A Hippo-YAP pathway activity-guided multimodal framework highlights NDRG1-associated tumor microenvironmental features in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) is characterized by marked molecular and histopathological heterogeneity, which limits the performance of conventional clinicopathological risk stratification systems. Hippo-YAP signaling is closely involved in HCC progression and tumor microenvironment regulation. However, how Hippo-YAP pathway activity is reflected in histopathological phenotypes and intercellular communication patterns remains insufficiently understood. We developed a Hippo-YAP pathway activity-guided pathomics framework by integrating H&E whole-slide image features, transcriptomic pathway activity, and survival information from the TCGA-LIHC cohort. Conventional pathomics features and deep learning features were extracted using CellProfiler and ResNet50, respectively. Hippo-YAP pathway activity was quantified by GSVA based on a published 22-gene Hippo-YAP signaling pathway-related gene set. Prognosis-related pathomics features were used to construct an exploratory survival risk model. Single-cell transcriptomic, spatial transcriptomic, immunohistochemical, and CommPath-based cell-cell communication analyses were further performed to characterize the cellular localization, spatial distribution, protein expression, and microenvironmental relevance of NDRG1. A set of pathomics features associated with Hippo-YAP pathway activity and overall survival was identified. A StepCox[forward] + RSF model was selected as an exploratory risk model. However, the model showed substantial performance degradation from the training cohort to the internal validation cohort, with the 3-year AUC decreasing from 0.940 to 0.676 and the C-index decreasing from 0.930 to 0.620. This discrepancy indicates overfitting and limited individual-level discriminative ability. Therefore, the risk score was used only as an internal exploratory stratification variable, and no clinical predictive utility is claimed. Network analysis identified NDRG1 as a candidate molecule linking Hippo-YAP-associated pathomics features with unfavorable prognosis. Multi-level validation independent of risk-score stratification showed that NDRG1 was mainly enriched in malignant tumor cells, relatively concentrated in tumor-core and tumor-edge regions, and showed increased protein expression in HCC tissues. CommPath analysis suggested that NDRG1-positive tumor cells may exhibit enhanced outgoing communication through complement-, metabolism-, extracellular matrix-, and lipid-related signaling programs. Furthermore, NDRG1 protein expression was increased in orthotopic tumor tissues and reduced in lenvatinib-treated tumor tissues. This study presents a Hippo-YAP pathway activity-guided pathomics framework for exploring the relationship among histopathological phenotypes, pathway activity, and tumor microenvironmental communication in HCC. Because of overfitting and weak internal validation performance, the risk model should not be interpreted as a clinically applicable prognostic tool. Instead, the main value of this study lies in identifying NDRG1 as a candidate bridge between Hippo-YAP-associated pathomics features and microenvironmental communication patterns. External validation and functional experiments are required before any clinical or prognostic application can be considered.