NOTCH1 drives sexually dimorphic immune responses in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma presents strong sexual dimorphism, being 2-3 times more frequent in males than in females; however, the role of sex in response to immunotherapies in HCC remains unknown. We demonstrate that NOTCH1, an understudied oncogene in HCC, elicits sexually dimorphic anti-tumor immunity and response to FDA-approved immunotherapies. Surprisingly, males harboring NOTCH1-driven tumors displayed enhanced anti-tumor immune responses, which, in mice, were mediated by dendritic and T cells. Conversely, females harboring NOTCH1-driven tumors presented immune evasion and resistance to immunotherapies through a defect in DC-mediated priming and activation of CD8⁺ T cells in mice, which was restored therapeutically with CD40 agonism. Mechanistically, the sexually dimorphic immunity was mediated by genes in the sex chromosomes but not by sex hormones. Together, our study unravels an unexpected association between NOTCH1 and sex in cancer immunity and highlights the potential of restoring the DC-CD8⁺ T cell axis with CD40 agonism to improve outcomes.

INTRODUCTION

Liver cancer, of which the most frequent type is hepatocellular carcinoma (HCC), is the fourth leading cause of cancer-related mortality worldwide, with over 750,000 new cases annually (1). There are several treatment options available for advanced HCC patients (2) and among those, immunotherapies have demonstrated the best clinical outcomes. The objective response rate (ORR) for atezolizumab (αPD-L1) plus bevacizumab (αVEGFA) is 29.8% – the highest among all therapies for HCC and an unprecedented clinical achievement (3). Similarly, two clinical trials combining αCTLA4 with αPD-1/PD-L1, tremelimumab plus durvalumab (4) and ipilimumab plus nivolumab (5), have shown ORRs of 20% and 34%, respectively. However, with approximately two-thirds of patients remaining unresponsive, there is an urgent need to identify biomarkers of response to enable better patient stratification and alternative combination immunotherapies for resistant patients. In some tumor types, response to immunotherapies is affected by sex (6-10). However, the precise role that sex plays in response to immunotherapies in HCC has not been established (11). This represents an outstanding gap in knowledge, given the strong sexual dimorphism characteristic of HCC, being 2-3 times more frequent in male than in female patients (12).

In an effort to identify precision therapies, novel candidate biomarkers of response to immunotherapies, and successful combination immunotherapies in resistant HCCs, we previously developed a mouse model strategy that enables the interrogation into how different genetic alterations in tumor cells affect anti-tumor immunity and response to immunotherapies (13). The model uses hydrodynamic tail-vein injection (HDTVi) of genetic elements (14) to overexpress oncogenes, delete or mutate tumor suppressor genes, and modulate immunogenicity specifically within hepatocytes (13). With this model, we demonstrated that β-catenin (encoded by CTNNB1) promotes immune escape and resistance to αPD-1 immunotherapy (13), a result that was validated in two separate HCC patient cohorts (15,16). Nonetheless, β-catenin-driven tumors were responsive to the combination immunotherapy atezolizumab plus bevacizumab (17). These findings are of key clinical importance, given that β-catenin is altered in one-third of HCC patients, rendering it one of the most important cancer drivers in the disease (18-20).

One understudied cancer driver in HCC is NOTCH1, which predominantly acts as an oncogene in HCC (21,22). Similar to β-catenin activation, NOTCH1 activation has been described in around 30% of HCC patients (21,22). We found that the dominant male incidence in HCC incidence was unexpectedly lost in the context of NOTCH1 overexpression, suggesting that tumor cells with NOTCH1-overexpression in females may have acquired an advantage. Furthermore, by generating a novel HCC mouse model driven by MYC overexpression and NOTCH1 activation, we demonstrate that NOTCH1 drives a sexually dimorphic anti-tumor immune response whereby tumors in females undergo immune evasion and resistance to analogs of FDA-approved immunotherapies for HCC patients, while tumors in males present enhanced anti-tumor immunity and response to such immunotherapies. Mechanistically, this NOTCH1-driven sexually dimorphic immune phenotype was mediated by differences in sex-chromosome genes. Moreover, the immune escape in females was due to impaired dendritic cell-mediated priming and activation of tumor antigen-specific CD8⁺ T cells, which was therapeutically reversed using a CD40 agonist. Taken together, NOTCH1 is a critical driver of HCC that contributes to sexually dimorphic immune responses that can be exploited therapeutically to improve immunotherapy efficacy.

RESULTS

NOTCH1 activation promotes sexually dimorphic immune responses

NOTCH1 is frequently upregulated in HCC patients (overexpressed in around one-third of HCCs) (21,22) (Fig. 1A) and its expression significantly correlates with genes in the NOTCH signaling pathway, indicating activation (Supplementary Fig. S1A). Interestingly, high-NOTCH1 HCCs (first tertile) were significantly associated with a lower incidence of HCC in males (51.28%) compared to low-NOTCH1 HCCs (second and third tertile; 75.32%) in the TCGA LIHC cohort (Fig. 1B) (18), suggesting that NOTCH1 may be less tumorigenic in males than in females. Similarly, when we assessed the TCGA LIHC cohort for the “atezolizumab bevacizumab response signature” (ABRS; first tertile) generated in the IMbrave150 study (17), we observed that NOTCH1 overexpression was significantly associated with high levels of this signature exclusively in males (Fig. 1C) suggesting that males with NOTCH1-expressing tumors may respond to immunotherapy. Further, NOTCH1 overexpression in the IMbrave150 cohort (17) in females showed a trend towards lower response to atezolizumab plus bevacizumab (defined as either complete response or partial response) (p=0.17) (Supplemental Fig. S1B). Together, these data suggest that NOTCH1 activation may be contributing to an unexpected sexual dimorphism in HCC, wherein tumors overexpressing NOTCH1 would be less tumorigenic and/or respond better to immunotherapy in males compared to females.

Fig. 1. Sex-dependent immune responses in NOTCH1-driven liver tumors.

A, Oncoprint of NOTCH1 overexpression (expression > 0.2 SD above the mean, 117 patients) in 348 HCC patients (TCGA LIHC cohort, in red). B, Percentage of male and female HCC patients with high-NOTCH1 (117 patients) or low-NOTCH1 (231 patients) tumors in the TCGA LIHC cohort. Chi-square test. C, Percentage of HCC patients with high (first tertile) or low (second and third tertiles) atezolizumab-bevacizumab response signature (ABRS) in males (left) or females (right) with high (first tertile) or low (second and third tertiles) levels of NOTCH1 expression. Post-hoc analysis with Chi-squared test. Tertiles were established to account for the whole HCC patient cohort, including sorafenib-treated patients. Numbers indicate number of patients within each category. D, Representative livers from E-H. Scale bar represents 1 cm. E-P, Survival curves in C57BL/6 WT (wild-type) female (E,G,I,J,M,N) or male (F,H,K,L,O,P) mice harboring MYC;sg-p53 (MYC-luc;sg-p53 or MYC-lucOS;sg-p53) tumors (E,F) or MYC;NICD1 (MYC-luc;NICD1 or MYC-lucOS;NICD1) tumors (G-P). Mice were treated with IgG control (I-P) or αCD8a depleting monoclonal antibodies (I-L), αPD-1 monoclonal antibody (M,O), or αVEGFR2 and/or αPD-L1 monoclonal antibodies (N,P). Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test corrected with Bonferroni for two (L) or three (N,P) comparisons.

Ns, not significant. * p < 0.05. ** p < 0.01. *** p < 0.001. **** p < 0.0001.

To dissect the association between NOTCH1 and sex, we made use of a novel mouse model of HCC that we recently developed (bioRxiv 2024:2024.10.23.619856) that is based on the HDTVi of genetic elements (14) to express activated NOTCH1 (pT3-EF1A-NICD1 (NICD), Notch1 IntraCellular Domain, containing the cytoplasmic domain initiating at amino acid 1753) (23), MYC (pT3-EF1A-MYC), and sleeping beauty 13 (SB13) transposase (CMV-SB13) into C57BL/6 mice (Supplementary Fig. S1C). SB13 transposase enables integration and stable expression of the oncogenes (14) while MYC drives the hepatocyte proliferation that characterizes liver regeneration in response to liver damage (24,25). As a control of efficient liver tumorigenesis (13), we performed HDTVi of a CRISPR-based vector expressing an sgRNA against p53 (sg-p53), pT3-EF1A-MYC, and CMV-SB13 (Supplementary Fig. S1C). To account for the potential role of the immune system in the observed phenotypes, we modulated the antigenicity of the tumors by expressing a non-antigenic luciferase (luc) (13) or a highly immunogenic version of luciferase (lucOS) that is fused to three well-characterized immunogenic T cell antigens: SIYRYYGL (SIY), SIINFEKL (SIIN, OVA257-264), and OVA323-339 (26). As previously shown (13), mice harboring antigenic MYC-lucOS;sg-p53 tumors had a significantly longer median survival than mice harboring non-antigenic MYC-luc;sg-p53 tumors, denoting enhanced anti-tumor immunity conferred by tumor antigen expression (Fig. 1D-F). The same immune phenotype was observed in females and males, excluding a role for sex in immune surveillance associated with p53 deletion. In contrast, mice harboring antigenic MYC-lucOS;NICD1 tumors (median survival, MS=38 days) had the same survival as mice harboring non-antigenic MYC-luc;NICD1 tumors (MS=36 days) in females (Fig. 1D,G), indicating immune escape, while tumor antigen expression led to a phenotype of enhanced anti-tumor immunity (MS=53 days) compared to non-antigenic MYC-luc;NICD1 tumors (MS=38 days) in males (Fig. 1D,H). The sexually-dimorphic immune responses were confirmed in an independent experiment (Supplementary Fig. S1D,E), solidifying our findings.

To ensure this sexually dimorphic immune response was not due to variances in transfection efficiency of vectors, MYC immunohistochemistry staining was performed one week after injection of the plasmids (a surrogate for transfection efficiency), which showed similarity in all experimental groups (luc versus lucOS; p53 versus NICD1; male versus female) (Supplementary Fig. S1F), excluding any possible confounding effect due to baseline tumoral load. Furthermore, tumor cell proliferation was similar in female and male NOTCH1-activated tumors (Supplementary Fig. S1G), excluding the potential role of proliferation differences in driving the sexually dimorphic anti-tumor immunity in this setting. To demonstrate that the immune escape observed in MYC;NICD1 tumors in females was not due to MYC overexpression but instead NOTCH1 activation, we injected MYC-luc and MYC-lucOS with SB13 and without NICD1 into female mice, which showed a more profound depletion of luciferase signal exclusively in the MYC-lucOS condition (Supplementary Fig. S1H,I). Similarly, deletion of p53 in MYC;NICD1 tumors was not able to restore immune surveillance in females (Supplementary Fig. S1J), further supporting the dominant role of NOTCH1 in driving immune escape in females. To further substantiate the sexually dimorphic immune phenotype related to NOTCH1, we depleted CD8⁺ T cells in the NOTCH1-driven tumor models. As expected, CD8⁺ T cell depletion (Supplementary Fig. S1K,L) had no effect in females regardless of tumor antigenicity (Fig. 1I,J) but significantly abolished the survival benefit conferred by tumor antigen expression in males harboring antigenic MYC-lucOS;NICD1 tumors (Fig. 1K,L). These data highlight the lack of an effective anti-tumor CD8⁺ T cell response in non-antigenic MYC-luc;NICD1 tumors. Further, these data suggest that the observed sexually dimorphic anti-tumor immune response is owed to an enhanced ability for males to generate an efficacious anti-tumor CD8⁺ T cell response to the tumour antigens in NOTCH1-driven tumors.

Given the role of sex in mounting immune responses, we next investigated its effect in the sensitivity to analogs of FDA-approved immunotherapy regimens. Female mice harboring MYC-lucOS;NICD1 tumors, which are more suited for immunotherapy studies than MYC-luc;NICD1 tumors lacking anti-tumor CD8⁺ T cell activity (Fig. 1I), were resistant to treatment with αPD-1 (Fig. 1M; Supplementary Fig. S1M), the combination of αPD-L1 and αVEGFR2 (Fig. 1N; Supplementary Fig. S1N), and the combination of αPD1 and αVEGFA (Supplementary Fig. S1O); the latter mimicking the standard-of-care for advanced HCC patients. In contrast, male mice harboring MYC-lucOS;NICD1 tumors were responsive to both treatment strategies (Fig. 1O,P; Supplementary Fig. S1M,N), again highlighting a sexually dimorphic phenotype mediated by enhanced anti-tumor immunity in males. Together, NOTCH1 activation enables augmented anti-tumor immunity of antigenic tumors in males compared with females, demonstrating sexually dimorphic immune responses.

Sex-chromosome genes dictate the sexually dimorphic immune phenotypes in NOTCH1-driven tumors

To understand the mechanism behind the sexually dimorphic immune phenotypes between males and females, we performed bulk RNA-sequencing (RNA-seq) and single-cell RNA-sequencing (scRNA-seq) of murine HCC tumors at humane end-point. Comparison of antigenic MYC-lucOS;NICD1 tumors from male and female mice by bulk RNA-seq of advanced tumors revealed a strong enrichment of specific Y-chromosome genes in males (Uty, Kdm5d, Ddx3y, Eif2s3y; Fig. 2A). scRNA-seq analysis of murine liver tumors, which could be uniquely clustered into 16 groups of immune, stromal, and malignant cells by UMAP (27) (Supplementary Fig. S2A-E), confirmed that most of these same genes from the Y chromosome (Uty, Ddx3y, Eif2s3y) were overexpressed in MYC-lucOS;NICD1 malignant cells from males compared to females (Fig. 2B,C). Similarly, these specific Y-chromosome genes (UTY, KDM5D, DDX3Y) were among the most differentially expressed in high-NOTCH1 tumors from males when compared to high-NOTCH1 tumors from females in the TCGA LIHC cohort (18) (Fig. 2D). While it is expected that comparison of tumors from males versus females will identify Y chromosome genes as some of the most differentially expressed, it is important to note that out of 200 genes located in the Y chromosome, only three (UTY, KDM5D, DDX3Y) were significantly differentially expressed in HCC patients and from those three, two (UTY and KDM5D) were recently shown to play a role in adaptive anti-tumor immunity (7), suggesting potential biological relevance.

Fig. 2. Mechanism of immune surveillance in males with NOTCH1 liver tumors.

A, Volcano plot of differentially expressed genes (DEGs) from bulk RNA-sequencing between male and female MYC;NICD1 whole murine tumors. Comparison of murine MYC-lucOS;NICD1 males (right, n = 4) versus females (left, n = 6). Representative DEGs of interest are highlighted. B, UMAP plot of subclustered cell populations as analyzed by single-cell RNA sequencing (scRNA-seq). Malignant cell populations color-coded by murine tumor model. C, Volcano plot of DEGs from scRNA-seq between male (n = 2) and female (n = 6) MYC;NICD1 malignant cells. D, Volcano plot of differentially expressed genes (DEGs) from bulk RNA-sequencing between male and female high-NOTCH1 HCC tumors (first tertile) obtained from TCGA LIHC cohort. Comparison of male (right, n = 60) versus female (left, n = 57) patients. Representative DEGs of interest are highlighted. E, Schematic of experiment in F. F, Survival curves in Four Core Genotypes mice mice harboring MYC-lucOS;NICD1 tumors. Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test corrected with Bonferroni for all comparisons. G, Number of CD8⁺ (left) or CD4⁺ (right) T cell clonotypes (left axis) and clonal distribution (individual bars) detected by single cell TCR sequencing in all tumor types (n = 4 female MYC-lusOS;sg-p53, n= 3 female MYC-luc;sg-p53; n = 3 female MYC-lucOS;NICD1, n=3 female MYC-luc;NICD1, and n = 2 male MYC-lucOS;NICD1). H-I, UMAP plot of subclustered cell populations as analyzed by single-cell RNA sequencing (scRNA-seq). Dendritic cell (DC) populations color-coded by DC subset (H) or by murine tumor model (I). J-K, Violin plots displaying the relative expression by scRNA-seq of selected DC identity genes across each indicated DC population aggregated across all tumor types (J) or selected DC activation genes separated by murine tumor model and sex (K). L-M, Survival curves in C57BL/6 WT male mice (L,M) or BATF3^−/− male mice (M) harboring MYC;NICD1 (MYC-luc;NICD1 or MYC-lucOS;NICD1) tumors. Mice were treated with IgG control or αCD4 depleting monoclonal antibodies (L). Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test corrected with Bonferroni for two (L,M) comparisons.

Ns, not significant. * p < 0.05. ** p < 0.01. *** p < 0.001. **** p < 0.0001.

To dissect the mechanism behind the sexual dimorphism in immune responses, we utilized the Four Core Genotype (FCG) mouse model (B6.Cg-Tg(Sry)2Ei Sry^dl1Rlb/ArnoJ), in which the sex determining region (SRY) of the Y chromosome is deleted and re-inserted into an autosome, enabling the determination of gonadal development to occur independently of sex chromosomes (28). This yields four genetically unique offspring: XY and XX gonadal males plus XY and XX gonadal females (Fig. 2E). By comparing XY and XX gonadal males to each other and comparing XY and XX gonadal females to each other, one can interrogate whether the observed effects are due to genes located in the sex chromosomes. In contrast, by comparing XX gonadal females and XX gonadal males to each other and comparing XY gonadal females and XY gonadal males to each other, one can interrogate whether the observed effects are due to sex hormones (Fig. 2E). As expected for MYC-lucOS;NICD1 tumors, XX gonadal females had significantly poorer median survival compared with XY gonadal males (Fig. 2F). Interestingly, however, XY gonadal females had significantly enhanced median survival, similar to that of XY gonadal males, while XX gonadal males had significantly poorer median survival, similar to XX gonadal females (Fig. 2F). This confirms the role of sex-chromosome genes, and not sex hormones, in driving the sexually dimorphic immune phenotypes in NOTCH1-driven tumors.

Anti-tumor immunity in males harboring antigenic MYC-lucOS;NICD1 tumors relies on the DC-T cell axis

Comparison of antigenic MYC-lucOS;NICD1 tumors from males and females by RNA-seq also revealed a significant enrichment of transcripts related to activated DCs (Cd40, Cd86) (29), CD4⁺ and CD8⁺ T cells (Cd4, Cd8a, Pdcd1), and important chemokines and chemokine receptors (Ccl5, Ccr7) in males (Fig. 2A), indicating a distinct immune phenotype. Similarly, MYC-lucOS;NICD1 tumors from males were significantly enriched in the Hallmark “interferon_gamma_response” signature (30) when compared to MYC-lucOS;NICD1 tumors from females (Supplementary Fig. S2F), suggesting enhanced CD8⁺ T cell activity. Moreover, we found higher expression of MHC Class I molecule H2-K1, beta-2-microglobulin (B2m), and TNF receptor superfamily member 21 (Tnfrsf21) in MYC-lucOS;NICD1 malignant cells from males when compared to those from females by scRNA-seq (Fig. 2C), indicating a higher potential for interaction with immune cells in males. Similarly, CD274 and IL12RB2 were overexpressed in high-NOTCH1 tumors from males when compared to high-NOTCH1 tumors from females in the TCGA LIHC cohort (18) (Fig. 2D). Moreover, high-NOTCH1 tumors from male patients were significantly enriched in the Hallmark “interferon_gamma_response” signature when compared to high-NOTCH1 tumors from females in the TCGA LIHC cohort (18) (Supplementary Fig. S2G). Similar results were obtained by focusing on tumors expressing high levels (first tertile) of NOTCH ligands (JAG1, JAG2, DLL1, DLL4) and comparing males versus females (Supplementary Fig. S2G). Interestingly, only HCC showed an enrichment of the Hallmark “interferon_gamma_response” signature in males versus females in a series of human tumor types (Supplementary Fig. S2H). In fact, other tumor types, such as cholangiocarcinoma, bladder cancer, or melanoma showed an enrichment of the Hallmark “interferon_gamma_response” signature in females (Supplementary Fig. S2H). Together, these results indicate the presence of an enhanced anti-tumor immune microenvironment in males harboring NOTCH1-driven HCCs.

To further interrogate these differences in the tumor immune infiltrates, we utilized our scRNA-seq analysis (Supplementary Fig. S2I). Clonal distribution of the TCR (T cell receptor) in CD8⁺ and CD4⁺ cells (Supplementary Fig. S2J) revealed an enrichment in both clonotypes and clonal expansion in MYC-lucOS;NICD1 tumors from males compared with females, similar to that seen in MYC-lucOS;sg-p53 tumors, which also undergo immune surveillance (Fig. 2G). On the other hand, MYC-lucOS;NICD1 tumors from females displayed similar numbers of clonotypes and clonal expansion to the non-antigenic MYC-luc;sg-p53 females and MYC-luc;NICD1 females (Fig. 2G). These data hint that males with NOTCH1-driven tumors are generating both enhanced and more diverse CD8⁺ and CD4⁺ T cell responses. As CD4⁺ T cells play a critical role in the licensing of DCs to elicit antigen-specific CD8⁺ T cell responses, we explored what differences could be seen in the DCs across the models and sexes. Sub-clustering of DCs revealed three distinct populations (Fig. 2H-J): cDC1 (Xcr1, Clec9a), cDC2 (Cd209a, Ccr7, Relb), and mature DCs enriched in immunoregulatory molecules (mregDCs; Ccr7, Il12b, Relb). Interestingly, these mregDCs displayed higher expression of Cd40, Cd274, and Flt3 transcripts in males with NOTCH1-driven tumors compared to females with NOTCH1-driven tumors (Fig. 2K), implying heightened activation. Given that these data hinted at enhanced cross-talk between CD4⁺ T cells, CD8⁺ T cells and DCs in male NOTCH1-driven tumors, we sought to assess the effect that depleting CD4⁺ T cells or cDC1s (through Batf3-KO mice) would have on the anti-tumor response in males. Tumor antigen expression in male mice harboring MYC-lucOS;NICD1 tumors did not confer a survival benefit upon depletion of CD4⁺ T cells with selective antibodies (Fig. 2L;Supplementary Fig. 2K) or in the absence of cDC1 (Fig. 2M). These data, coupled with the CD8⁺ T cell depletion (Fig. 1L), confirm that the enhanced anti-tumor immunity observed in males harboring immunogenic NOTCH1-driven tumors is mediated by CD4⁺ T cells, DCs, and CD8⁺ T cells.

NOTCH1-driven tumors from females lead to impaired DC-T cell axis

To better understand the immune escape mechanism in females harboring NOTCH1-driven tumors, we performed exhaustive comparison of the tumor immune microenvironment in female mice injected with MYC-lucOS;NICD1 and MYC-lucOS;sg-p53, the latter serving as a sex-matched positive control for immune surveillance in HCC. During early tumorigenesis (i.e. between week one and two after HDTVi) both antigenic MYC-lucOS;sg-p53 and MYC-lucOS;NICD1 livers from female mice harbored similar quantities of both cDC1 and cDC2 (Fig. 3A; Supplementary Fig. 3A-I); however, the activation status of these cDC subsets was impaired in the MYC-lucOS;NICD1 livers compared to MYC-lucOS;sg-p53 livers as indicated by less PD-L1, CD80, and CD86 expression (Fig. 3B-D; Supplementary Fig. 3C-I). One week after HDTVi, both liver infiltrating CD4⁺ T cells (Fig. 3E) and cytotoxic CD8⁺ T cells (Fig. 3F) were significantly less proliferative in MYC-lucOS;NICD1 compared with MYC-lucOS;sg-p53 females. Consequently, three weeks post-HDTVi, there was a significant reduction in the T cell compartment in MYC-lucOS;NICD1 compared with MYC-lucOS;sg-p53 females, with fewer number of total CD3⁺ T cells (Fig. 3G), CD4⁺ T cells (Fig. 3H), CD8⁺ T cells (Fig. 3I), and tumor antigen-specific CD8⁺ T cells (Fig. 3J) in addition to sustained poor proliferation of cytotoxic CD8⁺ T cells (Fig. 3F). Furthermore, significantly fewer tumor-infiltrating CD8⁺ T cells exhibited co-expression of PD-1, CD39, TIM3, Ki-67, and granzyme B in MYC-lucOS;NICD1 livers at week three (Fig. 3K), indicating less antigen-experience, activation, proliferation, and cytolytic capacity. Similar observations were made for tumor antigen-specific CD8⁺ T cells (Fig. 3L,M). These data demonstrate an early impairment of CD4⁺ T cell proliferation, potentially driving a defect in the CD4⁺ T cell-mediated activation and licensing of DCs to cross-present antigen to CD8⁺ T cells. Consequently, DCs have impaired activation and fewer CD8⁺ T cells with cytotoxic potential are observed within livers of female mice harboring MYC-lucOS;NICD1 tumors.

Fig. 3. NOTCH1-driven tumors in females lead to impaired CD8⁺ T cell priming and activation.

A-M, Quantification of the absolute number (A,G-J,L,M), percentage (E,F,K) of indicated cell populations, or expression level of indicated cell surface markers (B-D; MFI, median fluorescence intensity). Flow cytometry analysis was performed one, two, or three weeks (A-M) after injection of vectors within the liver (A-M). Mann Whitney test was performed comparing MYC-lucOS;sg-p53 (red) versus MYC-lucOS;NICD1 (purple) tumors. Each dot represents one mouse. Control livers from non-tumor bearing mice (green) were included as internal controls to detect issues with sample processing/staining and/or data acquisition with the flow cytometer. N, Schematic of experiment shown in O. Pre-activated (plate-bound αCD3 + IL2 + αCD28) OT-I CD8⁺ T cells were transferred retro-orbitally into mice at day 17 after injection with the vectors. O, Survival curve in C57BL/6 WT (wild-type) female mice harboring MYC-lucOS;NICD1 tumors. Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test. P, Schematic of experiment shown in Q-X. Naïve OT-I CD8⁺ T cells (N) were transferred retro-orbitally into mice at day 11 after injection with the vectors. Q-X, Quantification of the percentage (Q-X) of indicated cell populations. Flow cytometry analysis was performed at day 13 (Q-X) after injection of vectors within the liver (X), peripheral blood (Q,R), or tumor draining lymph nodes (S-W). Mann Whitney test was performed comparing MYC-lucOS;sg-p53 (red) versus MYC-lucOS;NICD1 (purple) tumors. Each dot represents one mouse. ACT, adoptive cell transfer; CTV, cell trace violet. Control livers from MYC-luc;sg-p53 tumors lacking antigens (black) were included as internal controls to detect issues with sample processing/staining and/or data acquisition with the flow cytometer.

Ns, not significant. * p < 0.05. ** p < 0.01. *** p < 0.001. **** p < 0.0001.

To assess if differences in circulating CD8⁺ T cells could be observed while tumours developed, blood was sampled at week one, two and three post HDTVi. Whilst the percentage of total circulating CD8⁺ T cells remained the same in females with MYC-lucOS;NICD1 and MYC-lucOS;sg-p53 tumors at all time points (Supplementary Fig. 3J), there was a spike in circulating tumor antigen-specific CD8⁺ T cells at week two in the MYC-lucOS;sg-p53 mice (Supplementary Fig. 3K), preceding the infiltration of these tumor-antigen specific CD8⁺ T cells into the livers of these mice at week three (Fig. 3J). Interestingly, the functionality of these circulating CD8⁺ T cells mirrored what was seen in the liver/tumor infiltrating T cells, with lower Ki67, Granzyme B, CD39 expression on T cells from MYC-lucOS;NICD1 female mice compared with MYC-lucOS;sg-p53 female mice (Supplementary Fig. 3K-O). Additionally, within the tumor-draining lymph nodes, similar levels of total cDCs as well as cDC1s (CD8a⁺) could be detected in mice harboring MYC-lucOS;sg-p53 and MYC-lucOS;NICD1 tumors (Supplementary Fig. 3P,Q). However, at all time-points assessed, antigen-experienced CD8⁺ T cells (PD-1⁺) were either significantly less cytolytic (granzyme b⁺) and/or activated (CD39⁺) in MYC-lucOS;NICD1 compared with MYC-lucOS;sg-p53 tumor-bearing mice (Supplementary Fig. 3R,S). These data further supports that T cell function is systemically impaired in females harboring MYC-lucOS;NICD1 tumors compared with MYC-lucOS;sg-p53 mice.

To further assess the mechanism underlying the poor anti-tumor CD8⁺ T cell response in the female mice harboring MYC-lucOS;NICD1 tumors, we utilized the OT-I CD8⁺ T cell system (Supplementary Fig. 3T). These OT-I CD8⁺ T cells (31) express a transgenic TCR specific for the SIINFEKL peptide that is expressed in tumors containing lucOS. Upon effective activation (either via ex vivo stimulation or by endogenous licensed cDCs presenting SIINFEKL peptide), these OT-I cells should be effectively recruited into livers harboring lucOS-expressing tumor cells and possess the capacity to lyse these cells in an antigen-specific manner. To assess if there was a defect in T cell infiltration into the tumors, ex vivo-activated OT-I CD8⁺ T cells were transferred on day 10 post-HDTVi into mice harboring MYC-lucOS;sg-p53 or MYC-lucOS;NICD1 tumors (Supplementary Fig. 3T). As a control for the enhanced number of CD8⁺ T cells, an additional group of mice were transferred activated non-specific CD8⁺ T cells isolated from wildtype C57Bl/6J mice (Supplementary Fig. 3T). As anticipated, transfer of the activated OT-I cells into mice harboring MYC-lucOS;sg-p53 tumors led to a significant increase in median survival when compared to mice injected with ex vivo-activated non-specific CD8⁺ T cells (Supplementary Fig. 3U). In mice harboring MYC-lucOS;NICD1 tumors, adoptive transfer of ex vivo-activated OT-I CD8⁺ T cells overcame the NOTCH1-driven immune escape, leading to a significant increase in median survival (Supplementary Fig. 3V). Similar results were obtained when adoptive transfer of ex vivo-activated T cells was performed on day 17 post-HDTVi (Fig. 3N,O), at which point there is substantial tumor burden in the mice. These results confirm that, once properly activated, tumor antigen-specific CD8⁺ T cells are effectively able to home to, infiltrate, identify, and eliminate MYC-lucOS;NICD1 tumors in females mice.

These data collectively demonstrate that infiltration and tumor cell recognition/killing capacity of tumor antigen-specific CD8⁺ T cells as well as tumor cell susceptibility to CD8⁺ T cell-mediated killing is intact in females with MYC-lucOS;NICD1 tumors. Further, the quantity of cDCs was comparable between MYC-lucOS;sg-p53 and MYC-lucOS;NICD1 livers but cDC activation and CD4⁺ T cell proliferation/quantity was attenuated in the MYC-lucOS;NICD1 mice. Thus, we hypothesized that the diminished presence and activation of tumor antigen-specific CD8⁺ T cells may be due to impaired licensing of cDCs by CD4⁺ T cells, and subsequently, poor DC-mediated CD8⁺ T cell priming and activation in the context of NOTCH1 activation. To test for the ability of endogenous DCs to prime and activate tumor antigen-specific CD8⁺ T cells, we performed adoptive transfer of naïve, non-activated OT-I CD8⁺ T cells into mice injected with MYC-luc;sg-p53 (negative control), MYC-lucOS;sg-p53 (positive control), or MYC-lucOS;NICD1, at day 11 after hydrodynamic injection. OT-I CD8⁺ T cells were pre-labelled with cell-trace violet (CTV) allowing their in vivo proliferation post-priming/activation to be analyzed two days later via flow cytometry, with fluorescent levels of this trace becoming lower with subsequent cell divisions (Fig. 3P). While there were similar levels of total circulating CD8⁺ T cells in the peripheral blood of all three models (Fig. 3Q), there was a reduced percentage of adoptively transferred OT-I T cells maintained in the blood of MYC-lucOS;sg-p53 and MYC-lucOS;NICD1 mice expressing tumor antigens by day two (Fig. 3R), indicating efficient recruitment of the naïve OT-I cells out of circulation in both antigenic tumor models. However, significantly higher proliferation (as indicated by less CTV retention and more Ki-67 expression) of the transferred OT-I T cells was observed in the tumor-draining lymph nodes of MYC-lucOS;sg-p53 mice (~ 90%) compared to MYC-lucOS;NICD1 mice (~ 60% with high variability amongst mice), while transferred OT-I T cells underwent very little division (around 1%) in MYC-luc;sg-p53 mice (Fig. 3S,T; Supplementary Fig. 3W). Furthermore, fewer of these transferred OT-I T cells in the tumor-draining lymph nodes expressed CD69, PD-1, or granzyme B in MYC-lucOS;NICD1 mice compared with MYC-lucOS;sg-p53 mice, revealing that they were less activated, antigen-experienced, and cytolytic (Fig. 3U-W). Additionally, of the transferred OT-I T cells that did divide and were present in the liver (CTV^loCD44⁺), fewer expressed granzyme B in MYC-lucOS;NICD1 mice compared with MYC-lucOS;sg-p53 mice, further implicating impaired activation (Fig. 3X). This assay directly assessed the ability of endogenous DCs to prime and activate these transferred naïve tumor antigen-specific OT-I CD8⁺ T cells, with the readout being their subsequent proliferation and activation status. These data demonstrate that while tumor antigen-specific CD8⁺ T cell priming and activation by endogenous DCs is not completely abolished within the MYC-lucOS;NICD1 mice, it is significantly impaired, which in turn leads to ineffective recruitment of properly activated tumor antigen-specific CD8⁺ T cells into the liver and, thus insufficient immune-mediated control of tumor burden.

Immunotherapies that stimulate DC-T cell axis are effective in NOTCH1-driven tumors in females

Given the impairment in DC activation and subsequent priming and activation of anti-tumor CD8⁺ T cell responses, we hypothesized that MYC-lucOS;NICD1 tumors could be responsive to a therapeutic strategy aimed at stimulating this critical step. We reasoned that CD40 agonism (CD40a) would overcome the impaired DC activation in this model. We further postulated that combination with αCTLA-4 and αPD-1, currently approved as a combination for HCC (4,5) and with shown efficacy in mouse models (32), would create an effective CD8⁺ T cell response that provides a sustained survival benefit. Treatment of mice harboring MYC-lucOS;NICD1 tumors with this triple combination therapy (CD40a, αCTLA-4, αPD-1) starting day 7 and on more advanced tumors on day 14 post-HDTVi led to a significant enhancement in overall survival (Fig. 4A,B; Supplementary Fig. 4A,B). Unsurprisingly, the T cell therapies (αCTLA-4 and αPD-1) had little effect on their own or in combination (Supplementary Fig. 4C,D). On the other hand, all treatments arms involving CD40 agonism displayed some efficacy, although the triple combination was the only treatment that significantly improved survival over control when correcting for multiple comparisons (Supplementary Fig. 4C,E). The triple combination immunotherapy (Supplementary Fig. 4A) had no effect on the survival of female mice harboring non-antigenic MYC-luc;NICD1 tumors (Supplementary Fig. 4F), highlighting the lack of immunogenicity of these non-antigenic tumors resulting in a failure of these immunotherapies to stimulate tumor-specific DC and T cell responses. Unsurprisingly, the triple combination immunotherapy was therapeutic in female mice harboring antigenic MYC-lucOS;sg-p53 tumors and male mice harboring antigenic MYC-lucOS;NICD1 tumors (Supplementary Fig. 4G,H), both of which respond to αPD-1, indicating that intrinsic anti-tumor CD8⁺ T cells can be further potentiated with the addition of CD40a to enhance DC activation.

Fig. 4. Enhancing DC-T cell axis in females harboring NOTCH1-driven tumors has therapeutic effects.

A-B, Survival curves in C57BL/6 WT (wild-type) females harboring MYC-lucOS;NICD1 tumors and treated with either IgG control or Triple Combo (CD40 agonist, αPD-1, and αCTLA-4) starting on day 7 (A) or day 14 (B) after injection with the vectors. Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test. C-I, Quantification of the absolute number of indicated cell populations (C-G,I) or expression level of indicated cell surface markers (H,I). Flow cytometry analysis was performed in the liver at day 19 after injection of vectors in the liver. Mice were treated with either IgG control or Triple Combo (CD40 agonist, αPD-1, and αCTLA-4) starting on day 7 after injection with the vectors. Mann Whitney test was performed comparing MYC-lucOS;NICD1 mice that received IgG (purple) to those that received triple combination (orange). Each dot represents one mouse. Healthy livers (H, from non-tumor bearing mice) were included as an internal control to detect issues with sample processing/staining and/or data acquisition with the flow cytometer. J-L, Survival curves in C57BL/6 WT (wild-type) (K), BATF3^−/−(J), or muMT (L) females harboring MYC-lucOS;NICD1 tumors and treated with either IgG control or Triple Combo (CD40 agonist, αPD-1, and αCTLA-4) starting on day 7. To deplete macrophages, mice were treated with αCSF1R monoclonal antibody starting at day 3 after injection with the vectors (K). Number of mice per group is shown as well as median survival in days (d). Log-rank Mantel-Cox test corrected with Bonferroni for two comparisons (J-L).

Ns, not significant. * p < 0.05. ** p < 0.01. *** p < 0.001. **** p < 0.0001.

Mechanistically (Supplementary Fig. 4I), while the triple combination therapy did not lead to a significant increase in the number of DCs present in the tumor microenvironment (Fig. 4C), it led to an increase in the total number of T cells, CD4⁺ T cells, and CD8⁺ T cells (Fig. 4D), compared with IgG control. Importantly, the number of SIINFEKL-specific CD8⁺ T cells was consistently but not significantly increased in the livers of treated mice compared with IgG control (Fig. 4E), likely due to heterogeneity observed within the murine tumors. The triple combination therapy promoted higher numbers of CD4⁺ T cells and CD8⁺ T co-expressing CD44, PD-1, and Ki-67, indicative of a more activated T cell phenotype (Fig. 4F,G). Additionally, in mice treated with the triple combination therapy, total CD4⁺ T cells as well as proliferating CD4⁺ T cells expressed higher levels of ICOS (Fig. 4H), a co-stimulatory molecule similar to CD28 expressed on T cells that binds to ICOSL expressed on antigen presenting cells (33). Furthermore, CD8⁺ T cells in the livers of mice treated with the triple combination therapy expressed higher levels of perforin and granzyme b, suggesting greater cytolytic capacity (Fig. 4I). These data reveal the presence of more activated, antigen experienced, and proliferative CD4⁺ and CD8⁺ T cells within the tumor microenvironment of mice harboring MYC-lucOS;NICD1 tumors treated with the triple combination therapy.

To identify the cell type involved in the therapeutic response of the triple combination therapy, we depleted the immune cell populations that typically express CD40. Depletion of cDC1s in BATF3-KO mice (Fig. 4J) led to a significant reduction in survival in female mice harboring MYC-lucOS;NICD1 tumors and treated with the triple combination immunotherapy. In contrast, depletion of macrophages with αCSFR1 (Fig. 4K) or B cells in muMT mice (Fig. 4L) had no effect in the response of female mice harboring MYC-lucOS;NICD1 tumors to the triple combination therapy. Indeed, CD40 was predominantly expressed in mregDCs (Supplementary Fig. 4J), suggesting that these cells may be the targets of the CD40 agonist. Finally, depletion of CD4⁺ T cells with αCD4 had no effect on therapeutic efficacy (Supplementary Fig. 4K), supporting that the mechanism of action of the CD40 agonism is downstream of CD4⁺ T cell help (i.e. CD4⁺ T cell-mediated licensing of DCs through expression of CD40L (Supplementary Fig. 4J), further reinforcing DCs as targets of the CD40 agonist and response to the triple combination therapy.

DISCUSSION

Several tumor types arising in non-reproductive organs show strong sexual dimorphism, being more frequent in men than in women (34): this is the case in HCC, which is 2-3 times more frequent in males than in females (12). Sexual dimorphism in HCC can be explained by differences in behavioral risk factors (men have traditionally been more exposed to HCC risks factors such as alcohol and smoking) and differences in biology (HCC is also more frequent in male mice and rats, which do not consume alcohol or tobacco) (12). Biological differences have mainly been attributed to opposing roles that sex hormones play in liver inflammation and subsequent tumorigenesis: androgen promotes liver tumorigenesis (35) while estrogen plays a protective anti-inflammatory and anti-tumoral role (36). Nonetheless, how sex-specific mechanisms interact with tumor genotype or affect response to immunotherapies has yet to be established.

Sexual differences in immunity have been known for a long time and can also be explained by sex hormones and sex chromosomes. While the effect of sex in immune responses is very complex, it is accepted that in mice and humans immune responses are stronger and more effective in females than in males (37). Sexual dimorphism in response to immunotherapies is a more recent concept and, in some indications, it appears response to immunotherapies may be worse in females (6). One seminal study demonstrated the importance of androgens and androgen receptor in CD8⁺ T cell activity (9). More recently, the role of sex chromosome genes has been established in driving sexually dimorphic immune responses. For instance, Y chromosome loss in tumor cells was shown to promote immune evasion in bladder cancer and loss of two Y chromosome genes, UTY (KDM6C) and KDM5D, was able to recapitulate the effects (7). In contrast, in the context of KRAS-mutant colorectal tumors, Y chromosome KDM5D expression in tumor cells was shown to impair anti-tumor immunity (8). In glioblastoma, expression of Kdm6a (Utx) through X chromosome inactivation escape in CD8⁺ T cells in females led to increased effector function and improved tumor control (10). Despite the proven involvement of sex chromosome genes in sexually dimorphic immune responses, their precise mechanism of action has yet to be elucidated.

Here, by analyzing HCC patient data and performing experiments in a novel mouse model of HCC, we have identified NOTCH1 activation as a driver of sexually dimorphic immune responses. In patients, high expression of NOTCH1 was associated with female sex and improved immune responses in males. In mice, activation of NOTCH1 led to immune surveillance and response to immunotherapies in males but immune escape and resistance to immunotherapies in females. Mechanistically, NOTCH1-driven tumors in males lead to immune surveillance that was mediated by the DC-T cell axis, which was, conversely, impaired in females. This was in part due to impaired DC activation and, subsequently, compromised priming and activation of efficacious anti-tumor CD8⁺ T cell responses. This, however, could be restored by CD40 agonism and subsequent DC reactivation. Finally, by utilizing a mouse model that separates gonadal sex from sex chromosomes (28), we demonstrated that sex differences were mediated by genes in the sex chromosomes and not by sex hormones.

Our study provides several innovative and relevant findings. First, we demonstrate that NOTCH1 activation interacts with female sex in HCC, reverting the archetypical male predominance that is associated with HCC, as it is 2-3 times more frequent in males than in females. Since the male-dominant sexual dimorphism in HCC occurs in the context of liver inflammation, it is possible that NOTCH1 activation is more prevalent in the absence of underlying inflammation. Second, we show that the female-specific immune escape driven by NOTCH1 activation is mediated by sex chromosome genes and not sex hormones: this discovery is again noteworthy as the male predominance is mainly mediated by the effect that sex hormones have in inflammation (36). Third, we demonstrated that NOTCH1-driven immune escape in females is owed to defective DC-mediated generation of effective anti-tumor CD8⁺ T cell responses. Fourth, we identify a therapeutic strategy, CD40 agonism plus αCTLA-4/αPD-1, that not only restores response in female NOTCH1-driven tumors but also enhances response in αPD-1 responsive tumors. Thus, CD40 agonism may have clinical utility when added to current FDA-approved immunotherapies for HCC to restore sensitivity in resistant patients or increase sensitivity in responders. Previous studies had established the potential of CD40 agonism as a strategy to improve response to immunotherapies in preclinical models of cholangiocarcinoma (38) or clinical studies in metastatic pancreatic cancer (39).

One limitation of the study is that we use strong model antigens to trigger T cell-dependent anti-tumor immune responses, which may only phenocopy those HCC patients with high tumor mutational burden and subsequent high antigenicity. However, both our previous study (13) and this current study demonstrate the great utility of using model antigens as it enables the identification of robust mechanisms of immune escape, as there is a large immunogenic hurdle to overcome. Further, these model antigens have high utility in mechanistic experiments (e.g. adoptive transfer experiments with OT-1 T cells). Another caveat is that our model is driven by an activated form of NOTCH1 (NICD1, which is nuclear), limiting the use of our model to NOTCH inhibitors with activity in the nucleus.

Future studies will identify the genes and/or antigens in the sex chromosomes that mediate anti-tumor immune responses orchestrated by NOTCH1 in HCC. Similar to previous reports (7), our findings based on transcriptomic differences between male and female tumors suggest that genes in the Y chromosome may be playing an anti-tumorigenic role. Indeed, loss of Y chromosome has been proposed in HCC as oncogenic (40). Furthermore, it will be critical to establish whether the sexually dimorphic phenotype is driven by the effect of sex chromosome genes in tumor cells, immune cells, or both. It is well known that certain immune cells, such as dendritic cells, can show sexual dimorphism in particular contexts (41,42). Future clinical studies will help elucidate the role that sex plays in response to immunotherapies and whether the interaction between specific genetic alterations (e.g. NOTCH1 overexpression) and sex affects this response. While the use of single and combination immunotherapies has revolutionized HCC clinical care and patient outcomes (43), most patients (around 70%) receiving immunotherapies still do not benefit. The discovery of predictive biomarkers of response and resistance to immunotherapy is, therefore, a pressing challenge, both to enable a precision medicine approach in liver cancer immunotherapy treatment as well as to design novel therapeutic strategies to overcome resistance. Several studies in HCC have identified potential biomarkers of response and resistance, including both gene signatures and specific genetic alterations (15-17,44-46). However, the identification of predictive biomarkers is hindered by the limited access to tumor material from HCC patients and the complexity of highly heterogenous patient cohorts, which present with both high inter-tumor heterogeneity and diverse etiologies. Studies in mouse models of HCC represent a complementary, reductionist approach to nominate potential biomarkers (46), with the advantage of minimal heterogeneity. This enables enhanced granularity to dissect underlying mechanisms and, therefore, the rational design and preclinical testing of novel combination therapies. Indeed, the unexpected association between NOTCH1-induced immune escape and sex had gone unnoticed in human studies, where the confluence of multiple variables (e.g. genetic alterations, etiology, sex, prior treatments, lifestyle, etc.) renders interrogation of all potential interactions extremely difficult.

MATERIALS & METHODS

Vector Use

The CMV-SB13 plasmid was kindly provided by Dr. Scott Lowe (MSKCC, New York). The pT3-EF1a-NICD1 plasmid (Addgene plasmid #46047) (23) was a kind gift from Dr. Xin Chen (University of Hawaii Cancer Center). The pT3-EF1a-MYC-IRES-luc (MYC-luciferase; Addgene plasmid #129775) and pT3-EF1a-MYC-IRES-lucOS (MYC-luciferase-OS; Addgene plasmid #129776) were previously generated (13) from the pT3-EF1a-MYC vector (MYC, Addgene plasmid #92046) (47) from Dr. Xin Chen (University of Hawaii Cancer Center) and Lenti-LucOS (Addgene plasmid #22777) (26) from Dr. Tyler Jacks (MIT). The px330-sg-p53 was previously published and validated (48). The px330 vector was a gift from Feng Zhang (Addgene plasmid # 42230; RRID:Addgene_58779). All constructs were verified by nucleotide Sanger sequencing and vector integrity was confirmed by restriction enzyme digestion.

Hydrodynamic Tail Vein Injection

A sterile 0.9% NaCl solution/plasmid mix was prepared containing DNA. 11.4 μg of pT3-EF1a-MYC-IRES-luciferase (MYC-luc), 12 μg of pT3-EF1a-MYC-IRES-luciferase-OS (MYC-lucOS), 10 μg of px330-sg-p53 (sg-p53), 10 μg of pT3-EF1a-NICD1 (NICD), and a 4:1 ratio of transposon to SB13 transposase-encoding plasmid was dissolved in 2 ml of 0.9% NaCl solution (Intermountain). Mice were injected with the 0.9% NaCl solution/plasmid mix into the lateral tail vein with a total volume corresponding to 10% of body weight in 5-7 seconds (14). Only the hepatocytes transfected with all three plasmids (transposon-based, CRISPR-based or transposon-based, and transposase-encoding) will have the potential to form tumors, since two independent genetic “hits” are necessary for malignant transformation in C57BL/6 mice (14). Vectors for hydrodynamic delivery were produced using the QIAGEN plasmid PlusMega kit (QIAGEN). Equivalent DNA concentration between different batches of DNA was confirmed to ensure reproducibility among experiments. To ensure consistent results within experiments, we use mice that are 4-6 weeks of age (females 5-6 weeks; males 4-5 weeks), around 19 grams of weight (females 17-19 grams; males 18-20 grams), and always include experimental controls to account for variations in tumor dynamics across experiments.

Mice

Wild-type 4–6-week-old C57BL/6 mice were purchased from Envigo for each in vivo experiment. Hydrodynamic tail vein injections were performed in mice six-weeks of age (females) and/or averaging 19 g of bodyweight (males). C57BL/6-Tg(TcraTcrb)1100Mjb/J (OT-I; Jackson Strain 003831), B6.SJL-Ptprc^a Pepc^b/BoyJ (CD45.1; Jackson Strain 002014), B6.129S(C)-Batf3tm1Kmm/J, (BATF3 KO, Jackson Strain 013755), B6.129S2-Ighmtm1Cgn/J, (muMT, Jackson Strain 002288) were obtained from Jackson. One male B6.Cg-Tg(Sry)2Ei Srydl1Rlb/ArnoJ, (Four Core Genotype (FCG), Jackson Strain 010905) breeder was kindly received from Dr. Lisa Satlin, MD at the Icahn School of Medicine at Mount Sinai. Homozygous C57BL/6-Tg(TcraTcrb)1100Mjb/J (OT-I; Jackson Strain 003831) mice were crossed with homozygous B6.SJL-Ptprca Pepcb/BoyJ (CD45.1; Jackson Strain 002014) mice. The resultant offspring, which are hemizygous for the OT-I transgene and CD45.1/2, enabling tracing of cells, were used for adoptive cell transfer (ACT) experiments. All mouse experiments were approved by the ISMMS Animal Care and Use Committee (protocol number IACUC-2014–0229). Mice were maintained under specific pathogen-free conditions, food and water were provided ad libitum and all animals were examined prior to the initiation of the studies to ensure that they were healthy and acclimated to the laboratory environment. Upon sacrifice of the mice with liver tumors, the livers and lungs were collected; samples from each organ were frozen, embedded in OCT (Fisher), and/or formalin-fixed and paraffin-embedded. The attrition rate in mouse experiments was lower than 10%: mice were only removed from the experiment if they had health concerns independent of the experiment (e.g. injuries, infection…).

Luciferase Detection

To assess transfection efficiency or tumor progression within a given experimental condition in vivo as well as stratify mice into various treatment groups assuring equal initial tumor load/transfection efficiency of plasmids administered by hydrodynamic tail-vein injection (HDTVi), bioluminescence imaging was performed using an IVIS Spectrum system (Caliper LifeSciences; purchased with the support of NCRR S10-RR026561-01). Mice were injected intraperitoneally with freshly prepared D-luciferin (150 mg/kg) (Thermo Scientific). Luciferase signal was quantified 5 minutes post-D-luciferin administration using Living Image software (Caliper LifeSciences). Mice were stratified into treatment cohorts such that each cohort had equivalent average luciferase signal before treatment start. Furthermore, mice were allocated in a way that ensured different experimental groups were represented in each cage, to minimize “cage effects”. Mice were only randomized after these two initial steps of stratification. Mice with luciferase signal a log of magnitude lower or higher than the average signal at the start of each study were excluded from the study.

Immunohistochemistry

Immunohistochemical staining of murine liver and lung samples was performed using 4 μm-thick sections from formalin-fixed and paraffin-embedded tissues (FFPE). Tissue deparaffinization and re-hydration was performed through three 5-minute serial xylene treatments followed by incubation in decreasing ethanol concentrations (100%, 100%, 90%, and 80%) for 3 minutes each, then incubated in distilled water for 5 minutes. Heated antigen retrieval in Target Retrieval Solution of the corresponding pH (Dako) was performed at 95°C for 15 minutes, allowed to cool on ice, followed by three 5-minute washes in Tris-Buffered Saline pH 7.4 (TBS; Fisher). Tissue sections were blocked for endogenous peroxidase activity with 3% hydrogen peroxide (Sigma Aldrich) for 10 minutes, then washed three times for 5 minutes with TBS. Serum-Free Protein Block (Agilent) with appropriate serum was applied to tissue sections for 30 minutes, then briefly rinsed with TBS. Primary antibody was diluted in Background reducing Antibody Diluent (Agilent) with appropriate serum and left to incubate overnight at 4°C in a humid chamber. Three 5-minute rinses in TBS with 0.04% Tween20 (TBST) were performed under agitation, followed by incubation in secondary antibody with appropriate serum for 30 minutes (ImmPRESS^® HRP Universal Antibody Polymer Detection Kit; Vector Labs), then another three 5-minute rinses in TBST were performed under agitation. DAB Peroxidase Substrate Kit (Vector Labs) was used, per manufacturer’s instructions, for antibody revelation followed by counterstaining with Mayer’s Hemalum solution (Sigma-Aldrich), then tissue dehydration in serial ethanol (90% 15 seconds, 100% 15 seconds, 100% 15seconds) and xylene (twice for 5 minutes) treatments was performed. Slides were mounted with Permount Media (Fisher), then scanned at 20x resolution using Hamamatsu, NanoZoomer S60. Appropriate positive and negative controls were included with each staining procedure. The following antibodies were used at indicated dilutions: MYC (Y69, abcam, dilution 1:100, pH6) and Ki67 (D3B5, Cell Signaling, dilution 1:200, pH6).

RNA Extraction, RNA-Sequencing, and Analysis

Total RNA from mouse liver tumors (15-25 mg) was isolated using Trizol^® (Invitrogen) followed by digestion with DNase I (Roche) and purification with the RNeasy Kit (Qiagen). RNA sequencing (poly-A selected, PE150, 40 million reads) was performed through Genewiz. Read counts for each transcript were measured using featureCounts (49). Differentially expressed genes were determined using DESeq2 (50) and a cut-off of 0.05 on adjusted p-value (corrected for multiple hypotheses testing) was used for creating gene lists. To perform GSEA, we used GENI platform (https://www.shaullab.com/geni) (51). For selected genes, Student´s t test was performed to test for differences in gene expression values (TPM, transcripts per million) and Benjamini-Hochberg correction methods was employed. The RNA-seq files can be found at GEO (GSE218377).

Tissue Dissociation for Single-Cell RNA Sequencing Analyses

First, mice were sacrificed and then the liver was perfused with PBS+EDTA 2mM through the inferior cava vein. Individual tumor nodules were visually inspected and selected to avoid healthy adjacent tissue and normalize for both size and level of expected necrosis across all samples. The selected tumor was then extracted from the liver using a sterile scalpel, weighed, and manually minced with sterile instruments. Minced tumor nodules were then enzymatically digested with 2.5mL enzymatic digestion mix (Miltenyi Biotech,Tumor dissociation Kit) in DMEM medium in a gentleMACS C Tube (Miltenyi Biotech) and then placed on gentleMACS Octo Dissociator (Miltenyi Biotech) for enzymatic and further mechanical dissociation with the “37°C_m_TDK1” program. Lysates were strained through a 70-μm cell strainer (Miltenyi) and resultant single cells were pelleted at 300g for 5 min at 4°C. From this point onward, cells were kept at 4°C. Elimination of red blood cells using “Red blood cell lysis solution” (Miltenyi Biotech) for 2 min on ice followed by elimination of dead cells using the “Dead Cell Removal” kit (Miltenyi Biotech) was performed according to manufacturer protocols. CD45 positive cells (immune cells) were separated from CD45 negative cells (tumor and non-immune stromal cells) using the CD45 microbeads mouse kit (Miltenyi Biotech) and each fraction was counted manually using a hemacytometer, then recombined at 1:1 ratio. This step was performed to ensure sufficient numbers of immune cells were sequenced from the bulk tumor samples. Only samples with a viability >80% were used.

Single-Cell RNA Sequencing Sample Processing

Single-cell RNA sequencing (scRNA-seq) sample processing was immediately performed after cell extraction/isolation by the HIMC core facility at the Icahn School of Medicine at Mount Sinai. Two single-cell suspensions were counted (Nexcelom Cellometer Auto 2000) and each loaded on one lane of the 10x Genomics NextGem 5’v2.0 assay as per the manufacturer’s protocol with a targeted cell recovery of 10,000 cells per lane. Gene expression (5’Gex) and TCR libraries were generated as per the 10x Genomics demonstrated protocol (https://assets.ctfassets.net/an68im79xiti/57JaTECQNBPSpyDz8oucdi/ced6aa8eaf73d6ee18dea8fdbd945faa/CG000331_Chromium_Next_GEM_Single_Cell_5-v2_UserGuide_RevD.pdf).

All libraries were quantified via Agilent Technologies 4200 Tapestation and KAPA library quantification kit (Roche). Gene expression libraries were sequenced at a targeted depth of 25,000 reads per cell and TCR libraries at a targeted read depth of 5,000 reads per cell. Libraries were sequenced on the Illumina NovaSeq S2 100 cycle kit with run parameters set to 26x10x10x90 (R1xi7xi5xR2). The scRNA-seq files can be found at GEO (GSE218377).

Single-Cell RNA Sequencing Data Pre-Processing, Dimensionality Reduction and Clustering

Libraries were processed with Cell Ranger v5.0.1 and aligned to the mm10 reference genome. Genes were filtered when expressing a cell number of <3. Cells with UMIs sum (nFeatureRNA) of <200 and, mitochondrion gene percentage of >10 were removed by package Seurat (v3.1.5) (27). A total of 87,227 cells passed the filter with an average of ~5800 cells per sample. We then performed dimensionality reduction and unsupervised clustering using Seurat functions. In brief, PCA was performed first, and all PCs were loaded for UMAP dimensionality reduction. To find clusters, the same PCs were imported into FindClusters, an SNN graph-based clustering algorithm, with resolution = 2.5. Next, the Wilcoxon Rank Sum test was performed to identify markers (FDR < 0.05) in each cluster (FindAllMarkers). Cell type annotations were then assigned based on these markers manually after comparing them with other datasets and guided by functional enrichments using the Enrichr database (https://maayanlab.cloud/Enrichr/enrich, version June 8, 2023) (52). Further, to find subpopulations and remove doublets inside these populations, each cluster was segregated and re-analyzed at higher resolution. For the scTCR-seq (single-cell TCR sequencing) analysis, T-cell receptor (TCR) filtered contig files, as generated from Cellranger v5.0.1, were initially linked to individual cell barcode identifiers. TCR sequences were grouped into clonotypes based on CDR3 (complementarity-determining region 3) sequence identity. Clonotypes, defined as groups of at least two cells sharing an identical TCR sequence, were then identified and quantified. Data were analyzed to assess clonal diversity, expansion, and distribution within the sample. TCR data were integrated with other single-cell data (e.g., gene expression) to correlate TCR clonotypes with cellular phenotypes and functional states.

Human HCC Sample Analysis

Gene expression profiling data of HCC patients was obtained from cBioPortal (53) TCGA LIHC dataset (n = 374 ) (18). Samples were stratified depending on high (first tertile) or low (second and third tertiles) NOTCH1 expression. Student´s t test was performed to test for differences in gene expression values (TPM, transcripts per million) and Benjamini-Hochberg correction methods was employed. To perform GSEA, we used GENI platform (https://www.shaullab.com/geni) (51) or XENA (https://xena.ucsc.edu/) (54) (for HCC and other tumor types). Information about response to atezolizumab plus bevacizumab clinical response (CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease) as well as NOTCH1 expression levels and sex (female or male) was obtained from the IMBrave150 clinical trial (n=119 patients treated with atezolizumab plus bevacizumab) (55). Regarding the analysis of this cohort, we defined “high-NOTCH1” tumors as those in the first tertile. The cohort includes both male and female patients treated with both “atezolizumab+bevacizumab” and “sorafenib”. However, we only included in the association studies those patients treated with “atezolizumab+bevacizumab”.

Treatments

Therapeutic treatments were initiated seven days after the hydrodynamic delivery of the plasmids. To confirm therapeutic relevance in therapies that conferred a survival advantage, treatments were initiated 14 days after the hydrodynamic delivery of the plasmids, a time point in which abundant microscopic lesions are present in the livers, maintaining the same scheduling of doses. Mice receiving different treatment conditions were housed together to reduce cage-specific effects. All monoclonal antibodies (mAbs) were administered intraperitoneally in 200 μL of sterile antibody diluent (BioXcell). In the experiments with PD-1 mAbs, three doses of either PD-1 (200 μg, clone RMP1–14, BioXcell) or IgG2a (200 μg, clone 2A3, BioXcell) were given intraperitoneally at days 7, 9 and 11. In the experiments with VEGFR2/VEGFA and PD-L1 mAbs, three doses of either VEGFR2 (200 μg, clone DC101, BioXcell), IgG1 (200 μg, clone HRPN, BioXcell), or VEGFA/IgG (kindly received from Genentech) were administered on days 7, 9, and 11, and three doses of PD-L1 (200 μg, clone 10F.9G2, BioXcell), IgG2b (200 μg, clone LTF-2, BioXcell), or PD-L1/IgG (kindly received from Genentech) were administered on days 8, 10, and 12. In the experiments with CD40 agonist, αPD-1, and αCTLA-4 mAbs, one dose of CD40 agonist (100 μg, clone FGK4.5/ FGK45, BioXcell) or IgG2a (200 μg, clone 2A3, BioXcell) was administered on day 7, and three doses of αPD-1 (200 μg, clone RMP1–14, BioXcell) plus αCTLA-4 (200 μg, clone 9D9, BioXcell) or IgG2a (200 μg, clone 2A3, BioXcell) plus IgG2b (200 μg, clone MPC-11, BioXcell) were administered on days 8, 11, and 14. In the T cell depletion experiments, αCD4 (200 μg, clone GK1.5, BioXcell) plus αCD8 (200 μg, clone 2.43, BioXcell) or IgG (400 μg, IgG2b, clone LTF-2, BioXcell) were administered at days 7, 9, 11, and 13, and then once weekly until the end of the experiment. In the macrophage depletion experiments, αCSF1R (400 μg, clone AFS98, BioXcell) was administered at days 3, 5, 7, 9, and then three times weekly until the end of the experiment.

Isolation of Immune Cells and Flow Analysis

The liver was perfused with 2 mM PBS-EDTA, dissected out, and approximately 1 g of liver was chopped and enzymatically digested using HBSS 1x (Corning) supplemented with 1 mg/ml collagenase IV (SIGMA) and DNase I 0.02 mg/ml (SIGMA) for 30 minutes at 37 °C with agitation. The digestion was neutralized with HBSS 1x and the liver suspension was strained through a 70 μm nylon mesh (Corning) and centrifuged at 50 g at room temperature for 3 minutes to generate a single-cell suspension. The hepatocyte pellet was discarded and the supernatant was centrifuged at 600 g at 4°C for 5 minutes to pellet the immune cells. After centrifugation, the pellet was re-suspended in 36% HBSS-percoll and centrifuged at 800 g at 4°C for 20 minutes without breaks or acceleration. The resultant pellet was incubated with ACK lysing buffer (Gibco) for 5 minutes at room temperature to lyse erythrocytes, then immediately neutralized with cell staining buffer (Biolegend). Immune cells were pelleted at 600 g at 4°C for 5 minutes. For testing successful T cell depletion, 50 μL of blood was obtained through cheek bleeding into an EDTA-coated tube to prevent clotting. The blood was incubated with ACK lysing buffer (Gibco) twice for 7 minutes, then 5 minutes at room temperature to lyse erythrocytes, then immediately neutralized with cell staining buffer (Biolegend). Immune cells were pelleted at 600 g at 4°C for 5 minutes. For tumor draining lymph node analysis, the portal and coeliac lymph nodes were carefully dissected without surrounding fat and gently dissociated through a 70 μm nylon mesh (Corning) using the back of a syringe. The strained cells were washed in 5 mL RPMI 1640 Medium, GlutaMAX^™ Supplement, HEPES (Gibco) with 10% FBS (Gibco) and Penicillin-Streptomycin (Gibco) at 600 g at 4°C for 5 minutes. The resultant immune cell pellet was used for analysis.

Immune cells were pelleted in 96-well plates, Fc receptors were blocked for 15 minutes at room temperature using Biolegend’s TruStain FcX PLUS anti-mouse CD16/32 antibody diluted in Cell Staining Buffer (Biolegend), per manufacturer instructions. Pentamer staining (SIINFEKL, ProImmune) was carried out at room temperature protected from light for 15 minutes diluted in Cell Staining Buffer (Biolegend). Cells were then stained with remaining extracellular antibodies diluted in Cell Staining Buffer (Biolegend) on ice protected from light for 30 minutes. If staining with more than one polymer dye-conjugated antibody was performed, Super Bright Complete Staining Buffer (eBioscience) was used per manufacturer’s instructions. Samples were subsequently incubated with a viability stain in PBS without BSA for 10 minutes at room temperature protected from light followed by fixation. Depending on intracellular stainings, fixation was completed using the Fixation Buffer (Biolegend; no intracellular stain), FoxP3/Transcription Factor Staining Buffer Set (eBioscience; nuclear intracellular stain), or BD Cytofix/Cytoperm (BD Bioscience; cytokine intracellular stain) according to manufacturer´s instructions. An LSRFortessa flow cytometer (BD Bioscience) provided by the Tisch Cancer Institute Flow Cytometry Core was used to acquire the samples. Cells were selected based on size (FSC) and granularity (SSC), doublets were excluded twice using SSC and FSC parameters, then live leukocytes were detected using a viability dye and CD45 stain. Data analysis was performed using FlowJo software (Tree Star). Absolute counts were obtained using CountBright Absolute Counting Beads (Invitrogen) per manufacturer’s instructions. Ultracomp eBeads (Invitrogen) were used for antibody compensation controls. Viability compensation controls were performed by heat-killing cells for 15 minutes at 95°C, then performing viability staining as described above. Normal livers (from non-tumor bearing mice) were included as internal controls to detect issues with sample processing/staining and/or data acquisition with the flow cytometer.

T Cell Isolation and Labeling for Adoptive Cell Transfer

Lymph nodes and spleens were harvested from the OT-I CD45.1/2 mice, placed in sterile complete media, and gently dissociated through a 70 μm cell strainer using the back of a syringe. The strained cells were washed in 10 ml complete media at 600 g at 4°C for 5 minutes and the resultant immune pellet was incubated with 1 ml ACK lysis buffer (Gibco) for 5 minutes at room temperature to lyse erythrocytes, then neutralized/washed in 5 ml complete media at 600 g at 4°C for 5 minutes. The pellet was resuspended in 1 ml sterile PBS and CD8⁺ T cells were negatively selected using EasySep Mouse CD8⁺ T Cell Isolation Kit (StemCell), per manufacturer’s instructions, in polystyrene tubes. Pure isolation of CD8⁺ T cells was confirmed through flow cytometry. For labeling of isolated CD8⁺ T cells prior to adoptive transfer, isolated cells were brought to 1 million cells/ml PBS and stained with CellTrace^™ Violet Cell Proliferation Kit (Invitrogen) per manufacturer’s instructions. Cells were resuspended in appropriate volume of sterile saline to inject 2-3 million cells in 150 μl per mouse retro-orbitally.

T Cell Activation for Adoptive Cell Transfer

As previously described (56), to activate CD8⁺ T cells in vitro, cell culture plates were pre-coated with 5 μg/ml αCD3e (BioXcell; clone 145-2C11) in sterile PBS for 2 hours at 37°C. Prior to plating the cells, the PBS was aspirated from the plates and the cells were resuspended at one million cells/ml in complete media supplemented with 1 μg/ml αCD28 (BioXcell; clone 37.51) and 20 ng/ml Recombinant Murine IL-2 (Peprotech). Cells were left in standard cell culture conditions for 48 hours to activate, then collected, counted, and washed in sterile PBS at 600 g at 4°C for 5 minutes. Cells were resuspended in appropriate volume of sterile saline to inject 150 μl per mouse retro-orbitally. Activation of CD8⁺ T cells was confirmed through flow cytometry. Complete media contained the following: RPMI 1640 Medium, GlutaMAX^™ Supplement, HEPES (Gibco) with 10% FBS (Gibco), Penicillin-Streptomycin (Gibco), MEM Non-Essential Amino Acids Solution 1X (Gibco), Sodium Pyruvate (Gibco), and 2-Mercaptoethanol (Gibco).

Statistical analysis

Data are expressed as mean ± standard deviation (SD) or median ± 95% confidence interval. Statistical significance was determined using Mann-Whitney U test (when n<10 or non-normal distribution) or Student´s t-test (n>10 and normal distribution). For paired comparisons we used the Wilcoxon test. For comparisons of more than two groups we used ANOVA or Kruskal-Wallis test. Group size was determined based on the results of preliminary experiments and no statistical method was used to predetermine sample size. Group allocation for treatments was performed to ensure equivalent baseline tumor load through luciferase signal and cage diversity within groups. Outcome assessment was not performed in a blinded manner. The differences in survival were calculated using the log-rank Mantle-Cox test. Prism 10 software (GraphPad Software) and R studio were used to create the graphs and for the statistical analysis. For multiple comparisons, Benjamini-Hochberg (Student´s t-test) or Bonferroni (Mantle-Cox test) corrections were used. Chi-square test was used to study association. Significance values were set at *p<0.05, **p<0.01, ***p<0.001, and ****<0.0001. Ns, not significant.

STATEMENT OF SIGNIFICANCE.

While HCC presents strong sexual dimorphism, the role of sex in response to immunotherapies remains elusive. With a novel HCC mouse model and validation in HCC patients, we demonstrate that NOTCH1 disrupts anti-tumor immunity specifically in females through a mechanism mediated by sex-chromosome genes, which is reversed with CD40 agonism.

Data Availability Statement

The data generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE218377.