 387 (47.8%) patients, respectively, while data on triglyceride concentration were available for 440 (54.4%), 383 (47.3%), and 351 (43.4%) patients, respectively. Average on-treatment hypercholesterolemia and hypertriglyceridemia were detected in 472 (78.9%) and 181 (32.3%) patients, respectively.

There was a moderate, positive correlation between baseline and on-treatment concentration of each of the three metabolites, while we found a strong, positive correlation between their average and maximum on-treatment concentration (Supplementary Table S3). Therefore, for subsequent evaluations we only considered the average concentration of each blood metabolite (rather than their maximum).

Association between dichotomized metabolic parameters and PFS

Patients who were hyperglycemic at baseline had nonstatistically significantly different PFS when compared with normoglycemic patients [median PFS (mPFS), 6.14 vs. 7.26 months, respectively; unadjusted HR, 1.18; 95% CI, 0.93–1.50; P = 0.168; Fig. 1A]. Similarly, there were no significant PFS differences between hyperglycemic and normoglycemic patients according to on-treatment glycemia (mPFS 6.97 vs. 7.13 months; unadjusted HR, 1.08; 95% CI, 0.91–1.28; P = 0.371; Fig. 1B).

Figure 1.

PFS represented through Kaplan–Meier curves according to baseline (A) and on-treatment (average) blood glucose (B) concentration.

Figure 1.

PFS represented through Kaplan–Meier curves according to baseline (A) and on-treatment (average) blood glucose (B) concentration.

Close modal

The impact of baseline and on-treatment cholesterol and triglyceride concentration according to the prespecified thresholds was nonstatistically significant as well. In particular, we did not find a significant association between baseline cholesterol or triglycerides levels and patient PFS (mPFS in hypercholesterolemic vs. normocholesterolemic patients: 7.95 vs. 7.82 months; unadjusted HR, 0.94; 95% CI, 0.78–1.12; P = 0.479; mPFS in hypertriglyceridemic vs. normotriglyceridemic patients: 5.75 vs. 7.95 months; unadjusted HR, 1.12; 95% CI, 0.89–1.41; P = 0.342; Supplementary Fig. S3A and S3B). Similarly, PFS was not statistically significantly different in hypercholesterolemic versus normocholesterolemic (mPFS of 7.59 vs. 6.21 months, respectively; unadjusted HR, 0.92; 95% CI, 0.75–1.12; P = 0.403) and in hypertriglyceridemic vs. normotriglyceridemic (mPFS: 7.95 vs. 7.20 months, respectively; unadjusted HR, 0.93; 95% CI, 0.78–1.12; P = 0.479) patients when on-treatment metabolite levels were considered (Supplementary Fig. S3C and S3D).

Impact of baseline and on-treatment glycemia as continuous variables on PFS

Then, we investigated in a multivariable model the impact of blood glucose concentration, as evaluated as a continuous variable, on patient PFS. To this aim, we first performed an exploratory analysis based on Random Forest algorithm (see Materials and Methods) to exclude clinically irrelevant variables (i.e., variables not associated with PFS). On the basis of this analysis, the following covariates were excluded: presence of lung metastases, bone metastases, lymph node metastases, central nervous system (CNS) metastases, or soft-tissue metastases; prior therapy with anthracyclines and/or taxanes; adjuvant chemotherapy; and adjuvant ET. The use of metformin was also excluded as a covariate for subsequent analyses (Supplementary Fig. S4A). The following predictors of PFS were instead selected for further evaluation in the multivariable model: patient age, BMI, Eastern Cooperative Oncology Group Performance Status (ECOG PS), line of everolimus-exemestane treatment, everolimus dosages, presence of visceral disease, presence of liver metastases, disease-free interval (as defined as the time between surgery of the primary tumor and tumor recurrence as metastatic disease), baseline and on-treatment glycemia, baseline and on-treatment cholesterolemia, and baseline and on-treatment triglyceridemia (Supplementary Table S4). Of note, the effect of metabolic parameters on patient PFS was nonlinear and, in the case of blood glucose, was characterized by a pattern of interaction between baseline and on-treatment glycemia (Supplementary Fig. S5A and S5B).

After selecting potentially relevant variables, we fitted a Cox regression model to assess the independent impact of these variables on patient PFS. In a first model, among metabolic variables we only included baseline and on-treatment blood glucose levels, along with their interaction. Missing metabolic data were imputed (see Materials and Methods). This model revealed a negligible impact of baseline glycemia on PFS, while there was a moderate association between high on-treatment glycemia and worse PFS (Table 2A). Notably, the impact of both baseline and on-treatment glycemia on PFS was largely attributable to the interaction between these two factors, as demonstrated by hierarchical statistical testing of model coefficients (Supplementary Table S5). We found similar results when cholesterol and triglyceride concentration was also included in the Cox model (Table 2B). In both multivariable models, more advanced everolimus-exemestane treatment line, worse ECOG PS and the presence of liver metastases were associated with worse PFS, while a reduction of everolimus dosage during the treatment course correlated with better PFS (Table 2A and B). To test the stability of the first model (Table 2A), we fitted another Cox model keeping the same structure, but only including complete data, that is, after excluding missingness, for a total number of 643 patients included. Of note, this analysis confirmed that the interaction between baseline and on-treatment glycemia is largely responsible for the observed association between blood glucose levels and patient PFS (Supplementary Table S6). To further confirm the robustness of these results, we performed a landmark analysis, in which we excluded patients experiencing disease progression during the first 3 months of everolimus-exemestane treatment (i.e., when on-treatment glycemia is evaluated). This analysis confirmed an impact of baseline and on-treatment glycemia on patient PFS (Supplementary Table S7). In all these models, patients undergoing precocious everolimus interruption or dose reduction had a lower risk of undergoing disease progression when compared with patients continuing everolimus until disease progression (Supplementary Fig. S6A). We asked whether this finding could be explained by a different duration of everolimus exposure (time to everolimus treatment interruption) in different patient subsets. Interestingly, patients undergoing everolimus dose reduction were exposed to everolimus for longer time intervals when compared with patients who received standard everolimus dosages until disease progression; in contrast, the length of everolimus exposure was significantly lower in patients undergoing precocious treatment interruption when compared with patients who did not interrupt everolimus, as well as when compared with patients undergoing everolimus dose reduction (Supplementary Fig. S6B). As expected, everolimus-induced grade 1/2 or grade 3/4 adverse events were significantly more common in patients undergoing treatment interruption/dose reduction (Supplementary Table S8). Removing the variable “everolimus interruption/dose variations” from the multivariable model confirmed the main study findings, including the interaction between baseline and on-treatment glycemia in affecting patient PFS (Supplementary Table S9).

Table 2.

Multivariable Cox proportional hazards models for PFS when considering only baseline and on-treatment blood glucose concentration as a metabolic variable (A) or after also including cholesterol and triglyceride levels (B). In both models, missing blood glucose measurements were imputed.

A. Imputed data/blood glucose only
VariablesHR95% CIP
Baseline glycemiaa Continuous 0.94 0.78–1.13 <0.001 
On-treatment glycemiaa Continuous 1.19 0.98–1.44 <0.001 
Line of EVE-EXE treatment Continuous 1.23 1.12–1.35 <0.001 
Age Continuous 1.15 0.94–1.41 0.157 
Disease-free interval Continuous 0.89 0.71–1.12 0.733 
BMI Continuous 1.04 0.84–1.27 0.251 
Everolimus interruption/dose reduction Reduction vs. full dose 0.78 0.67–0.91 <0.001 
 Interruption vs. full dose 0.40 0.30–0.53  
ECOG PS 1 vs. 0 1.31 1.11–1.55 0.005 
 2 vs. 0 1.46 0.65–3.25  
Visceral disease Yes vs. No 1.18 0.97–1.42 0.093 
Presence of liver metastases Yes vs. No 1.32 1.07–1.63 0.010 
B. Imputed data/all metabolic parameters 
Baseline glycemiaa Continuous 0.92 0.77–1.10 <0.001 
On-treatment glycemiaa Continuous 1.19 0.98–1.45 <0.001 
Baseline cholesterol Continuous 1.10 0.91–1.32 0.398 
Average cholesterol Continuous 0.89 0.74–1.07 0.206 
Baseline triglycerides Continuous 1.16 0.95–1.40 0.323 
Average triglycerides Continuous 0.95 0.77–1.18 0.901 
Line of EVE-EXE treatment Continuous 1.24 1.13–1.36 <0.001 
Age Continuous 1.15 0.93–1.40 0.185 
Disease-free interval Continuous 0.91 0.72–1.14 0.811 
BMI Continuous 1.01 0.82–1.24 0.212 
Everolimus interruption/dose reduction Reduction vs. full dose 0.78 0.67–0.91 <0.001 
 Interruption vs. full dose 0.38 0.28–0.52  
ECOG PS 1 vs. 0 1.31 1.11–1.54 0.006 
 2 vs. 0 1.46 0.65–3.27  
Visceral disease Yes vs. No 1.20 0.99–1.45 0.059 
Presence of liver metastases Yes vs. No 1.31 1.05–1.62 0.015 
A. Imputed data/blood glucose only
VariablesHR95% CIP
Baseline glycemiaa Continuous 0.94 0.78–1.13 <0.001 
On-treatment glycemiaa Continuous 1.19 0.98–1.44 <0.001 
Line of EVE-EXE treatment Continuous 1.23 1.12–1.35 <0.001 
Age Continuous 1.15 0.94–1.41 0.157 
Disease-free interval Continuous 0.89 0.71–1.12 0.733 
BMI Continuous 1.04 0.84–1.27 0.251 
Everolimus interruption/dose reduction Reduction vs. full dose 0.78 0.67–0.91 <0.001 
 Interruption vs. full dose 0.40 0.30–0.53  
ECOG PS 1 vs. 0 1.31 1.11–1.55 0.005 
 2 vs. 0 1.46 0.65–3.25  
Visceral disease Yes vs. No 1.18 0.97–1.42 0.093 
Presence of liver metastases Yes vs. No 1.32 1.07–1.63 0.010 
B. Imputed data/all metabolic parameters 
Baseline glycemiaa Continuous 0.92 0.77–1.10 <0.001 
On-treatment glycemiaa Continuous 1.19 0.98–1.45 <0.001 
Baseline cholesterol Continuous 1.10 0.91–1.32 0.398 
Average cholesterol Continuous 0.89 0.74–1.07 0.206 
Baseline triglycerides Continuous 1.16 0.95–1.40 0.323 
Average triglycerides Continuous 0.95 0.77–1.18 0.901 
Line of EVE-EXE treatment Continuous 1.24 1.13–1.36 <0.001 
Age Continuous 1.15 0.93–1.40 0.185 
Disease-free interval Continuous 0.91 0.72–1.14 0.811 
BMI Continuous 1.01 0.82–1.24 0.212 
Everolimus interruption/dose reduction Reduction vs. full dose 0.78 0.67–0.91 <0.001 
 Interruption vs. full dose 0.38 0.28–0.52  
ECOG PS 1 vs. 0 1.31 1.11–1.54 0.006 
 2 vs. 0 1.46 0.65–3.27  
Visceral disease Yes vs. No 1.20 0.99–1.45 0.059 
Presence of liver metastases Yes vs. No 1.31 1.05–1.62 0.015 

Note: The HR for continuous variables is expressed as the HR of disease progression related to the interquartile range (interval between the 75th and 25th quantiles). Bold indicates statistical significance (P < 0.05).

Abbreviations: BMI, body mass index; CI, confidence interval; EVE-EXE, everolimus-exemestane; ECOG PS, Eastern Cooperative Oncology Group performance status.

aIncluding nonlinear and interaction terms.

Role of the interaction between baseline and on-treatment glycemia on PFS

The presence of an interaction between baseline and on-treatment glycemia makes results of Cox models poorly interpretable, in particular, with respect to the HRs that summarize the impact of individual variables on PFS. To dissect the pattern of interaction between baseline and on-treatment glycemia, we plotted log relative hazards according to on-treatment blood glucose concentrations (80–270 mg/dL range) at three different levels of baseline blood glycemia, namely, 85, 95, and 125 mg/dL, which correspond to the 10th, 50th, and 90th distribution quantiles, respectively. For baseline glycemia of 85 mg/dL, we found a 4-fold increase in log relative hazard for increasing on-treatment blood glucose levels (Fig. 2A). At a level of baseline glycemia of 95 mg/dL, we observed a similar pattern, with a 2-fold increase in log relative hazard for increasing on-treatment blood glucose levels (Fig. 2B). Finally, the log relative hazard curve was flat at the level of 125 mg/dL baseline glycemia (Fig. 2C). These data indicate that an increase of blood glucose concentration during everolimus-exemestane therapy might be associated with an increased risk of disease progression in patients with normal glycemia at baseline, but not in patients who are already hyperglycemic before treatment initiation.

Figure 2.

A–C, Curves showing the impact of on-treatment glycemia on hazard for disease progression, according to baseline glycemia. Curves were drawn at the 10th (A), 50th (B), and 90th (C) percentile of the baseline (85, 95, 125). D, Contour plot model describing the impact of baseline glycemia (y axis), on-treatment glycemia (x axis), and predicted patient PFS (z axis, corresponding to the color scale).

Figure 2.

A–C, Curves showing the impact of on-treatment glycemia on hazard for disease progression, according to baseline glycemia. Curves were drawn at the 10th (A), 50th (B), and 90th (C) percentile of the baseline (85, 95, 125). D, Contour plot model describing the impact of baseline glycemia (y axis), on-treatment glycemia (x axis), and predicted patient PFS (z axis, corresponding to the color scale).

Close modal

Because the log relative hazard metric does not have immediate clinical translation, we used a contour plot to illustrate the predicted 1-year PFS as a joint effect of baseline and on-treatment blood glucose concentration, while keeping the remaining factors at their average level. As shown in Fig. 2D, most points—each point representing an individual patient—lie in a wide yellow area of the plot, which corresponds to approximately 30% 1-year PFS probability (i.e., the average PFS in the whole patient population). Of note, point patients with lower baseline glycemia and undergoing an increase of their glycemia during the everolimus-exemestane treatment, which correspond to the red area in the bottom right corner of the plot (roughly delimited by the 25% level curve), were associated with the lowest PFS, while point patients with higher baseline glycemia and lower on-treatment glycemia (top left corner) corresponded to the best PFS.

To illustrate the impact of the interaction between baseline and on-treatment glycemia in a more intuitive way, we compared PFS Kaplan–Meier curves of patients who were normoglycemic at baseline (<100 mg/dL) and became diabetic (≥126 mg/dL) during everolimus-exemestane therapy with PFS Kaplan–Meier curves of other patient subsets. Patients with normal baseline blood glucose levels who became diabetic during the treatment (Group A) had significantly worse PFS when compared with the remaining patients (Group B; mPFS, 6.34 vs. 7.33 months; unadjusted HR, 1.42; 95% CI, 1.01–1.99; P = 0.040; Fig. 3A). Also within these two different cohorts, metformin use was not associated with significantly different PFS (Supplementary Fig. S4B and S4C).

Figure 3.

A and B, Kaplan–Meier curves representing patient PFS according to baseline glycemia (normal vs. high) and on-treatment diabetic status (yes vs. no).

Figure 3.

A and B, Kaplan–Meier curves representing patient PFS according to baseline glycemia (normal vs. high) and on-treatment diabetic status (yes vs. no).

Close modal

Among patients who experienced early diabetes during everolimus-exemestane therapy, we also compared the PFS of patients with normal baseline glycemia (Group A) and patients who were already hyperglycemic at baseline (i.e., plasma glucose concentration in the 100–125 mg/dL range, Group B); interestingly, the former had significantly worse PFS when compared with the latter patients (mPFS, 6.34 vs. 10.32 months; unadjusted HR, 1.76; 95% CI, 1.15–2.69; P = 0.008; Fig. 3B).

The mTORC1 inhibitor everolimus in combination with exemestane is an effective treatment for patients with HR+/HER2 aBC progressing on/after prior NSAI therapy (8). Hyperglycemia/diabetes and hyperinsulinemia are common AEs in patients treated with everolimus or other PI3K/AKT/mTORC1 axis inhibitors (8–10), and could reduce the efficacy of these agents by reactivating the IR/PI3K/AKT/mTORC1 pathway (19). Here, we conducted a large, multicenter study, namely, EVERMET, to investigate the impact of baseline and on-treatment blood glucose concentration on PFS in patients with HR+/HER2 aBC treated with everolimus-exemestane.

We found that both baseline and on-treatment glycemia, when evaluated as continuous variables, are associated with patient PFS, and this association is mainly attributable to their interaction. In detail, patients with normal baseline glycemia who experienced hyperglycemia/diabetes during everolimus-exemestane treatment had significantly worse PFS when compared with the remaining patients, and in particular when compared with patients who were already hyperglycemic at baseline and experienced on-treatment hyperglycemia/diabetes. The robustness of the study results was confirmed by a parallel multivariable model in which we also included other important metabolic parameters that are modulated by everolimus-exemestane therapy, that is, triglycerides and cholesterol, as well as by a landmark analysis that excluded patients undergoing disease progression during the first 3 months of everolimus-exemestane treatment.

Among variables that were consistently associated with better patient PFS in multivariable models was the precocious interruption or dose reduction of everolimus, which were both associated with an increased incidence of treatment-induced adverse events, as expected. To explain this association, we hypothesized that patients undergoing everolimus interruption/dose reduction had been exposed to longer duration of everolimus treatment which, in turn, might have conditioned interruption/dose reduction on the one hand, and longer clinical benefit on the other hand. To test this hypothesis, we compared the duration of everolimus therapy in patients undergoing/not undergoing everolimus interruption or dose reduction. Of note, everolimus treatment exposure was significantly longer in patients undergoing everolimus dose reduction when compared with patients continuing everolimus at full dosage until disease progression, while it was significantly lower in patients who precociously interrupted everolimus therapy. On the basis of results of these analyses, we conclude that everolimus dose reduction might have contributed to longer drug exposure which, in turn, might have resulted in higher clinical benefit from everolimus. On the other hand, the observed PFS prolongation in patients undergoing precocious everolimus interruption could reflect higher systemic and intratumor exposure to the drug during the first months of treatment, which could justify an increased incidence of treatment-related adverse events on the one hand, and higher treatment efficacy and longer PFS on the other hand.

As per clinical protocol, we initially evaluated the potential impact of baseline or on-treatment hyperglycemia, as defined as blood fasting glucose concentration ≥126 mg/dL, on patient PFS. In the primary analysis, we did not find a statistically significant association between hyperglycemia and the risk of disease progression. When interpreting these results in the light of the final study findings, we should consider that: (i) in the primary analysis, we only evaluated the effect of metabolic variables at one timepoint (baseline or on-treatment glycemia), while we did not take into account the impact of their interaction on PFS, and (ii) both baseline and on-treatment glycemia are continuous variables, while in the primary analysis we evaluated them as dichotomous. In clinical studies, dichotomizing continuous variables is a common tool that is used to identify parameter thresholds that can be used to allocate patients in different classes of risk, thus favoring decision processes by physicians. However, dichotomization of continuous variables can be misleading for several reasons: (i) commonly used thresholds may not be appropriate for the specific clinical context (for instance, the 126 mg/dL threshold, which is used for the diagnosis of diabetes mellitus, might fail to distinguish between patients with cancer more or less likely to benefit from a specific antitumor therapy, and (ii) even if appropriate thresholds are found for specific clinical contexts, dichotomization may be misleading in the case of nonmonotonic or nonlinear relationships between metabolite concentration and clinical outcomes, as was the case of the association between blood glucose levels and patient PFS in our study. For these reasons, the impact of metabolic factors on clinical outcomes could be more reliably assessed when these variables are evaluated as continuous rather than dichotomous variables, and by using interactive, longitudinal models.

To explain the interaction between baseline and on-treatment glycemia in affecting patient PFS, we hypothesize that higher baseline blood glucose and insulin levels could be associated with higher baseline activation of the PI3K/AKT/mTORC1 axis in cancer cells and, potentially, with higher tumor cell sensitivity to everolimus-induced inhibition of mTORC1 regardless of on-treatment glycemia/insulinemia. On the other hand, tumors arising in patients with normal baseline glycemia/insulinemia might display lower baseline activation of the PI3K/AKT/mTORC1 axis; in conditions of normal extracellular blood glucose/insulin concentration, these tumors could maintain some sensitivity to everolimus-exemestane, while the occurrence of precocious everolimus-induced hyperglycemia and hyperinsulinemia could result in a boost of PI3K/AKT/mTORC1 activation, and in cancer cell resistance to the treatment. While this hypothesis needs to be confirmed by preclinical and prospective clinical studies, our findings indicate that blood glucose and, potentially, insulin concentration does not affect HR+/HER2 breast cancer cell response to pharmacologic mTORC1 inhibition per se, but their effect could be strongly influenced by the metabolic environment in which the tumor grew before the treatment, and in particular by baseline blood glucose/insulin concentration.

If confirmed by future prospective studies, our findings could have relevant clinical implications. Indeed, in the subgroup of patients with normal baseline glycemia, preventing or promptly reversing everolimus-induced hyperglycemia or diabetes could improve everolimus-exemestane efficacy. To this aim, specific dietary and/or pharmacologic interventions capable of preventing everolimus-induced hyperglycemia/diabetes should be considered in patients with HR+/HER2 aBC treated with everolimus-exemestane, especially if they are normoglycemic at baseline. Regarding dietary approaches, a low intake of refined carbohydrates and sugars could be recommended to patients initiating everolimus-exemestane treatment. As for pharmacologic approaches, metformin or other antidiabetic medications should be promptly initiated if dietary interventions are insufficient to keep blood glycemia below the diabetic threshold during the first months of treatment. Of note, because everolimus-induced hyperglycemia tends to spontaneously resolve during the course of the treatment (23), blood glucose levels should be more intensively monitored to prevent or to promptly manage everolimus-induced hyperglycemia/diabetes during the first 3 months of therapy, when a nonirrelevant proportion of disease progression events occur (15.1% of patients in the EVERMET study). At the same time, our results indicate that patients who are hyperglycemic at the time of everolimus-exemestane initiation could achieve poor, if any benefit from blood glucose reduction during everolimus-exemestane treatment; in these patients, a tight control of patient glycemia and, in case, the reversal of everolimus-induced diabetes could be potentially less impactful on tumor-related outcomes, while antidiabetic treatments should be primarily used to prevent diabetes-induced symptoms and complications.

Because hyperglycemia and hyperinsulinemia are class effects of PI3K/AKT/mTORC1 axis inhibitors, results of our study could also apply to other clinical contexts in which these compounds are used. For instance, the PI3K inhibitor alpelisib has been recently approved by the FDA and European Medicines Agency in combination with fulvestrant for the treatment of postmenopausal women and men with HR+/HER2 aBC progressing on/after prior AI therapy (9). Similar to everolimus, alpelisib can cause hyperglycemia and hyperinsulinemia, which could reduce its efficacy (19). Although the widespread use of alpelisib in the daily treatment of patients with HR+/HER2 aBC bearing PIK3CA-mutated tumors might be limited by several factors, including the lack of an extensive tumor genomic profiling in several cancer centers, the suboptimal safety profile of alpelisib, and recent labels limiting alpelisib use in Europe and Italy to patients previously treated with single-agent ET (which has now been replaced by ET plus CDK 4/6 inhibitor-based combinations as a standard-of-care first-line therapy), the alpelisib-fulvestrant combination remains a potentially useful therapy that could be used in up to 40% of all patients with HR+/HER2 aBC. Therefore, because the incidence of severe (grade 3 or 4) hyperglycemia is common with alpelisib (actually more common than with everolimus) despite the precocious use of metformin in the SOLAR-1 study (24), exploring strategies to prevent or promptly manage alpelisib-induced hyperglycemia/diabetes is a clinically relevant issue, especially for patients with normal baseline blood glucose levels.

In recent years, metformin has been extensively investigated in both preclinical and clinical setting for its potential direct (cell autonomous) or indirect (through its impact on systemic metabolism) antitumor effects (25–27). Because metformin acts by reducing glucose production in the liver and at the same time by sensitizing peripheral tissues to the effects of insulin, it has been considered a good candidate drug to be combined with everolimus-exemestane for the treatment of patients with HR+/HER2 aBC. Quite disappointingly, one recent prospective study showed modest clinical efficacy of upfront everolimus-exemestane plus metformin combination in overweight/obese postmenopausal women with HR+/HER2 aBC (27), and similarly negative results emerged from a preclinical study in which metformin was used in combination with PI3K inhibitors in mouse models of HR+/HER2 breast cancer (19). In line with these data, in our study we did not find a significant association between metformin use and PFS in patients with HR+/HER2 aBC treated with everolimus-exemestane. This evidence, together with the potential pharmacokinetic interactions between everolimus and metformin in patients with advanced cancers (28) and the risk of increasing the incidence of diarrhea, indicate that metformin might be not an ideal drug to be used in combination with everolimus.

Conversely, specific dietary interventions, such as ketogenic diets or cyclic calorie-restricted, low-carbohydrate, low-protein diets, collectively referred to as fasting-mimicking diets (FMD), which reduce blood glucose/insulin concentrations and do not have overlapping toxicities with everolimus, have been shown to inhibit the PI3K/AKT/mTORC1 pathway synergistically with ETs or PI3K inhibitors in preclinical in vivo experiments (29, 30). In one study, high-fat ketogenic diets were found to be more effective than metformin in reducing PI3K inhibitor–induced hyperglycemia and hyperinsulinemia, and demonstrated additive or synergistic in vitro and in vivo antitumor activity in combination with PI3K inhibitors (19). More recently, cyclic FMDs showed synergistic antitumor activity with standard ETs plus/minus CDK 4/6 inhibitors in preclinical models of HR+/HER2 breast cancer, with initial promising results also in patients with cancer (29). Of note, the synergistic activity between ET and FMD was mediated by FMD-induced reduction of blood insulin/insulin-like growth factor-1 levels, which results in increased PTEN expression and consequent inhibition of the PI3K/AKT/mTORC1 pathway in cancer cells. Because ketogenic diets and FMD are potentially safe and feasible interventions in well-selected cancer patient populations, combining them with everolimus or other inhibitors of the PI3K/AKT/mTORC1 pathway could produce highly synergistic antitumor effects, while at the same time improving the tolerability of these drugs.

The following are major strengths of this study: (i) this was the first, large multicenter study to show an interaction between baseline and on-treatment blood glucose concentration in affecting the PFS of patients with HR+/HER2 aBC treated with the everolimus-exemestane combination; (ii) the large sample size and the multicenter nature of the study make our data robust (in this respect, PFS data in the whole population of patients enrolled in the EVERMET study are consistent with data reported in the experimental arm of the BOLERO-2 trial and in previous real-world data studies; refs. 8, 31, 32); (iii) we enrolled a high number of patients receiving the same treatment in a relatively short time interval (5 years), thus excluding a significant role of relevant changes in clinical practice of HR+ breast cancer treatment; (iv) at least one blood glucose measurement at baseline and during the first 3 months of everolimus-exemestane therapy was available for the majority of patients; and (v) the main study findings were confirmed in different multivariable models and also by a landmark analysis.

The main limitation of this study consists in the retrospective design and the consequent missing data, which could in part limit the reliability of our findings; nonetheless, the main study findings were confirmed after removing patients with incomplete data from the analysis, thus adding robustness to our results. Moreover, the study was negative as for its primary endpoint, and the lack of a control arm does not establish definitive causal associations between metabolic toxicities and treatment efficacy.

In conclusion, patients with normal baseline blood glucose concentration are at higher risk for disease progression if they experience precocious hyperglycemia/diabetes during everolimus-exemestane treatment. Prospective clinical trials are needed to investigate the impact of dietary or pharmacologic interventions aimed at preventing or precociously reversing everolimus-induced increase of blood glucose concentration on the clinical outcomes of patients with HR+/HER2 aBC.

C. Vernieri reports research funding from Roche and an advisory role at Novartis. L. Moscetti reports personal fees from Novartis, Eli Lilly, and Roche during the conduct of the study. C. Molinelli reports personal fees from Novartis and Eli Lilly outside the submitted work. M. Palleschi reports personal fees from Novartis outside the submitted work. L. Gerratana reports personal fees from Lilly outside the submitted work. F. Puglisi reports personal fees from Amgen, Daiichi Sankyo, Eli Lilly, Ipsen, MSD, Novartis, Pierre-Fabre, Pfizer, and Seagen; grants and personal fees from AstraZeneca and Roche; and grants from Eisai outside the submitted work. N. La Verde reports personal fees from Novartis, Roche, p MSD, Celgene, GSK, and Gentili, as well as grants from Eisai outside the submitted work. G. Arpino reports grants and personal fees from Roche; grants, personal fees, and nonfinancial support from Novartis and Pfizer; personal fees and nonfinancial support from Lilly and Eisai; personal fees from AstraZeneca, Amgen, and MSD; and nonfinancial support from Ipsen outside the submitted work. A. Rocca reports personal fees from Novartis, Lilly, and Pfizer, as well as nonfinancial support from Roche outside the submitted work. D. Generali reports personal fees from Novartis, Lilly, and Istituto Gentili, as well as grants from Eisai outside the submitted work. F. Montemurro reports personal fees from Roche, Novartis, Pfizer, Eli Lilly, Pierre Fabre, and Seagen outside the submitted work. G. Curigliano reports personal fees from Roche, Pfizer, Novartis, Daiichi Sankyo, Merck, Lilly, Veracyte, Foundation One, AstraZeneca, Seagen, and Ellipsis outside the submitted work. L. Del Mastro reports personal fees from Eli Lilly, Novartis, MSD, Genomic Health, Pierre Fabre, Daiichi Sankyo, AstraZeneca, Seagen, Eisai, and Ipsen, as well as personal fees and nonfinancial support from Roche and Pfizer outside the submitted work. V. Guarneri reports personal fees from Eli Lilly, Novartis, Roche, and MSD outside the submitted work. F. de Braud reports personal fees from Roche, Pfizer, BMS, Merck, MSD, Servier, Sanofi, Nerviano Medical Sciences, EMD Serono, Novartis, Incyte, and Tesaro outside the submitted work. No disclosures were reported by the other authors.

C. Vernieri: Conceptualization, resources, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing. F. Nichetti: Conceptualization, resources, data curation, formal analysis, visualization, writing–original draft, project administration, writing–review and editing. L. Lalli: Formal analysis, visualization, methodology. L. Moscetti: Resources, investigation, writing–review and editing. C.A. Giorgi: Resources, data curation, investigation. G. Griguolo: Resources, investigation. A. Marra: Resources, data curation, investigation. G. Randon: Resources, data curation, investigation. C.G. Rea: Resources, data curation, investigation. F. Ligorio: Writing–review and editing. S. Scagnoli: Data curation, investigation, project administration. C. De Angelis: Resources, investigation. C. Molinelli: Resources, data curation, investigation. A. Fabbri: Resources, investigation. E. Ferraro: Data curation, investigation. D. Trapani: Resources, data curation, writing–review and editing. A. Milani: Resources, investigation. E. Agostinetto: Resources, data curation, investigation, writing–review and editing. O. Bernocchi: Data curation, investigation. G. Catania: Resources, investigation. A. Vantaggiato: Resources, investigation. M. Palleschi: Resources, investigation. A. Moretti: Resources, investigation. D. Basile: Resources, investigation. M. Cinausero: Resources, investigation. A. Ajazi: Resources, investigation. L. Castagnoli: Resources, investigation. S. Lo Vullo: Formal analysis, methodology, writing–review and editing. L. Gerratana: Resources, investigation. F. Puglisi: Resources, investigation, writing–review and editing. N. La Verde: Resources, investigation, writing–review and editing. G. Arpino: Resources, writing–review and editing. A. Rocca: Resources, investigation. M. Ciccarese: Resources, investigation. R. Pedersini: Resources, investigation. A. Fabi: Resources, investigation. D. Generali: Resources, investigation, writing–review and editing. A. Losurdo: Resources, investigation. F. Montemurro: Resources, investigation. G. Curigliano: Resources, investigation, writing–review and editing. L. Del Mastro: Resources, methodology, project administration. A. Michelotti: Resources, writing–review and editing. E. Cortesi: Resources, writing–review and editing. V. Guarneri: Resources, data curation, investigation. G. Pruneri: Resources, visualization. L. Mariani: Data curation, formal analysis, supervision, validation, visualization, methodology, writing–review and editing. F. de Braud: Conceptualization, supervision, writing–review and editing.

We would like to thank the “Associazione Italiana per la Ricerca sul Cancro” (AIRC; MFAG 2019-22977; to principal investigator, C. Vernieri) and the Scientific Directorate of Fondazione IRCCS Istituto Nazionale dei Tumori for funding our research. We would also thank Monica Milano and Pietro Indelicato for useful suggestions in study design and data collection.

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.

1.
Cancer Genome Atlas Network. 
Comprehensive molecular portraits of human breast tumours
.
Nature
2012
;
490
:
61
70
.
2.
Goncalves
MD
,
Hopkins
BD
,
Cantley
LC
. 
Phosphatidylinositol 3-kinase, growth disorders, and cancer
.
N Engl J Med
2018
;
379
:
2052
62
.
3.
Kandoth
C
,
McLellan
MD
,
Vandin
F
,
Ye
K
,
Niu
B
,
Lu
C
, et al
Mutational landscape and significance across 12 major cancer types
.
Nature
2013
;
502
:
333
9
.
4.
Millis
SZ
,
Ikeda
S
,
Reddy
S
,
Gatalica
Z
,
Kurzrock
R
. 
Landscape of phosphatidylinositol-3-kinase pathway alterations across 19784 diverse solid tumors
.
JAMA Oncol
2016
;
2
:
1565
73
.
5.
Schiff
R
,
Massarweh
SA
,
Shou
J
,
Bharwani
L
,
Mohsin
SK
,
Osborne
CK
. 
Cross-talk between estrogen receptor and growth factor pathways as a molecular target for overcoming endocrine resistance
.
Clin Cancer Res
2004
;
10
:
331S
6S
.
6.
Yamnik
RL
,
Digilova
A
,
Davis
DC
,
Brodt
ZN
,
Murphy
CJ
,
Holz
MK
. 
S6 kinase 1 regulates estrogen receptor alpha in control of breast cancer cell proliferation
.
J Biol Chem
2009
;
284
:
6361
9
.
7.
Yamnik
RL
,
Holz
MK
. 
mTOR/S6K1 and MAPK/RSK signaling pathways coordinately regulate estrogen receptor alpha serine 167 phosphorylation
.
FEBS Lett
2010
;
584
:
124
8
.
8.
Baselga
J
,
Campone
M
,
Piccart
M
,
Burris
HA
 3rd
,
Rugo
HS
,
Sahmoud
T
, et al
Everolimus in postmenopausal hormone-receptor-positive advanced breast cancer
.
N Engl J Med
2012
;
366
:
520
9
.
9.
Andre
F
,
Ciruelos
E
,
Rubovszky
G
,
Campone
M
,
Loibl
S
,
Rugo
HS
, et al
Alpelisib for PIK3CA-mutated, hormone receptor-positive advanced breast cancer
.
N Engl J Med
2019
;
380
:
1929
40
.
10.
Paplomata
E
,
Zelnak
A
,
O'Regan
R
. 
Everolimus: side effect profile and management of toxicities in breast cancer
.
Breast Cancer Res Treat
2013
;
140
:
453
62
.
11.
Juric
D
,
Janku
F
,
Rodon
J
,
Burris
HA
,
Mayer
IA
,
Schuler
M
, et al
Alpelisib plus fulvestrant in PIK3CA-altered and PIK3CA-wild-type estrogen receptor-positive advanced breast cancer: a phase 1b clinical trial
.
JAMA Oncol
2019
;
5
:
e184475
.
12.
Ellard
SL
,
Clemons
M
,
Gelmon
KA
,
Norris
B
,
Kennecke
H
,
Chia
S
, et al
Randomized phase II study comparing two schedules of everolimus in patients with recurrent/metastatic breast cancer: NCIC Clinical Trials Group IND.163
.
J Clin Oncol
2009
;
27
:
4536
41
.
13.
Huffman
TA
,
Mothe-Satney
I
,
Lawrence
JC
 Jr
. 
Insulin-stimulated phosphorylation of lipin mediated by the mammalian target of rapamycin
.
Proc Natl Acad Sci U S A
2002
;
99
:
1047
52
.
14.
Fraenkel
M
,
Ketzinel-Gilad
M
,
Ariav
Y
,
Pappo
O
,
Karaca
M
,
Castel
J
, et al
mTOR inhibition by rapamycin prevents beta-cell adaptation to hyperglycemia and exacerbates the metabolic state in type 2 diabetes
.
Diabetes
2008
;
57
:
945
57
.
15.
Chresta
CM
,
Davies
BR
,
Hickson
I
,
Harding
T
,
Cosulich
S
,
Critchlow
SE
, et al
AZD8055 is a potent, selective, and orally bioavailable ATP-competitive mammalian target of rapamycin kinase inhibitor with in vitro and in vivo antitumor activity
.
Cancer Res
2010
;
70
:
288
98
.
16.
Houde
VP
,
Brule
S
,
Festuccia
WT
,
Blanchard
PG
,
Bellmann
K
,
Deshaies
Y
, et al
Chronic rapamycin treatment causes glucose intolerance and hyperlipidemia by upregulating hepatic gluconeogenesis and impairing lipid deposition in adipose tissue
.
Diabetes
2010
;
59
:
1338
48
.
17.
Fruman
DA
,
Chiu
H
,
Hopkins
BD
,
Bagrodia
S
,
Cantley
LC
,
Abraham
RT
. 
The PI3K pathway in human disease
.
Cell
2017
;
170
:
605
35
.
18.
Pizzuti
L
,
Marchetti
P
,
Natoli
C
,
Gamucci
T
,
Santini
D
,
Scinto
AF
, et al
Fasting glucose and body mass index as predictors of activity in breast cancer patients treated with everolimus-exemestane: the EverExt study
.
Sci Rep
2017
;
7
:
10597
.
19.
Hopkins
BD
,
Pauli
C
,
Du
X
,
Wang
DG
,
Li
X
,
Wu
D
, et al
Suppression of insulin feedback enhances the efficacy of PI3K inhibitors
.
Nature
2018
;
560
:
499
503
.
20.
Schemper
M
,
Smith
TL
. 
A note on quantifying follow-up in studies of failure time
.
Control Clin Trials
1996
;
17
:
343
6
.
21.
Ishwaran
H
,
Kogalur
UB
,
Blackstone
EH
,
Lauer
MS
. 
Random survival forests
.
Ann Appl Stat
2008
;
2
:
841
60
.
22.
Stekhoven
DJ
,
Buhlmann
P
. 
MissForest–non-parametric missing value imputation for mixed-type data
.
Bioinformatics
2012
;
28
:
112
8
.
23.
Buszewska-Forajta
M
,
Rachon
D
,
Stefaniak
A
,
Wawrzyniak
R
,
Konieczna
A
,
Kowalewska
A
, et al
Identification of the metabolic fingerprints in women with polycystic ovary syndrome using the multiplatform metabolomics technique
.
J Steroid Biochem Mol Biol
2019
;
186
:
176
84
.
24.
Vernieri
C
,
Corti
F
,
Nichetti
F
,
Ligorio
F
,
Manglaviti
S
,
Zattarin
E
, et al
Everolimus versus alpelisib in advanced hormone receptor-positive HER2-negative breast cancer: targeting different nodes of the PI3K/AKT/mTORC1 pathway with different clinical implications
.
Breast Cancer Res
2020
;
22
:
33
.
25.
Elgendy
M
,
Cirò
M
,
Hosseini
A
,
Weiszmann
J
,
Mazzarella
L
,
Ferrari
E
, et al
Combination of hypoglycemia and metformin impairs tumor metabolic plasticity and growth by modulating the PP2A-GSK3beta-MCL-1 axis
.
Cancer Cell
2019
;
35
:
798
815
.
26.
Arrieta
O
,
Barron
F
,
Padilla
MS
,
Aviles-Salas
A
,
Ramirez-Tirado
LA
,
Arguelles Jimenez
MJ
, et al
Effect of metformin plus tyrosine kinase inhibitors compared with tyrosine kinase inhibitors alone in patients with epidermal growth factor receptor-mutated lung adenocarcinoma: a phase 2 randomized clinical trial
.
JAMA Oncol
2019
;
5
:
e192553
.
27.
Yam
C
,
Esteva
FJ
,
Patel
MM
,
Raghavendra
AS
,
Ueno
NT
,
Moulder
SL
, et al
Efficacy and safety of the combination of metformin, everolimus and exemestane in overweight and obese postmenopausal patients with metastatic, hormone receptor-positive, HER2-negative breast cancer: a phase II study
.
Invest New Drugs
2019
;
37
:
345
51
.
28.
Molenaar
RJ
,
van de Venne
T
,
Weterman
MJ
,
Mathot
RA
,
Klumpen
HJ
,
Richel
DJ
, et al
A phase Ib study of everolimus combined with metformin for patients with advanced cancer
.
Invest New Drugs
2018
;
36
:
53
61
.
29.
Caffa
I
,
Spagnolo
V
,
Vernieri
C
,
Valdemarin
F
,
Becherini
P
,
Wei
M
, et al
Fasting-mimicking diet and hormone therapy induce breast cancer regression
.
Nature
2020
;
583
:
620
4
.
30.
Vernieri
C
,
Nichetti
F
,
Raimondi
A
,
Pusceddu
S
,
Platania
M
,
Berrino
F
, et al
Diet and supplements in cancer prevention and treatment: clinical evidences and future perspectives
.
Crit Rev Oncol Hematol
2018
;
123
:
57
73
.
31.
Cazzaniga
ME
,
Airoldi
M
,
Arcangeli
V
,
Artale
S
,
Atzori
F
,
Ballerio
A
, et al
Efficacy and safety of everolimus and exemestane in hormone-receptor positive (HR+) human-epidermal-growth-factor negative (HER2-) advanced breast cancer patients: new insights beyond clinical trials. The EVA study
.
Breast
2017
;
35
:
115
21
.
32.
Jerusalem
G
,
Mariani
G
,
Ciruelos
EM
,
Martin
M
,
Tjan-Heijnen
VC
,
Neven
P
, et al
Safety of everolimus plus exemestane in patients with hormone-receptor-positive, HER2-negative locally advanced or metastatic breast cancer progressing on prior non-steroidal aromatase inhibitors: primary results of a phase IIIb, open-label, single-arm, expanded-access multicenter trial (BALLET)
.
Ann Oncol
2016
;
27
:
1719
25
.