Advanced precision modeling reveals divergent responses of hepatocellular carcinoma to combinatorial immunotherapy.

Dear Editor Combinatorial immunotherapy has provided patients with advanced hepatocellular carcinoma (HCC) the potential for long-term survival. However, sustained responses are seen only in a minority of patients [1]. Thus, there is an unmet need for precision modeling to understand the different responses and uncover predictive biomarkers for treatment stratification. Here, we investigated the responses of HCC to combinatorial immunotherapy in two genetic mouse models, N90-CTNNB1OE;TP53KO and MycOE;TGFαOE mice, which produce overexpression of an activated form of β-catenin (CTNNB1) together with deletion of tumor protein 53 (TP53) or overexpression of the MYC oncogene together with transforming growth factor α (TGFα), respectively. Mutations in these pathways are frequently present in human HCC. We found that the two models of HCC displayed remarkably distinct immune landscapes (Figure 1A, Supplementary Figures S1-S2). Multi-locular tumors developed rapidly (less than 3 months) in both models (Figure 1A, Supplementary Figure S1E). Remarkably, the Myc-driven MycOE;TGFαOE tumors were immunologically “cold” and exhibited high proliferation rates, while N90-CTNNB1OE;TP53KO tumors were immunologically “hot” and showed elevated levels of vascularization and immune cell infiltration (Supplementary Figures S1-S2). Given the increased expression of vascular endothelial growth factor (VEGF) and programmed cell death-ligand 1 (PD-L1) in the N90-CTNNB1OE;TP53KO model (Supplementary Figure S2), we investigated if there are distinct responses of the two HCC models to combinatorial immunotherapy. We administered lenvatinib, a multi-tyrosine kinase inhibitor that inhibits both the VEGF and tumor fibroblast growth factor receptor pathways together with anti-programmed cell death protein 1 (PD-1) antibodies to N90-CTNNB1OE;TP53KO or MycOE;TGFαOE mice with fully established tumors (Supplementary Figure S3A). Remarkably, the combinatorial immunotherapy significantly reduced the burden of large tumors ( ≥ $ \ge \;$ 2 mm-size) compared with IgG isotype control in N90-CTNNB1OE;TP53KO animals, but not in the MycOE;TGFαOE -driven model (Figure 1A, Supplementary Figure S3). Thus, immunologically “hot” N90-CTNNB1OE;TP53KO tumors responded more to combinatorial immunotherapy than immunologically “cold” MycOE;TGFαOE tumors. Next, we investigated the hepatic response to combinatorial immunotherapy in both models in detail by immunofluorescence staining with the B cell marker CD45R and the CD8+ T marker CD8A in combination with Ki67 to assess proliferation and T and B cell activation status. Strikingly, we found that immune cell clusters in the N90-CTNNB1OE;TP53KO model contained tertiary lymphoid structures (TLSs) with B cell follicle and T cell zones after combinatorial immunotherapy (Figure 1B; Supplementary Figure S4). Notably, of the 93 TLSs observed in our study, 88 were located in the tumor periphery, while only 5 were present within the tumor itself (data not shown). A large proportion of follicular B cells were Ki67 positive (74.8% for treatment versus 27.8% for IgG isotope control) in TLSs of CTNNB1OE;TP53KO mice, indicating their high activation state (data not shown). In contrast, the Myc/TGFα-driven tumors exhibited very few TLSs (Figure 1B). Collectively, the less proliferative N90-CTNNB1OE;TP53KO tumors were responsive to combinatorial immunotherapy, which was associated with an augmented TLS response. To better understand the distinct responses of the two HCC models to combinatorial immunotherapy, we employed NanoString technology to perform immune transcriptomic profiling of N90-CTNNB1OE;TP53KO- and MycOE;TGFαOE-driven tumors with or without combinatorial immunotherapy. Notably, the expression of genes relevant to the TLS response was highly elevated in the N90-CTNNB1OE;TP53KO-driven tumor model (Figure 1C, Supplementary Figure S5). We next focused on ten differentially expressed TLS genes previously recognized as crucial for TLS formation, maturation, or both [2-8]. Among these genes are B-cell lymphoma 6 (Bcl6), myocyte-specific enhancer factor 2C (Mef2c), and chemokine [C-X-C motif] ligand 13 (Cxcl13) (Supplementary Figure S5D and S5F), which are markers of TLS, while the seven other genes have been implicated in TLS initiation. Next, we derived TLS-maturation and TLS-initiation scores based on the expression levels of these genes and found both to be significantly higher in N90-CTNNB1OE;TP53−/− versus MycOE;TGFαOE samples prior to and post combinatorial immunotherapy (Figure 1D, Supplementary Figure S5E). Thus, increased TLS initiation and maturation in the livers of N90-CTNNB1OE;TP53−/− mice were associated with their higher sensitivity to combinatorial immunotherapy. Among the TLS genes, only Bcl6, a master regulator for germinal center maturation, and Cxcl13, a marker gene for activated TLS [7], exhibited increased mRNA expression after combinatorial immunotherapy in the N90-CTNNB1OE;TP53KO model (Supplementary Figure S5F). When we evaluated how the activation of these two genes relates to TLS activation, we found that Bcl6 was positively correlated with TLS activation, while the opposite was true for Cd276 (Supplementary Figure S5G). Next, we investigated human HCC gene expression data and discovered a highly significant, positive correlation between CTNNB1 and BCL6 (r = 0.527, P < 0.001) or MEF2C transcript levels (r = 0.504, P < 0.001) (Supplementary Figure S5H), and a highly significant, positive correlation between hypoxia- inducible factor (HIF)1A and BCL6 (r = 0.405, P < 0.001) (Supplementary Figure S5I). HIF1A-binding motif analysis suggested that HIF1A bound specifically to the BCL6 promoter (Supplementary Figure S5J-K). Taken together, these results demonstrate that TLS gene signatures are distinguishing features of the immunologically hot N90-CTNNB1OE;TP53−/− model. Finally, we examined which gene ontology pathways are significantly enriched in response to combinatorial immunotherapy. Not surprisingly, we found that pathways such as adaptive and innate immune responses, TNF signaling, IFN-γ and NF-κB signaling, among others, were significantly enriched (Supplementary Figure S6A). To assess if T-cell exhaustion and T-cell exclusion associate with response and resistance to combinatorial immunotherapy, we derived gene sets to compute T-cell exhaustion and T-cell exclusion scores. When we analyzed features from our transcriptome data that differentiate the two genetic models post combinatorial immunotherapy, we found that expression of Cd276 and the T-cell exclusion gene signature were enriched in the tumor microenvironment of the immunologically cold MycOE;TGFαOE model compared to the CTNNB1OE;TP53KO model (Figure 1C). However, all other features, including the aforementioned TLS initiation and maturation scores and the T-cell exhaustion scores, were enriched in the CTNNB1OE;TP53KO model, both prior to and post combinatorial immunotherapy (Figure 1C, Supplementary Figure S6B). Next, we performed disease-free survival analysis in human HCC patients stratified by the TLS initiation scores calculated using gene expression data. As shown in Figure 1E, high TLS initiation scores were significantly associated with longer disease-free survival. Conversely, low T-cell exclusion scores predicted longer disease-free survival (Supplementary Figure S6C). A model summarizing our findings is shown in Figure 1F. In sum, liver cancer driven by activation of the β-catenin pathway and loss of p53 (the CTNNB1OE;TP53KO model) is an example of immunologically “hot” HCC, which benefits significantly from combinatorial immunotherapy. The TLS-related features we derived from transcriptomics data can predict the efficacy of combinatorial immunotherapy, while immune exclusion predicts immunotherapy failure, exemplified by the highly proliferative but treatment-resistant MycOE;TGFαOE model. Thus, our study suggests that future stratification based on TLS and T cell exclusion features prior to treatment can predict the response to combinatorial immunotherapy. Jinping Liu: Investigation, Visualization, Writing-Original draft; Lan Cheng: Investigation; Hilana El-Mekkoussi: Investigation; Charles-Antoine Assenmacher: Formal analysis; Michelle Y. Y. Lee: Formal analysis; Danielle R. Jaffe: Investigation; Kaisha Garvin-Darby: Investigation; Ashleigh Morgan: Investigation; Elisabetta Manduchi: Formal Analysis; Jonathan Schug: Formal analysis; Klaus H. Kaestner: Writing – Original Draft, Funding acquisition, Supervision. We thank Drs. S. Shapira (University of Pennsylvania) and K. Wangensteen (Mayo Clinic) for critical reading of the manuscript. We also thank D. Getz, L. Robillard, Drs. R. Dusek and W. Wu (University of Pennsylvania) for the suggestions of administrating Lenvatinib and anti-PD-1 combinatorial immunotherapy. The authors declare that they have no competing interests. This work was supported by the National Institutes of Health grant R01-CA-249929-06A1 to KHK. We acknowledge support from the Molecular Pathology & Imaging Core (MPIC) of the UPenn Center for Molecular Studies in Digestive and Liver Diseases (P30 DK050306), the Comparative Pathology Core, the Wistar Institute Genomics Facility, the Functional Genomics Core of the UPenn Diabetes Research Center (P30 DK019525). Animal studies were approved by the Institutional Animal Care and Use Committee of the University of Pennsylvania under protocol # 805623. Not applicable. Plasmids used for the derivation of the mouse models are available upon request. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.