Insulin feedback is a targetable resistance mechanism of PI3K inhibition in glioblastoma

Abstract

Background: Insulin feedback is a critical mechanism responsible for the poor clinical efficacy of phosphatidylinositol 3-kinase (PI3K) inhibition in cancer, and hyperglycemia is an independent factor associated with poor prognosis in glioblastoma (GBM). We investigated combination anti-hyperglycemic therapy in a mouse model of GBM and evaluated the association of glycemic control in clinical trial data from patients with GBM.

Methods: The effect of the anti-hyperglycemic regimens, metformin and the ketogenic diet, was evaluated in combination with PI3K inhibition in patient-derived GBM cells and in an orthotopic GBM mouse model. Insulin feedback and the immune microenvironment were retrospectively evaluated in blood and tumor tissue from a Phase 2 clinical trial of buparlisib in patients with recurrent GBM.

Results: We found that PI3K inhibition induces hyperglycemia and hyperinsulinemia in mice and that combining metformin with PI3K inhibition improves the treatment efficacy in an orthotopic GBM xenograft model. Through examination of clinical trial data, we found that hyperglycemia was an independent factor associated with poor progression-free survival in patients with GBM. We also found that PI3K inhibition increased insulin receptor activation and T-cell and microglia abundance in tumor tissue from these patients.

Conclusion: Reducing insulin feedback improves the efficacy of PI3K inhibition in GBM in mice, and hyperglycemia worsens progression-free survival in patients with GBM treated with PI3K inhibition. These findings indicate that hyperglycemia is a critical resistance mechanism associated with PI3K inhibition in GBM and that anti-hyperglycemic therapy may enhance PI3K inhibitor efficacy in GBM patients.

Keywords: glioblastoma; hyperglycemia; insulin; metformin; phosphatidylinositol 3-kinase.

Graphical Abstract

Figure 1.

PI3K inhibitor-driven insulin feedback is reduced by anti-hyperglycemic therapy in mice. Serum glucose levels were measured in mice treated with BKM-120 (60 mg/kg) (A) or GDC-0084 (15 mg/kg) (B) along with metformin (200 mg/kg) and a regular or ketogenic diet. P values by 1-way ANOVA for vehicle v. BKM-120 (<.001), metformin (ns), BKM-120 + metformin (<.01), metformin + ketogenic diet (<.05), ketogenic diet (ns), BKM-120 + ketogenic diet (ns), and BKM-120 + ketogenic diet + metformin (ns). P values by 1-way ANOVA for vehicle v. GDC-0084 (ns), metformin (<.05), metformin + ketogenic diet (<.0001), GDC-0084 + ketogenic diet (<.0001), and GDC-0084 + ketogenic diet + metformin (<.0001). Insulin (C) and C-peptide (D) levels were measured in mice treated with BKM-120 and GDC-0084 as in A and B. Mice fed a regular or ketogenic diet were treated with BKM-120 (15 mg/kg) followed 30 minutes later by ¹⁸F-fluorodeoxyglucose, and PET-MR imaging was then performed (E) and quantified in F. **P < .01, ****P < .0001.

Figure 2.

Insulin rescues PI3K inhibitor-induced growth suppression and signaling, 667 cells were treated with 0.5 μM BKM-120 (A) or 0.4 μM GDC-0084 (B) along with 10 ng/µL insulin, and cell proliferation was measured by Cell Titer Glo assay. (C) 667 cells were treated with 0.25 μM BKM-120 along with 20 μg/mL insulin, and PI3K-mTOR signaling was analyzed by western blot. (D) The glioma cells 308, 810, and 1228 were treated with 0.25 μM GDC-0084 along with 10 μg/mL insulin, and PI3K-mTOR signaling was analyzed by western blot. **P < .01, ***P < .001, ****P < .0001.

Figure 3.

Metformin improves the efficacy of PI3K inhibition in an orthotopic GBM mouse model, (A) Mice bearing orthotopic 667 GBM xenografts were treated with GDC-0084 (15 mg/kg) and/or metformin (200 mg/kg), and overall survival was measured from the time of treatment initiation. (B) Mice underwent MR brain imaging at 8–10 days after treatment initiation, with results quantified in C. (D) Tumor tissue was harvested from mice at the survival endpoint and processed for immunohistochemistry with the indicated antibodies. Percent expression and H-scores were quantified using Halo software (E). Met, metformin. *P < .05, **P < .01, ***P < .001, ****P < .0001. Scale bars = 20 μm. GBM, Glioblastoma.

Figure 4.

Buparlisib induces hyperglycemia and insulin receptor phosphorylation in tumor tissue from GBM patients. Baseline and average on-treatment glucose (A), C-peptide (B), and HbA1c (C) values were obtained from patients treated on the Phase 2 study of buparlisib for recurrent GBM. (D) Logistic regression of the association between average glucose values on trial and progression-free survival. (E) Tumor tissue from patients in the study and from historical controls was subjected to immunohistochemistry for phosphorylated insulin receptor. (F) H-scores from tumor tissue were quantified using Halo software. **P < .01, ****P < .0001. Scale bars = 20 μm. GBM, Glioblastoma.

Figure 5.

Buparlisib increases T-cell and microglia abundance in tumor tissue from GBM patients. (A) Tumor tissue from 20 untreated patients (57 ROIs) and 9 patients treated with buparlisib was subjected to imaging mass cytometry using 38 epitopes. (B) Cell counts, area, and cell density for each collected ROI. (C) UMAP of cell expression color-coded with cell-type labels. (D) Heatmap of normalized cluster expression used to label cell types. The left heatmap shows cell-type identification markers, and the right heatmap shows functional markers. The bar plot on the right shows cell counts for each group. (E) Cell-type proportion of each ROI as a box plot in control and buparlisib-treated patients. (F) Sample control image of proliferative GBM ROI. GFAP and DNA is expressed in all cells in the ROI. Purple cells are Ki67+ GBM cells. (G) Sample visualization illustrating Ki67, CD8, and CD163 expression in control and treatment groups. *P < .05, **P < .01, ****P < .0001. Scale bars represents 200 μm. GBM, Glioblastoma; GFAP, Glial fibrillary acidic protein; ROI, Region of interest; UMAP, Uniform Manifold Approximation and Projection.