Intermittent PI3Kδ inhibition sustains anti-tumour immunity and curbs irAEs

Abstract

Phosphoinositide 3-kinase δ (PI3Kδ) has a key role in lymphocytes, and inhibitors that target this PI3K have been approved for treatment of B cell malignancies^1-3. Although studies in mouse models of solid tumours have demonstrated that PI3Kδ inhibitors (PI3Kδi) can induce anti-tumour immunity^4,5, its effect on solid tumours in humans remains unclear. Here we assessed the effects of the PI3Kδi AMG319 in human patients with head and neck cancer in a neoadjuvant, double-blind, placebo-controlled randomized phase II trial (EudraCT no. 2014-004388-20). PI3Kδ inhibition decreased the number of tumour-infiltrating regulatory T (T_reg) cells and enhanced the cytotoxic potential of tumour-infiltrating T cells. At the tested doses of AMG319, immune-related adverse events (irAEs) required treatment to be discontinued in 12 out of 21 of patients treated with AMG319, suggestive of systemic effects on T_reg cells. Accordingly, in mouse models, PI3Kδi decreased the number of T_reg cells systemically and caused colitis. Single-cell RNA-sequencing analysis revealed a PI3Kδi-driven loss of tissue-resident colonic ST2 T_reg cells, accompanied by expansion of pathogenic T helper 17 (T_H17) and type 17 CD8⁺ T (T_C17) cells, which probably contributed to toxicity; this points towards a specific mode of action for the emergence of irAEs. A modified treatment regimen with intermittent dosing of PI3Kδi in mouse models led to a significant decrease in tumour growth without inducing pathogenic T cells in colonic tissue, indicating that alternative dosing regimens might limit toxicity.

Fig. 1. PI3Kδi drives anti-tumour immunity but causes significant irAEs.

a, Swimlane plot depicting treatment regimen, intervals and occurrence and grade of irAEs in PI3Kδi-treated (top) and placebo-treated (bottom) patients; patients with partial response or complete pathological response are highlighted in magenta. Vertical dashed lines show average duration of treatment. b, d, Volcano plots of whole-tumour RNA-seq analysis (b) or bulk RNA-seq analysis of purified tumour-infiltrating CD8⁺ T cells (d) comparing patients treated with AMG-319 to those treated with placebo. DEGs between pre- and post-treatment samples are highlighted in red and were called by DEseq2; adjusted P values were calculated with the Benjamini–Hochberg method. Depicted are transcripts that changed in expression by more than 0.75-fold and had an adjusted P value of ≤0.1 (b) or <0.05 (d). c, Median cell count of FOXP3⁺ cells in pre- and post-treatment samples of placebo- or AMG319-treated patients. AMG-319-treated patients have been further stratified into patients for whom the interval between stopping of treatment and immunohistochemistry assessment was more than four days (long interval (LI)) or less than one day (short interval (SI)). P = 0.015 for SI. Data are mean ± s.e.m.; two-tailed Wilcoxon matched-pairs signed rank test (c). Differential expression analysis (b, d) was performed using DESeq2 (v1.24.0). AE, adverse event; NS, not significant. Source data

Fig. 2. PI3Kδi affects distinct T_reg cell subtypes.

a, b, Uniform manifold approximation and projection (UMAP) plots single-cell transcriptomes and T cell receptor (TCR) sequence data of FOXP3⁺CD4⁺ T cells in placebo-treated control mice (a; n = 3 mice) and PI-3065-treated mice (b; n = 3 mice). Circle size indicates degree of clonal expansion. c, d, Violin plots showing Seurat-normalized expression levels (colour scale depicts percentage of expressing cells) of highlighted genes in the indicated clusters from a, b. The centre line depicts the median, edges delineate the 25th and 75th percentiles and whiskers extend to minimum and maximum values. e, Scatter plots showing Seurat-normalized expression levels of highlighted genes in colonic T_reg cells in placebo-treated and PI-3065-treated mice. The dashed line indicates the expression cut-off; numbers indicate the frequency in each quadrant. f, RNA velocity analysis visualized by UMAP, depicting likely developmental trajectories of T_reg cells from a, b. Arrows indicate velocity streamlines. Source data

Fig. 3. PI3Kδi exacerbates colitis.

a, Mice were fed either a control diet or a diet containing PI-3065 for the duration of the experiment and were additionally treated with 2.5% dextran sulfate sodium (DSS) from day 14 to day 20. Change in body weight is shown relative to body weight before treatment on day 0. n = 10 mice per group, P < 0.0001. b, Representative sections from haematoxylin and eosin (H&E) histology scans and colitis scoring from zinc-formalin-fixed colonic tissue sections from mice treated with placebo or PI3Kδi in a. n = 10 mice (placebo) and n = 9 (PI3Kδi) (one mouse died before the experimental endpoint); P < 0.0001 for inflammation, extent, crypt damage and overall colitis scoring; representative samples from the H&E staining are highlighted in magenta. c, d, Mice were inoculated subcutaneously with B16F10-OVA cells and fed either a control diet or a diet containing PI-3065. Infrequent dosing, PI3Kδi for 2 days followed by 5 days off drug; intermittent dosing, PI3Kδi for 4 days followed by 3 days off drug; continuous dosing, PI3Kδi for the duration of the experiment. Tumour volume (c) and flow-cytometric analyses of cell frequencies (d) of mice treated as indicated. n = 6 mice (placebo), n = 7 mice (intermittent dosing), n = 8 mice (continuous dosing) and n = 8 mice (infrequent dosing). Placebo versus intermittent dosing (c), P = 0.0023; placebo versus continuous dosing (c), P = 0.0059; placebo versus continuous dosing, P = 0.003; continuous dosing versus intermittent dosing and infrequent dosing (left), P = 0.0003; placebo versus continuous dosing, P = 0.0005; for continuous dosing versus intermittent dosing, P = 0.0001; placebo versus infrequent dosing (d; third from left), P < 0.0001; placebo versus continuous dosing, P = 0.0086; placebo versus intermittent dosing, P = 0.045; continuous dosing versus intermittent dosing and infrequent dosing, P < 0.0001. Data are mean ± s.e.m.; two-tailed Mann–Whitney test (a–c) or one-way ANOVA comparing the mean of each group with the mean of each other group followed by Dunnett’s test (d). Data are representative of at least two independent experiments. Source data

Fig. 4. Continuous dosing drives pathogenic T_C17 responses.

Mice were inoculated subcutaneously with B16F10-OVA cells and fed either a control diet or a diet containing PI-3065 inhibitor, with treatment conditions as in (Fig. 3c, d). a, Seurat clustering visualized by UMAP of CD8⁺ T cells in colonic tissue at day 18 after tumour inoculation of mice treated as indicated. Pie charts depict the percentage of each cluster under the different treatment conditions.b, Heat map comparing gene expression of cells in all clusters. Depicted are transcripts that change in expression by more than 0.5-fold with adjusted P values of ≤0.05. DEGs were called by MAST analysis; adjusted P values were calculated with the Benjamini–Hochberg method. Sell is also known as Cd62l; Itgae is also known as Cd103. c, Seurat-normalized expression of indicated genes in the different clusters. d, Clone size of cells in indicated clusters in UMAP space. e, Euler diagrams show the clonal overlap between CD8⁺ T cells in the different clusters. f, RNA velocity analysis visualized by UMAP depicting likely developmental trajectories of CD8⁺ T cells. Arrows indicate velocity streamlines. T_C cells, cytotoxic T cells. Source data

Extended Data Fig. 1. Trial schematic, pharmacokinetic & pharmacodynamic assessments and tumor response evaluation.

a, Trial schematic of the placebo-controlled randomized phase II study. b, Consort workflow, 36 patients were screened, of which 33 were recruited and randomly allocated to the placebo control arm or AMG319 drug-treatment arm; 30 patients ultimately received at least one dose of either AMG319 or placebo. Of the 21 patients that were treated with AMG319, 6 patients received daily doses of 300 mg and 15 patients (2 patients withdrew consent prior to receiving the first dose) received daily doses of 400 mg. An initial biopsy was taken before trial initiation and surgical resection of tumors was performed 4–6 weeks after the first dose of treatment. Pre- on- and post-treatment blood samples were collected for further analysis. c, d, Assessment of the level of AKT phosphorylation in B cells at indicated time points pre-dose and 4h after treatment with AMG319, data from one representative patient are shown. d, Plasma concentrations of AMG319 in placebo-controlled and drug-treated patients at indicated time points, n = 9 patients for the Placebo group and n = 18 patients for the AMG319 group. Highlighted in red are patients who were either on-treatment or had only recently (2 days prior to analysis) or briefly discontinued treatment. e, Waterfall plot depicting the change in tumor volume from screening to pre-surgery measured by MRI scan in patients treated with AMG319 (blue bars) or placebo (green bars), shown are all patients for which MRI scans have been performed. Data in (d) are mean +/− S.E.M.

Extended Data Fig. 2. Single-cell RNA-seq analysis reveals substantial oligoclonal expansion of tumor-infiltrating CD8⁺ T cells post-treatment.

a, Analysis of smart-seq2 single-cell RNA-seq data of sorted tumor-infiltrating CD3⁺ T cells from patients 20, 30, 32, 33, 34 and 35 displayed by UMAP analysis. b, Violin plots depicting the Seurat normalized expression of differentially expressed highlighted genes in CD4⁺ T cells (left) or CD8⁺ T cells (right) of the 6 patients pertaining to a, the center line depicts the median, edges delineate the 25th and 75th percentiles and whiskers depict minimum and maximum values. c, Percentages of non-expanded and expanded CD8⁺ and CD4⁺ T cell clones in pre- versus post-treatment samples. d, e, flow-cytometric analyses of the frequency of and expression of activation markers in circulating T_REG cells in placebo-treated (d) and AMG319-treated patients (e), P = 0.0098 for the frequency of circulating T_REG cells (baseline vs d8 and baseline vs d22), P = 0.0234 for ICOS⁺ T_REG cells, at indicated time points; the blue lines depict patients with grade 3/4 irAEs and the red dots indicate the patients with CR/PR. Data are mean +/− S.E.M. Significance for comparisons were computed using two-tailed Wilcoxon matched-pairs signed rank test between baseline and d8 or d22 respectively. Source data

Extended Data Fig. 3. PI3Kδ-inhibition induces are pro-inflammatory tumor microenvironment.

Mice were inoculated s.c. with B16F10-OVA cells and fed either a control diet or a diet containing the PI-3065 PI3Kδ inhibitor for the indicated treatment period. Tumor volume (a) and flow-cytometric analyses of cell frequencies (b–g) of mice treated as indicated; P = 0.003 (a), P = 0.0076 (b), P = 0.0279 (c), P = 0.0172 (d), P = 0.0013 (e), P = 0.0435 (f), n = 9 mice for the Placebo group and n = 10 mice for the PI3Kδ group for a-f g, shown are representative contour plots of intratumoral CD8⁺ T cells depicting the indicated markers. h, Tumor volume of Rag1^−/− or CD8^−/− mice treated as indicated, n = 6 mice/group for Rag1^−/− and n = 5 mice/group for CD8^−/− mice. i–k, flow-cytometric analyses of T_REG cell frequencies (b) in indicated organs of mice treated as indicated, n = 6 mice for the Placebo group and n = 5 mice for the PI3Kδ group. Shown are representative contour plots of FoxP3-expressing (RFP⁺) CD4⁺ T cells in indicated organs; P = 0.0043 (spleen), P = 0.0043 (tumor), P = 0.0043 (colon). Not significant, P = 0.1234; *P = 0.0332; ***P = 0.0002; and ****P < 0.0001. Data are mean +/− S.E.M and statistical significance for comparisons was computed using two-tailed Mann-Whitney test; data are representative of at least two independent experiments. DEGs in (b–e) were called using MAST and adjusted p-values were calculated with the Benjamini-Hochberg method. Source data

Extended Data Fig. 4. T_REG cells in different tissues exhibit unique transcriptomic signatures.

a, Analysis of 10x single-cell RNA-seq data displayed by UMAP analysis. Seurat clustering of FoxP3⁺CD4⁺ T cells in spleen (left), tumor (middle) and colon (right). b, Heatmap comparing gene expression of cells in all clusters. Depicted are transcripts that change in expression more than 0.5-fold and adjusted P value of ≤ 0.05. c–e, Volcano plots of single-cell RNA-seq analysis of placebo-treated control mice and PI-3065-treated mice in spleen (c) tumor (d) and colon (e). Highlighted are transcripts with a >0.5 log2 fold change. f, Bar charts depicting the proportion of cells in each cluster. Bars are colorized based on cells in indicated treatments making up the cluster. g, Violin plots showing normalized expression levels (log₂(CPM+1)) of highlighted genes in cluster the colonic clusters pertaining to Fig. 2a, b, the center line depicts the median, edges delineate the 25th and 75th percentiles and whiskers depict minimum and maximum values.

Extended Data Fig. 5. Colonic ST2 T_REG cells exhibit features of superior suppressive capacity.

a, Curtain plot highlighting selected genes in each cluster with average transcript expression (color scale) and percent of expressing cells (size scale). b, Flow-cytometric analyses depicting the expression of ST2 in CD8⁺ T cells or CD4⁺ T_REG cells in representative zebra plots (left), the frequency of ST2⁺ T_REG cells (n.s., P = 0.0549), FOXP3⁺ cells (P = 0.0007), and the ratio of CD8⁺ T cells to ST2 T_REG cells (P = 0.0047) in placebo-treated (n = 6 mice) and PI3Kδi-treated mice (n = 8 mice) c, Violin plots showing normalized expression levels (log₂(CPM+1)) of highlighted genes in indicated clusters pertaining to Fig. 2a, b, the center line depicts the median, edges delineate the 25th and 75th percentiles and whiskers depict minimum and maximum values. d–g, Flow-cytometric analyses of CD8⁺ T cell frequencies in colon (d) P = 0.0025, spleen (e) P = 0.0013 and of the expression of PD-1 (f) P = 0.0025 and ICOS (g) P = 0.0025 on colonic CD8⁺ T cells, n = 7 mice for the Placebo group and n = 5 mice for the PI3Kδ group. Not significant, P = 0.1234; *P = 0.0332; ***P = 0.0002; and ****P < 0.0001. Data (b, d–g) are mean +/− S.E.M and statistical significance for comparisons was computed using two-tailed Mann-Whitney test; data are representative of at least 2 independent experiments. Source data

Extended Data Fig. 6. Continuous dosing drives pathogenic T_H17 responses.

Mice were inoculated s.c. with B16F10-OVA cells and fed either a control diet or a diet containing the PI-3065 PI3Kδ inhibitor, treatment conditions as in (Fig. 3c, d). a, depicted is Seurat clustering, visualized by UAMP, of CD4⁺ T cells in colonic tissue at day 18 after tumor inoculation of mice treated as indicated, pie charts depict the percentage of each cluster in the different treatment conditions. b, Heatmap comparing gene expression of cells in all clusters. Depicted are transcripts that change in expression more than 0.5-fold and adjusted P value of ≤ 0.05, DEGs were called by MAST analysis, adjusted p-values were calculated with the Benjamini-Hochberg method. c, Seurat-normalized expression of Ifng (top left), Il17a (top right), Foxp3 (bottom left) and Il4 (bottom right) in the different clusters. d, Clone size of cells in indicated clusters in UMAP space. e, Euler diagrams show the clonal overlap between the CD4⁺ T cells in the different clusters. f, RNA velocity analysis visualized by UMAP depicting likely developmental trajectories of CD4⁺ T cells, arrows indicate velocity streamlines.