Humoral IgG1 responses to tumor antigens underpin clinical outcomes in immune checkpoint blockade

PCs are enriched in clinical responders to PD-1 blockade

To test the hypothesis that B cells or PCs have a key role during immunotherapy and treatment response, we examined specimens from eight sets of independent clinical trial cohorts: two newly generated datasets from our own investigator-initiated trials in patients with HCC treated with neoadjuvant PD-1 blockade, with or without radiation, for discovery and validation, respectively. We also analyzed published datasets of a combination of PD-1, PD-L1, CTLA-4 and VEGF-A blockade, including large cohorts such as IMbrave150 and The Cancer Genome Atlas (TCGA) (Fig. 1a). No sex-associated differences were observed in any of the analyzed outcomes, consistent with the balanced representation of male and female patients in the cohort.

First, we profiled approximately 30,000 B cells and PCs derived from 1.2 million single-cell transcriptomes of the tumor microenvironment, uninvolved adjacent liver and the draining lymph nodes resected from 38 patients with early-stage HCC of the discovery cohort¹⁸. Of these, 27 patients received neoadjuvant ICB in the form of an anti-PD-1 blocking antibody, and 11 were untreated controls (Extended Data Fig. 1a,b). We resolved six distinct B cell states (naive, memory, activated, class-switched, class-unswitched and atypical) alongside plasmablasts and three PC populations with discrete transcriptional programs (Fig. 1b). Naive B cells were characterized by the expression of IGHD and CD19 and were decreased in tumor compared to adjacent tissue (false discovery rate (FDR) < 0.05); memory B cells showed MYC, CD69 and NR4A1 associated with an activation phenotype and with tumor enrichment, whereas PCs were characterized by MZB1 and JCHAIN (Fig. 1c). Naive B cells dominated the B cell fraction of the immune cells across the tissue compartments, followed by memory B cells, relative to total B cells and PCs (Extended Data Fig. 1c–f). As expected, canonical and cluster-specific genes were expressed in more than 70% of cells (Extended Data Fig. 1g,h). Pathological responses to neoadjuvant PD-1 were defined as more than 50% necrosis of tumor at the time of surgery, followed by differential cell abundance showing an enrichment of all PC phenotypes in the tumor among ICB responders (FDR < 0.05), which was not as pronounced in adjacent uninvolved liver (Fig. 1d,e). Conversely, non-responders were enriched in unswitched memory B cells in tumor (FDR < 0.05) but less so in normal tissues (Fig. 1d,e).

Preexisting IgG1 PCs associated with clinical response are expanded after PD-1 blockade

We hypothesized that skewing of IgG1-producing PCs is linked to clinical response, given their overrepresentation in single-cell data from responders (Extended Data Fig. 1i). To test this, we used complementary bulk RNA-seq of paired pretreatment biopsies and posttreatment resected tumors from the discovery cohort. Principal component analysis showed that gene expression profiles were markedly different between responders and non-responders (Fig. 1f,g and Extended Data Fig. 1j–k).

In posttreatment tumors, IGHG1 emerged as one of the top upregulated genes. Its expression was already elevated in pretreatment samples from responders and increased significantly after therapy, whereas it remained unchanged or decreased in non-responders (FDR < 0.01; Fig. 1h,i). Projection of the single-cell signatures onto bulk data emphasized the increase of PCs in responders (FDR < 0.01) (Fig. 1j). Thus, these results suggest that skewing toward an IgG1 signature may exist at baseline (pretreatment) in anti-PD-1 responders, which was significantly amplified after ICB treatment and was highly associated with response.

ICB responders have clonally expanded IgG1 PCs trafficking between the tumor and draining lymph node

Using a combination of BCR-seq with scRNA-seq, we investigated clonal expansion and immunoglobulin isotype (Fig. 2a). We found that plasmablasts and PCs were largely of the IgG1 and IgG2 subclass (FDR < 0.01), whereas other memory and naive B cells were an admixture of IgM, IgD, IgA and, to a much lesser degree, the other IgG subclasses (Fig. 2b, top). When stratifying by ICB response, IgG1 and IgG2 PCs were almost exclusive to responders, whereas IgM, IgA and IgD dominated in non-responders (Fig. 2b, bottom, and Extended Data Fig. 2a,b). Irrespective of ICB response, 26% of total B cells and PCs were expanded (more than one cell per clone) (Extended Data Fig. 2d,e). Responders exhibited marked intratumoral IgG1⁺ PC and plasmablast expansion, unlike non-responders, who showed memory B cell expansion (Fig. 2c–e and Extended Data Fig. 2c,f,g). IGHG1, MZB1, JCHAIN and XBP1 were among the top genes enriched in clonally expanded cells from responders (Fig. 2f). In responders, clonally expanded cells were enriched for IgG1⁺ PCs, whereas non-expanded cells expressed MS4A1 (encoding CD20) (Fig. 2g). Together, these data suggest a tumor-specific clonal expansion of IgG1 antibody-producing cells in ICB responders.

Next, we investigated whether B cell differentiation and clonal expansion occur at the primary tumor site or in draining lymph nodes. In lymph nodes, IgM was the dominant isotype (Extended Data Fig. 2h). Although unique CDR3 sequence overlap among lymph node, tumor and adjacent liver samples was limited within individual patients (Extended Data Fig. 2i), clonal tracking revealed significantly larger IgG1⁺ and IgG2⁺ tumor PC clones in responders compared to non-responders (FDR < 0.01; Fig. 2h). The shared CDR3 clonotypes across cellular compartments (lymph node, naive, B memory and plasmablast/PC) revealed that clonally expanded cells with the same CDR3 sequence could be found at the lymph node as well as the tumor site (Fig. 2i and Supplementary Table 1), thus suggesting trafficking of these expanded clones between the tumor and the draining lymph node. Further inspection of the bulk sequencing data validated the expansion of IGHG1 phenotype, including evidence of the same CDR3 barcodes before and after treatment in both responders and non-responders (Extended Data Fig. 2j).

Tumor microenvironment spatial distribution reveals PC-driven immunity in responders and memory B cell dysfunction in poor outcomes

Using multiplex immunohistochemistry (mIHC) and computational tools¹⁹, we examined the spatial distribution of the B cells (CD20⁺), PCs (MZB1⁺) and other immune cells in the tumor and adjacent liver in 17 mIHC biopsies (6 responders and 11 non-responders) from our discovery cohort. We observed that among anti-PD-1 responders, the MZB1⁺ PCs were highly infiltrative throughout the tumor parenchyma compared to the PCs in non-responders (Fig. 3a,b). Next, we used a radial binning approach to define cell communities or immune aggregates in an unsupervised fashion (Fig. 3c). Here, responders showed enrichment of CD3⁺CD8⁺ T cells, CD68⁺ macrophages and MZB1⁺ PCs. Conversely, CD20⁺ B cells were found within the lymphoid aggregates or within the stromal compartment of the tumor rather than admixed with tumor parenchyma, reinforcing PC expansion as a hallmark of effective ICB response (Fig. 3d).

Next, we applied an single-cell variational inference-based autoencoder to integrate the spatial transcriptome cohort³ and then used CellCharter to identify spatial clusters constituted by multiple cell types (Fig. 3e,f). In ICB responders, we saw a marked enrichment of IgG1⁺ PCs confined to an immune-rich T cell/B cell/vascular fibroblast hybrid niche, with significant upregulation of IL-4 and IL-13 signaling (Fig. 3g,h). Pathway analysis of these clusters revealed a pro-BCR/TCR signaling hub, marked by MZB1, IL2RG, FCER2, PRDM1 and CD79A, consistent with a highly immunogenic, PC-driven microenvironment (Fig. 3h,i). By contrast, non-responders harbored focal accumulations of CD27⁺ memory and dysfunctional B cells within fibroinflammatory stromal regions. These niches were accompanied by regulatory T, T helper 17 (T_H17) and monocyte signatures, collectively defining an immunosuppressive ‘stromal memory/exhausted B cell reservoir’ (Fig. 3g–j).

IgG1 PCs in tumor and lymph nodes increase after radiation therapy and PD-1 checkpoint blockade

To validate these findings in an independent cohort¹⁸, we analyzed B cells and PCs from patients with HCC treated with neoadjuvant stereotactic radiation followed by anti-PD-1 therapy (biopsy, lymph node, tumor and adjacent normal) (Fig. 1a and Extended Data Fig. 3a–c). In bulk sequencing, differential gene expression between responders and non-responders was most pronounced in pretreatment biopsies and posttreatment tumors but minimal in lymph nodes and adjacent normal tissues (Extended Data Fig. 3d–h). After treatment, total immunoglobulin increased, with responders showing selective IgG1 enrichment (Extended Data Fig. 3i,j). The bulk data also showed an increase in IGHG1, IGHG2 and IGHG3 expression in tumor tissue of responders (Extended Data Fig. 3k). As expected, we observed higher plasmablast and PC levels in responder biopsies (FDR < 0.2) (Extended Data Fig. 3l–n). Interestingly, the increase in IGHG1−IGHG4 and decrease in IGHM were also observed in responders across tissues (FDR < 0.2) (Extended Data Fig. 3o). Clonal size analysis showed that the largest clones were identified in lymph nodes, followed by pretreatment and posttreatment tumor tissue (Extended Data Fig. 3p). Shared clonotypes were observed between all three posttreatment tissue types and pretreatment biopsies (Extended Data Fig. 3q).

Consistent with our findings, validation cohort scRNA-seq revealed significantly enriched plasmablasts in tumor tissue (FDR < 0.05) (Extended Data Fig. 4a). Activated B memory, class-switched and unswitched cells were also enriched in responders’ adjacent normal tissues (FDR < 0.05) (Extended Data Fig. 4b). In tumor tissue of responders, there was a minor increase in PCs (FDR < 0.2), whereas, in non-responders, there was a clear increase in B memory cells (FDR < 0.01) (Extended Data Fig. 4c). However, differential expression analysis recapitulated IGHG1 as the top marker of clinical response in both tumor and adjacent liver (Fig. 4a,b). Contrary to our discovery cohort, this independent validation cohort showed similar B cell and PC numbers between adjacent normal and tumor tissue (Extended Data Fig. 4e). Notably, most of the sequenced cells were already clonally expanded (Extended Data Fig. 4f–h). However, once again, the IgG1 isotype dominated across all PC and non-naive B cell phenotypes in tumor tissues of responders, whereas non-responder B cells and PCs were more prevalent for IgM and IgA (Fig. 4c). Regardless of tissue, differential abundance analysis showed that IGHG1 isotype, activated memory B cells, PCs and plasmablasts were associated with response, whereas IgM and IgG2 in memory B cells were enriched in non-responders (Fig. 4d,e). Although clonal expansion sizes were similar between B cells and PCs, tracking shared CDR3 clones revealed greater memory-to-plasma compartment transitions in responders, with exceptionally large IGHG1 clones (Fig. 4f,g). Together, these results show that IgG1 PCs account for more than 50% of total isotypes identified in responders and for less than 30% in non-responders (Fig. 4h).

IgG1 skewing is associated with response in other ICB-responsive tumors

To further extend our observations, we analyzed public datasets and found that patients with melanoma who responded to PD-1 with or without CTLA-4 blockade³ also showed expansion of PCs after treatment and a decrease in non-responders, despite a larger non-responder PC compartment before treatment, potentially due to low number of cells in this study (Fig. 4i–k). Furthermore, inspection of the isotype also showed IgG1 enrichment in PCs of responders (Fig. 4i–k and Extended Data Fig. 5e). Next, we validated our findings in advanced HCC treated with anti-PD-1 and anti-VEGF-A therapies in two independent cohorts of unresectable HCC: IMbrave150 (ref. ²⁰) and Cappuyns et al.²¹ (Fig. 4l,m). In both datasets, responders exhibited a skewing toward the IgG1 isotype, with IgG1 emerging as the dominant subclass. Another independent cohort of patients with HCC and intrahepatic cholangiocarcinoma (ICCA)²² showed that HCC tumors had a low tumor diversity score, linked to favorable outcomes, whereas ICCA tumors exhibited a high tumor diversity score, linked to a more aggressive phenotype and worse outcomes (6-month survival for ICCA versus 26-month survival for HCC). Notably, HCC samples had greater PC abundance, consistent with better progression-free survival in HCC than in ICCA (Fig. 4n,o). Together, these findings suggest that IgG1 PCs are strongly linked to ICB response.

ICB responders produce IgG antibodies against cancer antigens

Given the enrichment of IgG1 PCs in ICB responders, we explored whether these patients potentially generated antitumor antibodies detectable in the patients’ serum. Thus, we investigated serum samples collected before and during treatment from our discovery cohort, testing against a panel of 20 tumor-associated antigens. This panel included common CTAs, mutational antigens and stem-cell-associated antigens. A higher proportion of responders (63%) exhibited antitumor IgG antibodies in their serum compared to non-responders (17%), with the IgG antibodies primarily belonging to the IgG1 subclass and targeting CTAs such as MAGE-A, GAGE7, PRAME and NY-ESO-1 (Fig. 5a,b). These data suggest that responders with circulating IgG antibodies against CTAs generally had higher titers than non-responders. Conversely, IgG titers in non-responders were less dynamic, showing minimal changes after anti-PD-1 treatment. IgG1 antibodies targeting CTAs can enhance antigen presentation by antigen-presenting cells and potentially prime CD8⁺ T cells through immune complexes and cross-presentation^23,24.

Next, to investigate T cell responses, we performed an ELISpot assay with CD8⁺ T cells isolated from pretreatment and on-treatment peripheral blood mononuclear cells (PBMCs), resected tumors and draining lymph nodes with detectable NY-ESO-1 antibodies. The responder’s CD8⁺ T cells showed significant IFNγ, whereas the CD8⁺ T cells from the non-responder, who had predominantly IgA, showed only minimal reactivity (Fig. 5c,d). Notably, the responder had circulating NY-ESO-1 antibodies before ICB treatment, but IFNγ production by CD8⁺ T cells was observed only in on-treatment samples, suggesting that antitumor B cell response may precede T cell response, as in previous reports^23,24.

To assess if serum autoantibodies target cancer antigens more than other autoimmune targets, we performed seromic profiling (IgG and IgA against approximately 20,000 antigens, including 186 CTAs; Supplementary Table 1) on 32 paired pretreatment and posttreatment samples from a discovery HCC neoadjuvant PD-1 cohort (8 responders and 8 non-responders). We observed that IgG autoantibodies and, to a lesser extent, IgA were enriched for CTAs in responders compared to non-responders (Fig. 5e).

Surprisingly, reactivity to CTA was enriched for response, as antibodies detected to another approximately 400 other known non-CTA tumor antigens (including p53) had similar prevalence in responders and non-responders for IgG and IgA (Fig. 5f). Notably, almost all antigen-specific antibodies were unique to individual patients and were found before treatment and after treatment, although some increases in reactivity were noted after treatment (Extended Data Fig. 4i,j). Looking in individual samples, only CTA-specific antibodies showed a significant increase with clinical benefit in total number of reactivities (on average, 2−3 hits in responders versus 0−1 hits in non-responders) (Fig. 5g). In parallel, we did not observe correlation between gene expression of CTAs and autoantibodies, suggesting that immunogenicity is more important than expression alone (Fig. 5h). Finally, the increase in autoantibodies against CTAs in responders (P < 0.05) in IgG was identified as specific for CTAs compared to IgA, autoantigens and other antigens (Fig. 5i,j). Together, these results support the notion that antibody production and reactivity against CTAs are indicators of clinical response to ICB, in parallel to an increase of IgG1 PCs.

Plasma IgG1 signature associated with improved survival in immunotherapy

To explore the relevance of IgG1 PC expression in patient survival, we used independent immunotherapy clinical trials (approximately 1,500 patients)^25,26,27. Notably, high IGHG1 expression was associated with improved overall survival in multiple datasets, including skin cutaneous melanoma (SKCM) (TCGA)²⁵ and POPLAR²⁶ and OAK²⁶ trials (patients with NSCLC treated with anti-PD-L1) (Fig. 6a). Notably, non-immunotherapy trials showed no effect of IGHG1 on survival (lung squamous cell carcinoma (LUSC) and liver hepatocellular carcinoma (LIHC)). These results indicate that chemotherapy-treated cancers have no clear link between clinical response and IGHG1.

B cells differentiate toward plasma IgG1 in clinical responders

Finally, we evaluated the immune cell contributions in cell−cell communication networks. IgG1 PCs, alongside specific macrophage and T cell subsets, were key drivers of interaction strength in responders (Fig. 6b). Responders showed stronger interaction scores among IgG1 PCs, plasmacytoid dendritic cells and macrophages, whereas non-responders had enriched interactions involving monocytes, regulatory T cells, NK cells and immature dendritic cells (Fig. 6c,d). These findings suggest IgG1-skewed PCs foster a favorable immunogenic microenvironment through enhanced immunostimulatory signaling, potentially improving immunotherapy outcomes. We also identified key pathways potentially driving plasma IgG1 differentiation, including IL-6, TNF, MK, CD70, BTLA, MIF, BAG and CypA, which were enriched as both incoming and outgoing signals (FDR < 0.05) (Extended Data Fig. 5a–d and Supplementary Table 2).

By integrating these signaling results with differential gene expression and pseudotime analysis, we identified genes strongly associated with clinical outcomes. Responders showed higher expression of genes related to PC differentiation, such as ERN1 and RRBP1, whereas non-responders showed increased CXCR4 (Fig. 6e,f and Supplementary Table 2). These patterns suggest that ICB induces B cell activation and class switching, leading to plasmablast expansion and differentiation into IgG1-secreting PCs (Fig. 6f,g). By contrast, non-responders accumulate fewer diverse memory cells, atypical B cells and non-IgG1 PCs. In summary, we identified critical gene programs that support sustained B cell activation and plasma IgG1 differentiation, both of which are associated with favorable clinical response.