Anti-PD-1 immunotherapy with nivolumab or pembrolizumab is given over months or even years1 with the intention of enhancing the immune response to cancer. The effects of long-term disruption of the PD-1 pathway on immune responses to other antigens including humoral responses to vaccines that involve PD-1-expressing Tfh remain poorly understood. To address this question, we enrolled adults with renal cell carcinoma or urothelial carcinoma who were receiving immunotherapy (Cohort 1; Supplementary Table 1) and were due to receive seasonal inactivated influenza vaccine. Participants were divided into two groups: non-anti-PD-1-based therapy (n = 10, median age = 65) or anti-PD-1-based therapy (n = 29, median age = 71.5). Given the importance of Programmed Death-Ligand 2 signaling in the humoral response26,27, we considered participants who received anti-Programmed Death-Ligand 1 as members of the non-anti-PD-1 group. In a second, independently generated cohort at a different institution, we enrolled adults with melanoma receiving immunotherapy (n = 30, median age = 61.5) and healthy adults not receiving immunotherapy (n = 27, median age = 33) (Cohort 2; Supplementary Table 2). Participants received influenza vaccination on the same day as immunotherapy infusion (median of cycle 12 of anti-PD-1 for Cohort 1 and cycle 7 for Cohort 2; Extended Data Fig. 1a,b). Blood was drawn on the day of vaccination (baseline), 1 week later and again 3–6 weeks after vaccination (late) (Extended Data Fig. 1a), as has been done previously21,22,25,28.

In healthy adults, CXCR5+ PD-1+ cTfh expressing the activation markers Inducible Costimulator (ICOS, CD278) and CD38 expand 1 week after influenza vaccination and this population contains influenza-specific CD4 T cells25. We therefore first asked whether anti-PD-1 immunotherapy alters the cTfh response to the vaccine. To avoid bias of using PD-1 itself in detection of cTfh due to treatment with therapeutic anti-PD-1 antibodies, we used a broader definition of non-naive CXCR5+ CD4 T cells that coexpressed ICOS and CD38 (ICOS+CD38+ cTfh) to identify responding cTfh in this study (Extended Data Fig. 1c–e). Following vaccination in Cohort 1, there was a 2.6-fold-induction of ICOS+CD38+ cTfh in patients with cancer receiving anti-PD-1 compared with a 1.1-fold-change in patients with cancer on therapies other than anti-PD-1 (P = 0.013, Wilcoxon test; Fig. 1b), demonstrating that anti-PD-1 therapy is associated with augmented cTfh responses to influenza vaccination.

a, cTfh were profiled for expression of ICOS and CD38 at baseline and at 1 week after influenza vaccine in Cohort 1. b, ICOS+CD38+ cTfh fold-change at 1 week in Cohort 1 (P = 0.01, two-sided Wilcoxon test; anti-PD-1 (n = 18) versus non-anti-PD-1 (n = 7)). c, Patients with melanoma were recruited in Cohort 2 and profiled following influenza vaccination. cTfh responses shown at baseline and 1 week. d, ICOS+CD38+ cTfh fold-change at 1 week compared with baseline in Cohort 2 (P = 0.05, two-sided Wilcoxon test; anti-PD-1 (n = 20) versus healthy (n = 27)). e, Plasmablast responses after influenza vaccination in Cohort 2. f, Kendall correlation between plasmablasts and ICOS+CD38+ cTfh at 1 week for Cohorts 1 and 2. g, Plasma CXCL13 over time (P = 2 × 10−3 for anti-PD-1 at 1 week compared with baseline, two-way ANOVA with Tukey’s post-test). Data are presented as mean value (point) ±s.e.m. (shaded area). h, Plasma CXCL13 fold-change at 1 week in Cohort 2 (P = 2.5 × 10−3, two-sided t-test; anti-PD-1 (n = 20) versus healthy (n = 27)). HC, healthy controls; TKI, tyrosine kinase inhibitor.

Source data

We examined an independent cohort of patients with melanoma receiving anti-PD-1 immunotherapy (Cohort 2; Supplementary Table 1). Again, anti-PD-1 therapy was associated with more robust increases in cTfh responses compared with healthy control participants 1 week after vaccination (P = 0.05, Wilcoxon test), with responses returning to baseline by the late time point (Fig. 1c,d and Extended Data Fig. 1f–h). We also confirmed that using PD-1 in the definition of cTfh resulted in similar observations (Extended Data Fig. 1i–k). The increase in the cTfh response was not associated with more recent initiation of anti-PD-1 therapy (Extended Data Fig. 1l) and was unlikely to be due to age-related differences between participants (Extended Data Fig. 1m,n). Further subsetting based on CCR6 and CXCR3, as described previously23, did not identify differences between treatment groups (Extended Data Fig. 1o). Finally, we considered other regulators of the B cell response. In particular, T follicular regulatory cells (Tfr) phenotypically resemble Tfh but express Foxp3 and can regulate Tfh–B cell interactions29. There was a greater decrease in the frequency of circulating Tfr (cTfr) in the anti-PD-1-treated participants than in the healthy adults in Cohort 2 (Extended Data Fig. 1p–r). However, this change could reflect dilution of cTfr by the ICOS+CD38+ CXCR5+ cTfh or perhaps be due to altered kinetics. Given the structure of the study, it was not possible to assess cTfh and cTfr responses at different time points. Nonetheless, these data indicate that anti-PD-1 treatment was associated with an augmented numerical response of cTfh in a subset of patients 1 week after influenza vaccination.

We next asked if the B cell response to vaccination was affected by anti-PD-1 therapy. Although the plasmablast frequency was not higher in the anti-PD-1 group following vaccination, there was a subset of anti-PD-1 participants with robust induction of plasmablasts 1 week after vaccination, (Fig. 1e and Extended Data Fig. 2a–d). Correlations have been observed previously between plasmablast and cTfh responses to influenza vaccination22,25. Here, the plasmablast response correlated with ICOS+CD38+ cTfh responses 1 week after vaccination in the anti-PD-1 group in both cohorts (Fig. 1f). These data suggest that anti-PD-1 therapy alters Tfh–B cell interactions, likely in secondary lymphoid tissues, resulting in higher frequencies of both cTfh and plasmablasts in the peripheral blood following vaccination.

One biomarker of GC activity in lymphoid tissue is plasma CXCL13 (ref. 30). We observed weak induction of plasma CXCL13 in healthy adults 1 week after influenza vaccination (Fig. 1g), consistent with previous studies showing that influenza vaccination induced only modest changes in this chemokine detectable in blood thought to be related to the relatively weak immune response to this vaccine30. In contrast, however, anti-PD-1-treated participants had substantial induction of plasma CXCL13 1 week after vaccine compared with baseline in both cohorts (Fig. 1g,h and Extended Data Fig. 2e–g). Although this CXCL13 in the blood did not correlate strongly with the cTfh or plasmablast response, there was a correlation with later antibody responses (Extended Data Fig. 2h). Collectively, these data reveal a robust enhancement of vaccine-induced cTfh, plasmablast and likely GC activity in the presence of anti-PD-1 therapy.

Humoral responses to influenza vaccination are typically assessed by hemagglutinin inhibitory (HAI) antibody titers, which are a correlate of protection following vaccination31. Thus, we investigated whether anti-PD-1 therapy impacted the induction of HAI titers. We focused on Cohort 2 because of the larger number of individuals available for analysis.

We first examined the fold-change in the HAI antibody titer at the 3–6-week time point compared with the baseline titer. Indeed, across all three strains of influenza included in the vaccine, neutralizing antibody titers increased by a median of 4-fold in the presence of anti-PD-1 compared with 2-fold in the absence of anti-PD-1 (Fig. 2a and Extended Data Fig. 3a). Most anti-PD-1-treated adults had antibody titers of >1:40 following vaccination and were thus seroprotected, similar to healthy adults, indicating no clinical deficiency in the outcome of vaccination in this setting (Fig. 2b). The absolute HAI titer was similar between the anti-PD-1 and non-anti-PD-1 at the late time point, and the difference in fold-induction reflected slightly lower baseline HAI titers in the anti-PD-1 group than in the non-anti-PD-1 group (Fig. 2c and Extended Data Fig. 3b). These data are consistent with a previous study showing increases in influenza-specific antibodies in patients vaccinated while on anti-PD-1 therapy15. In Cohort 1, there were similar trends, but these differences in HAI titers in Cohort 1 did not reach statistical significance, perhaps because of a smaller number of participants, more varied treatment regimens or the effects of previous cancer therapy (Extended Data Fig. 3c,d). In this setting, correlations between cTfh or B cell responses and HAI titers were not apparent (Extended Data Fig. 3e). The differences in antibody observed in Cohort 2 were unlikely to be due to the differences in age between the two groups, since independent analyses did not reveal age-associated differences in influenza-specific neutralizing antibodies at baseline (Extended Data Fig. 3f). Together, these data show that the increases in cTfh, plasmablast and CXCL13 responses were associated with quantitatively increased antibody responses in the setting of anti-PD-1 compared with participants not receiving anti-PD-1 therapy.

a, HAI as fold-change at the late time point compared with baseline for each strain. Nominal P values from two-sided t-test comparisons are shown (anti-PD-1 (n = 24) versus healthy (n = 27 for H1N1, n = 15 for H3N2 and FluB)). b, Seroprotection for each strain, shown as the proportion of participants who achieved an HAI titer of 1:40 or higher 21–42 d after vaccination. c, H1 inhibition titers determined for the H1N1 (*P = 0.01; anti-PD-1 baseline versus anti-PD-1 late; two-way ANOVA with Tukey’s), H3N2 (*P = 0.03; anti-PD-1 baseline versus anti-PD-1 late; two-way ANOVA with Tukey’s) and influenza B strains (***P = 7.1 × 10−4; healthy baseline versus healthy late; two-way ANOVA with Tukey’s). Data are presented as mean value (point) ± s.e.m. (shaded area). d, Proportion of anti-H1 IgG1 antibodies galactosylated (***P = 1.8 × 10−12, two-way ANOVA with Tukey’s; anti-PD-1 versus healthy at baseline). Data are presented as mean value (point) ± s.e.m. (shaded area). e, Proportion of anti-H1 antibodies galactosylated at baseline (P = 3.4 × 10−9, two-sided t-test; anti-PD-1 (n = 30) versus healthy (n = 27)). f, Sialylation for anti-H1 IgG1 antibodies (**P = 7.1 × 10−3, two-way ANOVA with Tukey’s; comparison of anti-PD-1 versus healthy at 1 week). Data are presented as mean value (point) ± s.e.m. (shaded area). g, Proportion of anti-H1 antibodies sialylated at baseline (P = 0.02, two-sided t-test; anti-PD-1 (n = 30) versus healthy (n = 27)). h, Affinity determined as EC50/Kd (*P = 0.02, two-way ANOVA with Tukey’s; anti-PD-1 at baseline versus late). Data are presented as mean value (point) ± s.e.m. (shaded area).

Source data

Antibodies provide protection by being present at sufficient quantities and by undergoing qualitative improvements during the T cell-dependent phase of induction in lymphoid tissues. However, because the data above revealed a quantitative impact of anti-PD-1 on induction of antibody following vaccination, but endpoint HAI titers were similar, we next wanted to investigate potential qualitative effects of anti-PD-1 treatment on antibody responses. For example, antibody effector function is dictated by two important components of the antibody Fc fragment, both the amino acid sequence (for example, IgG subclass IgG1–4) and the composition of the conserved N-linked glycan at asparagine position 297 (ref. 32). We previously demonstrated that changes to anti-hemagglutinin (HA) glycoforms, in particular increased total sialylation of influenza-specific IgG, drove B cell affinity selection and determined the efficacy of influenza vaccination33. We therefore investigated how disruption of the PD-1 pathway impacted antibody glycosylation and subsequent affinity. We focused on the antibody response to hemagglutinin (H1), the primary target of the antibody response in vaccinated individuals34. IgG subclass and glycan distribution of antigen-specific IgG were evaluated using mass spectrometry33. Overall, anti-H1 antibodies were enriched for the IgG1 subclass, with less IgG2 and IgG3/4 in the setting of anti-PD-1 compared with controls (Extended Data Fig. 4a; P = 9 × 10−3, two-way analysis of variance (ANOVA) with Sidak’s post-test). We next investigated how anti-PD-1 treatment impacted antibody glycosylation. Afucosylation of antibodies results in enhanced antibody-dependent cytotoxicity32, but was not affected by anti-PD-1 treatment (Extended Data Fig. 4b). Sialylation and galactosylation are two additional Fc glycosylation events that can regulate antibody function. Sialylated antibodies conferred improved protection to influenza in vivo33 and galactosylation is a prerequisite for sialylation32. Indeed, sialylation and galactosylation of anti-H1 antibodies were correlated for both anti-PD-1-treated and nontreated groups (Extended Data Fig. 4c). However, the anti-PD-1-treated patients had lower total galactosylation (Fig. 2d,e) and sialylation (Fig. 2f,g) of anti-H1 antibodies than non-anti-PD-1-treated participants at baseline and after vaccination, and these observations were consistent across all immunoglobulin subclasses (Extended Data Fig. 4d–f). In Cohort 1, anti-PD-1-treated participants had lower galactosylation and sialylation of anti-H1 antibodies at baseline (Extended Data Fig. 4g–j). In previous studies, we found similar changes in glycosylation between antigen-specific and total antibodies33,35,36. We therefore focused here on anti-H1 antibodies because these are a major mechanism of vaccine-induced protection. However, we observed similar changes in glycosylation patterns on anti-H1 antibodies of all antibody subclasses examined, suggesting a broad impact on antibody glycosylation (Extended Data Fig. 4d,f). Altogether, these data identify an effect of anti-PD-1 therapy on sialylation of anti-H1 antibodies at baseline and after vaccination, effects that might be predicted to impact the quality of influenza-specific antibodies.

To test whether altered antibody glycosylation in the setting of anti-PD-1 treatment had functional consequences, we next examined antibody affinity. Sialylation of antigen-specific IgG and immune complexes contributes to affinity maturation through FcγRIIb-mediated modulation of GC B cell selection favoring higher affinity clones33. Thus, lower sialylation in the context of anti-PD-1 was predicted to impact subsequent antibody affinity. Indeed, baseline antibody affinity was lower in patients on anti-PD-1 compared with healthy controls (Fig. 2h and Extended Data Fig. 4k). Using a second assay that estimates affinity based on binding to low- versus high-density antigen33, we confirmed that anti-PD-1-treated participants had lower affinity IgG1 at baseline (Extended Data Fig. 4l–n). Collectively, these data identify both quantitative and qualitative changes in antibody responses to influenza vaccine associated with blockade of PD-1. Because some of these changes were present before yearly vaccination, these data suggest perhaps an underlying impact of anti-PD-1 therapy on influenza-specific immune memory, in addition to the effects on the responses provoked by the vaccination studied here.

Given these changes in antibody, we next investigated whether anti-PD-1 treatment was associated with changes in B cell subsets that responded to vaccination, including ASCs (gated as CD19+IgD−CD71+CD20lo) and activated B cells (ABCs; gated as CD19+IgD−CD71+CD20hi)37. Even before vaccination, anti-PD-1 therapy was associated with a higher frequency of circulating ASCs and reciprocally reduced ABC frequencies (Extended Data Fig. 4o,p). However, whereas ABCs only weakly correlated with ICOS+CD38+ cTfh, at baseline and after vaccination, ASC frequencies were positively correlated with ICOS+CD38+ Tfh in anti-PD-1 patients before and after vaccination (Extended Data Fig. 4q,r). Taken together, these data indicate that anti-PD-1 treatment is associated with alterations in B cell subsets at baseline and in vaccine-induced antibody quantity and quality. Moreover, these alterations before vaccination suggest that treatment with anti-PD-1 may alter the baseline Tfh–B cell and humoral immune set point in these individuals.

To further interrogate the effects of anti-PD-1 on immune responses to influenza vaccination, we performed transcriptional profiling on sorted ICOS+CD38+ cTfh, ABCs, ASCs or naïve B cells (gated as CD19+IgD+CD27lo) from participants in Cohort 2. Sample distribution in t-distributed stochastic neighbor embedding (t-SNE) space was driven primarily by cell subset and minimally by treatment or timing relative to vaccination (Extended Data Fig. 5a–e). However, at 1 week post vaccination, transcriptional differences were readily apparent in ICOS+CD38+ cTfh from anti-PD-1-treated adults compared with ICOS+CD38+ cTfh from healthy adults including upregulation of genes such as MKI67 and ESPL1, indicating recent proliferation38, and lower expression of genes including IFI44L, OASL and TNFRSF1B (Fig. 3a and Supplementary Table 5), suggesting reduced response to interferon signaling39,40. A similar upregulation of ESPL1, MKI67 and other cell cycle genes was observed in anti-PD-1-induced ‘burned-out’ CD8 T cells in patients with cancer41. Gene ontology (GO) term enrichment highlighted proliferation and cell cycle in the transcriptional signatures of ICOS+CD38+ cTfh from anti-PD-1-treated participants following vaccination compared with leukocyte activation, migration, as well as cytokine production and signaling in the control group (Fig. 3b and Extended Data Fig. 5f). Indeed, Ki67 protein expression 1 week after vaccination correlated with the fold-change in the ICOS+CD38+ cTfh (Extended Data Fig. 5g), supporting the notion that anti-PD-1 therapy was associated with an enhanced proliferative cTfh response after immunization.

a, Top and bottom 50 genes from differential expression analysis of ICOS+CD38+ cTfh at 1 week after vaccination. b, GO analysis shown for genes enriched in anti-PD-1 (peach) and those enriched in healthy (blue) for ICOS+CD38+ cTfh. P values from hypergeometric test with Benjamini–Hochberg P value correction. c, Top and bottom 50 genes from differential expression analysis of ASCs at 1 week after vaccination. d, GO analysis shown for genes enriched in anti-PD-1 (peach) and those enriched in healthy (blue) for ASCs. P values from hypergeometric test with Benjamini–Hochberg P value correction. e, GSEA for MSigDB Hallmark gene sets used to compare ICOS+CD38+ cTfh at 1 week for anti-PD-1 and healthy. Positive enrichment scores denote enrichment towards the anti-PD-1 cohort. f, GSEA for Hallmark gene sets for ASCs at 1 week for anti-PD-1 and healthy. Positive enrichment scores denote enrichment towards the anti-PD-1 cohort. g, GSEA shown for the Mitotic Spindle gene set for ICOS+CD38+ cTfh, ASCs and ABCs at 1 week after vaccine. Normalized enrichment score (NES) shown. h, GSEA for the G2M checkpoint gene set with NES is shown. i, GSEA plot shown for the IL-2/STAT5 pathway in the MSigDB Hallmark database for ICOS+CD38+ cTfh. FDR, false discovery rate.

Source data

Similarly, ASCs from vaccinated anti-PD-1-treated participants upregulated genes such as AURKB, BUB1 and ESPL1, consistent with greater proliferation relative to ASCs from healthy adults (Fig. 3c and Extended Data Fig. 5h), and had terms for ‘cell cycle’ and ‘mitotic cell division’ by GO (Fig. 3d and Extended Data Fig. 5i). ASCs from the anti-PD-1 group also had higher expression of CCL2 and IL18RAP, but lower expression of IFNGR2, DUSP1 and KCNA3. Moreover, upregulation of ARID1A in ICOS+CD38+ cTfh and SMARCD1 in ASCs from the anti-PD-1 group also suggested a potential impact on SWI/SNF complex use in these cells responding to vaccination in the absence of normal PD-1 signaling. These proliferation-related genes were not observed before vaccination (Extended Data Fig. 5i,j).

We next performed gene set enrichment analysis (GSEA)42 for both cTfh and ASCs to identify pathways altered due to anti-PD-1 at 1 week after vaccination. For the ICOS+CD38+ cTfh, we identified four gene sets that were enriched in the anti-PD-1 participants, all of which were associated with proliferation, whereas 27 gene sets were enriched in the controls (Fig. 3e). Indeed, enrichments of gene sets associated with proliferation were also identified in the ASC and ABC subsets of anti-PD-1 participants 1 week after vaccination (Fig. 3f–h and Extended Data Fig. 6a,b). These data indicate that anti-PD-1 drove greater activation and proliferation of cTfh and B cell populations during the vaccine response.

In addition to the genes and pathways upregulated in the presence of anti-PD-1, genes and pathways downregulated or that failed to be induced following influenza vaccination in anti-PD-1 patients were also of interest. Indeed, anti-PD-1 was associated with relative reduction in several pathways of direct relevance for cTfh biology in vaccine-responding ICOS+CD38+ cTfh. First, the STAT5/Blimp-1 axis is a negative regulator of Tfh differentiation and function, in part through antagonism of the Tfh-driving transcription factor Bcl6 (refs. 43,44). Here, anti-PD-1 treatment was associated with downregulation of IL-2/STAT5 signaling in activated cTfh (Fig. 3i) consistent with the enhanced cTfh response in a subset of anti-PD-1 patients. Moreover, enrichment for the IL-2/STAT5 gene set was negatively correlated with ICOS+CD38+ cTfh frequency in anti-PD-1 patients but not healthy controls, consistent with a regulatory effect of this pathway on Tfh activation and/or expansion that is revealed by anti-PD-1 treatment (Extended Data Fig. 6c). Other Tfh-regulating pathways, such as TGF-β signaling and apoptosis, were also downregulated in the setting of anti-PD-1 treatment (Fig. 3e). Downregulation of regulatory signaling pathways such as IL-2/STAT5 signaling that normally impede Tfh activity in the setting of anti-PD-1 may help explain the greater influenza vaccine-induced cTfh and antibody responses in the setting of anti-PD-1 (Fig. 2).

A second relevant transcriptional imprint of anti-PD-1 treatment was downregulation of genes involved in the TNF/NFkB pathway. These changes were of interest at least in part because TNF and NFkB signaling are regulators of cTfh biology45. In the setting of anti-PD-1 treatment, signatures of TNF/NFkB-related genes in ICOS+CD38+ cTfh were highly biased to the control participants (Fig. 3e), suggesting that anti-PD-1 treatment impaired efficient use of TNF/NFkB signaling in vaccine-responding activated cTfh. A similar picture emerged for ASCs (Fig. 3f), suggesting a coordinated set of biological pathways impacted by anti-PD-1 therapy for both cTfh and ASCs. Thus, these data indicate that anti-PD-1 may alter the humoral, cTfh and likely GC responses by dysregulating control of proliferation and altered sensing of key cytokine circuits needed for proper Tfh control of humoral immunity.

A major limitation of current checkpoint blockade in cancer immunotherapy is the development of irAEs12,13. Although the underlying mechanisms have been unclear, one possibility is dysregulated GC-dependent immune responses including CD4 T cells (for example, Tfh) and antibody responses46,47. Thus, we hypothesized that differences in Tfh biology during anti-PD-1 therapy may offer insights into the mechanisms of irAE.

To test this idea, we focused on participants in Cohort 2 where anti-PD-1 was predominantly used as monotherapy, and considered irAEs irrespective of timing to influenza vaccination (Extended Data Fig. 7a and Supplementary Table 6). We considered whether messenger RNA profiling of ICOS+CD38+ cTfh at baseline could identify characteristics of those who had an irAE versus those who did not have any irAEs. Patients with irAEs had slightly more transcripts for genes associated with cellular activation, such as PDCD1 and ICOS, relative to patients who did not have irAEs, although these differences were not statistically significant (Extended Data Fig. 7b,c). However, we observed higher expression of ICOS protein and PD-1 protein, using αIgG4 as a proxy for PD-1 (ref. 2), in anti-PD-1-treated adults with irAEs compared with anti-PD-1-treated adults who did not have irAEs (Extended Data Fig. 7d–f). Transcriptional analysis of ICOS+CD38+ cTfh revealed 18 genes upregulated in activated cTfh from patients with irAEs, including KIF2C, MKI67 and BIRC5, compared with patients without irAEs (Fig. 4a, Extended Data Fig. 7g and Supplementary Table 7). Consistent with these proliferation-associated genes, GO analysis also revealed enrichment for terms related to cell cycle and cellular activation in patients with irAEs (Extended Data Fig. 7h). Other genes of potential interest were also revealed, including the cytokine IL32, the splicing factor XAB2 and CD52, the target of alemtuzumab used to treat chronic lymphocytic leukeamia and multiple sclerosis48, which were all more highly expressed in activated cTfh from anti-PD-1-treated adults with irAEs. In contrast, higher ITGA5, CD82 and FOS were found in cTfh from anti-PD-1-treated adults without irAEs. These data indicated that, before influenza vaccination, ICOS+CD38+ cTfh in anti-PD-1-treated adults with irAEs were more activated than in anti-PD-1-treated adults without irAEs and anti-PD-1 therapy may engage additional biologically relevant pathways.

a, Anti-PD-1 participants in Cohort 2 were subgrouped based on irAEs (Y, yes; N, no). Differential expression analysis was performed on ICOS+CD38+ cTfh at baseline, and all genes with Padj < 0.05 are shown. b, GSEA for Hallmark gene sets in the setting of irAE. c. Fold-change in cTfh responses after influenza vaccine, calculated as the frequency at 1 week relative to baseline (P = 0.06, two-sided t-test; irAE ‘Yes’ (n = 12) versus ‘No’ (n = 7)).

Source data

GSEA identified enrichment of G2M checkpoint and E2F targets in ICOS+CD38+ cTfh at baseline from participants with irAEs compared with those without irAEs (Fig. 4b), further supporting the idea that ICOS+CD38+ cTfh from patients with irAEs were more activated, even before influenza vaccination, compared with patients without irAEs. This observation is consistent with other reports that identified T cell activation as a predictor of irAEs49. In addition to proliferation-associated pathways, multiple other irAE-associated biologic pathway changes in cTfh were revealed, including downregulation of IL-2/STAT5, TNF/NFkB, IL6/JAK/STAT, IFN-γ and TGF-β signaling, consistent with the results above. Taken together, these data support the notion of an anti-PD-1-mediated rewiring of cTfh biology in patients with irAEs that includes increased activation and proliferation concomitant with blunting of cytokine pathway signaling.

These observations provoked the hypothesis that the post-influenza vaccine cTfh response might distinguish anti-PD-1 patients with irAEs from those without irAEs because of an underlying change in the global set point of Tfh regulation. To test this idea, we examined the magnitude of the ICOS+CD38+ cTfh increase 1 week followi