IL27 Signaling Serves as an Immunologic Checkpoint for Innate Cytotoxic Cells to Promote Hepatocellular Carcinoma

Hepatocellular carcinoma (HCC) is the most common form of liver cancer, with a poor survival rate and limited treatment options (1, 2). Although innovative therapeutic strategies targeting the tumor microenvironment and immune contexture provide hope to many patients with cancer (3, 4), HCC remains poorly responsive to therapies and continues to be a highly lethal disease (2, 5). For example, despite wide clinical use of T-cell activation–based immunotherapies in several cancer types (6), checkpoint inhibitors have only recently demonstrated potentially promising results in HCC (7). Preventive approaches aimed at hepatitis B vaccination and hepatitis C eradication could potentially curb new HCC cases, but liver cancers caused by environmental toxins and fatty liver disease are clearly on the rise (8, 9).

Chronic liver inflammation induced by infections, alcohol, or obesity drives chronic injury and promotes compensatory proliferation of transformed hepatocytes facilitating HCC formation and growth (10). Although anticancer immune responses generally play an important tumor-restrictive role, it remains to be determined how they are regulated by the inflammatory entities in HCC (11).

Irrespective of its etiology, HCC development is accompanied by the accumulation of immune cells that contribute to cancer progression via the production of proinflammatory cytokines such as IL6, IL1, TNF, and IL17 (12). Cytokine signaling enables HCC growth by activating proto-oncogenic transcription factors such as NF-κB and STAT3 in transformed hepatocytes and HCC cells (10, 12–14). Conversely, the presence of cytotoxic or IFNγ-producing T cells suppresses tumor growth (15). Moreover, the liver microenvironment is uniquely enriched in innate lymphocytes, including natural killer (NK) cells and innate lymphoid cells type 1 (ILC1), capable of tissue immunosurveillance and antitumorigenic functions potentially affecting HCC tumorigenesis (16, 17). During HCC development, the number and activation state of liver NK cells gradually decline due to yet unidentified mechanisms possibly related to specific signals originating from the cancer cells and the tumor microenvironment (18). The mechanisms that drive functional suppression of NK cells and hence undermine important innate anticancer immune responses remain incompletely understood.

IL27 is a member of the IL6/IL12 cytokine superfamily and is an important regulator of immune responses (19). IL27 is composed of two subunits, IL27p28 and IL27EBI3. The IL27 receptor (IL27R), composed of two chains, WSX-1/IL27RA and gp130, is expressed by multiple immune cell subsets, including NK cells, and by some nonhematopoietic cells (20). IL27R transmits signals primarily via STAT1 and STAT3 activation (19), among which STAT3 is a hepatocyte-intrinsic driver of HCC (21). IL27 was shown to have a broad anti-inflammatory role in infectious and chronic immune-mediated diseases (22–24). IL27RA ablation leads to elevated production of proinflammatory cytokines, including IL17A and IL6 (23, 25), indicating that in inflammation-driven cancers such as HCC, IL27 could potentially play a tumor-restrictive, anti-inflammatory role. In vitro and in vivo experiments revealed that IL27 enhances LAG3, TIM3, PD-1, and TIGIT-inhibitory molecule expression in T cells (26). Although immunomodulatory effects of IL27 were shown in various pathophysiologic models (27–31), the context-dependent and cell type–specific underlying mechanisms remain incompletely understood. The protective role of IL27 and its receptor signaling in cancer has been suggested based on its anti-inflammatory role in inflammatory diseases and studies using subcutaneous cell line transplants (32–34) or spontaneous cancer development in a context of p53 deficiency (35); nevertheless, the role of IL27R signaling in cancer development in vivo has not been comprehensively addressed.

Here we found that genetic loss of Il27ra surprisingly suppressed liver cancer development in two different faithful in vivo models of HCC: diethylnitrosamine (DEN) carcinogen–driven and nonalcoholic steatohepatitis (NASH)–driven HCC. Analysis of several independent human HCC cohorts revealed that expression of IL27RA or increased serum protein levels of the IL27 cytokine subunit EBI3 are associated with poor survival and enriched in patients with a more advanced stage of the disease. Mechanistically, we found that IL27R signaling limited the accumulation and activation of innate cytotoxic cells (NK and ILC1) in tumors, and Il27ra genetic ablation relieved the inhibition and resulted in infiltration of more activated cells that correlated with reduced HCC burden. Moreover, IL27R signaling repressed the expression of NK cell–activating stress ligands on tumor cells. Inactivation of IL27R signaling, in turn, relieved this suppressive effect, most notably in bona fide cytotoxic NK-cell populations and “cytotoxic-like” ILC1 populations, and enhanced cytotoxic antitumor immune responses in vivo. The dependence of IL27R signaling–driven immunoregulatory effects in HCC on innate cytotoxic cells was further established by NK1.1⁺ cell depletion, which reverted the effect of IL27R inactivation on HCC. Both IL27 cytokine and IL27R were amenable to pharmacologic blockade resulting in reduced HCC, associated with heightened NK1.1⁺TCRβ⁻ cell accumulation and upregulation of cytotoxic molecules, establishing the importance of this ligand–receptor pair in HCC biology and providing a rationale for their therapeutic neutralization. Taken together, our data suggest that inactivation of IL27R signaling enhances innate cytotoxic cell cytotoxicity, providing new therapeutic opportunities in liver cancer as well as preventive approaches in high-risk patients with NASH and liver fibrosis who are likely to progress to HCC.

Although the role of IL27R signaling has been investigated in infectious and inflammatory diseases (36–40), unequivocal evidence about its possible contribution to tumor development in vivo is missing. We assessed the protein levels of IL27 cytokine in human serum and found that the IL27EBI3 subunit was significantly elevated in patients with HCC compared with healthy controls (Fig. 1A). These increased levels of IL27EBI3 were similarly observed in hepatitis B or C virus–positive or –negative patients and were independent of alpha-fetoprotein (AFP) levels (Supplementary Fig. S1A and S1B). Moreover, the probability of survival negatively correlated with the serum level of IL27EBI3, where patients within the highest tertile group of IL27EBI3 protein demonstrated the poorest overall survival, whereas patients within the lowest tertile showed the best overall survival (Fig. 1B). Next, we evaluated the impact of IL27RA receptor expression in human HCC tumors on disease-free survival using The Cancer Genome Atlas (TCGA) data. Patients with high IL27RA expression exhibited poor disease-free survival and presented with more advanced stages of HCC (Fig. 1C and D). We also obtained similar results in previously described independent “SNU” (41) and “LCI” (42) HCC cohorts (Fig. 1E–G). These correlative expression data in human HCC samples may suggest a potentially cancer-promoting role of IL27R signaling in HCC. Such a role was nevertheless surprising, as active IL27R signaling was previously shown to reduce the expression of IL6, IL17A, and other inflammatory cytokines deemed protumorigenic in HCC (12, 23). The analysis of a published single-cell RNA sequencing (scRNA-seq) data set of CD45⁺ cells from human HCC (43) revealed that IL27RA is expressed by most of the tumor-infiltrating immune cells, whereas IL27EBI3 and IL27P28 expression is mostly restricted to myeloid lineages (Fig. 1H–J). Similarly, IHC staining showed that at the protein level, IL27RA was expressed by infiltrating immune cells as well as cancer cells in human HCC tumors (Supplementary Fig. S1C).

To directly evaluate the role of IL27R signaling in HCC, we used a genetic approach in the well-established mouse model of HCC driven by the carcinogen DEN (10, 44, 45). To exclude the potentially confounding influence of microbiota, we used cagemate and littermate controls. Male Il27ra^+/− or Il27ra^−/− mice were injected with 25 mg/kg of DEN i.p. at day 15 after birth, and tumor development was assessed at 10 months of age. We found that HCC tumorigenesis was markedly reduced in Il27ra^−/− mice compared with Il27ra^+/− controls (Fig. 1K and L), supporting the results from human cohorts. The body weight of tumor-bearing mice was not affected by Il27ra deficiency (Supplementary Fig. S1D). Serum level of ALT, a marker of liver damage reflecting HCC formation, was significantly lower in Il27ra^−/− mice compared with their Il27ra^+/− tumor-bearing counterparts, suggesting a preserved liver function in the absence of IL27R (Fig. 1M). No difference was detected in serum levels of albumin, globulin, or total bilirubin (Supplementary Fig. S1E). Consistent with the limited tumor burden, HCC tumors from Il27ra-deficient mice were also characterized by lower expression of the inflammatory marker lipocalin-2 (Lcn2; Fig. 1N), proliferation markers Ki-67 (Fig. 1O) and cyclin D1 (Ccnd1; Fig. 1P), and reduced activation of kinase ERK1/2 (Supplementary Fig. S1F). No difference in fibrosis was detected between genotypes (Fig. 1Q; Supplementary Fig. S1G), likely reflecting the inflammation-driven nature of the DEN model being less dependent on fibrosis (46). To determine if limited proliferation is a broader representation of IL27R-dependent liver regenerative capacity, we compared liver regeneration in Il27ra-deficient and Il27ra-sufficient mice. Partial hepatectomy of two thirds of liver tissue revealed no significant differences between Il27ra^−/− and control Il27ra^+/− mice in the ability to regenerate liver mass (Supplementary Fig. S2A). Stimulation of HCC cells derived from DEN tumors with recombinant IL27 (rIL27) or short hairpin RNA knockdown of Il27ra did not affect cell growth in vitro as demonstrated by clonogenic assay (Supplementary Fig. S2B–S2D), despite IL27-mediated induction of STAT3 phosphorylation in HCC cells (Supplementary Fig. S2E). These data suggest that IL27R likely regulates HCC development independent of its effects on liver regeneration or cell-autonomous effects on hepatocyte and cancer cell proliferation.

To further assess the contribution of IL27R signaling to HCC in cancer versus immune and stromal cells, we first orthotopically implanted DEN-derived HCC cells with knockdown of Il27ra or control into the liver of Il27ra-sufficient mice. The lack of IL27R on tumor cells in the context of the IL27R-sufficient immune environment did not affect tumor growth (Supplementary Fig. S2F and S2G). Conversely, we transplanted Il27ra-sufficient HCC cells into the livers of Il27ra^−/− or control mice and observed a substantial reduction of tumor growth in Il27ra^−/− mice (Supplementary Fig. S2H and S2I). Collectively, our data demonstrate that IL27R signaling in the microenvironment promotes HCC in the DEN carcinogen–induced model in vivo.

HCC progression is regulated by various immune cells accumulating in tumor and nontumor tissue (12). In order to elucidate potential mechanisms by which IL27R signaling controls HCC, we first conducted gene expression analysis on isolated DEN-induced HCC tumors from Il27ra-deficient and Il27ra-sufficient animals. Among the most notable changes, we found an increase in NK cell–specific markers and overall upregulation of the NK-mediated cytotoxicity pathway in tumors of Il27ra^−/− mice (Fig. 2A). We also found the enrichment in NK-cell signature genes to be the highest among other immune cell signatures in the absence of IL27R (Fig. 2B and C). Moreover, we found an upregulation of IL15, a cytokine driving NK-cell differentiation toward a more mature, activated state (47, 48), that was especially pronounced in tumors of Il27ra^−/− mice (Fig. 2D), suggesting a potential role of IL27R signaling in the control of NK cells in HCC. Flow cytometry analysis of immune cell composition in tumor and adjacent nontumor tissue did not reveal any significant differences in numbers of CD8 or CD4 T cells as well as monocytes or neutrophils (the latter collectively representing the intratumoral myeloid-derived suppressor cell population) in tumors, whereas we detected an increased presence of neutrophils in nontumor tissues and a reduction of macrophages in tumors of Il27ra^−/− mice (Supplementary Fig. S3A–S3C). In agreement with the data obtained in mRNA gene expression analyses, we found a significant increase in the percentage of NK1.1⁺TCRβ⁻ cells in tumors of IL27R-deficient mice compared with IL27R-sufficient controls (Fig. 2E and F), which was unlikely due to changes in NK-cell proliferation, as no difference in BrdU incorporation was detected (data not shown). The NK1.1⁺TCRβ⁻ population consists of at least two different cell lineages: bona fide NK cells and ILC1 (49, 50). Although these two lineages were proposed to develop via distinct pathways, they have a lot of phenotypic and functional similarities with some degree of organ- and disease-related specificity (51), complicating the unequivocal discrimination between these subsets (49, 50, 52). To better characterize the NK1.1⁺TCRβ⁻ population and further distinguish NK cells and ILC1, we first performed FACS staining for Eomes, Perforin, CD49a, and CD49b markers. We found that the NK1.1⁺TCRβ⁻ population in nontumor and tumor liver tissue contains both NK cells (Eomes⁺Perforin⁺CD49b⁺NK1.1⁺TCRβ⁻) and ILC1 (Eomes⁻Perforin⁻CD49a⁺NK1.1⁺TCRβ⁻) and that bona fide NK cells outnumbered ILC1 (Fig. 2G and H; Supplementary Fig. S3D). Although both populations had been detected, CD49b⁺ NK cells but not CD49a⁺ ILC1 cells were found to be significantly increased in tumors of DEN-treated Il27ra^−/− mice (Fig. 2G and H).

Analysis of human HCC gene expression data using CIBERSORT for NK-cell proportion estimations confirmed a significant but relatively small negative correlation between the presence of NK cells and IL27RA and IL27EBI3 expression in tumors (Fig. 2I). Moreover, analysis of published human HCC scRNA-seq data (53) also showed a negative correlation between IL27RA expression and NK cell activation signature (Fig. 2J). These data altogether suggest that the expression of cytokine IL27 or its receptor negatively correlates with the presence of NK cytotoxic cells in both mouse and human HCC.

To further gain insights into the role of IL27R signaling in the control of ILC and NK-cell function and cytotoxicity in HCC, we performed the flow cytometry analysis of NK1.1⁺ cells from nontumor and tumor tissues and found that the ablation of Il27ra led to the upregulation of granzyme B (Fig. 3A). qRT-PCR analysis of FACS-sorted NK1.1⁺TCRβ⁻ cells from HCC tumors and adjacent normal tissues revealed the upregulation of Gzmb and Tnfsf10 mRNA, implying enhanced activation and cytolytic potential of these cells (Fig. 3B). We also observed an upregulation of Cxcr6 in NK1.1⁺TCRβ⁻ cells from Il27ra^−/− mice (Fig. 3B), a chemokine receptor that is characteristically expressed on hepatic innate cytotoxic cells and known to regulate their accumulation (54, 55).

To further elucidate the role of IL27R signaling in NK and other cells in HCC, we performed scRNA-seq of FACS-sorted Live/Dead⁻CD45⁺ NK1.1⁺ and NK1.1⁻ cells from nontumor and tumor tissues of DEN-treated Il27ra^−/− and Il27ra^+/− mice using droplet-based 10X Genomics Chromium scRNA-seq (Supplementary Fig. S4A and S4B). The resulting quality-controlled single-cell atlas included 41,772 cells that were clustered based on Seurat's graph-based clustering approach (56) and visualized using a uniform manifold approximation and projection (UMAP) plot. Cluster annotation and identification were corroborated using overlapping marker genes from Immgen and previous transcriptional profiling of mouse NK cells and ILCs (ref. 57; Supplementary Fig. S4C). The combined analysis of all captured CD45⁺NK1.1⁺ and CD45⁺NK1.1⁻ cells revealed the presence of 34 clusters comprising at least 19 different immune cell types (Supplementary Fig. S4D). Similar to the data from human HCC (Fig. 1H–J), Il27ra expression was observed on many hematopoietic cells, with the highest expression on NK- and various T-cell subsets, whereas Il27Ebi3 cytokine expression was restricted mostly to myeloid cell subsets (Supplementary Fig. S4E). As a heightened accumulation of NK1.1⁺ cells was detected in tumors of Il27ra^−/− mice, we further concentrated on scRNA-seq analysis of Live/Dead⁻CD45⁺ NK1.1⁺ cells. Clustering analysis revealed the presence of NK cells, ILC1, and NKT cells among 9,511 sorted NK1.1⁺ cells (Fig. 3C), among which we identified two clusters of ILC1 (1,642 cells) and seven clusters of NK cells (4,376 cells; Fig. 3D). Cell type–specific gene expression analysis further confirmed the specificity of the clusters (Supplementary Fig. S5A–S5E). Permutational multivariate analysis of variance (PERMANOVA) analysis of gene expression between Il27ra^−/− and Il27ra^+/− tumor samples for NK and ILC1 clusters showed the most significant changes in NK clusters 3 and 5, whereas less significant changes were found in ILC1 clusters 2 and 4 (Fig. 3E). Therefore, we decided to further elucidate transcriptional changes among NK cells and ILC1 and their dependence on IL27R signaling. First, we performed trajectory analysis that identified three potential trajectories of NK-cell differentiation that start from the least differentiated (or immature) cluster, 7. Two of the trajectories progress toward clusters 1, 3, and 5 (Fig. 3F; Supplementary Fig. S6A). Importantly, initial recruitment of NK cells was elevated (cluster 7) in tumors of Il27ra^−/− mice, whereas cell numbers at the intermediate stages (clusters 0, 6, 8) were depleted at the expense of increased numbers of NK cells progressing through the trajectories toward terminal clusters 1, 3, and 5 in the absence of IL27R signaling (Fig. 3G). The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of differential gene expression (DEG) along the trajectory demonstrates upregulation of several metabolic pathways and NK-cell cytotoxicity genes along the differentiation trajectory toward more mature clusters (Fig. 3H). Indeed, changes in cellular metabolism have been previously linked to enhanced NK-cell activation and cytotoxicity (58, 59). In order to gain insights into the cluster identity along the trajectories, we first performed pseudotemporal ordering of single-cell transcriptomes, which revealed that NK-cell differentiation trajectories indeed move toward more mature cytotoxic effectors (clusters 1, 3, and 5), as suggested by the upregulation of cytotoxic genes Gzmb and Prf1, downregulation of “immature” genes Emb, Kit, and Il7r, and no changes in IFNγ (Fig. 3I; Supplementary Fig. S6B). DEG analysis showed that intratumoral NK cells from Il27ra^−/− mice were more activated (Orai1, Nfatc1, Litaf, Irf8, Ifngr1, Ifnar, Tnfsf1a, Tnfsf1b, Il21r, Dgat1, and Nfil3 were upregulated; whereas Ccl3, Ccl4, Atf3, Dusp2, Zfp36, and Car2 were downregulated; refs. 60–63). They also expressed lower levels of inhibitory receptors (Klrc1, Klrc2, and Klre1) known to interact with HLA-E, which regulates NK-cell antitumor activity (63). The expression of various motility genes (Tubb2a and Tubb4b; ref. 64) was enhanced, and S1pr5 was downregulated, suggesting activation and retention of NK cells in the tissue. Lastly, the expression of cytotoxicity genes (Gzmb, Prf1, and Serpinb9; ref. 65) was also enhanced (Fig. 3J). Overall, these data imply that ablation of IL27R signaling enhances the activation and cytotoxicity of NK cells and induces their differentiation toward more mature subsets. On the contrary, IL27 signaling deters the acquisition of maturity and cytotoxicity by NK cells, thereby playing an immunoregulatory role.

Among ILC1, we identified two clusters (Fig. 3D) and established the trajectory of their differentiation from cluster 4 to cluster 2 (Fig. 3K; Supplementary Fig. S6C), as the absence of IL27R increases the differentiation of “cytotoxic-like” ILC1 (cluster 2) at the expense of “conventional” ILC1 in cluster 4 (Fig. 3L). Pseudotemporal ordering of single-cell transcriptomes revealed the upregulation of Gzmb, but not Prf1, whereas the expression of markers of immature cells (Il18r1, Tcf7 and Kit) was reduced during the progression from cluster 4 to cluster 2. No difference in Ifng was detected (Fig. 3M; Supplementary Fig. S6D). However, we found an upregulation of Gzmb, Gzma, Gzmc, Serpin6b, Fasl, Il21r, Dgat1, Cd7, Ccr5, Cd160, Itga1 and Nfil3 and downregulation of Atf3, Dusp2, Zfp36, Hif1a and Id2 genes, similar to recently described (65) transcriptional changes during ILC1 differentiation toward “cytotoxic-like” ILC1 (Fig. 3N). KEGG pathway analysis further confirmed upregulation of TNF signaling and “NK-cell cytotoxic pathways” in ILC1 (Fig. 3O). Therefore, IL27R signaling represses the emergence of the “cytotoxic” profile in a subset of ILC1 similar to its function in NK cells. Ridgeline plot visualization of gene expression across multiple clusters of tumor NK cells and ILC1 further demonstrated the upregulation of representative cytotoxicity genes Gzmb, Faslg and Prf1 and chemokine receptor Ccr5 in NK clusters 1, 3, and 5 in tumors of Il27ra^−/− mice compared with Il27ra^+/− controls (Supplementary Fig. S5B). Although the loss of IL27R also induced cytotoxic gene expression in ILC1 (cluster 2), the level of their expression in ILC1 cells was lower compared with NK cells, especially for a key cytotoxic marker, Prf1 (Supplementary Fig. S5B). The expression of antitumorigenic Ifng, which also functionally defines ILC1, was similar between NK and ILC1 populations and did not depend on IL27R signaling (Fig. 3I and M; Supplementary Fig. S6B and S6D).

Altogether, scRNA-seq analysis of tumor NK1.1⁺ revealed that in HCC, NK cells outnumber ILC1, IL27R signaling negatively regulates terminal differentiation of intratumoral NK cells and ILC1 toward a cytotoxic phenotype, and ablation of IL27R signaling alleviates this differentiation block and promotes the expression of the cytotoxic program.

NK-cell activation is controlled by multiple activating and inhibitory receptors, which regulate their “active” versus “inactive” state (66). NKG2D activating receptor engagement by “stress molecule” ligands on target cells increases NK-cell activity (54, 67–70), whereas inhibitory receptor Ly49C engages MHC class I (H-2K^b and H-2D^b) on target cells and dampens NK-cell activation (67, 71, 72). We found an upregulation of activating receptor NKG2D (Fig. 4A and B) and downregulation of inhibitory receptor Ly49C (Fig. 4C) specifically on CD49b⁺ NK cells from HCC tumors of Il27ra^−/− mice. These data are in agreement with scRNA-seq results showing downregulation of multiple inhibitory receptors, including Klrc1, Klrc2, and Klre1, in NK cells from Il27ra^−/− mice (Fig. 3J). No significant changes in NKG2D or Ly49C expression were detected in blood or splenic NK1.1⁺CD49b⁺ NK cells or liver NK1.1⁺CD49a⁺ ILC1, suggesting organ- and cell-type specificity (Supplementary Fig. S7A–S7C). Next, we used conventional flow cytometry markers of NK-cell maturation and found an elevated presence of terminally differentiated mature cytotoxic CD11b⁺CD27⁻ NK cells (73) in HCC tumors from Il27ra^−/− mice, whereas less mature CD11b⁺CD27⁺ NK cells were found in tumors of Il27ra^+/− controls (Fig. 4D) similar to what we observed in scRNA-seq.

To further understand the functional implication of IL27 signaling in the regulation of NK-cell activity, we performed ex vivo cytotoxicity and degranulation assays. Sorted NK cells were cocultured with YAC-1 target tumor cells followed by the analysis of killing efficiency and degranulation of CD49a⁺ ILC1 or CD49b⁺ NK cells (Fig. 4E–G). We found that the cytotoxicity of both liver- and spleen-derived cells was elevated in the absence of IL27R, whereas liver NK cells were more efficient killers (Fig. 4F). The ability of CD49b⁺Il27ra^−/− cells to degranulate was also heightened (Fig. 4G). Next, we performed in vivo cytotoxicity assay with RMA-S (sensitive to NK killing) and RMA (insensitive) cell lines, which were labeled with fluorescent dyes and injected intraperitoneally at a 1:1 ratio into naive Il27ra^−/− or Il27ra^+/− mice. FACS analysis of the peritoneal lavage 48 hours after cell administration revealed a substantial reduction of RMA-S NK-sensitive cells in Il27ra^−/− mice compared with Il27ra^+/− controls, indicating more efficient NK cell–mediated cytotoxicity (Fig. 4H and I). This further supported our ex vivo observations and demonstrated enhanced NK-cell cytotoxicity in the absence of IL27R signaling.

To test whether IL27 can directly regulate NK cells, sorted NK and ILC1 cells from the spleen and liver of wild-type mice were stimulated in vitro with rIL27. IL27 suppressed the expression of various cytotoxic and activating genes including Gzmb, Faslg, Ifng, and Klrk1 (Fig. 4J) in CD49b⁺ NK liver cells. CD49a⁺ ILC1 demonstrate different responses to rIL27 stimulation and did not downregulate cytotoxic genes Gzmb and Klrk1 in contrast to CD49b⁺ cells (Supplementary Fig. S7D). Splenic CD49b⁺ NK cells showed lower responsiveness to rIL27 stimulation (Supplementary Fig. S7E), confirming site and organ specificity.

In order to further characterize the link between immediate liver injury, IL27R signaling, and NK-cell accumulation, we analyzed livers from mice subjected to acute carcinogen-induced liver injury. Eight-week-old Il27ra^+/− and Il27ra^−/− mice were administered 100 mg/kg DEN, and gene expression in the liver was analyzed 48 hours later by qRT-PCR. We found strongly elevated expression of Cxcl9 and Cxcl10 chemokines as well as Prf1 and Tnfsf10 in the livers of DEN-treated Il27ra^−/− mice (Supplementary Fig. S7F and S7G), indicating that IL27R controls innate cytotoxic cell recruitment and cytotoxicity during both early and late stages of HCC development and may counteract early immunosurveillance exerted at the level of tumor seeds.

Taken together, these data suggest that IL27R signaling is implicated into the control of innate cytotoxic cell accumulation, activation, and cytotoxicity in the liver during HCC development, affecting NK cells, which are most abundant, and also ILC1.

In addition to the acquisition of a mature functional cytotoxic program, NK cells require the engagement of activating receptors on NK cells with corresponding stress-induced ligands on target cells and a loss of inhibitory signals from MHC-I complexes. In particular, NK-cell activation is dependent on RAE-1 and H60 families of activating “stress” ligands (74, 75) and is further enhanced by MHC-I downregulation, a “missing self” signal on target cells (76). The analysis of “stress” ligand expression in normal or tumor tissues of mice with DEN-induced HCC revealed a significant upregulation of Raet1 and H60b expression in tumors of Il27ra^−/− mice (Fig. 4K), which was confirmed by flow cytometry and IHC (Fig. 4L and M). Conversely, rIL27 was able to suppress Raet1 and H60b expression in HCC cells (Fig. 4N). Furthermore, a significant downregulation of surface MHC-I expression was observed on tumor cells in DEN-treated Il27ra^−/− mice compared with Il27ra^+/− controls (Fig. 4O), along with a reduction in Tap1 expression, a regulator of MHC class I peptide loading (Fig. 4P). These data imply that IL27R signaling promotes the expression of MHC-I and represses the upregulation of NK-triggering “stress” ligands on tumor cells, thereby serving as an indirect and direct immunoregulator of NK-cell activity and NK cell–mediated antitumor responses.

Although the incidence of liver cancer caused by hepatitis B and C declines due to efficient vaccines and therapies (77), obesity-driven HCC is on the rise (9, 78). Obesity drives the development of NASH, strongly promoting HCC (79). To complement our studies performed in the carcinogen-induced DEN model, we next sought to generalize our observations using a NASH-dependent model of HCC. We crossed Il27ra^−/− and MUP-uPA mice in which the uPA (urokinase plasminogen activator) transgene is controlled by the mature hepatocyte-specific promoter (MUP). MUP-uPA expression combined with Western diet (WD) feeding drives strong fibrosis, steatosis, and spontaneous HCC development, producing an overall faithful model of NASH-driven HCC similar to humans (80–82). MUP-uPA⁺Il27ra^−/− and MUP-uPA⁺Il27ra^+/− control mice were fed the WD starting at 8 weeks of age for a total period of 8 months. Similar to observations in the DEN-induced HCC model, IL27R-deficient MUP-uPA⁺ mice were largely protected from cancer development (Fig. 5A and B), whereas no significant difference in body weight was found (Supplementary Fig. S8A). Tumors from MUP-uPA⁺Il27ra^−/− mice were characterized by the reduction of Ccnd1 and Lcn2 expression (Fig. 5C and D), implying reduced proliferative capacity of tumors and inflammation in the absence of IL27R signaling. Augmented fibrosis is associated with fatty liver disease and is a driver for HCC development (83). We found a reduction of fibrosis in livers of MUP-uPA⁺Il27ra^−/− mice compared with controls as determined by Van Gieson and Trichrome (Fig. 5E and F) and anti–α-SMA IHC stainings (Supplementary Fig. S8B). These data suggest that IL27R signaling is implicated in the control of liver fibrosis and NASH-driven HCC.

Analysis of published scRNA-seq data from livers of mice developing NASH (84) further confirmed the cell-specific pattern of IL27R and IL27 cytokine subunit expression (Supplementary Fig. S8C–S8F), similar to what we observed in mouse and human HCC (see Fig. 1H–J; Supplementary Fig. S4D). No changes were observed in CD4 or CD8 T cells (Fig. 5G), but an elevated accumulation of NK1.1⁺TCRβ⁻ cells in tumors of MUP-uPA⁺Il27ra^−/− mice (Fig. 5H) was detected by flow cytometry. Among NK1.1⁺TCRβ⁻ cells, CD49b⁺NK1.1⁺ (NK cells) were significantly upregulated in tumors of Il27ra^−/− mice (Fig. 5I) along with a heightened expression of Cxcr6 and Gzmb (Fig. 5J). Consistent with the results in the DEN model, we detected an elevated expression of Raet1 and H60b in tumors of IL27R-deficient MUP-uPA⁺ mice (Fig. 5K), whereas MHC-I and Tap1 were downregulated (Fig. 5L and M). These data further advanced a functional and mechanistic link between IL27R signaling and regulation of natural cytotoxic cell–mediated antitumor immunity and tumorigenesis in an independent model of HCC of different etiology.

Next, we sought to establish the “linear” mechanistic link between IL27R signaling, NK-cell activity, and HCC. We first tested whether NK cells are essential for the antitumor effect of IL27R ablation. We depleted NK (and ILC1) cells using the anti-NK1.1 antibody in DEN-treated Il27ra^−/− and Il27ra^+/− mice for a total of 5.5 months prior to tumor development evaluation. The efficiency of NK1.1⁺ cell depletion was confirmed by FACS analysis of blood, nontumor, and tumor tissues (Supplementary Fig. S9A and S9B). The depletion of NK1.1⁺ cells enhanced tumor growth in Il27ra^−/− mice and eliminated the differences between the Il27ra^−/− and Il27ra^+/− cohorts (Fig. 6A and B), suggesting that innate cytotoxic cells are one of the key functional immune populations regulated by IL27R in the context of HCC tumorigenesis. The depletion of NK1.1⁺ cells also augmented liver fibrosis in both IL27R-deficient and IL27R-sufficient mice, eliminating the difference between genotypes (Fig. 6C; Supplementary Fig. S9C and S9D).

NKp46 is a natural cytotoxicity receptor expressed by NK cells essential for their activation (85). FACS analysis of liver NK cells confirmed elevated NKp46 expression in Il27ra^−/− mice compared with Il27ra^+/− controls (Fig. 6D). Moreover, in vitro stimulation of sorted CD49b⁺ NK cells with rIL27 suppressed NKp46 (Ncr1) expression, suggesting that IL27R signaling could be at least partially involved in the regulation of this cytotoxicity receptor and therefore may affect the functional activity of NK cells (Fig. 6E). We, therefore, took advantage of the NKp46-GFP reporter strain (Ncr1^gfp), in which the Ncr1 gene is replaced by green fluorescent protein (GFP; ref. 86). Heterozygous or homozygous disruption of NKp46 in these mice was shown to affect expression levels of NKp46, affecting cell activation and differentiation (87). We crossed Ncr1^gfp to Il27ra^−/− mice to assess HCC development. Ncr1^+/gfpIl27ra^−/− and Ncr1^+/gfpIl27ra^+/− mice were administered with DEN as described above, and tumor development was analyzed at 10 months of age. Similar to NK-cell antibody depletion, reduced NK-cell activation capacity in Ncr1^+/gfp mice negated the antitumor effect of IL27R deficiency and resulted in a similar tumor load and number of tumors formed in both IL27R-deficient and IL27R-sufficient mice (Fig. 6F and G; see Fig. 1K and L for comparison). Importantly, heterozygous Ncr1^+/gfp status did not affect tumor development in IL27R-sufficient mice. Therefore, we conclude that not only the presence but also the activation of cytotoxic NK cells and possibly ILC1 are required for the antitumor effect of IL27R deficiency.

In order to test the therapeutic potential of IL27/IL27R signaling inhibition, we used an IL27 neutralizing antibody (SRF381), which binds to IL27p28 and blocks IL27-induced signaling (Supplementary Fig. S10A–S10C). MUP-uPA⁺ mice were fed a WD for 8 months and treated with the SRF381 antibody or isotype control for the last 3.5 months. We found a significant reduction of HCC in mice that received SRF381 compared with IgG2a isotype controls (Fig. 7A–C). Administration of the anti-IL27 antibody also suppressed fibrosis as determined by Trichrome and anti–α-SMA staining (Fig. 7D and E). Flow cytometry analysis revealed no alterations in CD8 or CD4 T cells (Fig. 7F) but an increased presence of NK1.1⁺ cells in tumors of mice that received anti-IL27 (Fig. 7G). Gene expression analysis showed upregulation of Gzmb, Tnfsf10, Klrk1, and Cxcr6 in nontumor and tumor tissue of mice that received SRF381 (Fig. 7H), suggesting that the neutralization of IL27 suppresses HCC via upregulation of innate cytotoxic mechanisms.

Furthermore, to assess the feasibility of therapeutic IL27R receptor blockade and to establish that both the IL27 ligand and receptor are essential, we performed pharmacologic inhibition of IL27R. Anti-IL27R or control antibody was administered to WD-fed MUP-uPA⁺ mice for the last 4 months of WD feeding. Similar to genetic models and IL27 neutralization, we found that pharmacologic blockade of IL27R also suppressed tumor growth (Fig. 7I–K), establishing an important role for the IL27 pathway in HCC tumorigenesis.

Overall, our data suggest that IL27R signaling controls HCC tumor development in vivo through a new immunologic checkpoint regulating innate cytotoxic NK cells and ILC1. Particularly, we found that IL27R signaling suppressed NK-cell accumulation, terminal differentiation, and activation by controlling the expression of the cytotoxic program, activating and inhibitory receptors on NK cells, as well as MHC-I and stress-induced ligands on cancer cells. Pharmacologic inhibition of IL27 or its receptor significantly suppressed HCC development and growth, implying that the IL27 pathway may serve as a promising therapeutic target for patients with this devastating disease.

Primary liver cancer is the fifth most common cancer worldwide, accounting for over 800,000 deaths each year (88). Despite efforts to curb HCC via the implementation of vaccines against hepatitis B and drugs against hepatitis C, the incidence of liver cancer is sharply on the rise because of the increased prevalence of fatty liver disease coupled with the epidemic of obesity and type II diabetes as well as increased exposure to environmental toxins and pollutants (88). Although chronic inflammation is known to contribute to liver cancer development, mechanisms of anticancer immune responses in HCC are not completely understood. As many patients with chronic liver disease are set to progress to HCC at some time point, approaches to curb HCC progression and new avenues for HCC therapies are in dire need.

IL27 and its receptor signaling have been implicated in the regulation of inflammation in various acute and chronic inflammatory diseases (23, 36, 38, 40, 89). The role of IL27 in cancer development remains controversial and has not been extensively investigated in faithful in vivo models. Although IL27 was originally described as a proinflammatory cytokine, subsequent studies demonstrated its largely anti-inflammatory role, placing IL27 into the clan of immunoregulatory cytokines (90). We and others have previously shown that inactivation of IL27R in chronic inflammatory diseases results in enhanced inflammation and production of IL6, IL17A, and other cytokines (23, 24, 39), suggesting that perhaps in HCC IL27 would play an anti-inflammatory role, limiting protumorigenic inflammation, and that IL27R signaling inactivation will result in heightened HCC development.

Here we found that elevated IL27RA mRNA expression correlates with more advanced stages of HCC and poor survival since the initial treatment in the TCGA, LCI, and SNU HCC patient cohorts. Moreover, we found that patients with HCC have elevated serum levels of the IL27EBI3 subunit compared with healthy controls, which is in line with a recent prospective cohort study in which serum IL27 was found to be predictive of de novo HCC formation in patients with liver disease (91). Furthermore, using two different in vivo models of HCC—a carcinogen-induced injury-promoted (DEN) model and a NASH-driven (MUP-uPA) model—we found that genetic inactivation of IL27R suppressed HCC.

The liver microenvironment is uniquely enriched for NK cells, where they also exert immune surveillance (16). NK cells are potent killers of senescent, infected, or cancerous cells and participate in the regulation of immune responses via the production of inflammatory cytokines in the normal liver (16). During HCC development, the number and activation of liver NK cells are gradually reduced due to the chronic exposure to a yet unidentified, presumably tumor-derived stimulus and reduction in the expression of NK cell–attracting chemokines (18). As obesity-induced NASH is one of the major drivers of HCC, it is important to note that NK-cell number and function are also reduced in fatty liver disease (92). The emerging mechanism of IL27R signaling promoting HCC relied on the inhibition of anticancer immune responses, particularly through controlling the accumulation and activation of NK1.1 innate cytotoxic cells, including NK and ILC1. Although some surface markers, such as CD49b on NK cells and CD49a and CD200R1 on ILC1, can distinguish between these two subsets of cells, the majority of cell-surface markers are similar, and, in most cases, only subtle changes in their expression allow for discrimination between these two populations (49). Although both cell populations are capable of performing cytotoxic and cytokine-producing functions, NK cells are believed to be more cytotoxic, whereas ILC1 are more potent producers of cytokines, including IFNγ. Recent work, however, demonstrated the ability of ILC1 to differentiate toward a more cytotoxic subpopulation (49, 65). Furthermore, NK cells and ILC1 are characterized by evident organ- and disease-specific phenotypes and responses, which further complicate the unequivocal, simultaneous analysis of their comparative physiologic roles (52). scRNA-seq of NK1.1⁺TCRβ⁻ cells revealed enhanced NK-cell and ILC1 differentiation toward more mature, more cytotoxic subpopulations in the absence of IL27R, implying that in the context of liver cancer, IL27R signaling plays an important suppressive role in the regulation of innate cytotoxicity. Although the presence of both ILC1 and NK cells has been detected in our models of liver cancer and IL27R signaling repressed cytotoxicity in both populations, CD49b⁺ NK cells outnumbered ILC1 and were elevated in the absence of IL27R signaling. Importantly, in HCC tumors, levels of the key antitumorigenic cytokine IFNγ were unchanged by IL27R status in NK cells or ILC1, arguing that a cytokine-producing function of NK cells and ILC1 is unlikely to be the key mediator of IL27R signaling in HCC. An upregulation of IL15, a cytokine driving NK-cell maturation and differentiation, in Il27ra^−/− mice could explain enhanced NK-cell maturation and cytotoxicity observed in naive and tumor-bearing mice, but IL27R-dependent mechanisms regulating IL15 require further investigation.

The depletion of NK1.1⁺ cells during HCC development demonstrated that the effect of IL27R signaling on HCC is indeed largely dependent on these cells despite specificity limitations of this approach (47), which affects NK, ILC1, and NKT cells. The role of NKT cells in HCC is not fully understood. NKT cells have been shown to play both anti- and protumorigenic roles in HCC, particularly depending on NASH-related or carcinogen-related etiology (93–95). Importantly, IL27R signaling ablation or neutralization uniformly reduced HCC in DEN-induced and NASH-induced models in agreement with poor prognosis regardless of viral or obesity status in patients with HCC. In addition, although being more restricted to NK and ILC1 cells, partial functional impairment of cell activation in Ncr1^+/gfp mice still resulted in enhancement of cancer development in Il27ra^−/− mice similar to observations with NK1.1⁺ cell depletion. These data support the importance of IL27R signaling in the control of natural cytotoxic cells, but additional work will be needed to address the possible involvement of NKT cells.

The execution of NK-cell cytotoxicity against target cells requires the engagement of activating receptors and is further regulated by the balance between activating and inhibitory receptor signaling (66). We observed that CD49b⁺ NK cells isolated from HCC tumors in the absence of IL27R signaling exhibit the upregulation of the activating receptor NKG2D and downregulation of the inhibitory receptor Ly49C (Fig. 4A). scRNA-seq data also confirmed a downregulation of multiple inhibitory receptors (Fig. 3J). Although IL27R signaling has been previously shown to regulate NK-cell function (96, 97), the outcomes were largely dependent on the context, the disease model used, and tissue localization of NK cells, and not specifically attributed to NK cells or ILC1. For example, IL27 together with IL12 and IL2 has been shown to promote NK-cell activity in vitro (96). However, our data demonstrate that in vitro stimulation of sorted liver CD49b⁺ NK cells with rIL27 downregulated gene expression of cytotoxic molecules such as granule components (Gzmb, Prf1), FasL (Faslg), and NKG2D (Klrk1), but it had a limited effect on splenic CD49b⁺ NK cells. Overall, these ex vivo and in vitro experiments supported our observations of heightened cytotoxicity of NK cells in Il27ra^−/− HCC tumors, suggesting a direct suppressive function of IL27R on NK cells that regulates cytotoxic programs and balances activating and inhibitory receptor signaling.

The expression of the so-called “stress ligands” on the surface of cancerous cells as well as the reduction of MHC-I expression to avoid T-cell immunity represents a “second signal” required to enable NK-cell cytotoxicity (74, 76). Our data revealed an elevated expression of NK cell–stimulating stress ligands (Raet1, H60b) and reduction in Tap1 MHC-I–processing molecule and surface MHC-I expression in HCC tumors of Il27ra^−/− mice. Moreover, the direct stimulation of HCC cells with rIL27 suppressed Raet1 and H60b expression, suggesting a potential mechanism of how IL27R may indirectly regulate NK-cell activation and cytotoxicity. Of note, no direct effect of IL27R signaling on the proliferation of normal or transformed hepatocytes had been observed in the cell-autonomous system where immune cells are not present, indicating that regulation of stress ligands phenotypically displays itself only within a complete tumor microenvironment harboring innate cytotoxic lymphocytes. Our data together with recently published work on the overexpression of IL27R in hepatocytes (98) pave the road to further examine additional roles of hepatocyte-specific IL27R signaling in HCC, as well as the possible involvement of IL27R signaling in other immune and nonimmune cell types.

Apart from the direct regulation of NK cells, IL27 can also regulate other cell types, which in turn could mediate NK-cell accumulation and activation via cell contact interactions (PD-1, MHC-I, and 2B4-CD48) or via the production of chemokines and cytokines affecting NK-cell recruitment (CXCL9 and CXCL10), differentiation, and/or maturation (IL15, IL12, IL18, IL10; ref. 99). Toward this end, we found that acute DEN administration triggered a heightened expression of Cxcl9 and Cxcl10 chemokines and increased expression of NK-cell effector molecules such as Prf1 and Tnfsf10 in the livers of IL27R-deficient mice 48 hours after DEN treatment. These observations are consistent with the scenario that IL27R signaling represses the NK cell–based cancer immunosurveillance, thereby enhancing HCC tumor initiation.

With the global epidemic of obesity and type II diabetes, NASH-driven HCC is set to surpass all other forms of HCC in incidence (8, 9). The MUP-uPA/WD HCC model faithfully resembles human HCC driven by fatty liver disease and fibrosis. We found that the inactivation of IL27R not only suppresses tumor development in this model but also strongly reduces the underlying fibrosis. Similar to the DEN model, MUP-uPA⁺Il27ra^−/− mice also showed enhanced innate cytotoxic cell activation, implying the existence of a common mechanism regulated by IL27R signaling. Stellate cells and fibroblasts constituting fibrotic masses in NASH and HCC are often senescent (100), and NK cells were shown to kill senescent fibroblasts, thereby regulating fibrosis in liver disease (101, 102). Enhanced NK-cell cytotoxicity was indeed linked to limited fibrosis observed in MUP-uPA⁺Il27ra^−/− mice. Likewise, NK-cell depletion resulted in enhanced fibrosis, especially evident in DEN-treated Il27ra^−/− mice.

Although several models of liver cancer are available, each of them has its advantages and limitations (80). Several earlier publications attempted to address the role of IL27 in cancer development, particularly utilizing cell transplantation approaches, overexpression of IL27 and IL27R, or rapid induction of multiple oncogenes in hepatocytes (32–34, 98). Although such models have clear advantages in terms of the fast and uniform tumor growth, they may be less perfect for studying mechanisms that rely on a competent tumor microenvironment and chronic inflammation, a key component regulated by cytokines in cancer. Our work here demonstrates that the ablation or pharmacologic neutralization of IL27R signaling limits liver cancer development in two different faithful models of HCC: carcinogen- and injury-promoted HCC (DEN) and NASH-driven HCC (MUP-uPA + WD). Along with human data on poor survival, advanced tumor stages, and lower NK cytotoxicity signatures in IL27RA^Hi HCC patients and heightened serum levels of IL27EBI3 in patients with HCC, our work implies that the IL27 pathway may play a tumor-promoting role in HCC that is generalizable across different models and types of HCC with different drivers. An essential component of this mechanism is mediated through IL27-dependent suppression of intratumoral innate cytotoxic cells, including NK cells and ILC1.

Taken together, our data uncover the important role of IL27R-mediated regulation of innate cytotoxic cells in mouse and human HCC, in which it controls NK-cell accumulation and activation as well as cytotoxicity of both NK cells and ILC1. This process is amenable to neutralization of IL27 or IL27R to suppress HCC, suggesting that inhibition of IL27 signaling represents a potential therapeutic strategy in HCC and liver diseases with a high risk of progression to HCC.

In the CHUM cohort, serum was derived from whole blood obtained prior to HCC resection in 130 patients. Clinical data were collected prospectively by the CHUM Hepatopancreatobiliary Biobank, accredited by the Canadian Tissue Repository Network (CTRnet). None of the patients had received chemotherapy or radiotherapy before surgery. In addition, serum was collected from 11 healthy donors without any medical history of cancer or hepatobiliary disease. Ethics approval for the study was obtained by the local institutional Ethics Board. Written informed consent was acquired from all of the patients.

Serum/plasma levels of EBI3 were determined with a custom sandwich Meso Scale Discovery (MSD) assay using EBI3-specific antibodies (SRF19557 as capture and SRF9D2 as detection). Briefly, MSD QUICKPLEX plates were prepared by coating each well with 40 μL of SRF19557 antibody diluted in PBS (0.5 μg/mL) and incubated overnight at 4°C. Plates were then washed and blocked (1% FBS, 0.01% Tween 20 in PBS) for 1 hour at room temperature (RT), followed by an additional wash. Individual samples from healthy controls or patients with HCC were diluted 1:2 in 0.1% FBS and 0.01% Tween 20 in PBS, and 25 μL of the sample was dispensed per well. The following control samples were tested on each plate: (i) pooled serum (diluted 1:2) from pregnant women (second and third trimesters) known to contain high levels of EBI3 (103); (ii) recombinant human IL27 (R&D Systems or PeproTech) tested at several concentrations; (iii) serum from healthy donors diluted 1:2; and (iv) diluent only. These samples were used as positive and negative controls to normalize values across plates. Samples were incubated for 2 hours at RT, followed by washing and a subsequent incubation with the detection antibody SRF9D2 (0.5 μg/mL) for 1 hour at RT. Following another wash, a goat anti-murine SULFO-TAG–labeled secondary antibody (diluted 1:1000, MSD) was added and incubated for 30 minutes at RT. Finally, plates were washed, and MSD read buffer was added to each well before reading on the MESO QUICKPLEX SQ120 instrument. Raw MSD values were used to calculate EBI3 levels extrapolated from a standard curve of rIL27 (R&D Systems). These EBI3 levels were then normalized across plates based on the values obtained from the pooled serum from pregnant women and a single concentration of rIL27 (10 μg/mL PeproTech) and represented as “normalized­ EBI3.”

TCGA gene expression data were analyzed using Cox regression analysis for the association of overall IL27RA expression with survival that does not require splitting patients into groups, but rather tests the association based on overall expression variation versus survival. Additionally, the top 25% expression cutoff was used to define the correlation between IL27RA expression and disease-free survival. Continuous expression of IL27RA was also used in multivariate Cox regression analysis with age, gender, and cancer grade as covariates. IL27RA was shown to be an independent factor associated with survival, showing P = 0.0179 in this combined model. CIBERSORT NK-cell proportion estimations were used to determine the correlation between IL27RA expression and presence of NK cells in the tumors.

We identified 233 paired HCC tumor and nontumor samples from the LCI cohort (GSE14520; ref. 42) and 70 HCC tumor samples from the SNU cohort (41). For the LCI cohort, IL27RA expression of tumor tissue was compared with its paired nontumor tissue, and tumors with the top 25% of IL27RA tumor-to-nontumor ratio were defined as “high” IL27RA tumors, whereas the rest were defined as “norm.” For the SNU cohort, where the information of nontumor tissues was not available, ROC curve analysis with the Youden index was used to identify the best cutoff value of IL27RA associated with overall survival, which was used to define “high” and “norm” groups of tumors based on IL27RA mRNA expression.

Il27ra^−/− (JAX#018078) and C57BL6/J (wild-type; JAX#000664) mice were purchased from the Jackson Laboratory and crossed to obtain Il27ra^+/− and Il27ra^−/− mice. Ncr1^gfp/gfp were obtained from Dr. Wayne Yokoyama (Washington University, St. Louis) and used to generate Il27ra^+/−Ncr1^+/gfp and Il27ra^−/−Ncr1^+/gfp. For the NASH model, Il27ra^−/− mice were crossed to MUP-uPA⁺ (104) to obtain MUP-uPA⁺Il27ra^+/− and MUP-uPA⁺Il27ra^−/− mice. All mice were on a C57BL/6 background. The genotyping was performed by standard PCR protocols. Mice were housed and bred under specific pathogen–free conditions in an AAALAC-approved barrier facility at Fox Chase Cancer Center (FCCC) or Cedars-Sinai Medical Center (CSMC). Littermate and cagemate controls were used for all experiments. All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at FCCC and CSMC and performed in compliance with all relevant ethical regulations for animal research. Studies characterizing the activity of SRF381 were conducted according to guidelines established and approved by the IACUC at Mispro Biotech.

To induce HCC development, 25 mg/kg of DEN (Sigma-Aldrich, N0258) was administered i.p. into 15-day-old male mice as described (45). Mice were maintained on autoclaved water and a regular chow diet. HCC development was analyzed at 10 months of age. For acute response, tissues were collected 48 hours after DEN (100 mg/kg) administration.

MUP-uPA⁺Il27ra^+/− and MUP-uPA⁺Il27ra^−/− female and male mice were fed the WD (Teklad, TD.88137) for 8 months beginning at 8 weeks after birth. HCC development was analyzed at 10 months of age.

Tumor number was calculated as the sum of all macroscopic tumors in the liver. Tumor load was calculated as a sum of diameters of all the tumors per animal. Serum ALT was measured by the ALT Activity Assay Kit (ab105134).

Anti-NK1.1 (200 μg/mouse; produced from PK136 hybridoma, FCCC cell culture facility) or IgG2a (200 μg/mouse; isotype control, FCCC cell culture facility) were injected i.p. weekly for 5.5 months starting from 4.5 months of age.

Anti-IL27 (500 μg/mouse; SRF381, Surface Oncology) or IgG2a isotype control (500 μg/mouse) antibodies were injected i.p. for the last 3.5 months prior to tissue collection at 10 months. The human monoclonal antibody SRF381 was selected for binding to recombinant human IL27, cross-reactivity to recombinant murine IL27, and for blocking murine IL27-mediated phosphorylation of STAT1 in splenocytes (see Supplementary Fig. S10). The Fc region of SRF381 was replaced with murine IgG2a. Anti-IL27R (150 μg/mouse; Amgen Inc.) or isotype control (150 μg/mouse) antibodies were injected i.p. for the last 4 months prior to tissue collection at 10 months.

Cell lines were obtained from ATCC (YAC-1 (RRID: CVCL_2244), RMA (RRID: CVCL_J385), and RMA-S (RRID: CVCL_2180). HCC cell lines were generated in the cell culture facility of the FCCC from tumors of DEN-injected Il27ra^−/−, Il27ra^+/−, and Il27ra^+/+ mice. All cell lines were used fewer than 10 passages before experiments were conducted. Cell lines were tested negative for Mycoplasma (2015) and have not been authenticated since the first acquisition.

For histologic analysis, a liver lobe (with tumors) was isolated and fixed in 10% buffered formalin (Fisher Healthcare, 23-245685) for 24 hours. Five-μm-thick sections were prepared and stained with hematoxylin (Sigma-Aldrich, HHS32) and eosin Y (Thermo Scientific, 6766007). All images were acquired with a Nikon Eclipse 80i microscope and EVOS Auto FL2.

A blinded pathology review of hematoxylin and eosin slides from a subset of DEN-induced tumors and MUP-uPA-induced tumors did not reveal any meaningful differences between IL27R-deficient and control mice. Tumors in Il27ra^−/− and control (Il27ra^+/−) mice showed a conventional moderately differentiated HCC morphology with predominantly trabecular architecture. Tumor cells exhibited monomorphic nuclei and contained abundant cytoplasm with variable amounts of intracytoplasmic globules (a finding that can be seen in human HCC) and minimal to mild steatosis. The immune cell infiltrate was sparse in both Il27ra^−/− mice and control mice. Occasional tumors exhibited prominent extramedullary hematopoiesis, but this was observed in both Il27ra^−/− mice and control mice. Tumors in MUP-uPA⁺Il27ra^−/− and MUP-uPA⁺Il27ra^+/− mice showed the classic features of the steatohepatitic variant of HCC that arises in NASH in humans, as expected from previous reports describing the histopathologic features of this model (80, 81). Tumors in both MUP-uPA⁺Il27ra^−/− and control mice showed extensive steatosis (mixed small and large droplet, but predominantly large droplet) and hepatocyte ballooning (swelling of tumoral cells with cytoplasmic degeneration and clumping of cytoplasmic contents). The nontumoral liver in both Il27ra^−/− and control mice showed typical features of human NASH, with abundant steatosis and hepatocyte ballooning, more prominent in the centrilobular region (zone 3). The immune cell infiltrate was sparse in both genotypes. Given the similarity in histopathologic features between the two groups, tumor sequencing was not pursued.

For IHC staining 5-μm-thick sections of livers containing tumors were deparaffinized by taking them through four changes in xylene and then washed by four changes in 100% ethanol followed by rehydration in tap water. Antigen retrieval was performed in 1× citrate buffer (Electron Microscopy Sciences, 64142-08) at 95°C for 1 hour, followed by 1 hour cooling down to RT. Then slides were rinsed in tap water for 3 minutes and dehydrated in 100% ethanol for 1 minute, followed by blocking in 3% H₂O₂ in PBS for 10 minutes. Slides were blocked with 5% goat serum in 1% BSA–PBS for 20 minutes, and then they were incubated with primary antibodies for Ki-67 (1:100; BioLegend, 151202, RRID:AB_2566621), p-ERK1/2 (1:400; Cell Signaling, 4370, RRID: AB_2315112), IL27R (34N4G11; Novus Biologicals, NBP2-19015, RRID:AB_2916313), α-SMA (1:500; Abcam, 124964, RRID:AB_11129103), and RAE-1 (R&D Systems, AF1136, RRID:AB_2238016) overnight at 4°C. After washing with 1% BSA–PBS, slides were incubated with secondary goat anti-rat and goat anti-rabbit biotinylated antibodies for 30 minutes at RT, followed by 30 minutes of incubation with streptavidin–horseradish peroxidase (HRP; 1:500; BD Pharmingen, 554066). For developing DAB substrate (Invitrogen) was applied for 3 minutes, followed by washing in water and counterstaining with hematoxylin solution (Sigma-Aldrich, HHS32). Excess of hematoxylin was removed by immersing slides in 0.25% ammonia water, followed by rinsing in water. Slides were mounted with coverslips using Permount mounting medium solution (Fisher Chemical, SP15). All images were acquired with a Nikon Eclipse 80i microscope or EVOS Auto FL2. Microsoft PowerPoint was used for one-step brightness adjustment for all images in parallel. Quantification was done using ImageJ (version 1.51).

For collagen staining, deparaffinized and rehydrated slides were stained for 5 minutes in Van Gieson solution (EMS, 26374-06), followed by dehydration by two changes in 100% ethanol and cleared by two changes of xylenes. All images were acquired with a Nikon Eclipse 80i microscope. Microsoft PowerPoint was used for one-step brightness adjustment for all images in parallel. Quantification was done using ImageJ (version 1.51).

For visualization of collagen fibers and histologic assessment of collagen deposition, deparaffinized and rehydrated slides were refixed in Bouin's solution for 1 hour at 56°C, rinsed in tap water for 5 minutes, followed by staining with Weigert's iron hematoxylin working solution for 10 minutes. Slides were rinsed in tap water for 10 minutes and stained in Biebrich scarlet-acid fuchsin solution for 10 minutes followed by washing in distilled water, differentiated in the phosphomolybdic-phosphotungstic acid solution for 10 minutes or until the collagen was not red, transferred to aniline blue solution for 5 minutes, rinsed briefly in distilled water and differentiated in 1% acetic acid solution for 2 minutes, followed by a wash in distilled water, dehydration, and clearing in xylene. All images were acquired with an EVOS Auto FL2 microscope. Microsoft PowerPoint was used for one-step brightness adjustment for all images in parallel. Quantification was done using ImageJ (version 1.51).

Mice were sacrificed by CO₂ inhalation, and livers were perfused with HBSS containing 2% of heparin (20 USP units/mL) to remove traces of blood. Livers were isolated, and nontumor and tumor tissues were dissected separately and incubated with a cocktail of digestion enzymes containing collagenase I (450 U/mL; Sigma-Aldrich, C0130) and DNase I (120 U/mL; Sigma-Aldrich, D4263) in HBSS (with Ca²⁺/Mg²⁺) for 40 minutes at 37°C with gentle shaking at 150 rpm. After incubation, cell suspension was filtered through a 70-μm cell strainer. Immune cells were enriched by density-gradient centrifugation over Percoll (GE Healthcare, 17-0891-01) at 1,000 × g for 25 minutes without brake (40% Percoll in RPMI 1640 and 80% Percoll in PBS). Leukocyte ring on a border of gradient and parenchymal cells on top were collected, washed, and stained. Spleens were isolated, mashed, and filtered through 70-μm cell strainers. Peripheral blood was collected by cardiac puncture, and erythrocytes were lysed by red blood cell (RBC) lysis buffer (15 mmol/L NH4Cl, 0.1 mmol/L NaHCO3, 0.1 mmol/L sodium EDTA) for 5 minutes at RT. The following antibodies were used to stain cells from liver, spleen, or blood: CD45-PerCP (30F-11; BioLegend, 103130, RRID:AB_893339), CD11b-Pacific Blue (M1/70; BioLegend, 101224, RRID:AB_755986), NK1.1-FITC (PK136; BioLegend, 108706, RRID:AB_313393), NK1.1-PE (PK136; BioLegend, 108708, RRID:AB_313395), TCRβ-Alexa Fluor 700 (H57-597; BioLegend, 109224, RRID:AB_1027648), CD4-APC/Cy7 (GK1.5; BioLegend, 100414, RRID:AB_312699), CD8a-APC (53-6.7; BioLegend, 100712, RRID:AB_312751), Ly6G-APC/Cy7 (1A8; BioLegend, 127624, RRID:AB_10640819), Ly6C-PE/Cy7 (HK1.4; BioLegend, 128018, RRID:AB_1732093), F4/80-APC (BM8; BioLegend, 123116, RRID:AB_893481), Granzyme B-Pacific Blue (GB11; BioLegend, 515408, RRID:AB_2562196), CD27-PE/Cy7 (LG.3A10; BioLegend, 124216, RRID:AB_10639726), Ly49C-Alexa Fluor 647 (4L03311; provided by Kerry Campbell, labeled with AF647), Ly49I-PE (YLI-90; eBioscience, 1943023, RRID:AB_466020), NKG2AB6-APC (16A11; BioLegend, 142808, RRID:AB_11124538), NKG2D-PE (CX5; BioLegend, 130208, RRID:AB_1227712), CD49a-PE/Cy7 (HMa1; BioLegend, 142608, RRID:AB_2749931), CD49b-APC/Cy7 (DX5; BioLegend, 108920, RRID:AB_2561458), CD11b-biotin (M1/70; BioLegend, 101204, RRID:AB_312787), CD31-PE (390; BioLegend, 102408, RRID:AB_312903), TER-119-biotin (TER-119; BioLegend, 116204, RRID:AB_313705), H-2Kb-PE/Cy7 (AF6-88.5; BioLegend, 116519, RRID:AB_2721683), H-2Kb/H-2Db-APC/Fire 750 (28-8-6; BioLegend, 114617, RRID:AB_2750197), and streptavidin-APC/Cy7 (BioLegend, 405208). All antibodies were used at a 1:50 dilution, and LIVE/DEAD Fixable Yellow Dead Cell Stain (Invitrogen, L34959) was used at 1:200.

Nontumor and tumor tissues were homogenized in TRIzol reagent (Invitrogen, 15596018) with 2.8 mm ceramic beads (OMNI Inter­national, 19-646-3) using Bead Ruptor (OMNI International). Total RNA was extracted using the Aurum Total RNA Fatty and Fibrous Tissue Kit (Bio-Rad, 7326870) according to the manufacturer's protocol. Sorted or treated cells were lysed in RLT Plus buffer (Qiagen, 157030074), and total RNA was isolated using the RNeasy Plus Mini Kit (Qiagen, 74136) according to the manufacturer's protocol. Complementary DNA was synthesized using iScript Reverse Transcription Supermix (Bio-Rad, 1708841) with random primers according to the manufacturer's protocol. qRT-PCR was performed with a Bio-Rad CFX 96 Connect Real-Time PCR Detection System using iTaq Universal SYBR Green Supermix (Bio-Rad, 1725124). The following primers were used: Rpl32 (FW 5′-TTCCTGGTCCACAATGTCAA-3′;REV 5′-GGCTTTTCGGTTCTTAGAGGA-3′), Lcn2 (FW 5′-ATTTCCCAGAGTGAACTGGC-3′;REV 5′-AATGTCACCTCCATCCTGGT-3′), Ccnd1 (FW 5′-CTGGCCATGAACTACCTGGA-3′;REV 5′-GTCACACTTGATCACTCTGG-3′), Cxcr6 (FW 5′-GAGTCAGCTCTGTACGATGGG-3′;REV 5′-TCCTTGAACTTTAGGAAGCGTTT-3′), Gzmb (FW 5′-CCACTCTCGACCCTACATGG-3′;REV 5′-GGCCCCCAAAGTGACATTTATT-3′), Prf1 (FW 5′-CTGCCACTCGGTCAGAATG-3′;REV 5′-CGGAGGGTAGTCACATCCAT-3′), Tnfsf10 (FW 5′-TCTGGTCCAGGGGTGTAAAG-3′;REV 5′-TGCTGACCTGCATTCATAGC-3′), Faslg (FW 5′-ACTCCGTGAGTTCACCAACC-3′;REV 5′-TTAAATGGGCCACACTCCTC-3′), Ifng (FW 5′-TGAACGCTACACACTGCATCTTG-3′; REV 5′-GACTCCTTTTCCGCTTCCTGA-3′), Klrk1 (FW 5′-TCAAGCCAGCAAAGTGGGAT-5′;REV 5′-GGACTCGAACAACGAACATTGG-3′), Raet1 (FW 5′-AGCACTTCACGTCACACCAG-3′;REV 5′-TATGGATACACCAACGGGCT-3′), H60b (FW 5′-AGCCTGAGAGAGCTTTCAGAA-3′;REV 5′-GGGTGTCAGAATTATGTTGGGAG-3′), Tap1 (FW 5′-GGACTTGCCTTGTTCCGAGAG-3′;REV 5′-GCTGCCACATAACTGATAGCGA-3′), Cxcl9 (FW 5′-TAGGCAGGTTTGATCTCCGT-3′; REV 5′-CGATCCACTACAAATCCCTCA-3′), Cxcl10 (FW 5′-CCTATGGCCCTCATTCTCAC-3′;REV 5′-CTCATCCTGCTGGGTCTGAG-3′), and Il15 (FW5’-AATCAGATACCGCAATGACCAC-3′; REV 5′-CAGAAGTTGTTTGGGATGGTGT-3′).

Tumor RNA (50 ng) was used for NanoString (Cancer Immuno­panel) in order to analyze the immune profile of the tumor microenvironment measuring the expression of 770 genes according to the manufacturer's protocol. The hybridization between target mRNA and reporter-capture probe pairs was performed at 65°C for 20 hours using the Applied Biosystems Veriti Thermal Cycler. All processing was carried out on a fully automated nCounter Prep Station. Excess of probes was removed, and probe–target complexes were aligned and immobilized in the nCounter cartridge, followed by the image acquisition and data processing by nCounter Digital Analyzer. The expression level of a gene was measured by counting the number of times the specific barcode for that gene was detected, and the barcode counts were then tabulated in a comma-separated value format. The raw digital count of expression was exported from nSolver v3.0 software. Statistically significant differentially expressed genes between genotypes were analyzed by KEGG pathway analysis.

For cell prediction analysis, the data were normalized in R and batch corrected using ComBat-seq in the R package sva (105). The resultant data sets were analyzed using gene set enrichment analysis (GSEA) and compared with a gmx file containing gene signatures of all cell types that we derived in-house from the Immgen database (www.immgen.org) and NK and ILC gene signatures from Robinette and colleagues (57). The GSEA was done in the GSEA software (106, 107). Normalized enrichment scores were visualized using the ggplot2 package in R to represent the abundance of different cell types in the data.

1,000 DEN-derived HCC cells were plated on a 0.1%-swine gelatin precoated 6-well plate in ACL-4 containing 20% FBS in triplicate per condition. Twenty-four hours later, the medium was changed to ACL-4 containing 5% FBS with or without rIL27 (200 ng/mL). On day 4, the medium was refreshed and cells were left to grow for an additional 3 days. On day 7 after the beginning of the treatment, the medium was aspirated, and cells were washed with HBSS (Ca2⁺/Mg2⁺ free) and fixed with 10% acetic and 10% methanol fixing solution for 15 minutes at RT. When the fixing solution was aspirated, plates were left to dry followed by adding 0.4% crystal violet staining solution for 20 minutes at RT. Rinsed-with-tap-water wells were scanned with an EPSON Perfection V600 Photo scanner.

For gene expression analysis, 500,000 DEN-derived HCC cells were plated on a 0.1%-swine gelatin precoated 6-well plate in ACL-4 containing 20% FBS in triplicate per condition. Twenty-four hours later, the medium was changed to complete DMEM with 5% FBS with or without rIL27 (200 ng/mL) for 3 hours at 37°C in a 5% CO₂ cell culture incubator.

Two-third partial hepatectomies were performed as previously described (108). Briefly, mice were anesthetized by isoflurane. Skin and abdominal walls were cut open, and left and right lobes were ligated sequentially with silk suture and excised. Then the abdominal wall was sutured, and the skin was clipped. Liver regeneration was analyzed in 8 days.

For western blot analysis, 500,000 DEN-derived HCC cells were plated on a 0.1%-swine gelatin precoated 6-well plate in ACL-4 containing 20% FBS in triplicate per condition. Twenty-four hours later, the medium was changed to complete DMEM without FBS with or without rIL27 (200 ng/mL) for 0, 15, and 30 minutes at 37°C in a 5% CO₂ cell culture incubator.

Treated HCC cells were washed with HBSS (Ca2⁺/Mg2⁺ free) and lysed in RIPA buffer supplemented with phosphatase and protease inhibitors (Sigma-Aldrich, PPC1010; 200 μL per 106 cells), followed by centrifugation at 15,000 × g for 10 minutes at 4°C to pellet not lysed cell debris. The supernatant was collected for analysis. Protein concentration was determined by the BCA Protein Assay Kit (Sigma-Aldrich, 1001491004) according to the manufacturer's protocol. Cell lysates (40 μg) were separated by 4% to 20% Tris-glycine MINI-PROTEAN TGX gels (Bio-Rad, 456-1094) and transferred to PVD membranes using Trans-Blot Turbo Transfer Pack (Bio-Rad, 1704156). Each membrane was washed with TBST (10 mmol/L Tris-HCl, 150 mmol/L NaCl, 0.1% Tween-20; pH 7.6) and blocked with 5% skim milk in TBST for 1 hour, followed by overnight incubation at 4°C with the appropriate primary antibody: phospho-STAT3 (D3A7; Cell Signaling Technology, 9145S, RRID:AB_2491009) and STAT3 (124H6; Cell Signaling Technology, 9139S, RRID:AB_331757). Loading was evaluated by staining with anti–β-actin–HRP direct conjugate antibody (AC-15; Abcam, ab49900, RRID:AB_867494; 1:50,000) for 1 hour at RT. Each membrane was washed, and primary antibodies were detected with a 1:5,000 dilution of HRP-conjugated rabbit anti-mouse IgG (Cell Signaling Technology, 7076S, RRID:AB_330924) or mouse anti-rabbit IgG (Cell Signaling Technology, 7074S, RRID:AB_2099233). Bands were developed using ECL Prime western blotting detection reagent (GE Healthcare, RPN2232) and visualized with an autoradiography film (LabScientific, XAR ALF 2025).

NK cells and ILC were enriched from the spleen and liver of wild-type male mice by negative selection (Stem Cell, 19855) followed by antibody staining using CD49a-PE (HMα1; BioLegend, 142604, RRID:AB_10945158) and CD49b-PE (DX5; BioLegend, 108908, RRID:AB_313415) and PE-positive selection (Stem Cell, 18755, 17656) using magnetic beads. Purified NK cells were plated at 250,000 cells per well of a 12-well plate in complete RPMI 1640 with 5% FBS with or without 25 ng/mL of rIL27 (BioLegend, 577404) for 12 hours, followed by gene expression analysis.

NK cells were isolated by negative selection (Stem Cell, 19855) from Il27ra^−/− and Il27ra^+/− mice and incubated with lymphoblast target cell line YAC-1 (RRID: CVCL_2244) in a 1:1 ratio at 10,000 per well of a 96-well plate in complete RPMI media for 90 minutes at 37°C. Supernatant was used to assay cytotoxic efficiency of NK cells using CytoTox cell-mediated cytotoxicity assay (Promega, G1780) according to the manufacturer's protocol. The absorbance of light was measured at 490 nm. Killing efficiency was calculated as a ratio of experimental values to a positive control (lysed target cells).

For the degranulation assay, NK cells after the incubation with YAC-1 were stained on ice for 30 minutes with the following antibody cocktail: NK1.1-FITC (PK136; BioLegend, 108706), CD49a-Cy7/PE (HMα1; BioLegend, 142608), CD49b-APC (DX5, BioLegend, 108910), CD45-PerCP (30F-11; BioLegend, 103130), TCRβ-Alexa Fluor 700 (H57-597; BioLegend, 109224), and CD107a-PE (1D4B; BioLegend, 121612, RRID:AB_2134487) and analyzed on a BD Aria II flow cytometer. LIVE/DEAD Fixable Yellow Dead Cell stain (Invitrogen, L34959) was used to exclude dead cells. CD107a was used as a marker of degranulation.

NK cell–mediated cytotoxicity was measured as previously described (109). Briefly, RMA (RRID: CVCL_J385) and RMA-S (RRID: CVCL_2180) T-cell lymphoma cells were labeled with Orange CMRA (Invitrogen, C34551) or CPD eFluor 650 (eBioscience, 65-0840-90) dyes, respectively, and 2 × 10⁵ cells of each cell line were mixed in a 1:1 ratio and injected i.p. to Il27ra^+/− or Il27ra^−/− mice. Forty-eight hours later, mice were sacrificed, and peritoneal lavage cells were collected and analyzed by FACS.

Six-week-old female Balb/c mice were injected with 20 μg of either empty vector or linked murine IL27 minicircle DNA (System Biosciences) in 2 mL 0.9% normal saline via the tail vein over the course of 5 seconds. Injected animals were transferred to an empty cage with a heating pad to recover for 5 minutes. Five days after injection, mice were injected i.p. with 1 mg SRF381 or isotype control antibodies. Twenty-four hours after antibody treatment, whole blood was collected into K2-EDTA tubes and spun for plasma separation.

Spleens were isolated from 6- to 8-week-old female Balb/c mice, and single-cell suspensions were prepared by mechanical disruption followed by RBC lysis in ACK buffer. Splenocytes (3 × 10⁵) were incubated with 10 μL of plasma collected from IL27 minicircle-expressing mice for 30 minutes at 37°C with shaking. Cells were fixed in 5% paraformaldehyde for 5 minutes at 37°C and then permeabilized with BD permeabilization buffer for 15 minutes at 4°C. Cells were stained with FITC-conjugated anti-CD3 and PE-conjugated anti-pSTAT1 Y701 for 1 hour at RT, followed by the analysis on a BD LSRFortessa X20 flow cytometer. pSTAT1⁺ cells were analyzed in the CD3⁺ T-cell and NK1.1⁺ populations using FlowJo_V10 Software.

A published scRNA-seq data set (GSE140228; ref. 43) was analyzed for expression of IL27RA, IL27P28, and IL27EBI3 in patients with HCC using BBrowser software from BioTuring (110). A published scRNA-seq data set (GEO125449; ref. 53), which included HCC cells (n = 788) and tumor-associated nonmalignant cells (n = 2,079), from six HCC patients was analyzed for the impact of HCC IL27RA expression on NK-cell activation. For HCC cells, we identified their IL27RA expression levels, which were used to define IL27RA⁺ tumor (any cells from the same HCC tumor with positive IL27RA expression) and IL27RA⁻ tumor. For the nonmalignant cells, we first used a collection of markers to exclude T cells (CD2, CD3E, CD3D, and CD3G), B cells (CD79A, SLAMF7, BLNK, and FCRL5), tumor endothelial cells (PECAM1, VWF, ENG, and CDH5), cancer-associated fibroblasts (COL1A2, FAP, PDPN, DCN, COL3A1, and COL6A1), tumor-associated macrophages (CD14, CD163, CD68, and CSF1R), and liver cells (EPCAM, KRT19, PROM1, ALDH1A1, and CD24). Subsequently, NK cells were identified by positive expression of KLRB1, KLRC3, KLRC4, and KLRD1, and these NK cells were then grouped according to the IL27RA status of their originated tumors. Lastly, the activation status of NK cells was estimated by the mean expression of NK-cell activation signature used by CIBERSORT (111) and compared between IL27RA⁺ and IL27RA⁻ tumors.

CD45⁺NK1.1⁺ and CD45⁺NK1.1⁻ cells from tumor and nontumor tissue of DEN-treated Il27ra^−/− and Il27ra^+/− mice were FACS sorted, followed by single-cell sorting. The single-cell droplets were generated with a chromium single-cell controller using Chromium Next GEM Single-Cell 3′ Kit v3.1 (10X Genomics, 1000121). Approximately 5,000 to 10,000 cells were collected to make cDNA at the single-cell level. cDNA was fragmented to ∼270 bp, and the Illumina adapters with index, barcode, and unique molecular identifiers were ligated to the fragmented cDNA. After PCR, purification, and size selection, the scRNA libraries were ∼450 bp and sequenced on a NovaSeq 6000 (Illumina) at Novogene. Fastq files were obtained for the bioinformatic analysis.

Using the 10X Genomics Cell Ranger pipeline, reads were aligned to the mouse genome (mm10), and the sequencing depth was subsampled to about 30,000 reads per cell. Further processing, normalization, batch correction, clustering, DEG analysis, and visualization were performed using the Seurat package and the ggplot2 package in R. Contaminant cells were identified and removed through combining the NK1.1⁺ and NK1.1⁻ sorted samples, clustering the cells, and classifying the cell type of each cluster. For the combined analysis, cells that had less than 200 or over 3,400 unique genes were filtered, and cells that had over 7% mitochondrial counts were filtered. Contaminant NK1.1⁻ cell types were computationally removed from single cell–sequenced NK1.1⁺ sorted samples, and NK1.1⁺ cell types were computationally removed from single cell–sequenced NK1.1⁻ sorted samples. Cells that had less than 300 or over 2,400 unique genes were filtered out, and cells that had over 5.5% mitochondrial counts were filtered out as well. Myeloid cell–type marker genes were also filtered out from the NK1.1⁺ data set.

For cell-type prediction for cluster classification, the total gene expression of each cluster was analyzed using GSEA and compared with a gmx file containing gene signatures of all cell types that we derived in-house from the Immgen database (www.immgen.org) and NK and ILC gene signatures from Robinette and colleagues (57). The GSEA was done in the fgsea package in R (112). Markers that define clusters (found via differential expression analysis), NK and ILC markers from McFarland and colleagues (52), and other known cell-type markers (such as Cd3, Prf1, and Gzmb) were used to supplement this analysis for cell prediction.

For DEG analysis, the Wilcoxon rank-sum test was used, and the adjusted P value was calculated based on Bonferroni correction using all features in the data set. Cluster markers were identified as differentially expressed genes in each cluster compared with all other clusters among genes that showed at least 0.25-fold difference (log-scale) between the two populations and were detected in at least 25% of either of the two populations. When comparing samples, differentially expressed genes were identified among genes that were detected in at least 1% of either of the two populations. Identified differentially expressed genes were passed to DAVID Functional Annotation Bioinformatics Microarray Analysis to search for enriched gene sets. PERMANOVA values were calculated using the adonis function of the vegan package in R, with 10,000 permutations and pairwise distances calculated using Bray-Curtis distance.

For trajectory analysis, the Slingshot package in R was used to create and visualize the trajectories (113). DEG analysis with the trajectories was performed and visualized using the tradeSeq package in R (114).

The Student two-tailed t test was used for comparison between two groups. Survival curve data were analyzed using the long-rank (Mantel–Cox) test. The Tukey test was used for multiple comparisons. Data were analyzed using the GraphPad Prism Software (Version 7.0). Data are presented as mean ± SEM; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. P < 0.05 was considered statistically significant.

The data generated in this study are available upon reasonable request from the corresponding author. scRNA-seq data generated in this study were deposited to the Gene Expression Omnibus with accession number GSE200040.

M. Rausch reports a patent for U.S. Patent No. 11,332,524 pending to Surface Oncology. S. Turcotte reports grants from Surface Oncology during the conduct of the study, as well as grants from Bristol Myers Squibb and Iovance Biotherapeutics, and grants and personal fees from Turnstone Biologics outside the submitted work. J. Stagg reports personal fees from Surface Oncology during the conduct of the study, as well as grants and personal fees from Surface Oncology and personal fees from Tarus Therapeutics outside the submitted work. K.S. Campbell reports grants from Janssen R&D Systems, Immune Oncology Biosciences, and Bristol Myers Squibb and personal fees from Horizon Pharma, Immunitas, and Tavotek outside the submitted work, as well as a patent for genetically modified human NK cell lines issued, licensed, and with royalties paid from ImmunityBio. J.A. Hill reports personal fees from Twentyeight-Seven Therapeutics outside the submitted work, as well as a patent for U.S. Patent No. 11,332,524 issued to Surface Oncology, a patent for WO 2019/183499 pending, and a patent for WO 2020/123011 pending. S.I. Grivennikov reports grants from the NCI/NIH during the conduct of the study. E.K. Koltsova reports grants from the NCI/NIH, the National Heart, Lung, and Blood Institute/NIH, and the W.W. Smith Charitable Trust and other support from Surface Oncology during the conduct of the study, as well as a patent for US20220089714A1 pending. No disclosures were reported by the other authors.

T. Aghayev: Formal analysis, investigation, visualization, meth­­odology, writing–original draft. A.M. Mazitova: Formal analysis, validation, investigation, visualization, methodology, writing–review and editing. J.R. Fang: For­mal analysis, validation, investigation, visualization, methodology, writing–review and editing. I.O. Peshkova: Formal analysis, vali­dation, investigation, visualization, methodology, writing–original draft. M. Rausch: Resources, validation, investigation, visualization, methodology, writing–review and editing. M. Hung: Data curation, investigation, visualization, methodology. K.F. White: Resources, validation, investigation, visualization, methodology. R. Masia: Formal analysis, visualization. E.K. Titerina: Validation, investiga­tion, visualization. A.R. Fatkhullina: Validation, investigation, visualization. I. Cousineau: Formal analysis, investigation. S. Turcotte: Formal analysis, investigation. D. Zhigarev: Formal analysis, investigation. A. Marchenko: Formal analysis, visualization. S. Khoziainova: Formal analysis, investigation. P. Makhov: Formal analysis, investigation, visualization. Y.F. Tan: Formal analysis, investigation, methodology. A.V. Kossenkov: Investigation, visualization, methodology. D.L. Wiest: Formal analysis, investigation, methodology. J. Stagg: Supervision, validation, investigation, visualization, methodology, writing–review and editing. X.W. Wang: Data curation, investiga­tion, visualization, methodology, writing–review and editing. K.S. Campbell: Resources, data curation, writing–review and edi­ting. A.K. Dzutsev: Data curation, formal analysis, supervision, investigation, visualization, methodology, writing–review and editing. G. Trinchieri: Resources, data curation, supervision, writing–review and editing. J.A. Hill: Resources, data curation, validation, investigation, methodology, writing–review and editing. S.I. Grivennikov: Conceptualization, data curation, supervision, methodology, project administration, writing–review and editing. E.K. Koltsova: Conceptualization, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We acknowledge the help of the FCCC and CSMC facilities. We thank Dr. Ofer Mandelboim (Hebrew University, Jerusalem) for access to Ncr1^gfp/gfp mice and Dr. Wayne Yokoyama (Washington University, St. Louis) for providing the mice after backcrossing to the C57BL/6 background and microsatellite mapping verification. We thank Amgen Inc. for providing the anti-IL27R antibody. The sera and associated clinical data from patients operated on for HCC were obtained from the Hepatopancreatobiliary Biobank of the Centre hospitalier de l'Université de Montréal, supported by the Université de Montréal Roger Des Groseillers Chair and the Institut du Cancer de Montréal. This work utilized the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov). This work was supported by NIH/NCI Cancer Center Support Grant P30 CA006927 to FCCC; W.W. Smith Charitable Trust, NIH R21 CA202396, R01 HL133669, and R01 HL149946 grants to E.K. Koltsova; and NIH R01 CA227629 and CA218133 grants to S.I. Grivennikov. This work was partially supported by Cedars-Sinai Cancer funds to E.K. Koltsova and S.I. Grivennikov. M. Hung and X.W. Wang were supported by grants (ZIA BC 010877, ZIA BC 010876, ZIA BC 010313, and ZIA BC 011870) from the Intramural Research Program of the Center for Cancer Research, NCI.

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.