Circulating immunoregulatory B cell and autoreactive antibody profiles predict lack of toxicity to anti-PD-1 checkpoint inhibitor treatment in advanced melanoma

Abstract

Background: The majority of patients with melanoma develop immune-related adverse events (irAEs), and over half do not respond to anti-PD-1 (Programmed cell death protein 1) checkpoint inhibitor (CPI) immunotherapy. Accurate predictive biomarkers for both response to therapy and development of irAEs are currently lacking in clinical practice. Here, we conduct deep immunophenotyping of circulating regulatory and class-switched B cell and antibody immune states in patients with advanced stage III/IV melanoma prior to and longitudinally during CPI.

Methods: Mass cytometry, serum antibody isotyping and immuno-mass spectrometry proteome-wide screening evaluations to identify autoreactive antibodies were undertaken to profile circulating humoral immunity features in patients and healthy subjects and interrogate pretreatment B cell and antibody signatures that predict toxicity and response to anti-PD-1 therapy. In paired blood samples pretreatment and post-treatment, these humoral immune response profiles were monitored and correlated with the onset of toxicity.

Results: We found increased circulating IL-10+ (Interleukin-10+) plasmablasts and double-negative (DN) B cell frequencies, higher PD-L1 (programmed death ligand 1), TGFβ (Transforming Growth Factorβ) and CD95 expression by B cells, alongside higher IgG4 and IgE serum levels in patients with stage III/IV melanoma. This suggests enhanced B regulatory and Th2 (Thelper2)-driven responses in advanced disease. Increased baseline frequency of DN2 B cells, plasmablasts, and serum IgE, IgA and antibody autoreactivity were observed in patients who did not develop irAE. During treatment, higher IL-10+class-switched memory B cell, plasmablast and IgG1, IgG3 and IgE, alongside reduced IgG2, IgG4, IgA and IgM levels, were observed. A reduction in autoantibodies targeting tubulins was observed during treatment. Increased frequency of class-switched memory B cells predicted improved survival, while reduced transitional and PD-L1+TGFβ+ naive B cell frequencies and higher IgG4 and IgE levels predicted lower survival, on anti-PD-1 therapy.

Conclusions: Distinct B cell and antibody reactivities in patients with advanced melanoma share features with extrafollicular B cell responses in autoimmune diseases, may be protective from irAE and help predict outcomes to anti-PD-1.

Keywords: Autoimmune; B cell; Humoral; Immune Checkpoint Inhibitor; Skin Cancer.

Figure 1. Peripheral blood B cells and antibodies exhibit regulatory features in patient with advanced stage III and IV melanoma versus healthy volunteer blood. Mass cytometry (CyTOF, 25 markers) analyses of B cell peripheral blood populations in treatment-naive advanced patients with stage III and IV melanoma and healthy volunteers (HVs) (total n=74; HVs n=22, patients with stage III/IV melanoma n=52) (A–F). (A–B) Heatmap showing normalized median scaled expression of B cell markers (X axis) within 20 unsupervised clusters (Y axis), generated by the FlowSOM algorithm (A); and 20 B cell clusters visualized in UMAP plot (B). (C) Comparison of the relative abundance of differentially expressed unsupervised B cell clusters in HV and patients with melanoma stratified by stage of disease. (D) The 20 B cell clusters identified in (A) were merged into 10 defined B cell clusters and visualized in a bar chart showing the relative proportion of merged clusters per sample. (E) Mean scaled expression of PD-L1, TGFβ and CD95 in B cells from HVs and patients with melanoma stratified by stage. (F) In-depth review of marker expression per B cell population illustrating high IL-10 and CD95 in the plasmablast population. (G) Violin plots comparing serum antibody isotype titres (Luminex multiplex immunoassay) in HVs and patients with melanoma stratified by stage, showing enhanced levels of IgG4 in patient blood, and in stage IV sera compared with HV sera (Mann-Whitney U test) (based on non-parametric distribution). *p<0.05; **p<0.01. Breg, regulatory B cell; IL-10, Interleukin-10; PD-1, Programmed cell death protein 1; PD-L1, Programmed death ligand 1; TGFβ, Transforming Growth Factorβ; UMAP, Uniform Manifold Approximation and Projection.

Figure 2. Baseline B cell phenotypes predict immune-related anti-PD-1-related adverse events. CyTOF (25 markers) analyses of B cell peripheral blood populations in treatment-naive advanced patients with stage III and IV melanoma (total n=18: no toxicity n=11, high-grade (grade 3 and above) toxicity n=7). (A) Heatmap showing normalized median scaled expression of B cell markers (X-axis) within 25 unsupervised B cell clusters (Y-axis) generated with the FlowSOM algorithm. (B) Comparative UMAPs of B cell phenotypes in patients with melanoma stratified by onset of “no toxicity” versus “high-grade toxicity” to checkpoint blockade. Clusters highlighted by boxes relate to clusters 4, 5, 6, 7, 9, 10, 11, 12 and 23 identified by FlowSOM clustering in (A). All samples are randomly down-sampled to 30,000 B cells/sample. (C) Principal Component (PC) dimensionality reduction plot illustrating centered log-ratios (CLRs) of proportions of B cell clusters across samples of patient with melanoma taken prior to anti-PD-1 treatment. Clusters described in (A) identify CLRs of cluster proportions across samples belonging to two cohorts: patients with melanoma who experience “no toxicity” to anti-PD-1 versus those that experience “high-grade toxicity”. PC1 largely separates the two cohorts. (D) Relative abundance of unsupervised B cell clusters that are statistically significantly differentially expressed between “no toxicity” and “high-grade toxicity” groups at baseline. Plasmablast Breg and double negative (DN) clusters were enriched at baseline in the no toxicity patients. *p<0.05; **p<0.01. (E) Comparisons of mean scaled expression of B cell markers in conditions of “no toxicity” versus “high-grade toxicity” reveal the Breg-associated and exhaustion marker CD95 is enriched in the no toxicity group at baseline, while PD-1 is enriched in the high-grade toxicity cohort. *p<0.05. (F) UMAPs comparing the expression of exhaustion marker CD95 across various B cell clusters between “no toxicity” and “high-grade toxicity” patients. Breg, regulatory B cell; CyTOF, cytometry by time of flight; IL-10, Interleukin-10; PC, Principal Component; PD-1, Programmed cell death protein 1; PD-L1, Programmed death ligand 1; TGFβ, Transforming Growth Factorβ; UMAP, Uniform Manifold Approximation and Projection.

Figure 3. Non-classical isotypes and autoreactive antibodies at baseline predict lack of toxicity to anti-PD-1. (A) Violin plots comparing pretreatment IgG1, IgG2, IgG3, IgG4, IgM, IgE, IgA titres in serum of patient with melanoma (n=18) as a proportion of total immunoglobulin concentration (Luminex multi-plex immunoassay). Patients were stratified according to those who subsequently did not develop toxicity (no toxicity, n=11) to checkpoint inhibitor and those with high-grade (grade 3 and above) toxicity (n=7). Statistical analysis performed using Mann-Whitney U test. *p<0.05; **p<0.01. (B) Immuno-mass spectrometry (IMS) of serum immunoglobulins (IgG) revealed higher baseline autoreactivity in patients with melanoma treated with anti-PD-1 checkpoint blockade who did not develop immune-related toxicity. Proteins pulled down by antibodies in patient serum that are statistically significantly different in the presence or absence of toxicity as quantified by IMS peak area. *p<0.05. (C) Heatmaps of antigen cellular/extracellular location and summary of known relevance of autoantigen in disease settings. (D) Volcano plots summarizing differential putative autoantibody states in no toxicity versus toxicity states, stratified by organ system. Each dot represents a protein target. The Y axis indicates fold-change (FC) between “no toxicity” (positive FC values) and “toxicity” (negative FC values). Dots above the horizontal dotted line indicate significant difference in peak areas in the cohorts (p<0.05), highlighted as red dots. PD-1, Programmed cell death protein 1.

Figure 4. Anti-PD-1 therapy enhances memory B cells and plasmablast-like populations and alters the profile of circulating antibody isotypes on treatment. (A) Decrease in B cells seen after 12 weeks of treatment with anti-PD-1 therapy in stage III/IV melanoma. Paired analysis (left) and fold-change (right) comparison of B cells at baseline versus timepoint A versus timepoint B after 12 weeks of treatment (baseline pretreatment (n=32); timepoint A (after 6–8 weeks on-treatment, n=31) and timepoint B (after 12 weeks on-treatment, n=15) (*p<0.05). (B) Regulatory IL-10 and PD-L1 marker expression was elevated on-treatment with anti-PD-1 therapy in a cohort of patients with melanoma (n=32 at baseline, n=31 timepoint A, n=15 timepoint B). Mean scaled expression of each marker was extracted using the CATALYST pathway; graphs depict fold-change from baseline for timepoints A and B. *p<0.05, ***p<0.001. (C) Peripheral blood B cell phenotyping at baseline and on-treatment with anti-PD-1 therapy (n=78 samples). Heatmap showing normalized median scaled expression of B cell markers (X-axis) within 25 B cell clusters (Y-axis) generated with FlowSOM algorithm. (D) Two B cell subpopulations (identified using FlowSOM clustering algorithm (C)) increased on-treatment: class-switched memory IgG-IL-10+CD95hi (Cluster 13, p=0.035); and plasmablast-like IL-10+CD95 hi population (Cluster 23, p=0.042) (paired analysis performed using Wilcoxon rank test, *p<0.05). (E) Paired statistical comparison of circulating antibody isotypes was performed comparing fold-change on-treatment and antibody isotype as percentage of total circulating antibody titer at timepoint A on-treatment. Statistical analysis using the Wilcoxon paired rank test. *p<0.05, **p<0.01, ***p<0.001. IL-10, Interleukin-10; PD-1, Programmed cell death protein 1; PD-L1, Programmed death ligand 1.

Figure 5. Anti-PD-1 therapy alters serum autoreactivity in melanoma. Autoantibodies to putative antigens were identified using immuno-mass spectrometry (IMS) screening of patient serum against 13,028 human proteins. (A–B) IMS of serum immunoglobulins (IgG) (proteins pulled down by IgG antibodies in patient serum indicative of the presence of autoantibodies) revealed shifts in autoreactivity in melanoma patients treated with anti-PD-1 therapy. Fold change in peak area at baseline compared with on-treatment serum samples was evaluated for candidate proteins. Statistically significant fold changes are illustrated at timepoint A (A) and timepoint B (B). At timepoint A (A), significant increases in putative autoantibody titres to three proteins H4C1, TG and AT1B1, and decreases in autoantibody titres to three antigens, including four tubulin proteins (TUBB2A, TUBB4A, TUBB, TUBB4B, ACTC1, DLST and COX4I1), were detected. At timepoint B (B), autoantibodies were increased for three proteins, HSPA5, CD5L and HSP90AA1. *p<0.05, **p<0.01. (C) Heatmaps of serum autoreactivity and immune toxicity to anti-PD-1 shown for each patient. Putative autoantibodies that show a significant fold change in peak area (from A) were analyzed in relation to organ-specific toxicity. Top heatmap depicts the incidence of organ-specific toxicity (yellow: patients who did not experience toxicity; green: specific toxicity event; gray: no specific toxicity event). Bottom heatmap illustrates whether fold change of each protein is increasing (red), decreasing (black) or not changing (gray) on treatment. Each column represents an individual patient. (D) Fold change in putative autoantibodies is stratified by the presence or absence of immune toxicity on treatment. Autoreactivity against H4C1 was differentially increased in patients who experienced toxicity compared with those who do not develop toxicity. Wilcoxon signed rank test. *p<0.05. PD-1, Programmed cell death protein 1.

Figure 6. Regulatory B cell subsets and antibodies predict lower overall survival in patients treated with anti-PD-1 therapy. (A) Kaplan-Meier curves for B cell phenotypes that significantly predict overall survival, extracted from unsupervised clusters identified in figure 1. *p<0.05, Gehan-Breslow-Wilcoxon statistical test. (B) Kaplan-Meier curves for B cell phenotypes that significantly predict overall survival, extracted from merged B cell clusters identified in figure 1. *p<0.05, Gehan-Breslow-Wilcoxon statistical test. (C) Kaplan-Meier curves illustrating correlation between antibody isotype and overall survival. Relative enrichment of serum IgG4 and IgE in patients with melanoma prior to checkpoint immunotherapy treatment was negatively associated with overall survival in those treated with anti-PD-1 therapy. IL-10, Interleukin-10; PD-1, Programmed cell death protein 1; PD-L1, Programmed death ligand 1.