Spatial multicellular coordination networks are associated with neoadjuvant immunotherapy response in hepatocellular carcinoma

Abstract The response to immune checkpoint blockade in hepatocellular carcinoma (HCC) remains highly variable, and the abundance of immune cells alone has been shown to be insufficient as a predictive biomarker. We analyse spatial transcriptomics data from seven neoadjuvant cabozantinib- and nivolumab-treated HCC patients to test whether the therapeutic outcome is associated with spatial coordination of multicellular programs rather than their presence. Modeling immune, stromal, and malignant transcriptional programs as continuous spatial fields, we introduce a graph-based framework that quantifies their organization using field energy and inter-program gradient coupling. Across tumors, immune exhaustion and cytotoxic programs display a comparable spatial structure irrespective of response, indicating that immune presence is not discriminatory. In contrast, responding tumors exhibit structured cancer-associated fibroblast organization and permissive malignant-stromal interfaces, whereas non-responders show pronounced malignant-CAF segregation and immune exclusion. Integrating three biologically motivated coupling terms, we define a Spatial Immune Coordination Index (SICI) that shows directionally consistent separation of response states across 21 of 27 coefficient parameterisations under sign-flip sensitivity testing, independent of immune abundance. Malignant–CAF coupling is the primary discriminating metric (Cohen’s $$d=2.40$$ , Cliff’s $$\delta =+1.00$$ , $$p_{\textrm{exact}}=0.029$$ , $$n=7$$ ). In an exploratory bulk RNA-seq analysis in TCGA-LIHC ( $$n = 533$$ ), coordination-informed composite scores show statistically significant but modest associations with overall survival. These hypothesis-generating findings suggest that spatial coordination topology is associated with the immunotherapy response in HCC and provide a principled graph-based framework for developing and evaluating future spatial biomarkers.