Immunogenomic profiling determines responses to combined PARP and PD-1 inhibition in ovarian cancer

Abstract

Combined PARP and immune checkpoint inhibition has yielded encouraging results in ovarian cancer, but predictive biomarkers are lacking. We performed immunogenomic profiling and highly multiplexed single-cell imaging on tumor samples from patients enrolled in a Phase I/II trial of niraparib and pembrolizumab in ovarian cancer (NCT02657889). We identify two determinants of response; mutational signature 3 reflecting defective homologous recombination DNA repair, and positive immune score as a surrogate of interferon-primed exhausted CD8 + T-cells in the tumor microenvironment. Presence of one or both features associates with an improved outcome while concurrent absence yields no responses. Single-cell spatial analysis reveals prominent interactions of exhausted CD8 + T-cells and PD-L1 + macrophages and PD-L1 + tumor cells as mechanistic determinants of response. Furthermore, spatial analysis of two extreme responders shows differential clustering of exhausted CD8 + T-cells with PD-L1 + macrophages in the first, and exhausted CD8 + T-cells with cancer cells harboring genomic PD-L1 and PD-L2 amplification in the second.

Conflict of interest statement

J.G. and Y.Z. were previously employed by TESARO, and now is a GSK employee. B.D. was an employee of Tesaro at the time of the work related to this manuscript. G.I.S. receives research funding from Merck & Co. S.S. is a consultant for RareCyte, Inc. P.K.S. is a member of the SAB or Board of Directors of Applied Biomath and RareCyte Inc and has equity in these companies. In the last 5 years, the Sorger lab has received research funding from Novartis and Merck. Sorger declares that none of these relationships are directly or indirectly related to the content of this manuscript. For the SigMA algorithm, a patent application titled “Systems and methods for classifying tumors” has been filed under the Patent Cooperation Treaty (PCT) by authors D.G. and P.P. on 18 September, 2019 following a provisional patent application filed on 24 September, 2018. U.M. has served as a consultant for Merck and has been the North America PI of the NOVA study funded by TESARO. P.K. and A.D.D’A. have served as consultants/members of advisory boards for Merck and Tesaro/GSK. Other authors have no competing interests to declare.

Figures

Fig. 1. Tumor mutational signature 3 positivity associates with prolonged progression-free survival with the combination of niraparib and pembrolizumab.

a SigMA identified a larger proportion of tumors positive for homologous recombination deficiency (HRD). The proportions of tumors positive (red) and negative (blue) for HRD as annotated by the BRCA1/2 mutation, Myriad HRD test, BROCA, RAD51, and SigMA. b Sig3 positivity is associated with clinical benefit as determined by either complete or partial response or stable disease. Correlations of HRD to clinical benefit (Fisher’s exact test). c Proportions of patients positive (red) or negative (blue) for Sig3 according to best objective response. PD progressive disease, SD Stable disease, PR partial response. d Sig3 associates with increased progression-free survival (PFS; Log-rank test). Kaplan–Meier graph for PFS for the combination of niraparib and pembrolizumab according to Sig3 status. All test were two-sided.

Fig. 2. Immune signatures associate with tumor regression and objective response to niraparib and pembrolizumab.

a In the chemo-naive samples, pathway scores for six biological pathways were higher in the responders. The red dots depict the mean difference of Nanostring pathway scores between responders (n = 4) compared to non-responders (n = 15), and the red lines show the 95% confidence interval (CI) for the difference. p < 0.05 was considered significant, r; effect size, Mann–Whitney U-test. b The Nanostring pathway scores of three pathways related to Type-I interferon signaling were significantly higher in the responders compared to non-responders (Mann–Whitney U-test). c In the chemo-exposed samples, heatmap of the Nanostring cell-type z-scores showed a higher signature of relative exhausted CD8 + T-cell-scores (black box) in the responders (CR, PR; n = 6) compared to non-responders (SD, PD; n = 16). d The inferred relative cell-type scores from Nanostring advanced analysis are presented as the specific cell-type score over (indicated by versus; vs.) the general cell-type score. The relative score for exhausted CD8 + T-cells over the total CD8 + T-cells was significantly higher in the responders compared to non-responders (p = 0.02, Mann–Whitney U-test), and positively correlated with the percentage of best tumor regression from baseline in patients with clinical benefit (p = 0.01, Spearman’s correlation, Rho 0.68; n = 13). The cell-type scores for total CD8 + T-cells or the relative cell-type score for Exhausted CD8 + T-cells over the total tumor infiltrating lymphocytes (TILs) were not significantly different in the responders compared to non-responders. e Immune score associated with objective response (p = 0.01, Fisher’s exact test). f Waterfall plot of best percent of tumor regression from baseline as annotated by Sig3 and Immune score (IS) showing that all patients with tumor regression (dashed line represents ≥30%) were positive for Sig3, Immune score or both. g Combined score of tumors being positive for Immune score, Sig3 or both associated with clinical benefit (p = 0.01, Fisher’s exact test). h Positivity the combined score significantly correlated with prolonged PFS (p = 0.002, Log-rank test). i None of the patients whose tumors were negative for the combined score achieved objective response (p = 0.06, Fisher’s exact test). All test were two-sided. No adjustment was made for multiple hypothesis testing (see materials and methods). Box plots are presented as the range (whiskers), center line as the median, bounds of box mark the highest and lowest quartiles, and the dashed line represents the mean.

Fig. 3. Interferon activation and proliferative state of CD8 + T-cells associate with response.

a A hierarchical clustering heatmap of the mean expression levels of single-cell quantification using tCycIF and annotated with Sig3 status, Nanostring Immune score, sample category (Sample cat), and confirmed best objective response (confirmed BOR) in 26 patients. b Cell-type calls visualized using semi-supervised UMAP dimensionality reduction reveals the clustering of tumor-stroma immune (upper panel), and the tumor microenvironment immune cell subpopulations (lower panel) into distinct clusters. c Chemo-exposed samples had a higher immune- and stromal infiltration in the tumor microenvironment. The proportions of immune (blue), tumor (gray) and stromal (yellow) cells of the 26 samples (rows) analyzed via single-cell imaging. d Mean pSTAT1 protein expression in exhausted CD8 + T-cells was higher in responders (n = 10) compared to non-responders (n = 16) and e in patients with clinical benefit (n = 20) compared to patients with no clinical benefit (n = 6). f pSTAT1 expression and g Ki67 levels in effector CD8 + T-cells associated with response. p < 0.05 was considered significant, r; effect size, Mann–Whitney U-test, all test were two-sided. h We observed increased pSTAT1 and Ki67 expression in the exhausted and effector CD8 + T-cells in the responders (upper row) compared to non-responders (lower row). Scale bar 50 µm. Box plots are presented as the range (whiskers), center line as the median, bounds of box mark the highest and lowest quartiles, and the dashed line represents the mean.

Fig. 4. Two extreme responders show differential spatial patterns of cellular interactions in the tumor microenvironment.

a The graph showing the percentage of tumor regression from baseline over time (weeks) in the first extreme responder; she achieved a PR lasting over 10 months with 87% tumor regression from the baseline. b The proportion of immune cell subpopulations out of total immune cells in the tumor microenvironment. The patient’s tumors immune infiltration was enriched in macrophages and exhausted CD8 + T-cells. c The normalized (z-score) PD-L1 expression according to the tumor microenvironment cell subpopulations. The highest PD-L1 expression was observed in CD163 + IBA1 + CD11b + Macrophages. Individual dots represent single cells. Box plots are presented as the range (whiskers), center line as the median, bounds of box mark the highest and lowest quartiles, and the dashed line represents the mean. d Multiplexed immunofluorescent images confirmed the high infiltration of macrophages and PD-1 positive exhausted CD8 + T-cells (higher row), and the high expression of PD-L1 in the macrophages (lower row). Scale bar 50 µm. e Spatial visualization of neighborhood (k = 10) composition shows increased interaction between PD-1+ exhausted CD8 + T (E.CD8 + T)-cells and PD-L1-positive macrophages (PD-L1 + M) shown in magenta, compared to PD-L1-positive tumor cells (PD-L1 + T) Scale bar 50 µm. fK-means clustering indicated that neighborhood clusters containing PD-1+ exhausted CD8 + T-cells contain PD-L1 + macrophages and not PD-L1-positive tumor cells. g The second extreme responder exhibits PD-L1 and PD-L2 amplification confirmed by FISH. Scale bar 10 µm. h Spatial visualization of neighborhood (k = 10) composition shows increased interaction between PD-1+ exhausted CD8 + T (E.CD8 + T)-cells and PD-L1-positive tumor cells (PD-L1 + T) shown in yellow, compared to PD-L1-positive macrophages (PD-L1 + M). Scale bar 50 µm. iK-means clustering indicated that neighborhoods containing most of the PD-1+ exhausted CD8 + T-cells cluster together with the PD-L1 + tumor cells and less with the PD-L1-positive macrophages. j High-resolution imaging of the two extreme responders using cyclic immunofluorescence. First row depicts the first patient, with exhausted CD8 + T-cells, tumor cells and macrophages in the tumor microenvironment (first column), with positive PD-L1 expression (second column) in the IBA1 + macrophages, most of which were also positive for CD163 (third column), and co-localization of the exhausted CD8 + T-cells with the PD-L1-positive macrophages (fourth column). In the second patient (second row), the exhausted CD8 + T-cells were spatially more next to the tumor cells (first column), while there was clear staining of PD-L1 also in the tumor cell compartment (second column, arrows), in addition to the macrophages (third column), and PD-1/PD-L1 the exhausted CD8 + T-cells. Scale bar 20 µm.

Fig. 5. Spatial interactions of exhausted CD8 + T-cells associate with response and mutational signature 3.

a A hierarchical clustering of significant interactions of other cell types towards exhausted CD8 + T-cells from neighborhood analysis using permutation test; positive fold-change being attraction and a negative as avoidance, with a two-sided test and a cutoff of p < 0.05. b In the 16 tumors with significant attraction, of IBA + CD11b + Macrophages towards exhausted CD8 + T-cells the attraction score was higher in the responders (n = 6) compared to non-responders (n = 10). The responders (n = 10) had a significantly higher fraction (z-score) of PD-L1-positive macrophages (c) and tumor cells (d) interacting with exhausted CD8 + T-cells compared to non-responders (n = 16). e The fraction (z-score) of PD-L1-positive tumor cells neighboring exhausted CD8 + T-cells was higher in the Sig3 + (n = 12) compared to Sig3– (n = 6) tumors. p < 0.05 was considered significant, r; effect size, Mann–Whitney U-test. f The model summarizing the findings in the present study. The responses to the combination of niraparib and pembrolizumab are determined by tumor mutational signature 3-positivity, associated with increased PD-L1 positivity in the macrophages and tumor cells, the interferon-primed exhausted CD8 + T-cells and activated effector CD8 + T-cells, and their spatial interactions in the tumor microenvironment. Box plots are presented as the range (whiskers), center line as the median, bounds of box mark the highest and lowest quartiles, and the dashed line represents the mean.