PD-1^hi Identifies a Novel Regulatory B-cell Population in Human Hepatoma That Promotes Disease Progression

The receptor programmed death-1 (PD-1) and its ligands, PD-L1 and PD-L2, deliver inhibitory signals that regulate the balance between immune activation, tolerance, and immunopathology (1, 2). Deficiency in PD-1 induces development of various autoimmune phenotypes in mice with different genetic backgrounds (3). In contrast, PD-1 is upregulated on T cells, which leads to exhaustion of these cells in most solid tumors (4, 5). Blocking the interaction between PD-1 and PD-L1 can restore T-cell function and result in tumor regression in both humans and mice (6–9). Nevertheless, it should be emphasized that, in addition to being expressed on exhausted T cells, PD-1 can be rapidly upregulated on B cells, natural killer T cells, monocytes, and dendritic cells in inflamed tissues (10–13), suggesting that PD-1–associated cancer therapy targets not only T cells but also the whole immune system. Thus, evaluating PD-1 expression and function in non–T-cell leukocyte infiltrate is essential for understanding the roles and potential suppressive mechanisms of PD-1 in tumor immunopathogenesis.

B cells consistently are abundant cellular components in tumors (14, 15), but the subset composition and the biologic role of B cells in human tumors are poorly understood. Several recent investigations in mouse cancer models have revealed the existence of protumorigenic regulatory B cells (Breg; refs. 16–18). Here, we identified a novel protumorigenic PD-1^hi B-cell subset in human hepatocellular carcinoma (HCC). Our results show that these cells represent about 10% of the entire B-cell population in tumor tissues from patients with advanced-stage disease, and they exhibit a unique CD5^hiCD24^−/+CD27^hi/+CD38^dim phenotype that differs from the phenotype seen in conventional peripheral Bregs (19, 20). TLR4-mediated BCL6 upregulation is critical for PD-1^hi regulatory B-cell subset induction by HCC environmental factors, but this process can be counteracted by IL4–induced STAT6 activation. Moreover, upon PD-1 triggering or exposure to PD-L1⁺ cells, PD-1^hi B cells can effectively suppress tumor-specific T-cell immunity and contribute to tumor growth in vivo via IL10 signals.

We used flow cytometry (FACS) to study the expression of PD-1 on B cells in 20 normal blood samples, six normal liver samples (tissue distal to a liver hemangioma) paired with blood samples, and 159 HCC-associated specimens (matched blood, peritumoral liver, and tumor tissue from 43 hepatitis B virus (HBV)–related, six hepatitis C virus (HCV)–related, and four non-HBV/HCV–related HCCs; Supplementary Tables S1 and S2). PD-1 was weakly expressed on a fraction of circulating B cells in both healthy donors and patients (Fig. 1A and B and Supplementary Fig. S1A). Although the percentage of PD-1⁺ B cells was increased in both normal and peritumoral liver tissues, the intensity of PD-1 expression was low (Fig. 1A–C). Interestingly, most tumor tissues contained a significantly greater proportion of PD-1⁺ B cells (21.3% ± 1.4%) that even exhibited a PD-1^hi phenotype (6.6% ± 0.5%; Fig. 1A–C). Analyzing the ratio of PD-1^hi to PD-1⁺ B cells in the samples suggested that the local environment of the tumor had the greatest potential to induce PD-1^hi B cells (0.336 ± 0.016 in tumors; Fig. 1D). Furthermore, the proportions of PD-1⁺ B cells, particularly PD-1^hi B cells in tumor tissues, were significantly correlated with disease progression in patients (Fig. 1E and F). More importantly, using 12 months as the cutoff, 2.6 times more patients with higher proportions of PD-1^hi B cells than those with lower percentages showed early recurrence (58% vs. 22%; Fig. 1G). In contrast, no relationship was found between patients' recurrence and the percentages of total PD-1⁺ B cells in tumor tissues (Fig. 1H). The patients with HCC who were or were not infected with HBV or HCV exhibited similar patterns of PD-1 expression in blood and peritumoral liver and tumor tissue (Fig. 1B–D), and no correlation was found between serum HBV DNA levels and frequency of PD-1^hi B cells in tumors (Supplementary Fig. S1B). Notably, we also detected a positive correlation between the percentage of tumor PD-1^hi B cells and levels of plasma IL10, but not IL1β, IL6, IL17, IL21, or TNFα (Supplementary Fig. S1C).

We subsequently examined the distribution of PD-1^+/hi B cells in HCC samples. Confocal microscopy showed that PD-1^+/hi cells were present throughout the tissue, and 51.8% ± 9.2% of the PD-1^+/hi cells in peritumoral stroma were CD79a⁺ cells that were regarded as B cells (Fig. 1I and Supplementary Fig. S1D). Indeed, the CD79a⁺ cells were mainly distributed in HCC peritumoral stroma (N = 53), and pronounced infiltration of these cells was also associated with advanced tumor–node–metastasis (TNM) stages in the patients (Fig. 1J–L). These data suggest a protumorigenic role of peritumoral stromal PD-1^hi B cells in HCC.

Inasmuch as the PD-1^hi B cells potently correlated with disease progression and plasma IL10 in patients with HCC, we compared the phenotypic and functional features of HCC-infiltrating PD-1^hi B cells and conventional IL10-producing Bregs. In general, over 50% of the tumor B cells displayed a CD45RO⁺IgD⁻IgM⁻IgG⁺ tissue–resident memory phenotype with increased CXCR3 and CXCR4, but radically reduced CD62L, CXCR5, and CCR6, although CD45RA⁺BTLA⁺HLA-DR⁺IgD⁺IgM⁺IgG⁻–naïve B cells dominated in paired blood samples (Fig. 2A and Supplementary Fig. S2A and S2B). Interestingly, the CD5⁺CD24^hiCD27⁺CD38^hi B cells that were considered as conventional peripheral Bregs (19, 20) were hardly detected in tumor tissues, whereas they dominated in peripheral blood (Supplementary Fig. S2C and S2D). Consistently, the PD-1^hi B cells from HCC tissues exhibited a unique CD5^hiCD24^−/+CD27^hi/+CD38^dim phenotype that differed from that of conventional peripheral Bregs (Fig. 2B and C and Supplementary Fig. S2E). In contrast, the PD-1^dim B cells from tumor tissues often expressed low/moderate levels of CD5 and CD27, but high intensities of CD24 and CD38 (Fig. 2B and C and Supplementary Fig. S2E). The PD-1^hi B cells also showed an activated form with substantial expression of CD69 and CD86 (Fig. 2A–C). Similar expression patterns were observed for IgA, IgD, IgG, and IgM among PD-1⁻, PD-1^dim, and PD-1^hi B cells in tumor tissues (Fig. 2A). Furthermore, we purified the CD5^hiCD27^hi/+ (>85% were PD-1^hi) and CD5^dim/− B cells from HCC tissues (Supplementary Fig. S2F) and stimulated these cells, as well as blood CD24^hiCD38^hi B cells, with CD40L together with CpG oligodeoxynucleotides (CpG ODN), anti-human IgM antibody, or a cocktail of anti-IgM, anti-IgG, and anti-IgA antibodies, stimuli combinations for IL10 production in conventional Bregs (20). The stimulated PD-1^hi B cells were completely incapable of producing IL10, although the blood CD24^hiCD38^hi B cells stimulated in the same manner did secrete large amounts of that cytokine (Fig. 2D and Supplementary Fig. S2G). The CD5^dim/− B cells only marginally produced IL10 (Fig. 2D and Supplementary Fig. S2G). Together, these data indicate that PD-1^hi B cells represent a novel protumorigenic subset of B cells that are not related to conventional Bregs. Of note, we also found that the tumor CD5^hiCD27^hi/+ B cells could not differentiate into immunoglobulin-secreting plasma cells (Supplementary Fig. S2H).

Having established the presence of a protumorigenic PD-1^hi B-cell subset in HCC, we initially performed experiments to investigate whether environmental factors contributed to induction of B-cell PD-1 expression. Culture supernatant from primary HCC tumor cells (HCC-SN) showed great potential to induce considerable numbers of PD-1^hi/+ B cells with higher intensities of CD5 and CD27 from healthy blood B cells or HCC tissue–derived CD5^dim/− (PD-1^dim/−) B cells, whereas culture supernatant from normal liver failed to augment PD-1 expression on B cells (Fig. 3A and B and Supplementary Fig. S3A–S3B). CD40L, which is mainly expressed by activated T cells, also triggered B-cell PD-1 expression, although to a lesser extent (Fig. 3C). By comparison, the cytokines IL1β, IL6, and IL10, whose expression is upregulated in HCC-infiltrating monocytes (21, 22), did not affect PD-1 expression (Fig. 3C). Unexpectedly, the Th2 cytokine IL4 completely abolished PD-1 upregulation that was induced by CD40L (Fig. 3C) or even by HCC-SN (Fig. 3D), though it did not hamper the PD-1 expression in B cells already expressing PD-1 (Supplementary Fig. S3C). In support, a marked reduction of IL4-producing Th cells was detected in human HCC tissues (Supplementary Fig. S3D and S3E). Of note, the Th1 cytokine IFNγ did not induce or affect the B-cell PD-1 expression (Fig. 3C).

We previously showed that environmental hyaluronan (HA) fragments from hepatoma cells induced protumorigenic myeloid cells via TLR4 activation (23), which implies that innate stimuli may contribute to induction of B-cell PD-1 expression. Indeed, HA fragments and Toll-like receptor (TLR) agonists Pam3CysSK4, lipopolysaccharide (LPS), and CpG ODN did induce a significant proportion of PD-1^+/hi B cells, and the effect of HA fragments could be reversed by blocking the TLR4 signal (Fig. 3E and F and Supplementary Fig. S3F). In parallel, adding an HA-specific blocking peptide (Pep-1) significantly impaired the upregulation of PD-1 on HCC-SN–exposed B cells (Fig. 3F). As expected, blocking the TLR4 signal successfully decreased the proportions of PD-1^+/hi B cells induced by HCC-SN, whereas blocking TLR2 had only a marginal effect (Fig. 3G and Supplementary Fig. S3G). In support of this finding in vitro, the C57BL/10J hepatoma-bearing mice had higher levels of tumor PD-1^hi B cells with phenotypic features similar to those isolated from human HCC tumors (Supplementary Fig. S3H), and both knockout of TLR4 and injection of Pep-1 significantly reduced the amounts of such cells in livers with hepatomas and partially suppressed tumor growth (N = 10; P < 0.01; Fig. 3H and I and Supplementary Fig. S3I).

BCL6 was implicated in follicular T helper cell PD-1 induction (24, 25), and reduced STAT6 activation was correlated with increased PD-1–expressing CD4⁺ T cells in follicular lymphoma (26). In HCC tissues, BCL6 was selectively upregulated in PD-1^hi/+ B cells, whereas STAT6 was deactivated in all tumor B cells (Fig. 4A and B). Thus, we examined whether those proteins are involved in the regulation of B-cell PD-1 expression. Upregulation of BCL6 was much more pronounced in B cells after exposure to HCC-SN (Fig. 4C), but STAT6 activation was linked only to IL4 stimulation (Fig. 4D). Factors that could trigger PD-1 expression effectively triggered BCL6 expression, whereas cytokines IL1β, IL4, IL6, and IL10, which failed to induce PD-1^+/hi B cells, did not affect BCL6 expression (Figs. 3C and 4D and Supplementary Fig. S3F). These data suggest positive regulation of BCL6 in B-cell PD-1 expression that may be hampered by STAT6 activation. Consistently, using a pSIF-H1-CopGFP-shBCL6 lentiviral vector to silence BCL6 largely attenuated the HCC-SN–mediated PD-1^+/hi B-cell induction (Fig. 4E and Supplementary Fig. S4A and S4B); in parallel, knockdown of STAT6 in primary B cells effectively rescued the IL4-elicited PD-1 inhibition (Fig. 4F).

To further probe the mechanisms involved in the induction of PD-1^+/hi B cells by BCL6, we examined the activation kinetics of the PI3K/AKT, MAPK, and NFκB pathways, which are all downstream of TLR signaling (27). The activation patterns of the MAPKs JNK, ERK, and p38, and the NFκB inhibitor IκBα in B cells exposed to multifarious stimuli coincided with the ability of the cells to express BCL6 (Fig. 4C, D and G and Supplementary Fig. S4C). Activation of these pathways was selectively enhanced in B cells stimulated with HCC-SN or TLR agonist Pam3CysSK4, LPS, or CpG (Fig. 4G and Supplementary Fig. S4C). Accordingly, using inhibitors to block the signal transduction of JNK, p38, and NFκB effectively impaired HCC-SN–induced BCL6 upregulation and subsequent PD-1 expression, whereas abolishing the phosphorylation of ERK had only a marginal effect (Fig. 4H).

We next investigated whether the PD-1 signal contributes to the protumorigenic effects of PD-1^hi B cells. Stimulation of purified tumor B cells with a PD-1–specific agonist (12) induced marked production of IL10, but not IL6 or immunoglobulins, as compared with control goat IgG (Fig. 5A and Supplementary Fig. S5A) or antibodies to human leukocyte antigen HLA-ABC and HLA-DR (Fig. 5B), which provides further evidence that expression of PD-1 by B cells and IL10 production were interrelated. Moreover, measuring PD-1–mediated IL10 production over time in tumor B cells revealed rapid accumulation of IL10 in 12 to 18 hours, which reached a plateau within 24 hours (Fig. 5C). Similar results were obtained using an ELISpot detection system to analyze IL10 production (N = 5; Fig. 5D). In addition, using the PD-1 agonist to stimulate PD-1–expressing B cells induced by HCC-SN also led to at least 5-fold augmentation of IL10 production (Supplementary Fig. S5B). In contrast, the paired peripheral B cells from patients with HCC undergoing the same stimulation secreted very little IL10 (Fig. 5A and D), suggesting that sufficient interaction between PD-1 and its agonist is essential for IL10 production in B cells. Corresponding to this, there was a sharp decline in IL10 expression by tumor B cells with decreasing intensity of the PD-1 stimulus (Fig. 5E).

It is noteworthy that the levels of IL10 produced by tumor B cells were selectively correlated with the proportions of PD-1^hi B cells but differed from the proportions of PD-1^dim B cells (Fig. 5F). We subsequently demonstrated that CD5^hiCD27^hi/+ B cells (>85% of which were PD-1^hi; Supplementary Fig. S2F), but not CD5^dim/− B cells (PD-1^dim/−), undergoing PD-1 triggering were the major source of IL10 production (N = 5; P < 0.05; Fig. 5G and H). Notably, stimulus with the TLR agonist LPS or CD40L plus anti-IgM, both of which elicited IL10 production in conventional Bregs, did not induce IL10 production in tumor-derived CD5^hiCD27^hi/+ B cells (Fig. 5I). These findings reveal a novel mechanism involving expression of IL10 mediated by the PD-1 signals.

We subsequently assessed the impact of interaction between PD-1 and its physiologic ligands PD-L1 and PD-L2 on IL10 production by tumor B cells. Tumor B cells cultured with mitomycin C–treated HEK293T–PD-L1 cells produced more IL10 than those cultured with HEK293T-mock cells, and the IL10 production could be significantly inhibited by adding a PD-L1 blocking antibody (P < 0.05; Fig. 6A and Supplementary Fig. S6A). In contrast, tumor B cells cultured with HEK293T–PD-L2 cells produced only small amounts of IL10 (Fig. 6A). Similar results were obtained when assessing HCC-SN–pretreated B cells cultured with HEK293T–PD-L1 cells (Supplementary Fig. S6B).

In HCC tumors, we have previously found that PD-L1 is expressed chiefly on peritumoral stromal monocytes and only weakly expressed on other stromal and hepatoma cells (28, 29). We found that most of such PD-L1⁺ cells were in proximity to infiltrating CD79a⁺ cells (Fig. 6B), suggesting that physical contact between PD-L1⁺ monocytes and PD-1–expressing B cells mediates regulatory functions. In an ex vivo system, coculture of B cells and mitomycin C–treated tumor-derived monocytes elicited significant IL10 production (P < 0.05; Fig. 6C and D). Blocking PD-L1 by preincubating monocytes with a PD-L1–specific antibody markedly reduced IL10 production (P < 0.05; Fig. 6C and D). Furthermore, high levels of IL10 were detected in cocultures of PD-1–expressing B cells and mitomycin C–treated PD-L1⁺ monocytes, the latter of which were induced by LPS or IFNγ, and this process was also suppressed by an anti–PD-L1 antibody (Fig. 6E). These results confirmed the key role of the PD-1–PD-L1 interaction in the induction of IL10 production by B cells during HCC pathogenesis.

The colocalization of CD79a⁺ B cells and CD8⁺ T cells in peritumoral stroma of HCC (Fig. 7A) prompted us to analyze the regulatory effects of PD-1^hi B cells on effector T-cell function. High infiltration of PD-1^hi B cells positively correlated with both reduced number and dysfunction of CD8⁺ T cells, though not with the presence of FOXP3⁺ regulatory T cells in HCC tumors (Supplementary Fig. S7A). T cells derived from HCC tissues mainly exhibited a CCR7⁻CD45RA⁻ effector memory phenotype and these effector T cells expressed higher levels of IL10R than the naïve T cells in tumors (Supplementary Fig. S7B). Accordingly, tumor-derived CD5^hiCD27^hi/+ (PD-1^hi) B cells, but not the CD5^dim/− B cells, undergoing PD-1 triggering effectively induced dysfunctional PD-1⁻ effector T cells with impaired capacities for production of proinflammatory IFNγ and cytotoxic granzyme B and perforin ex vivo, and these effects could be attenuated by shielding the IL10R in the T cells (P < 0.05; Fig. 7B and C). We subsequently established an autologous mouse model to investigate the effect of PD-1^hi B cells on tumor-specific T cell cytotoxicity. PD-1⁻CD8⁺ T cells derived from mouse Hepa1-6 hepatoma effectively elicited cell death of hepatoma cells ex vivo, which was largely attenuated by adding Hepa1-6 hepatoma–derived CD5^hi (PD-1^hi) B cells plus mitomycin C–treated HEK293T-mouse PD-L1 cells (Fig. 7D). As expected, shielding either IL10R on CD8⁺ T cells or PD-1 on CD5^hi B cells significantly abolished the suppressive role of PD-1^hi B cells (Fig. 7D).

Considering that the absolute numbers of PD-1^hi B cells in mouse hepatoma tissues were not as pronounced as those in human HCC tissues (Supplementary Fig. S7C), we chose the adoptive PD-1^hi B-cell transfer model to investigate the PD-1^hi B-cell function in vivo (Supplementary Fig. S7D). To effectively promote mouse hepatoma progression, adoptive transfer of tumor CD5^hi (PD-1^hi) B cells from at least three hepatoma-bearing mice was needed (Supplementary Fig. S7E). Adoptive transfer of CD5^hi B cells, but not CD5⁻ B cells, could significantly promote tumor growth and impair CD8⁺ T-cell infiltration and function in liver in recipient mice (Fig. 7E–I). Consistent with the ex vivo findings, such CD5^hi B cell–mediated tumor growth and CD8⁺ T-cell reduction and dysfunction were significantly reversed by shielding the PD-1 signal before adoptive transfer of CD5^hi B cells or by injecting an antibody against IL10R or PD-L1 (Fig. 7E–I). These data together indicate that eliciting the PD-1 signal in B cells by PD-L1 defeats tumor-specific T-cell immunity in vivo via IL10 and thereby contributes to tumor growth.

Here, we identified a novel protumorigenic PD-1^hi B-cell subset and applied multiple complementary strategies to map the phenotype, mechanisms of induction, biologic function, and clinical relevance of those cells in the tumor microenvironment of patients with HCC.

Studies in human peripheral blood have identified a subset of Bregs that exhibit a CD5⁺CD24^hiCD27⁺CD38^hi phenotype (19, 20), and it has been suggested that these cells play a crucial part in regulating T-cell responses by releasing IL10 (19). In our investigation, the PD-1^hi B cells were established as a protumorigenic B-cell subset in tumor tissues, and they displayed a unique CD5^hiCD24^−/+CD27^hi/+CD38^dim phenotype that differed from that of conventional peripheral Bregs. Along this line, we also found that the classic combined stimuli for conventional Bregs (20) failed to induce IL10 production in PD-1^hi B cells. Interestingly, triggering PD-1 caused the PD-1^hi B cells to produce large amounts of IL10 that potently suppressed tumor-specific T-cell immunity. This reveals that PD-1^hi B cells represent a novel Breg subset that has not been previously defined in tumor tissues. In addition, by analyzing the content of CD24^hiCD38^hi conventional Bregs in HCC tumors, we established that PD-1^hi B cells constitute the dominant IL10-producing Breg subset in human HCC tumors.

TLRs have been linked to antigen-presenting cell (APC) maturation and proinflammatory cytokine production (30), but our results show that these signals also participate in PD-1^hi B cell–triggered T-cell dysfunction and cancer progression. More precisely, we observed that blockade of TLR4 activation effectively abolished the primary HCC-SN–mediated induction of PD-1^+/hi B cells. This finding agrees with that of recent studies showing that activation of TLR by microbial products led to upregulation of PD-1 on monocytes and B cells (10, 12). Notably, the TLR signals, particularly TLR9–CpG ODN interaction, also play an essential role in the terminal differentiation of B cells into plasma cells (31), but this process requires BCR engagement and costimulation via CD40L (32). In addition, several recent studies have revealed that T cell–derived IL21 also participates in the terminal differentiation of B cells (32). These findings together suggest a complicated process of B-cell maturation in human tumors. In other words, insufficient conditions for B-cell maturation in tumors, e.g., TLR agonist or CD40L alone, may be exploited to create PD-1–mediated immune privilege. Thus, if B cells locate in a niche that is suitable for their maturation, they may acquire functions with antitumor properties. This notion is supported by several clinical investigations indicating that the presence of B cells in tertiary lymphoid structures is associated with protective immunity in patients with colorectal, ovarian, and oral squamous cell cancers (33–35).

BCL6 is vital for activation and proliferation of B cells in the germinal center (36). However, using a pSIF-H1-CopGFP-shBCL6 lentiviral vector, we identified an immunosuppressive role of BCL6 in tumors and found that inhibiting the upregulation of BCL6 effectively attenuated HCC-SN–elicited generation of PD-1^hi Bregs. Indeed, other studies have shown that Bcl6^−/− mice develop exaggerated allergic responses (37), which indirectly reflects the function of BCL6 in controlling immune activation. More unexpectedly, we found that STAT6 activation induced by the Th2 cytokine IL4 could entirely abolish BCL6-mediated PD-1^+/hi B-cell induction, although it did not affect BCL6 upregulation. These findings are compatible with several recent studies showing that BCL6 was implicated in follicular T helper cell PD-1 induction and that reduced STAT6 activation was correlated with increased PD-1–expressing CD4⁺ T cells in follicular lymphoma (24–26). In addition, we also observed that the prevalence of IL4-producing T cells was lower in the tumor samples than in the paired peritumoral liver and blood samples, which supports our finding that PD-1^hi Bregs selectively accumulated in tumor regions.

Binding of PD-1 by PD-L1 triggers production of IL10 during tumor progression. The elevated expression of the two components of the PD-1–PD-L1 axis on B cells and monocytes/macrophages in the peritumoral stroma during HCC progression increases the likelihood of engagement of this pathway, that is, the probability that PD-1^hi B cells will encounter PD-L1⁺ cells such as monocytes/macrophages. The ensuing production of IL10 serves to control cytotoxic T-cell activation and help avoid hyperimmune activation. Notwithstanding, the prolonged and chronic exposure of immune cells to regulatory signals such as IL10 and PD-1 during tumor progression results in T-cell dysfunction (38). Thus a fine-tuned collaborative action between different types of immune cells in peritumoral stroma limits host response (cytotoxic T-cell response) to the malignancy. Furthermore, PD-1 expression on T cells and binding of that receptor by PD-L1 on APCs is a major mechanism underlying T-cell exhaustion during tumor progression (39–41). Our results identify a new mechanism by which PD-1–PD-L1 interaction induces immune dysfunction during tumor progression. It is plausible that these two mechanisms can synergize, as shown in our previous (28, 29) and present experiments.

At present, subset composition and function of B cells in human cancer is largely unclear. PD-1^hi B cells exhibited an IL10-producing function and accumulated in the peritumoral stroma of HCC. The frequencies of PD-1^hi B cells to total B cells, as well as the absolute numbers of B cells in the peritumoral stroma, were positively associated with advanced TNM stages in patients. In support, transfer of enriched PD-1^hi B cells to tumor-bearing mice drastically promoted disease advancement. Consistent with this, although not directly related to the peritumoral PD-1^hi subset, B cells have been found to inhibit antitumor immune responses in mice and induce immune privilege by expansion of regulatory T cells (16, 17). Interestingly, in the intratumoral regions of human HCC, Garnelo and colleagues observed that increased levels of B cells correlated with better survival of patients (42), suggesting existence of antitumorigenic B-cell subset(s) in HCC intratumoral regions. Thus, B cells may exhibit both antitumorigenic and protumorigenic functions, owing to their distribution and subset compositions in the designated environments. Studying the composition and functions of infiltrating B cells may help us better understand their roles in tumor pathogenesis.

Our results provide important new insights into possible manipulation of B cell–mediated immunosuppression in human tumors. Soluble factors derived from cancer cells promote PD-1 expression on B cells by triggering TLR4 activation. After interacting directly with PD-L1⁺–activated monocytes, the PD-1^hi B cells operate via IL10-dependent pathways to induce T-cell dysfunction and thereby create conditions that are conducive to tumor progression (Supplementary Fig. S8). Accordingly, immunotherapies that interfere with PD-1 and IL10 should be part of the arsenal used to restore immune function in patients with cancer. Thus, it is possible that studying the mechanisms that can selectively modulate functional activities of inflammatory stroma cells can provide a novel strategy for anticancer therapy (43–45).

Liver and HCC samples were obtained from patients who underwent curative resection at the Cancer Center of Sun Yat-sen University (Guangzhou, China; Supplementary Tables S1 and S2). None of the patients had received anticancer therapy before the sampling, and individuals with concurrent autoimmune disease, HIV, or syphilis were excluded. Paired fresh samples of blood (taken on surgery day), peritumoral liver tissue (taken 0.5–2 cm distal to the tumor site), and tumor tissue from 53 patients with HCC who underwent surgical resections between December 2010 and November 2012, 11 patients with HCC who underwent surgical resections between July 2014 and September 2014, 9 patients with HCC who underwent surgical resections between March 2015 and May 2015, and 5 patients with HCC who underwent surgical resections in January 2016 were used to isolate peripheral and tissue-infiltrating leukocytes. Paraffin-embedded or frozen tumor samples from the same patients who underwent surgical resections between December 2010 and November 2012 were used for immunohistochemistry and immunofluorescence analysis. Samples of normal tissue were obtained distal to liver hemangiomas (Supplementary Table S1). Clinical stages were classified according to the guidelines of the International Union against Cancer. All samples were anonymously coded in accordance with local ethical guidelines (as stipulated by the Declaration of Helsinki). Written informed consent was obtained from the patients, and the protocol was approved by the Institutional Review Board of Sun Yat-sen University.

Peripheral mononuclear leukocytes were isolated by Ficoll density gradient centrifugation, and fresh tissue-infiltrating mononuclear leukocytes were obtained as previously described (5). The whole isolation process or digestion buffer would not affect the survival of CD45⁺ mononuclear cells (Supplementary Fig. S2A). Purification of B cells, T cells, or monocytes from the leukocytes was achieved with a MACS column purification system (Miltenyi Biotec). CD24^hiCD38^hi blood B cells, CD5^hiCD27^hi/+ (PD-1^hi) or CD5^dim/− (PD-1^dim/−) tumor B cells, and CD45RA⁻CCR7⁻PD-1⁻ tumor effector T cells were further sorted using FACS (Moflo, Beckman Coulter). CD24^hiCD38^hi B cells of >90% purity and >90% viability, PD-1^dim/− B cells of >98% purity and >90% viability, and PD-1^hi B cells of >85% purity and >90% viability were used for subsequent functional experiments. Sorted B cells (1.5 × 10⁵) in 150 μL were left untreated, stimulated for 24 hours with 1 μg/mL CD40L (R&D Systems) plus 5 μg/mL anti-human IgM, or a cocktail of antihuman IgM, IgA, and IgG (5 μg/mL for each; Jackson ImmunoResearch Laboratories), or plus 2 μg/mL CpG ODN; or exposed for 24 hours to 10 μg/mL anti–PD-1 agonistic antibody, control goat IgG (R&D Systems), or anti–HLA-DR or anti–HLA-ABC antibody (eBioscience).

Details of staining are described in Supplementary Methods. Data were acquired on a Gallios flow cytometer (Beckman Coulter) and analyzed with FlowJo software (6.0.0.0). The fluorochrome-conjugated antibodies and isotype controls used are listed in Supplementary Tables S3 and S4. The isotype controls for representative data in Supplementary Fig. S1A and Fig. 2B are shown in Supplementary Fig. S9A and S9B.

Paraffin-embedded human HCC or mouse hepatoma samples were cut into 5-μm sections, which were processed for immunohistochemistry as previously described (5). Following incubation with an antibody against human CD79a or mouse CD8, the sections were stained in an Envision System (DakoCytomation). Evaluation of immunohistochemical variables was performed by two independent observers who were blinded to the clinical outcome. For immunofluorescence analysis, frozen sections of human HCC were stained with rabbit anti-CD79a plus goat anti–PD-1; rabbit anti-CD79a plus mouse anti-CD20; rabbit anti-CD79a plus mouse anti-CD3; rabbit anti-CD79a plus mouse anti–PD-L1; or rabbit anti-CD79a plus mouse anti-CD8, followed by Alexa Fluor 488–conjugated anti-goat IgG plus Alexa Fluor 555–conjugated anti-rabbit IgG or Alexa Fluor 488–conjugated anti-rabbit IgG plus Alexa Fluor 555–conjugated anti-mouse IgG (Molecular Probes). Positive cells were quantified using ImagePro Plus software or detected by confocal microscopy. The antibodies and isotype controls used are listed in Supplementary Tables S5 and S4. The representative isotype staining for CD79a, CD8, PD-1, and PD-L1 is shown in Supplementary Fig. S9C.

Tumor B-cell subsets sorted by FACS were incubated in RPMI-1640 supplemented with 10% FBS and 1 μg/mL CD40L (R&D Systems) plus 50 ng/mL recombinant IL21 (PeproTech). Culture supernatants were collected on day 6 for detection of immunoglobulins.

Details are given in Supplementary Methods.

Culture supernatants were acquired by culture of completely digested HCC tumor or hemangioma liver biopsy specimens. All specimens were from individuals without concurrent autoimmune disease, HBV, HCV, HIV, or syphilis. The digested tumor or liver cells were washed in medium containing polymyxin B (20 μg/mL; Sigma-Aldrich) to exclude endotoxin contamination. Thereafter, 10⁷ digested cells were resuspended in 10 mL of complete medium and cultured in 100-mm dishes. After 2 days, the supernatants were harvested, centrifuged, and stored at −80°C.

Details are given in Supplementary Methods.

Healthy blood B cells were left untreated or exposed to TLR agonist (Invivogen), cytokine (R&D Systems), primary liver-SN or primary HCC-SN, or intermediate-HA fragments (50–200 kDa) prepared as previously described (21) for 3 days or indicated times. Thereafter, the cells were harvested and stained with fluorochrome-conjugated antibodies and then analyzed by FACS. In some experiments, B cells were pretreated with 1 μg/mL CD40L, 10 ng/mL IL4 (R&D Systems), 20 μg/mL blocking antibody against TLR2 or TLR4 (eBioscience), 100 μg/mL HA-specific blocking peptide (Pep-1, GAHWQFNALTVR) or a control peptide (Pep-C, WRHGFALTAVNQ), or a specific inhibitor of the AKT (AIP2, 10 μmol/L), JNK (SP 600125, 5 μmol/L), ERK (U0126, 20 μmol/L), NFκB (BAY 11-7082, 5 μmol/L), or p38 (SB 203580, 20 μmol/L) signal (Sigma-Aldrich), and subsequently exposed to indicated stimuli. In other experiments, healthy blood B cells were transduced with a pSIF-H1-CopGFP-shRNA lentiviral vector for 4 hours, washed twice, and then exposed to IL4, primary HCC-SN, or LPS. Transduction and silencing efficiency was examined after 36 hours (Supplementary Fig. S4A and S4B), and PD-1 expression was assessed within GFP⁺ cells on day 3.

The mouse hepatoma cell line Hepa1-6 was obtained in January 2014 from the Cell Bank of Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China) within 6 months of the experiments. The cells were authenticated by short tandem repeat profiling and were confirmed to be Mycoplasma negative before use. For the in vivo assay, 5-to-6-week-old female C57BL/10ScNJ (TLR4 knockout; ref. 46) and C57BL/10J mice were purchased from the Nanjing Biomedical Research Institute of Nanjing University, and a Hepa1-6 hepatoma model was established as previously described (22). After 25 days, tumor tissues were harvested and digested for isolation of tumor-infiltrating leukocytes. B-cell PD-1 expression was determined by FACS. In some experiments, C57BL/10J mice bearing hepatoma underwent peritoneal injection of Pep-1 or Pep-C (25 mg/kg) every 3 days. Mice were randomly grouped and observers were blinded to the grouping. The animal use protocols were reviewed and approved by the Institutional Animal Care and Use Committee of Sun Yat-sen University.

Mock, PD-L1, or PD-L2 HEK293T transfectants, tumor-derived monocytes, or LPS- or IFNγ-induced PD-L1 monocytes were primary pretreated with 10 μg/mL mitomycin C for 20 minutes to abolish the growth or cytokine production activity of cells. Thereafter, purified tumor B cells were left untreated or cultured for 24 hours with indicated mitomycin C–treated cells in the presence or absence of 10 μg/mL PD-L1–blocking antibody or a control antibody (eBioscience). The production of IL10 by tumor B cells was determined by ELISA or ELISpot Assay.

In another set of ex vivo experiments, FACS-sorted PD-1⁻ effector T cells from tumors were left untreated or were incubated for 24 hours with FACS-sorted autologous tumor B cells (5:1) supplemented with 25 μg/mL tumor mass lysate in the absence or presence of 10 μg/mL PD-1 agonistic antibody (R&D Systems) or anti–IL10-receptor (IL10R) antibody (BioLegend). Production of IFNγ was determined by ELISpot.

Proteins from cells were extracted as previously described (23). The antibodies used are listed in Supplementary Table S5.

ELISpot assays were performed using commercial kits (BD Pharmingen) according to the manufacturer's instructions. The images were scanned with an ELISpot Reader (CTL), and spot numbers were counted manually.

PD-1^hi and PD-1^−/dim B cells were harvested from mice bearing Hepa1-6 hepatoma for 25 days (step 1 in Supplementary Fig. S7D). A total of 1 × 10⁵ sorted PD-1^−/dim or PD-1^hi B cells or PD-1^hi B cells pretreated with 10 μg/mL blocking antibody against PD-1 (Bio X Cell) in 100 μL buffered saline were injected into the peritoneum of recipient mice in the presence or absence of antibody against PD-L1 or IL10R (both from Bio X Cell; 5 mg/kg) every 5 days after 1-day inoculation of Hepa1-6 hepatoma. The recipient mice were sacrificed after 16 days (step 2 in Supplementary Fig. S7D), and tumors were photographed and measured and then fixed in formaldehyde before being subjected to further processing and IHC procedures. In another case, part of tumor tissues was used for isolation of infiltrating leukocytes, and the function of infiltrated CD8⁺ T cells was examined by FACS.

In the ex vivo experiment, FACS-sorted PD-1⁻CD8⁺ T cells alone or together with enriched PD-1^hi B cells derived from C57BL/10J Hepa1-6 hepatoma were added in the culture of Hepa1-6 cells (10:1) in the presence or absence of mitomycin C-treated mouse PD-L1–expressing HEK293T cells, supplemented with rat IgG or blocking antibody against PD-1 or IL10R (Bio X Cell). T-cell cytotoxicity was detected by flow cytometry.

Results are expressed as the mean ± SEM. Group data were analyzed by ANOVA followed by the Student t test or log-rank test. Correlations between parameters were assessed by Pearson correlation analysis and linear regression analysis. All data were analyzed using two-tailed tests, and P < 0.05 was considered statistically significant. No statistical method was used to predetermine sample size. No animal data were excluded.

No potential conflicts of interest were disclosed.

Conception and design: X. Xiao, X.-M. Lao, M.-M. Chen, D.-M. Kuang

Development of methodology: X. Xiao, M.-M. Chen, X.-F. Li, L. Zheng, D.-M. Kuang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Xiao, X.-M. Lao, M.-M. Chen, R.-X. Liu, Y. Wei, F.-Z. Ouyang, D.-P. Chen, X.-Y. Zhao, Q. Zhao, X.-F. Li, C.-L. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Xiao, X.-M. Lao, M.-M. Chen, D.-M. Kuang

Writing, review, and/or revision of the manuscript: X. Xiao, D.-M. Kuang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X.-M. Lao, L. Zheng, D.-M. Kuang

Study supervision: D.-M. Kuang

The authors thank Ms. Patricia Ödman for linguistic revision of the manuscript.

This work was supported by project grants from the National Natural Science Foundation of China (81422036, 31470855, and 81171982), the Guangdong Natural Science Funds for Distinguished Young Scholars (S2013050014639), the Foundation for the Author of National Excellent Doctoral Dissertation of PR China (201230), and the Fundamental Research Funds for the Central Universities (15lgjc09).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.