Neoadjuvant PD-1 blockade induces T cell and cDC1 activation but fails to overcome the immunosuppressive tumor associated macrophages in recurrent glioblastoma

Abstract

Primary brain tumors, such as glioblastoma (GBM), are remarkably resistant to immunotherapy, even though pre-clinical models suggest effectiveness. To understand this better in patients, here we take advantage of our recent neoadjuvant treatment paradigm to map the infiltrating immune cell landscape of GBM and how this is altered following PD-1 checkpoint blockade using high dimensional proteomics, single cell transcriptomics, and quantitative multiplex immunofluorescence. Neoadjuvant PD-1 blockade increases T cell infiltration and the proportion of a progenitor exhausted population of T cells found within the tumor. We identify an early activated and clonally expanded CD8+ T cell cluster whose TCR overlaps with a CD8+ PBMC population. Distinct changes are also observed in conventional type 1 dendritic cells that may facilitate T cell recruitment. Macrophages and monocytes still constitute the majority of infiltrating immune cells, even after anti-PD-1 therapy. Interferon-mediated changes in the myeloid population are consistently observed following PD-1 blockade; these also mediate an increase in chemotactic factors that recruit T cells. However, sustained high expression of T-cell-suppressive checkpoints in these myeloid cells continue to prevent the optimal activation of the tumor infiltrating T cells. Therefore, future immunotherapeutic strategies may need to incorporate the targeting of these cells for clinical benefit.

Fig. 1. Neo-aPD1 increases the proportion and number of CD3+ T cells among tumor-infiltrating immune cells in recurrent glioblastoma patients.

a Schematic of CD45+ cell isolation from tumor tissue. b The number of patients whose tumor-infiltrating CD45+ cells were analyzed using CyTOF and/or scRNAseq. c A UMAP projection of the CD45+ tumor-infiltrating cells analyzed with CyTOF on the common 22 marker panel (Supp. Data 1, n = 1,067,057 cells from 63 total patients: 26 GBM.new, 17 GBM.rec, 20 GBM.pembro). d Select marker intensities on the UMAP projections of all CD45+ cells. e A UMAP projection of CD45+ tumor-infiltrating cells analyzed with scRNAseq (n = 156,766 cells from 40 total patients: 14 GBM.new, 12 GBM.rec, and 14 GBM.pembro). f The percentage of tumor-infiltrating CD45+ cells belonging to T and myeloid cell clusters from (c) across different tumor groups (blue circles: GBM.new, black squares: GBM.rec, red triangles: GBM.pembro; each dot represents a different patient). g CD3 and CD14 gating in the CyTOF data separated by different tumor groups. Colors are the same as f. h The number of CD3+ T cells per mg of tumor across different tumor groups. The highest value for each condition was treated as outlier and removed. Colors are the same as f. i Representative 20x images of multiplex immunofluorescence staining of a GBM.rec (left) and GBM.pembro (right) tumor samples (blue: DAPI, red: CD14, cyan: CD4, white: CD8). Images were collected and analyzed for 8 GBM.rec and 12 GBM.pembro patients. j The number of CD4+/CD8+ cells per mm² of tissue section for GBM.rec (black squares) and GBM.pembro (red triangles) samples. P values were calculated using a two-sided Wilcoxon rank-sum test. Source data are provided as a Source Data file or in Supp. Data 2.

Fig. 2. Single-cell RNAseq analysis of intratumoral T cells shows transcriptional changes with neo-aPD1 therapy.

a A UMAP projection of the lymphoid compartment of the tumor samples analyzed using scRNAseq (n = 14,322 cells from 40 total patients: 14 GBM.new, 12 GBM.rec, and 14 GBM.pembro). b Differentially expressed genes in GBM.new (blue), GBM.rec (black), and GBM.pembro (red) patient samples as computed by Seurat. c The fraction of lymphoid cells with detected expression of the indicated genes. d, f Pseudotime projection of CD4 (d) and CD8 (f) T cells by Monocle 2. e, g The expression of CD4 (e) and CD8 (g) T cell marker genes across the different cell fates in d and f, respectively. (top) Proportion of the cells from in each fate, separated by tumor groups, are shown. h The percentage of all cells expressing at least one transcript for CCL5 and XCL1/2. P values were calculated using a two-sided Wilcoxon rank-sum test. Source data are provided as a Source Data file or in Supp. Data 3. In all boxplots, the median is indicated by the line within the box and the 25th and 75th percentiles indicated by the lower and upper bounds of the box. The upper and lower lines above and below the boxes represent the whiskers.

Fig. 3. The phenotype and clonal distribution of intratumoral vs. peripheral T cells in recurrent glioblastoma patients.

a A UMAP projection of the lymphoid compartment of the peripheral blood of recurrent GBM patients analyzed using scRNAseq (n = 56,444 cells from 5 GBM.rec and 8 GBM.pembro patients and two healthy donor controls). b The proportions of each lymphoid clusters in a in GBM.rec (black) and GBM.pembro (red) samples. c MSigDB Hallmark genesets showing significant overlap with the genes upregulated in the GBM.pembro PBMC (FDR values, two-sided fisher exact test). d T cell clone sizes as estimated by the TCRβ clones detected in the PBMCs (top) and TILs (bottom) using TRUST4. e, f, g Heatmaps showing the STARTRAC analysis of the shared TCRs among the PBMC clusters (e), across the PBMC and TIL clusters (f), and among the TIL clusters (g). Shared clones across two clusters indicates potential transition from one cluster to the other (non-directional). The higher the fraction of the shared TCR, the more likely that the T cells in the two clusters transition from one to the other. Source data are provided as a Source Data file or in Supp. Data 4 and 5. In all boxplots, the median is indicated by the line within the box and the 25th and 75th percentiles indicated by the lower and upper bounds of the box. The upper and lower lines above and below the boxes represent the whiskers. The gene set enrichment P values were calculated using two-sided Fisher exact test with FDR adjustment for multiple comparisons.

Fig. 4. Neo-aPD1 induces multiple immunosuppressive TAMs and monocyte populations.

a A UMAP projection of the myeloid compartment of the 40 patient samples analyzed using scRNAseq (n = 72,492 cells from 40 total patients: 14 GBM.new, 12 GBM.rec, and 14 GBM.pembro). b The expression levels of various marker genes of each cluster in a. c The proportion of the myeloid cells in each cluster across multiple tumor groups (blue: GBM.new, black: GBM.rec, red: GBM.pembro). d Differentially upregulated genes in GBM.pembro samples across the macrophage clusters. Colors the same as c. e MSigDB Hallmark genesets showing significant overlap with the union of the genes upregulated in the GBM.pembro group across all myeloid clusters (FDR values, two-sided fisher exact test). P values were calculated using a two-sided Wilcoxon rank-sum test unless otherwise noted. Source data are provided as a Source Data file or in Supp. Data 6. In all boxplots, the median is indicated by the line within the box and the 25th and 75th percentiles indicated by the lower and upper bounds of the box. The upper and lower lines above and below the boxes represent the whiskers. The gene set enrichment P values were calculated using two-sided Fisher exact test with FDR adjustment for multiple comparisons.

Fig. 5. Neo-aPD1 therapy increases the proportion of intratumoral migratory DCs.

a A UMAP projection of the DC compartment analyzed using scRNAseq (n = 2960 cells from 40 total patients: 14 GBM.new, 12 GBM.rec, and 14 GBM.pembro). b The normalized expression of the marker genes of different DC clusters. c Differentially upregulated genes in GBM.pembro samples in one or more DC clusters (blue: GBM.new, black: GBM.rec, and red: GBM.pembro). d Pseudotime projection of all DCs as analyzed by Monocle 2. Colors are as in c. e Heatmap showing the expression of DC/monocyte marker genes in each DC fate as defined in d. (top) The proportion of DCs from each tumor groups in each fate. Source data are provided in Supp. Data 9. In all boxplots, the median is indicated by the line within the box and the 25th and 75th percentiles indicated by the lower and upper bounds of the box. The upper and lower lines above and below the boxes represent the whiskers.

Fig. 6. Increased engagement of additional immune checkpoints may limit neoadjuvant anti-PD1-driven T cell activation.

a The total absolute (left) or normalized (right) number of CellChat-inferred interactions among the population in each tumor group (blue: GBM.new, black: GBM.rec, red: GBM.pembro, n = 156,766 cells from 40 total patients: 14 GBM.new, 12 GBM.rec, and 14 GBM.pembro). b Signaling pathways with significant differences in the overall information flow (sum of all normalized interaction) between GBM.rec and GBM.pembro. c The overall signaling patterns between CellChat curated pathways and our defined cell clusters. The interaction strength reflects the sum of all normalized interaction in each pathway. The top bar graph compares the sum of total interaction strength per cell type while the bar graph on the right summarize the interaction strength per pathway. Of note, myeloid cell interactions dominate the bulk of the interaction across all tumor groups. d, e The normalized expression of CXCL (d) and XCL pathway genes with significant expression in indicated tumor groups. The genes shown are known ligand receptor pairs (CXCL16-CXCR6, CXCL9/10/11-CXCR3, and XCL1/2-XCR1). f The inferred XCL1/2-XCR1 signaling network among the cell populations represented by the nodes. Color code as in e, size of node represents the size of the population, edge width represents the pathway specific interaction strength. g, h Same as f for CD86-CTLA4-CD28 signaling (g) or CD226-TIGIT-NECTIN (h) pathways. Both T cell checkpoints were inferred to be significantly engaged after neo-aPD1. Source data are provided as a Source Data file.