Spatial transcriptomics analysis of neoadjuvant cabozantinib and nivolumab in advanced hepatocellular carcinoma identifies independent mechanisms of resistance and recurrence

Abstract

Background: Novel immunotherapy combination therapies have improved outcomes for patients with hepatocellular carcinoma (HCC), but responses are limited to a subset of patients. Little is known about the inter- and intra-tumor heterogeneity in cellular signaling networks within the HCC tumor microenvironment (TME) that underlie responses to modern systemic therapy.

Methods: We applied spatial transcriptomics (ST) profiling to characterize the tumor microenvironment in HCC resection specimens from a prospective clinical trial of neoadjuvant cabozantinib, a multi-tyrosine kinase inhibitor that primarily blocks VEGF, and nivolumab, a PD-1 inhibitor in which 5 out of 15 patients were found to have a pathologic response at the time of resection.

Results: ST profiling demonstrated that the TME of responding tumors was enriched for immune cells and cancer-associated fibroblasts (CAF) with pro-inflammatory signaling relative to the non-responders. The enriched cancer-immune interactions in responding tumors are characterized by activation of the PAX5 module, a known regulator of B cell maturation, which colocalized with spots with increased B cell marker expression suggesting strong activity of these cells. HCC-CAF interactions were also enriched in the responding tumors and were associated with extracellular matrix (ECM) remodeling as there was high activation of FOS and JUN in CAFs adjacent to the tumor. The ECM remodeling is consistent with proliferative fibrosis in association with immune-mediated tumor regression. Among the patients with major pathologic responses, a single patient experienced early HCC recurrence. ST analysis of this clinical outlier demonstrated marked tumor heterogeneity, with a distinctive immune-poor tumor region that resembles the non-responding TME across patients and was characterized by HCC-CAF interactions and expression of cancer stem cell markers, potentially mediating early tumor immune escape and recurrence in this patient.

Conclusions: These data show that responses to modern systemic therapy in HCC are associated with distinctive molecular and cellular landscapes and provide new targets to enhance and prolong responses to systemic therapy in HCC.

Trial registration: ClinicalTrials.gov NCT03299946.

Keywords: Hepatocellular carcinoma; Immunotherapy; Neoadjuvant therapy; Spatial transcriptomics; Therapeutic resistance; Tumor recurrence.

Fig. 1

Spatial transcriptomics analysis of HCC samples treated with cabozantinib and nivolumab. A Experimental workflow. B Hematoxylin and eosin (H&E)-stained images of the samples profiled and the spatial clusters for responders. C H&E and spatial clusters for non-responders. D Tumor-, immune-, and cancer-associated fibroblast (CAF) composition in each responder sample as determined by spatial transcriptomics. E Tumor-, immune-, and CAF proportions in non-responders samples. To quantify the cell proportions of tumor, immune, and CAF in each patient, the multiple clusters that were classified as the same type were merged into one major category

Fig. 2

Differential expression analysis of A tumor clusters (subset of spots classified as a tumor in the clustering) from responders versus B tumor clusters from non-responders across all patients. C Volcano plot of the differential expression analysis showing the most differentially expressed genes in responders (red dots) and non-responders (blue dots), showing the upregulation of immune genes in responders relative to the upregulation of hepatocellular markers among non-responders. D Pathway enrichment analysis between responders (red dots) and non-responders (blue dots) reveals activation of immune-related pathways in responders’ tumors, while non-responders' tumors have activation of proliferation and metabolic pathways

Fig. 3

Intercellular interaction analysis. A HCC-immune interaction analysis identified the activation of the PAX5 network. B, C HCC-CAF interaction analysis pointed to the activation of the FOS and JUN networks. D PAX5 is a transcription factor that regulates B cell activity, and the PAX5 network identified co-localizes with the distribution of B cells as determined by the spatial distribution of B cell markers. E FOS and JUN are transcription factors that can regulate genes involved in extracellular matrix remodeling, and the networks regulated by these genes colocalize with CAF marker genes

Fig. 4

Intra-sample heterogeneity analysis with spatial transcriptomics. A Responder sample with remarkable heterogeneity with two tumor regions, one immune-rich (cyan) and one immune-poor (dark blue). B Each of these regions presents distinct transcriptional profiles as determined by clustering analysis. C–E The immune-rich tumor region highly expresses immune-related pathways that were initially observed to be enriched across responders’ samples. F–H Proliferation and metabolic pathways, which are enriched across non-responders’ tumors, are expressed in high levels at the immune-poor region. I-J The immune-rich tumor region expresses high levels of the antigen processing and presentation machinery genes, which is associated with the more efficient attraction of immune cells

Fig. 5

Intercellular interaction analysis in the context of intra-sample heterogeneity. A Intercellular interaction analysis in the immune-rich tumor region (cancer-immune) reveals activation of PAX5. B, C FOS and JUN are the active networks from the interaction analysis at the immune-poor tumor region (HCC-CAF). D PAX5 activation co-localizes with the expression of B cell markers concentrated at the borders of the immune-rich tumor region. E FOS and JUN networks are co-expressed with CAF markers adjacent to the immune-poor tumor region

Fig. 6

Cancer stem cell detection in hepatocellular carcinoma heterogeneous sample. A Cancer stem cell markers are highly expressed at the immune-poor tumor region of the responder sample with intrasample heterogeneity as noted by the spatial expression and the levels of expression are significantly different between the distinct tumor regions. B Heatmap with the expression of the CSC genes in the TCGA HCC samples. C Correlation heatmap with the Spearman correlation coefficient between expression of CSC markers and T cell proportions (CIBERSORT) in HCC samples from TCGA showing a negative correlation between T CD8 cell proportions and high expression of some CSC genes