Single-cell atlas of tumor cell evolution in response to therapy in hepatocellular carcinoma and intrahepatic cholangiocarcinoma Lichun Ma1,#, Limin Wang1,#, Subreen Khatib1, Ching-Wen Chang1, Sophia Heinrich1, Dana Dominguez1, Marshonna Forgues1, Julián Candia1, Maria O. Hernandez2, Michael Kelly2, Yongmei Zhao2, Bao Tran2, Jonathan M. Hernandez3, Jeremy L. Davis3, David E. Kleiner4,5, Bradford J. Wood5,6, Tim F. Greten5,7,*, and Xin Wei Wang1,5,* 1Laboratory of Human Carcinogenesis, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland 20892; 2Frederick National Laboratory for Cancer Research, Leidos Biomedical Research, Inc., Frederick, Maryland 20701; 3Surgical Oncology Program, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland 20892; 4Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland 20892; 5Liver Cancer Program, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland 20892; 6NIH Center for Interventional Oncology, Bethesda, Maryland 20892; 7Thoracic and GI Malignancies Branch, Center for Cancer Research, National Cancer Institute, Bethesda, Maryland 20892 # equal contribution * To whom correspondence should be addressed Tumor evolution is a key feature of tumorigenesis and plays a pivotal role in driving intratumor heterogeneity, treatment failure and patients’ prognosis. In this study, we aimed to construct a single-cell atlas in liver cancer and to determine tumor evolution in response to treatment. Here we performed single-cell transcriptomic analysis of 46 hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA) from 37 patients enrolled for interventional studies at the NIH Clinical Center, with 16 biopsies collected before and after treatment from 7 patients. We surveyed about 57,000 malignant and non-malignant cells and determined tumor cell transcriptomic clusters to track their cellular states by developing a machine learning-based consensus clustering method. We determined tumor cell relationship in each tumor using RNA velocity and reverse graph embedding methods. We studied tumor evolution using longitudinal samples from 4 patients with > 15 malignant cells in each biopsy. We validated our findings in bulk transcriptomic data from 488 patients with HCC and 277 patients with iCCA. We found that malignant cells mainly formed patient-specific clusters, consistent with previous findings, with evidence of both intratumor and intertumor heterogeneity. We observed an increase in tumor cell state heterogeneity based on tumor cell transcriptomic clusters as a surrogate of functional clonality, which was tightly linked to patients’ prognosis. An increased functional clonality was accompanied by a polarized immune cell landscape including an increased pre-exhausted T-cell subtypes. We found that osteopontin expression was tightly associated with tumor cell evolution and microenvironmental reprogramming. We demonstrated experimentally that osteopontin blockage leads to a reduced cellular biodiversity in organotypic culture models of HCC cells. In addition, we developed a user-friendly online interface for single-cell atlas of liver cancer useful for researchers. Our study offers insight into the collective behavior of tumor cell communities in liver cancer as well as potential drivers for tumor evolution in response to therapy.