Single-cell and spatial architecture of primary liver cancer

Primary liver cancer (PLC) poses a leading threat to human health, and its treatment options are limited. Meanwhile, the investigation of homogeneity and heterogeneity among PLCs remains challenging. Here, using single-cell RNA sequencing, spatial transcriptomic and bulk multi-omics, we elaborated a molecular architecture of 3 PLC types, namely hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC) and combined hepatocellular-cholangiocarcinoma (CHC). Taking a high-resolution perspective, our observations revealed that CHC cells exhibit internally discordant phenotypes, whereas ICC and HCC exhibit distinct tumor-specific features. Specifically, ICC was found to be the primary source of cancer-associated fibroblasts, while HCC exhibited disrupted metabolism and greater individual heterogeneity of T cells. We further revealed a diversity of intermediate-state cells residing in the tumor-peritumor junctional zone, including a congregation of CPE+ intermediate-state endothelial cells (ECs), which harbored the molecular characteristics of tumor-associated ECs and normal ECs. This architecture offers insights into molecular characteristics of PLC microenvironment, and hints that the tumor-peritumor junctional zone could serve as a targeted region for precise therapeutical strategies.


Characteristic
Of 31 TNK sub-clusters, we identified 7 CD4 + T cell sub-clusters, 15 CD8 + T cell subclusters, 2 CD3 + CD4 -CD8 -T cell sub-clusters, 1 mucosal-associated invariant T cell (MAIT) sub-cluster, 3 NK cell sub-clusters and 3 NKT cell sub-clusters (Supplementary Figure 6c), of which cell diversity was consistent with prior study 2 .We found 3 of 7 CD4 + T cell sub-clusters (C5, C10 and C13) and only 1 of 15 CD8 + T subclusters (C16) were tumor-derived clusters.Those tumor-derived T cells showed a higher proliferation ratio and greater patient heterogeneity.We observed that 5 CD8 + and 4 CD4 + T cell sub-clusters were mainly derived from HCC, and 3 CD8 + T cell subclusters from CHC, however, there was no specific ICC-derived sub-clusters (Supplementary Figure 7a).To gain deeper insight into this result, we analyzed the clonal diversity of T cell receptor (TCR) repertoire which was simultaneously sequenced corresponding to our 5' scRNA-seq libraries, hinting the clonotypes of T cells in ICC more diverse than those in HCC (p<0.05) and CHC (p>0.0.5).This might explain why ICC lacks its derived heterogeneous sub-clusters (Supplementary Figure 6d).
We next employed pseudotime trajectory to construct the possible developmental trajectory of CD4 + and CD8 + T cells.Generally, it displayed a process from naïve to effector to exhausted T cells in pseudotime trajectory, in agreement with prior studies 2 (Supplementary Figure 6e, Supplementary Figure 7b).Of note, we did not find subclusters of naïve CD8 + T cells, partly due to our strategy of unsupervised cell collection.However, we identified 7 CD8 + T cell sub-clusters (C2, C17, C18, C23, C32, C36 and C37) in the middle of trajectory, consistent with their intermediate states of functionality (Supplementary Figure 6e).As a representative of them, CD8 + C2 marked with GZMK was the largest intermediate-state sub-cluster (11.35% of all TNK supraclusters, 15663/137949) that was universal in 3 types of PLCs across all patients.It expressed both exhausted markers (PDCD1 and LAG3) and cytotoxic markers (GZMA and GZMK), in line with previous reports on HCC 2 (Supplementary Figure 6c).Clonal analysis, which focused on identical TCRs from the same ancestry but shared by different sub-clusters, revealed that CD8  6f).To further investigate whether intermediate-state functionality of C2 might connect to its spatial distribution pattern, we employed MNN and MIA algorithm, and it suggested that intermediate-state C2 were more likely to be located in J and S zones of P129TP1 and P129TP2 (Supplementary Figure 7c).These results indicate that the pervading intermediate-state cell populations in PLCs might link with their spatial distribution, and play a significant role in cell transition.Further exploration is needed in terms of their molecular mechanism and therapeutic potential.
In terms of B cells (Supplementary Figure 6a.g),22 related sub-clusters were assigned to 9 sub-clusters of memory B cells, featuring in activation of adaptive immune system, and 13 sub-clusters of plasma B cells, featuring in complement activation and protein processing in endoplasmic reticulum, as previously reported 2 .Through clonal analysis with BCR sequences, the majority of memory B cells was found to be monoclonal, while the majority of plasma B cells to be polyclonal.Moreover, abundant identical BCRs between memory and plasma B cells hinted dynamic cell type switching among them (Supplementary Figure 6h.i,Supplementary Figure 7e).Notably, we observed the largest sub-cluster of B cells (C1, 4,450 cells) identified as STAG3 + CD27 -memory B cells, and pseudotime trajectory displayed cells from C1 were located in the middle of trajectory, with memory and plasma B cells at opposite ends (Supplementary Figure 7g).Meanwhile, C1 presented multi-directional flows toward other sub-clusters in RNA velocity (Supplementary Figure 6i).These results indicate that these B cells are at the early stage of cell differentiation.C1 in a monoclonal state displayed higher clonal diversity than plasma B cell and other memory B cell sub-clusters, partly consistent with the recently reported CD45RB + CD27 -early memory B cell population that had intermediate levels of BCRs diversity and mutational burden 4 .We then investigated whether C1 might relate to its spatial distribution.C1 was found mainly derived from the metastatic lymph node (P126N, 98%), and a small proportion of C1 was collected from tumor and peritumor tissues, but rare from peripheral blood.The metascape enrichment analysis found that C1 was related to lymphocyte activation and active metabolism such as peptide chain elongation (Supplementary Figure 7h).These indicate that certain cells of C1 might be affected by local stimuli residing in TME.Together, these results suggest that C1 as a sub-cluster with huge cell populations more closely links to early memory B cells, retaining the potential for cell transition to effector cells and serving as a potential target for treatment.
With respect to myeloid cells, 23 related sub-clusters were assigned into sub-clusters of

Supplementary Figure 1. Unsupervised processing with optimal settings and spatial spot class prediction. a,
Effects of clustering based on different K-means, principle components (PCs) and resolution.Normalized mutual information was employed to assess the clustering robustness with different parameters, which set the baseline as components =50, k-means =30, resolution =0.6.b, Bar plot showing the number of cells from P121-P127.S, 1 month postoperative peripheral blood; O, postoperative (1 month) peripheral blood from P123 was collected twice in 5 day intervals for batch effect detection.c, Heatmap presenting predicted spatial distribution of sub-clusters from the major cell types using MNN and MIA algorithm.The horizontal axis indicates the clusters of ST spots, and the vertical axis indicates the sub-clusters of major cell types.Red color shows higher possibility of cells located in clusters of ST spots.T, tumor zone; P, peritumor zone; J, tumor-peritumor junctional zone; S, stroma zone; F, fatty infiltrated zone; n-F, non-fatty infiltrated zone.d, Bar plot showing the proportion of spot class based on different ST slices (left), spatial zone of P129TP1 (middle) and P129TP2 (right).doublet, two or more cells in one spot; singlet, one cell in one spot; uncertain, uncertain in one spot.e, Bar plot showing the spot number of doublet with different cell types.