Single Dose of Phosphatidylinositol 3-Kinase Inhibitor Alpelisib Induces Insulin Resistance in Healthy Adults: A Randomized Feasibility Study

Studying the mechanistic contribution of insulin resistance (IR) to type 2 diabetes complications has been challenging because of its indolent progression and multifactorial nature and the paucity of safe and effective methods for examining it in the experimental setting (1). Investigators have sought to circumvent the problem of chronicity by acutely inducing IR, including via infusion of fatty acid emulsions (2), controlled hyperalimentation (3), or medications such as corticosteroids (4). Although these strategies have yielded valuable insights, none have evaded all the pitfalls listed above, and a dearth of uniform protocols for their application limits the generalizability and reproducibility of the associated findings.

We sought to test a novel approach to modeling IR in humans: direct blockade of the insulin signaling pathway at the level of phosphatidylinositol 3-kinase (PI3K) using alpelisib, an orally administered small-molecule inhibitor of the PI3K p110α catalytic subunit that is U.S. Food and Drug Administration (FDA) approved to treat certain breast cancers (5). The mechanism of action of alpelisib is specific, its short duration of action allows for tight temporal control of the intervention, and its fixed-dose format allows for uniform application across studies (6). We conducted a double-blind placebo-controlled randomized feasibility trial to test the hypothesis that a single dose of alpelisib would be sufficient to produce biochemical evidence of IR (i.e., a rise in serum insulin with or without an accompanying rise in plasma glucose) in a diverse group of healthy volunteers. The reliability and tolerability of single-dose alpelisib to provoke IR would determine its feasibility as a strategy for the mechanistic study of IR in humans.

This double-blind randomized placebo-controlled single-site feasibility trial was conducted in accordance with Good Clinical Practice and approved by the Columbia University Irving Medical Center (CUIMC) Institutional Review Board. Participants were recruited and data collected between May and December 2023. Inclusion criteria were healthy adults age 18–65 years who had BMI of 18.0–26.9 kg/m² and could provide informed consent. Healthy was defined as no evidence of active or chronic disease arising from comprehensive medical history, physical examination, electrocardiogram, and standard blood tests. Exclusion criteria included documented weight change ≥3% in the previous 6 months, fasting plasma glucose ≥100 mg/dL or hemoglobin A_1c ≥5.7%, excessive use of alcohol, and active use of tobacco or illicit substances. Women of reproductive potential were required to use highly effective contraception and screened for pregnancy before inclusion. Participants were required to abstain from alcohol and strenuous exercise from 48 h before the screening visit through 24 h after the investigational agent dose.

Participants and investigators were blinded to study group assignment. Random assignment to parallel placebo and alpelisib groups was performed by an independent biostatistician using a computer-based random number generator; the allocation ratio was 1:1 without blocking. Based on insulin levels in a healthy population (7), we estimated that accruing five participants per group would allow detection of a doubling of fasting serum insulin at α = 0.05 and β = 0.20.

We used the standard 300-mg cancer treatment dose of alpelisib (Novartis, Basel, Switzerland). The CUIMC Research Pharmacy dispensed doses of alpelisib and placebo (microcrystalline cellulose) identically overencapsulated in Capsugel (Lonza, Morristown, NJ) to mask treatment allocation. An oral glucose tolerance test (OGTT) was performed using NERL Trutol (Thermo Fisher Scientific, Waltham, MA), containing 75 g d-glucose dissolved in 10 oz water. Meal replacement was provided as BOOST Plus (Nestlé Health Sciences, Vevey, Switzerland), comprising 50% carbohydrate, 35% fat, and 15% protein. The caloric content of each participant’s meal replacements was calculated based on basal energy needs, as estimated by the modified Harris-Benedict equation (8) without physical activity correction.

The study design, illustrated in Fig. 1A, included two visits over 4 days. Participants first attended a two-part screening visit to determine eligibility. On the day before the main study visit, participants consumed a standardized diet consisting only of BOOST Plus divided into three portions to be consumed at 08:00, 13:00, and 18:00. They were instructed to take the investigational agent dose (i.e., alpelisib or placebo) at home at 23:00 along with two saltine crackers to maximize absorption (9). The following morning, 3 days after screening, participants had blood drawn for the prespecified primary end points of the study: fasting plasma glucose and serum insulin/C-peptide. Next, participants underwent a 75-g 3-h OGTT with collection of blood samples for glucose and insulin/C-peptide, the secondary end points of the study, at 15, 30, 60, 90, and 120 min.

Plasma glucose was measured by automated analyzer by the NewYork-Presbyterian Hospital/CUIMC Clinical Core Laboratory. Serum insulin and C-peptide were measured by Associated Research and University Pathologists using a quantitative chemiluminescent assay. Plasma alpelisib levels were measured via gas chromatography–mass spectrometry by the CUIMC Biomarkers Core Laboratory following method development using pure alpelisib (MedChem Express, Monmouth Junction, NJ).

All data were analyzed by intention to treat and are presented as mean ± SD. The 4-point Matsuda index was calculated using a publicly available online tool (https://mmatsuda.diabetes-smc.jp/4points.html). Incremental area under the curve (AUC) from OGTT was calculated by the trapezoid method above baseline. Alpelisib versus placebo comparisons were performed by two-sample independent t test with Holm-Šídák correction for multiple comparisons (GraphPad Prism 10). Multiple comparisons testing was performed separately for primary and then for secondary end points; both raw and adjusted P values (P_adj) are presented in figures. Because of an i.v. catheter malfunction, data are missing for the 90-min OGTT time point for one placebo-treated participant. AUC calculation elided the missing time point for the affected participant (i.e., trapezoid spans 60–120 min rather than 60–90/90–120 min); reanalysis of the data entirely excluding this participant did not materially affect the outcome (data not shown).

The data generated during the current study are available from the corresponding author on reasonable request. No applicable resources were generated or analyzed during the current study.

A total of 23 potential participants enrolled in the study to undergo screening, of whom 13 were eligible for participation and 11 completed the study (Fig. 1B). Baseline characteristics were evenly distributed between the groups (Table 1). Alpelisib levels were checked in fasting plasma ∼10 h after ingestion of the dose (Fig. 1C). The drug was undetectable (<0.1 ng/mL) in the plasma of participants treated with placebo; the mean plasma level in the alpelisib-treated group was 701.22 ± 424.04 ng/mL, but alpelisib was detectable in the plasma of only four of the five participants in the group. (Although protocol nonadherence is the most likely explanation, quick metabolism or poor intestinal absorption is a formal possibility.) All participants’ data were included in our intention-to-treat statistical analysis.

There was a nonsignificant trend toward increased fasting glucose in those receiving alpelisib (plasma glucose 84 ± 5 mg/dL after placebo vs. 93 ± 11 mg/dL after alpelisib; P_adj = 0.10) (Fig. 2A). By contrast, serum insulin levels were significantly higher with alpelisib than with placebo (23 ± 12 vs. 5 ± 3 μU/mL, respectively; P_adj = 0.02) (Fig. 2B), as were serum levels of C-peptide (3.5 ± 1.5 vs. 1.1 ± 0.5 ng/mL; P_adj = 0.02) (Fig. 2C). These metrics translate into HOMA-IR scores of 5.4 ± 3.1 after alpelisib versus 1.1 ± 0.6 after placebo (P_adj = 0.02) (Fig. 2D). Note that the alpelisib arm data point shown as a square (rather than circle) depicts data from the participant whose drug level was undetectable. By exploratory post hoc analysis, exclusion of the data from this participant would increase mean fasting plasma glucose from 93 ± 11 (as shown above) to 96 ± 10 mg/dL (P_adj = 0.03 vs. placebo), serum insulin from 23 ± 12 to 27 ± 4 μU/mL (P_adj <0.01), and serum C-peptide from 3.5 ± 1.5 to 4.1 ± 0.5 ng/mL (P_adj <0.01).

During OGTT, AUC for plasma glucose was 9 ± 4 g/dL × min after alpelisib and 7 ± 2 g/dL × min after placebo (P_adj = 0.36) (Fig. 3A and B). Similarly to the fasting state, insulin levels ran higher during OGTT after alpelisib than placebo, with incremental AUC values of 22 ± 15 and 7 ± 2 mU/mL × min, respectively (P_adj = 0.05) (Fig. 3C and D). C-peptide followed suit, with incremental AUC values of 1.2 ± 0.4 μg/mL × min for alpelisib and 0.6 ± 0.2 μg/mL × min for placebo (P_adj = 0.02) (Fig. 3E and F). The Matsuda index value was 2.8 ± 1.5 for alpelisib and 9.3 ± 1.6 for placebo (P_adj = 0.05) (Fig. 3G). Again, the square-denoted point from the alpelisib group participant whose drug level was undetectable fell obviously out of line with the other four on every measure. Exploratory post hoc exclusion of this participant’s data raised the mean incremental AUC for plasma glucose, insulin, and C-peptide to 11 ± 2 g/dL × min (P_adj <0.02 vs. placebo), 26 ± 14 mU/mL × min (P_adj = 0.01), and 1.3 ± 0.3 μg/mL × min (P_adj <0.01), respectively.

Alpelisib treatment had no effect on pulse or blood pressure (data not shown). No serious adverse events occurred, and all adverse events reported were grade 1 and self-limited. Adverse events occurred in two of six participants receiving placebo; hunger, headache, and insomnia were each reported by one of six. In the alpelisib arm, four of five participants reported adverse events: hunger in two of five, and nausea, nocturia, and thirst each in one of five. The report of nocturia and thirst was made by the participant whose plasma alpelisib level was undetectable. No diarrhea or rash were reported.

Alpelisib induces hyperinsulinemic hyperglycemia in a majority of patients with baseline euglycemia after 1–2 weeks of daily use (5,10), but we are not aware of published clinical trial data on the metabolic impact of shorter treatment durations. We therefore sought to determine if a single dose of the PI3K inhibitor alpelisib is sufficient to induce IR acutely in healthy volunteers; its clear effect and good tolerability seem to validate its feasibility as a tool for the mechanistic study of the downstream effects of IR. Although there was no significant difference in fasting plasma glucose or glucose tolerance, insulin levels rose markedly in response to alpelisib, during both fasting and OGTT, strongly suggesting induction of IR (11). So clear was the insulin-desensitizing effect of alpelisib that it remained statistically significant despite one participant with euinsulinemia in the alpelisib arm whose plasma drug level was undetectable. The durability of our findings by intention-to-treat analysis reinforces the real-world utility of the approach, including in an outpatient setting. Our confidence in a direct insulin-secretagogue effect of alpelisib in the pancreas is much lower, particularly because PI3K inhibition in animal models tends to decrease insulin secretion (12). We intend to study alpelisib treatment under pancreatic clamp conditions to definitively settle this question in humans.

Alpelisib offers several advantages beyond its excellent reliability and tolerability: it has a single clearly defined mechanism of action, is orally active, and is FDA approved and commercially available (13). However, use of alpelisib for acute modeling of IR also has drawbacks. First, because of the potential toxicities (10) and high cost of the drug, we tested the effect of only a single dose. The metabolic impact of alpelisib is well known from the oncology literature to be short lived; its half-life is 8–9 h, it does not accumulate, and its hyperglycemic effect dissipates entirely over 24–72 h (6,14). Therefore, a single dose would not allow study of more indolent processes, such as bile acid turnover (15) or adipose tissue inflammation (16). Second, we do not yet know whether or to what extent the insulin-desensitizing effect of alpelisib manifests differentially across tissue in humans; the alpelisib-treated state may reflect the interplay of variegated disruptions in insulin signaling. Third, the pure IR induced by alpelisib is likely to resemble that in individuals with genetic loss of PI3K or insulin receptor function (i.e., glucose intolerance without hepatic triglyceride dysmetabolism) rather than the hepatic steatosis–associated selective IR more commonly seen in metabolic syndrome (17–19). This tool may therefore best be used for investigating research questions involving wholesale insulin-signaling dysfunction, akin to the use of insulin receptor–knockout mice (20). In this case, alpelisib may offer additional insights when used in strategic combination with other interventions. For instance, we are currently conducting an expanded follow-on study of the impact of alpelisib on glucose and lipid kinetics in both healthy volunteers and individuals with typical IR (ClinicalTrials.gov identifier NCT06354088). On the whole, our data suggest that single-dose alpelisib is a reliable and well-tolerated means of inducing acute IR in healthy volunteers and is likely to be a useful resource in human-based mechanistic studies of IR.

Acknowledgments. The authors thank Dr. Rajasekhar Ramakrishan for providing the study’s randomization scheme, Dr. Judith Korner for serving as the study’s independent medical safety monitor, Dr. Renu Nandakumar and the Biomarkers Core Laboratory team for developing the mass spectrometric alpelisib assay, Sonia Dobson for performing outpatient phlebotomy and electrocardiograms, and the inpatient care team of the Irving Institute for Clinical and Translational Medicine for nursing services during OGTT. All are affiliated with CUIMC.

Funding. This research study was supported by the Columbia University Department of Medicine; the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health (NIH) (P30DK063608, K12DK133995 to J.R.C. outside the submitted work); and the National Center for Advancing Translational Sciences, NIH (UL1TR001873).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Duality of Interest. J.W. serves as an advisor to GlaxoSmithKline and Galectin Therapeutics. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.R.C. conceived, designed, and led the study; analyzed the data; and wrote the manuscript. N.B. and Z.D.S. executed study procedures and assisted with data analysis and manuscript preparation. J.W., H.N.G., and B.L. provided expert consultation on the design, execution, and data analysis of the study and critically edited the manuscript. J.R.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.