Radiation and dual checkpoint blockade activate non-redundant immune mechanisms in cancer

Abstract

Immune checkpoint inhibitors result in impressive clinical responses, but optimal results will require combination with each other and other therapies. This raises fundamental questions about mechanisms of non-redundancy and resistance. Here we report major tumour regressions in a subset of patients with metastatic melanoma treated with an anti-CTLA4 antibody (anti-CTLA4) and radiation, and reproduced this effect in mouse models. Although combined treatment improved responses in irradiated and unirradiated tumours, resistance was common. Unbiased analyses of mice revealed that resistance was due to upregulation of PD-L1 on melanoma cells and associated with T-cell exhaustion. Accordingly, optimal response in melanoma and other cancer types requires radiation, anti-CTLA4 and anti-PD-L1/PD-1. Anti-CTLA4 predominantly inhibits T-regulatory cells (Treg cells), thereby increasing the CD8 T-cell to Treg (CD8/Treg) ratio. Radiation enhances the diversity of the T-cell receptor (TCR) repertoire of intratumoral T cells. Together, anti-CTLA4 promotes expansion of T cells, while radiation shapes the TCR repertoire of the expanded peripheral clones. Addition of PD-L1 blockade reverses T-cell exhaustion to mitigate depression in the CD8/Treg ratio and further encourages oligoclonal T-cell expansion. Similarly to results from mice, patients on our clinical trial with melanoma showing high PD-L1 did not respond to radiation plus anti-CTLA4, demonstrated persistent T-cell exhaustion, and rapidly progressed. Thus, PD-L1 on melanoma cells allows tumours to escape anti-CTLA4-based therapy, and the combination of radiation, anti-CTLA4 and anti-PD-L1 promotes response and immunity through distinct mechanisms.

Conflict of interest statement

The authors declare no competing financial interests. Code for computational analysis is available upon request.

Figures

Extended Data Figure 1. Patients and mice treated with RT + anti-CTLA4 for melanoma

a) Twenty-two stage IV melanoma patients (M stage indicated) were stratified by treatment site of a single index metastasis, which was the irradiated tumor. Two dosing levels (DL) for stereotactic body radiation (SBRT) were in each stratum. b) Waterfall plot of the RECIST % change from baseline of unirradiated tumors annotated to indicate metabolic responses by PET/CT (hatches above plot) and response of the irradiated index tumor as measured by CT and PET/CT (hatches below plot). RECIST criteria do not include irradiated tumors. Legend shows color-codes for response after CT or PET/CT (parenthesis). PMD: progressive metabolic disease; SMD: stable metabolic disease; PMR: partial metabolic response; CMR: complete metabolic response. White hatches indicate no imaging obtained. See Extended Data Table 2. c) Survival (right) and total tumor growth (bottom) after RT with either concurrent or sequential anti-CTLA4 compared to anti-CTLA4 (C4) or RT alone. The regimens and the standard regimen used for all other melanoma experiments are illustrated (left). The p-values for tumor growth are compared to anti-CTLA4. d) Survival after RT and/or anti-CTLA4 with or without T cell depletion (n=5–10) using anti-CD8 (CD8). Shown are overall p-values. The p-value for RT + anti-CTLA4 with and without anti-CD8 is p=0.005. Control is an isotype-matched antibody. e) Three mice with CRs were rechallenged with B16-F10 tumors. Shown is a representative mouse. Arrow indicates location of regressed tumor and vitiligo-like condition represented by non-pigmented fur (observed in approximately 50% of mice with CRs). Time line starts from original tumor implantation (day 0) and values above marks are days after first rechallenge. Recurrence occurred only after anti-CD8 treatment and second rechallenge.

Extended Data Figure 2. Tumor cells resistant to RT + anti-CTLA4 upregulate PD-L1 but not other candidate inhibitory receptor pathways

a) Unirradiated tumor growth (left: normalized, right: raw values) for mice implanted with Res 177 (n=21), Res 499 (n=25), and B16-F10 (n=18) melanoma cells and treated with RT + anti-CTLA4. For normalization, volumes were divided by average of untreated controls (V/Vcont) to account for differences in growth between untreated tumor types. The p-values are for comparisons with B16-F10 tumors. b) Corresponding tumor volumes of unirradiated or irradiated index tumors at day 21 (blue line is mean). c) Clonogenic survival for Res 499 and B16-F10 cells (n=2). d) Selection of immune variables that robustly predict resistance to RT + anti-CTLA4 using minimal depth (MD). A variable was selected if its MD was less than a threshold value for significance. Shown are bootstrap distributions of MD values (left) and % bootstrap models for which the MD for the indicated variable was significant (right). Bootstrap mean +/− SD for the out-of-bag prediction error rate is listed on top. e) Volcano plot of differentially expressed genes from resistant tumors. Horizontal black line is 5% false-discovery rate and dotted green line is fold-change cut-off. Ligands for select inhibitory receptors are indicated. See SI Table 1. f) Unirradiated tumor volumes (day 26–29) and g) survival after RT + anti-CTLA4 for mice with bilateral tumors from TSA breast cancer cells (n=25) or from the Res 237 subline selected to be resistant (n=21). h) Expression of candidate T cell inhibitory receptor ligands on B16-F10 and Res 499. Interferon-gamma (IFNg) responsiveness was tested. i) Boxplots show distribution of % positive CD8+CD44+ T cells for the indicated inhibitory receptor compared to IgG control. j) PD-L1 surface expression for CRISPR PD-L1 homozygous knockout Res 499 and wild type control cells. IFNg was used to induce PD-L1 and confirm abrogated response.

Extended Data Figure 3. Addition of PD-L1/PD-1 blockade antagonizes resistance to RT + anti-CTLA4, and optimal response to checkpoint blockade requires RT

a) Change in % CD8+CD44+ T cells after RT and checkpoint blockade vs. change in the degree of reinvigoration of exhausted T cells measured by % PD-1+Eomes+ T cells that are Ki67+GzmB+. Values are subtracted from average of untreated control. b) Growth of Res 499 tumors after RT + anti-CTLA4 (C4) with and without addition of anti-PD-L1 (PDL1). Shown are index and unirradiated tumors from n=25 mice in each group. The p-value is for comparison to RT + anti-CTLA4. c) Proportion of CRs (yellow) for mice with Res 499 tumors. d) Total tumor growth (index + unirradiated) for B16-F10 tumors after the indicated treatment that includes anti-PD-1 (PD1) or anti-PD-L1. The p-values are for comparisons to RT + anti-CTLA4 (n=18, n=5 for others). Pie charts show % CRs (yellow). e) Survival of mice after RT + anti-CTLA4 + anti-PD-1. Shown is the overall p-value, and f) the two-way comparisons that include those from Fig. 2d. g) Proportion of mice with CRs (yellow) after RT + anti-PD-L1 or anti-PD-1 that survived 90+ days after tumor rechallenge at day 60 (n=12). h) Survival of mice with bilateral Res 237 breast cancer tumors treated with RT + anti-CTLA4 with (n=16) or without (n=21) anti-PD-L1. i) Proportion of CRs (yellow) for mice with Res 237 or TSA breast cancer tumors. j) Survival of mice with pancreatic tumors from a cell line derived from KPC mice (KrasLSL-G12D/+;p53LSL-R172H/+;Pdx-1-Cre) (n=5 for each group). Select treatment groups are labeled on the plot for clarity. Overall p-value is shown.

Extended Data Figure 4. TCR clonotypes associated with RT are not observed in random clones from post-treatment blood and have distinct CDR3 features

a) Boxplot of the bootstrap variance explained by multivariable RF regression model for effect of RT, anti-CTLA4, and/or anti-PD-L1 on immune variables from TILs. b) K-means clustering (k=2) was used on the average CDR3 amino acid features of randomly sampled clones from post-treatment blood after anti-CTLA4, anti-PD-L1, and/or RT. Membership into each cluster was determined and the p-value for separation into treatment groups with and without RT was calculated. Boxplot shows log10 p-values from 1000 random iterations. Comparison to the p-value from the observed data (red dotted line) gives a simulated p < 0.001. c) Log10 p-values for separation into treatment groups with and without RT vs. cut-off value used to select the most frequent clones. The 0.05 significance level is indicated (red dotted line). d) Average % occupancy in the CDR3 of the most frequent T cell clonotypes after RT +/− checkpoint blockade (+RT, red line) or checkpoint blockade alone (NoRT, orange line) by contiguous short amino acid sequences of length three (3-tuples) belonging to e) subsets with distinct treatment-related amino acid properties. These properties are characterized by Atchley factors, which measure 1) PAH: accessibility, polarity, and hydrophobicity, 2) PSS: propensity for secondary structure, 3) MS: molecular size, 4) CC: codon composition, and 5) EC: electrostatic charge. Shown (right) are the average values of each Atchley factor for amino acids that comprise the 3-tuples from the indicated subset (red) compared to all unselected 3-tuples (blue). Boxplots (left) show the proportion of 3-tuples from each of these subsets that are found in the CDR3s of the five most frequent clones after treatment. Compared to pre-treatment samples (Pre-tx), subset 6 is associated with RT +/− checkpoint blockade (+RT) or checkpoint blockade alone (NoRT). Subset 1 is primarily associated with checkpoint blockade alone, and subset 16 is primarily associated with RT +/− checkpoint blockade.

Extended Data Figure 5. Peripheral T cell exhaustion, reinvigoration, CD8/Treg ratio, and tumor PD-L1 predict response to RT + immune checkpoint blockade

a) Heat map showing the relative proportions of PD-1+ CD8 T cells that are Ki67+GzmB+ or Eomes+ and the CD8/Treg ratio for each sample (columns) subtracted from the average values of untreated controls. Black hatches indicated CR and treatment with RT + anti-CTLA4 (C4) +/− anti-PD-L1 (P1). From these data, a multivariable RF predictor for CR was developed. Boxplot shows bootstrap distributions of variable importance scores (more predictive variables have higher values), and of b) minimal depth (MD), a statistic to measure predictiveness. Bar plot shows % bootstrap models for which the MD for the indicated variable was significant. Bootstrap mean +/− SD for the out-of-bag prediction error rate is listed on top. c) Probability of CR vs. change (treated vs. untreated control) in CD8/Treg ratio for mice with a high (blue dots) or low (red dots) change in % PD1+ splenic CD8 T cells that are Eomes+. d) Heat map similar to (a) except using T cells from peripheral blood. e) Percent peripheral blood PD-1+ CD8 T cells that are Eomes+ vs. Ki67+GzmB+ after RT + checkpoint blockade. Values are subtracted from average of untreated controls. Each circle represents a mouse. Probability of CR (proportional to circle size), prediction error rate, and quadrant boundaries are estimated from the RF model. f) Representative contour plots examining splenic CD8 T cells from B16-F10 or Res 499 tumors for PD-1 and Eomes (top), followed by examination of the PD-1+Eomes+ subset for Ki67 and GzmB (bottom). g) Ratios of PD-1+Eomes+ splenic CD8 T cells that are Ki67+GzmB+ (reinvigorated) compared to Ki67−GzmB− (exhausted) from mice with Res 499 tumors.

Extended Data Figure 6. Melanoma PD-L1 is associated with T cell exhaustion, response, and survival for patients treated on clinical trial of RT + anti-CTLA4

a) Representative images (right) for patients with biopsies showing PD-L1 staining on tumor cells classified as PD-L1lo (top), 2+ (middle), or 3+ (bottom). Scores of 2+ and 3+ are classified as PD-L1hi. The arrow indicates PD-L1 staining on macrophages. An isotype antibody negative control and positive controls are shown (left). b) Changes in % Ki67+GzmB+ in PD-1+ CD8 T cells after RT + anti-CTLA4 vs. PD-L1 status on melanoma cells from all patients with available pre- and post-treatment blood. c) Changes in % Ki67+GzmB+ in PD1+Eomes+ CD8 T cells (left) or in PD1+ CD8 T cells (right) vs. macrophage PD-L1 status. d) Hazard ratio and 95% CI for PFS from a Cox regression model using PD-L1 status on tumor cells and macrophages. e) Model for non-redundant mechanisms and resistance to RT and immune checkpoint blockade.

Figure 1. RT + anti-CTLA4 promotes regression of irradiated and unirradiated tumors and is inhibited by PD-L1 on tumor cells

a) Waterfall plot of unirradiated tumors after RT to a single index lesion with anti-CTLA4. Dashed lines are thresholds for PD (red) and PR (blue). * Patients with new lesions. ** Clinical progression without imaging. b) PET/CT images of irradiated (white arrows) and unirradiated (yellow arrows) tumors from patient PT-402. c) PFS and OS for all patients (dashed lines: 95% CI). d) B16-F10 tumor growth after RT to the index tumor (n=8), anti-CTLA4 (C4) (n=9), anti-CTLA4 and RT to the index tumor (n=18), or no (control) treatment (n=9). The p-values are comparisons with control. Pie chart shows %CRs (yellow). See Fig. 2d for survival. e) Heat map showing relative abundance of immune cells or their ratios from tumors that are resistant (black hatch) or sensitive to RT + anti-CTLA4. Boxplot shows bootstrap importance scores for each variable. Higher values (red) are more predictive. f) Change in T cell subsets or their ratio after RT + anti-CTLA4 for sensitive parental (Sen) or resistant (Res) tumors. Values are subtracted from average of untreated controls. Red line is mean. g) Heat map of resistance gene signature and PD-L1 across human melanoma. p < 0.001 by gene set enrichment analysis. h) Expression of PD-L1 on Res 499 compared to B16-F10 melanoma cells and of Res 237 compared to TSA breast cancer cells. Isotype control (IgG). i) Total tumor volume from PD-L1 knockout (KO) or control (WT) Res 499 and corresponding survival.

Figure 2. Addition of PD-L1 blockade reinvigorates exhausted T cells and improves response to RT + anti-CTLA4

a) Representative contour plot of CD8 TILs from B16-F10 or Res 499 tumors after RT and anti-CTLA4 (C4) +/− anti-PD-L1 (P1) examined for PD-1 and Eomes (top row), followed by examination of the PD-1+Eomes+ subset for Ki67 and GzmB (bottom row). Schema shows exhaustion and reinvigoration markers. b) Proportion of PD-1+Eomes+ CD8 T cells that are either Ki67−GzmB− or Ki67+GzmB+. c) Changes in T cell subsets and their ratio from Res 499 tumors. d) Survival of mice with B16-F10 tumors (n=18 for RT+C4, n=5 for others). Shown are overall p-values.

Figure 3. RT, anti-CTLA4, and anti-PD-L1 have distinct effects on the TCR repertoire, Tregs, and T cell exhaustion

a) Heat map of changes in the frequency of immune cells or their ratios from B16-F10 tumors. Black hatches indicate treatment. Bar plots show bootstrap importance scores (mean +/− SE) that assess changes in immune parameters predicted by treatment type (read row-wise). Higher values (yellow) represent stronger association. b) T cell subsets and their ratios. c) Frequency distribution (dashed line is 0.5%) and d) boxplot of diversity index (0: clonal, 1: fully diverse) for most frequent TCR clonotypes found in TILs of unirradiated B16-F10 tumors after RT and/or anti-CTLA4. Boxplot summarizes data for mice treated with anti-CTLA4 (NoRT) or RT +/− anti-CTLA4 (+RT). e) Representative contour plots and f) ratios examining PD-1+Eomes+ splenic CD8 T cells from mice with B16-F10 tumors for Ki67+GzmB+ (reinvigorated) or Ki67−GzmB− (exhausted) subsets. g) TCR clonal frequency in post-treatment blood vs. TILs (top row) or vs. pre-treatment blood (bottom row). Quadrant boundaries are top 5% quantiles from the control. Clones below detection in pre-treatment blood are assigned upper bounds (blue). h) Maximum clonal frequency in post-treatment blood (dot) of the most frequent TCR clonotypes found in TILs. i) Distances to cluster centroids for the average CDR3 amino acid features of the five most frequent clones in pre- or post-treatment blood from mice treated with (red) or without (orange) RT. Membership into two clusters (circles and squares) determined by k-means.

Figure 4. Tumor PD-L1 and T cell exhaustion and reinvigoration can predict response in mice and patients

a) Percent PD-1+ CD8 T cells that are Eomes+ vs. Ki67+GzmB+ after RT combined with checkpoint blockade. Values are subtracted from average of untreated controls. Each circle represents a mouse. Probability of CR (proportional to circle size), prediction error rate, and quadrant boundaries are estimated from an RF model. b) Percent Eomes+PD-1+ CD8 T cells in post-treatment blood vs. change in % PD-1+Eomes+ CD8 T cells that are Ki67+GzmB+ after treatment. Each circle represents a patient. PFS is proportional to circle size and quadrant boundaries are average values for patients under the mean PFS. Concordance index of the RF model is 0.59. c) Contour plot of peripheral blood CD8 T cells from patients PT-102 and PT-402 examined for PD-1 and Eomes (top row), followed by examination of the PD-1+Eomes+ subset for Ki67 and GzmB (bottom row). d) PD-L1 staining from corresponding tumor biopsies. e) Change in % Ki67+GzmB+ in PD-1+Eomes+ CD8 T cells vs. PD-L1 status of melanoma cells from all patients with available pre- and post-treatment blood. f) RECIST response, g) PFS, and OS stratified by PD-L1 status of melanoma cells.