Genomic and the tumor microenvironment heterogeneity in multifocal hepatocellular carcinoma

Abstract

Background and Aims:

Ambiguous understanding of tumors and tumor microenvironments (TMEs) hinders accurate diagnosis and available treatment for multifocal hepatocellular carcinoma (HCC) covering intrahepatic metastasis (IM) and multicentric occurrence (MO). Here, we characterized the diverse TMEs of IM and MO identified by whole-exome sequencing at single-cell resolution.

Approach and Results:

We performed parallel whole-exome sequencing and scRNA-seq on 23 samples from 7 patients to profile their TMEs when major results were validated by immunohistochemistry in the additional cohort. Integrative analysis of whole-exome sequencing and single-cell RNA sequencing found that malignant cells in IM showed higher intratumor heterogeneity, stemness, and more activated metabolism than those in MO. Tumors from IM shared similar TMEs while distinct TMEs were noticed in those from MO. Furthermore, CD20+ B cells, plasma cells, and conventional type II dendritic cells (cDC2s) were decreased in IM relative to MO while T cells in IM exhibited a more terminally exhausted capacity with a higher proportion of proliferative/exhausted T cells than that in MO. Both CD20 and CD1C correlated with better prognosis in multifocal HCC. Additionally, MMP9+ tumor-associated macrophages were enriched across IM and MO, which formed cellular niches with regulatory T cells and proliferative/exhausted T cells.

Conclusions:

Our findings deeply decipher the heterogeneous TMEs between IM and MO, which provide a comprehensive landscape of multifocal HCC.

INTRODUCTION

HCC is the fourth most lethal cancer worldwide, of which >50% are initially diagnosed as multifocal HCC with effective limited therapy.¹ When surgical resection is potentially curative for early HCC and improvements in systemic and immune therapies are available, multifocal tumors are still stubborn and cause poor outcomes with high recurrence rates.²

The multifocal lesions of HCC can be simultaneously or separately attributed to either intrahepatic metastasis (IM) developing from the primary tumor or multicentric occurrence (MO), always sharing different clonal origins.³ MO can be detected in early stages while metastasis has usually occurred before diagnosis for IM.⁴ Overwhelming evidence has revealed tumor heterogeneity across multiple lesions of multifocal HCC recently. IM and MO could carry distinct clonal architecture, genetic susceptibility, and tumor evolutionary trajectory.^5,6 DNA damage repair alterations could be involved in the tumor evolution of multifocal HCC, resulting in higher tumor mutational burden and tumor neoantigen burden.¹ The tumor microenvironments (TMEs) are complex milieu where tumor cells, stromal, and immune cells interact intimately, manipulating tumor growth and its response to immunity. Small tumors harbored higher immune cell infiltration than large tumors in patients with multifocal HCC.⁷ Besides, T-cell and M2 macrophage were less and more infiltrated in IM than MO, respectively, while inhibitory immune checkpoints were higher expressed in MO than IM.⁵ This evidence above generally investigated the tumor heterogeneity and TMEs in multifocal HCC and paved the way for deciphering discrepancy between IM and MO. However, accurate interactions and effective features among the TMEs between IM and MO remain intangible. Emerging technologies including single-cell RNA-sequencing (scRNA-seq) have facilitated better understanding of the heterogeneity of diverse cell states in solid tumors,^8,9 providing a silver lining for probing personalized therapies for multifocal HCC.

In the current study, we performed whole-exome sequencing (WES) and scRNA-seq on 23 samples from 7 patients with multifocal HCC, aiming to unveil the TMEs across IM and MO clearly. WES supported to distinguish IM from MO while scRNA-seq illustrated the landscape of TMEs in IM and MO. Then, cellular fractions and transcriptional functions between IM and MO were compared. We found that T cells related to proliferation and exhaustion were abundant in IM, where B cells, plasma cells, and conventional type II DCs (cDC2s) were absent relatively when compared with MO, as validated in the additional cohort. Immunohistochemistry (IHC) staining based on the additional cohort confirmed that both CD20 and CD1C correlated with better survival in multifocal HCC. MMP9+ tumor-associated macrophages (TAMs) were enriched in IM/MO and interacted with T cells above closely. Intratumoral cellular niches, including proliferative and exhausted T cells, MMP9+ TAMs, and regulatory T clusters (Tregs), were also identified by multicolor immunofluorescence. This evidence recognized different TMEs and defined a detailed regulatory network across IM and MO, enabling the exploration of effective therapies for patients with IM or MO.

METHODS

Human sample collection

In the initial cohort, 7 patients diagnosed with multifocal HCC were enrolled between March 2022 and March 2023 at the Qilu Hospital (Jinan, China) and the Shandong Provincial Hospital (Jinan, China), and a total of 23 samples from their tumors and normal tissues adjacent to the tumor (NAT) were collected for scRNA-seq, whole-exome sequencing and/or IHC/multicolor immunofluorescence assay. In the validation cohort, 10 multifocal patients with HCC with tumors and NAT available were enrolled from the Shandong Provincial Hospital (Jinan, China), and a total of 30 samples were prepared for WES and IHC. Besides, 100 samples from 47 patients with multifocal HCC were available for CD20 staining, and 72 samples from 35 patients with multifocal HCC were available for CD1C staining. All treatment-naive patients underwent curative resection. Samples were reviewed by 2 experienced pathologists. Diagnosis for each sample and clinical information for each patient were stored in Supplemental Table S1, http://links.lww.com/HEP/J661. The study adhered to the principles of the Declaration of Helsinki and was approved by the ethics committee of Shandong University (ECSBMSSDU2023-2-97) and performed with written informed consent from all patients.

Statistics

Statistical analyses and presentations were performed under R (v.4.2.0) and Python (v.3.9.0). All statistical tests used are described in the Figure 2, Figure 5 and Figure 6 legends.

Data availability

The raw WES and scRNA-seq data supporting the findings of this study have been deposited at the Genome Sequence Archive at the National Genomics Data Center (Beijing, China) under accession code (GSA-Human: HRA006293) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human.

Code availability

No new algorithms were developed for this study. Codes mainly used in this study are available at GitHub (https://github.com/YonghengYang/mHCC).

A full description of other methods can be found in the Supplemental Materials and Methods, http://links.lww.com/HEP/J661

RESULTS

Evolutionary trajectory and genomic heterogeneity of IM and MO

To illuminate the heterogeneity of malignant cells and the TMEs across IM and MO, 16 tumor samples and 7 normal tissues adjacent to the tumor (NAT) from 7 patients with multifocal HCC were prepared for scRNA-seq and WES (Figure 1A, supplemental table 1, http://links.lww.com/HEP/J661). Clinically, each patient bore 2 or 3 separate tumors. Based on Couinaud surgical anatomy, liver can be divided into 8 segments across caudate, left, quadrate, and right lobes.¹⁰ Most of the tumors in the initial cohort were located at right lobes ranging from segment V to VIII (Figure 1B and Supplemental Figure S1A, http://links.lww.com/HEP/J661). Histologic differences were noticed across different regions of each patient (Figure 1C). Of note, one sample (segment V, T2 from P5) was diagnosed as a nodular lesion in liver cirrhosis with steatosis and thus included in the NAT for the subsequent analysis as no significant mutation was detected (Supplemental Figure S1A, http://links.lww.com/HEP/J661).

FIGURE 1.

Genomic landscape of 7 patients with multifocal HCC. (A) A brief scheme describing tissue collection, preparing for scRNA-seq and WES, and data analysis. (B) Locations of specimens based on the Couinaud surgical anatomy with the percentages of the shared and private somatic mutations across diverse tumor regions from each patient with multifocal HCC and phylogenetic trees labeled (P1, P2, P3, P4, P6, P7). (C) Representative CT images and histopathology of tumors from IM and MO (P3 and P4). (D) Oncoprint of mutated genes detected significantly in tumor regions from 7 multifocal patients with HCC. (E) Representative heatmaps portraying CNV gain and loss in IM and MO patients (P1 and P2). Abbreviations: CNV, copy number variation; IM, intrahepatic metastasis; MO, multicentric occurrence; WES, whole-exome sequencing.

WES analysis showed that most of the tumors harbored classical TP53 and NBPF10 mutations (62.5% and 37.5%, respectively) (Figure 1D). Here, trunk events were described as mutation rates shared by multiple tumors in each patient, whereas private mutations would not. Diverse extents of private mutations were noticed in each tumor (Figure 1B and Supplemental Figure S1B, http://links.lww.com/HEP/J661). Moreover, patient-wise phylogenetic tree was constructed based on the somatic mutations to define IM or MO and evaluate the clonal evolution and genomic heterogeneity of multifocal HCC.⁷ Trunk events were nearly absent in P2, P4, P6, and P7, implying that those tumors in each patient were independent thus meaning MO while T1/T2 from P1 and T1/T2 from P3 shared proportional mutations in each patient respectively, suggesting IM (Figure 1B and Supplemental Figure 1A, http://links.lww.com/HEP/J661). As expected, copy number variations (CNVs) analysis revealed the extensive heterogeneity of tumors from IM and MO with diverse extents of gain or loss segments noticed. In addition, profiles of CNVs across tumors in IM were more similar than those in MO (Figure 1E and Supplemental Figure S2, http://links.lww.com/HEP/J661). Here, WES recognized different origins and revealed diverse evolutionary trajectories of IM and MO.

A glance at the TMEs between IM and MO at single-cell resolution

Overall, 195,425 high-quality single cells were grouped into 9 cell lineages annotated by canonical marker genes, including natural killer and T cells, epithelial cells, myeloid cells, endothelial cells, B cells, fibroblast cells, proliferative T cells, plasma cells and mast cells (Figure 2A, B and Supplemental Figure S3A, http://links.lww.com/HEP/J661). First, the fractions of different cell types were calculated to characterize the TMEs discrepancy between IM and MO, tumors (IM/MO) and NAT, respectively (Figure 2C, D and Supplemental Figure S3B, C, http://links.lww.com/HEP/J661). It showed that the frequencies of myeloid cells, fibroblasts, and proliferative T cells were increased in IM/MO when compared with NAT (fold change (fc) = 2.17, p-value = 0.0649 for myeloid cells; fc = 3.88, p-value = 0.0414 for fibroblasts and fc = 1.76, p-value = 0.0167 for proliferative T cells) (Figure 2D). Subsequently, abundances of these cell types across IM, MO and NAT were further compared.

FIGURE 2.

Single-cell transcriptome profiling of multifocal HCC. (A) Uniform manifold approximation and projection (UMAP) plot of 9 main cell clusters from 23 samples of 7 patients with multifocal HCC. (B) Violin plots showing the expression levels of canonical marker genes of major cell types. (C) Stacked bar plots showing the percentages of main cell clusters in each sample. (D) Quantification of the proportions of myeloid cells, fibroblasts, and proliferative T cells between tumors (IM/MO) (n=15) and NAT (n=8). Two-sided Wilcoxon test. (E) Quantification of the proportions of major cell types across IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (F) Representative images of immunohistochemistry staining of CD3, MKI67, and CD20 across IM, MO, and NAT, respectively. Random 10 regions from each slide were examined for positive comparison. Two-sided Wilcoxon test. (G) Representative images of immunohistochemistry staining of CD20 across IM and MO patients in the additional cohort (n=10). (H) Quantification of CD20+ cell fractions between IM and MO across 20 samples from 10 patients in the additional cohort with representative images of immunohistochemistry staining listed (×40). (I) Kaplan-Meier estimation of overall survival time in patients with multifocal HCC by IM (n=5) and MO (n=12). Log-rank test. (J) Kaplan-Meier plot showing better clinical outcomes in multifocal HCC with the higher expression of CD20 (n=47). Abbreviations: IM, intrahepatic metastasis; MO, multicentric occurrence; NAT, normal tissues adjacent to the tumor; NK, natural killer.

Remarkably, the fractions of CD20+ B cells and plasma cells were higher in MO than IM (fc = 16.67, p-value = 0.022 for B cells; fc = 13.00, p-value = 0.012 for plasma cells, respectively) (Figure 2E). In line with their importance in response to immune checkpoint blockade (ICB) treatment,¹¹ these data suggested that CD20+ B cells and plasma cells were lacking in tumor metastasis. Furthermore, myeloid cells were also increased in IM compared with MO and NAT (IM vs. MO: fc = 2.58, p-value = 0.01; IM vs. NAT: fc = 4.25, p-value = 0.008), indicating reprogramming of myeloid cells during tumor metastasis (Figure 2E). Considering that a similar tendency of raising across NAT, MO, and IM was observed in proliferative T cells (MO vs. NAT: fc = 1.77, p-value = 0.075; IM vs. MO: fc = 1.53, p-value = 0.472) (Figure 2E), proliferative T cells were more likely to represent a T-cell cluster related to energy or tolerance in IM and MO as those T cells in IM/MO showed higher anergy or exhaustion signature scores, but lower TCR signaling scores than those in NAT¹² (Supplemental Figure 3D, http://links.lww.com/HEP/J661). IHC staining validated the opposite abundance of proliferative T cells (CD3+, MKI67+) and B cells (CD20+) between MO and IM (Figure 2F). To further validate this, IHC staining of CD20 and WES was performed in 20 samples from 10 patients from the additional cohort (Supplemental Figure 4A–E, http://links.lww.com/HEP/J661). IM does harbor lower intensity of CD20+ B cells than that in MO (Figure 2G, H and Supplemental Figure S5A, http://links.lww.com/HEP/J661). IM was also associated with inferior outcomes (Figure 2I). It also showed that the higher expression of CD20 was associated with a survival advantage in the multifocal HCC cohort (n=47) (Figure 2J and Supplemental Figure 5B, http://links.lww.com/HEP/J661). Besides, the abundance of natural killer cells and T cells was increased gradually across IM, MO, and NAT (MO vs. IM: fc = 1.24, p-value = 0.104; NAT vs. MO: fc = 1.13, p-value = 0.31), indicating natural killer cells and T cells descended during tumorigenesis and tumor metastasis. Similarly for endothelial cell, it also augmented across IM, MO and NAT (MO vs. IM: fc = 1.26, p-value = 0.647; NAT vs. MO: fc = 2.12, p-value = 0.836), consistent with the previous report that liver sinusoidal endothelial cells reduced during HCC development.¹³ This evidence revealed that IM and MO embraced diverse TMEs as certain cell types were distributed differently across IM, MO, and NAT.

Regarding epithelial cells, most of them showing either high hepatic or biliary scores were identified as malignant cells (Figure 3A), as they formed sample-specific clusters regardless of patients which indicated biological differences across tumor samples (Figure 3B).¹⁴ Only samples with considerable malignant cells detected were used for the subsequent analysis.¹⁵ The inferred CNVs were applied to differentiate malignant cells and non-malignant cells and disclosed patterns that CNVs were diverse across samples from each MO patient but more similar in those from each IM patient, consistent with results from WES (Figure 3C and Supplemental Figure S6, http://links.lww.com/HEP/J661). Most patient-wise clusters expressed GPC3, a highly tumor-specific antigen (Figure 3D).¹⁶ T1 and T2 from P3 were negative for GPC3, but expressed CD24, a marker on the surface of cancer stem cells (CSCs) associated with “don’t eat me” signals, showing a higher malignant cell stemness in P3 grouped into IM.¹⁷ Notably, T3 from P2 was negative for both GPC3 and CD24 but positive for Vimentin corresponding to that it was diagnosed as sarcomatoid HCC (Figure 3D). Indeed, intratumor heterogeneity (ITH) is associated with cancer cell evolution and patient outcome.¹⁸ Thus, ITH_GEX depicting gene expression profiles was calculated to determine the ITH between IM and MO.¹⁹ Malignant cells in IM seemed to embrace higher ITH_GEX scores than those in MO (median: 0.200 vs. 0.158), suggesting that a thornier and more invasive phenotype in IM (Figure 3E). Furthermore, gene set variation analysis showed that malignant cells from IM and MO shared different signatures. Some pathways related to antitumor response, including interferon-γ response, apoptosis, and inflammatory response, were loss in IM but activated in MO, while glycolysis and fatty acid metabolism both of which were indispensable for tumor metastasis were highly enhanced in IM, not MO (Figure 3F).²⁰ Besides, malignant cells showed higher stem scores in IM than MO (Figure 3G). In total, those evidences demonstrated that malignant cells in IM harbored high ITH, partial stemness, and activated metabolism and thus were more intricate and trickier than those in MO.

FIGURE 3.

Landscape of malignant cells from multifocal HCC. (A) Boxplots showing hepatic scores and biliary epithelial scores between epithelial cell cluster and nonepithelial cell clusters. Two-sided t test. (B) tSNE plot of epithelial cells from IM, MO, and NAT colored by patients and sources. (C) Landscape of the inferred single-cell copy number variation profiles in epithelial cells classified by samples. (D) Violin plot showing the expression levels of GPC3, CD24, and Vim among samples from each patient (P1, P2, P3, and P4). (E) ITHGEX scores of epithelial cells between IM and MO. (F) Different pathway activities were estimated by gene set variation among samples from each patient with genes of interest labeled. (G) Violin plot showing stem scores calculated by EPCAM, KRT19, PROM1, ALDH1A1, CD24, ANPEP, CD44, ICAM1, CD47, and SOX9 between IM and MO. Abbreviations: IM, intrahepatic metastasis; MO, multicentric occurrence; NAT, normal tissues adjacent to the tumor; TME, tumor microenvironment; tSNE, T-distributed Stochastic Neighbor Embedding.

Programs related to exhaustion and proliferation in CD8 T cells changed across IM, MO, and NAT

Generally, 110,218 high-quality T cells were identified into 26 clusters covering five types, including CD4+ T cells, CD8+ T cells, natural killer (NK) cells, mucosal-associated invariant T cells (MAIT), and proliferative T cells (Figure 4A and Supplemental Figure S7, http://links.lww.com/HEP/J661). For the whole T cells, pathways linked to cell proliferation, including cell cycle and DNA replication were upregulated in IM and MO compared with NAT (Figure 4B, left), while programs, including HIF-1 signaling pathway and NF−kappa B signaling pathway both of which related to CD8+ T-cell exhaustion, were enhanced in T cells from IM relative to MO (Figure 4B, right).^12,21

FIGURE 4.

NK/T-cell profile in multifocal HCC at the single-cell resolution. (A) UMAP plots of T-cell clusters across IM, MO, and NAT colored by main cell types and fine cell types. (B) KEGG pathways enriched in T cells whose functions were compared by IM/MO and NAT, IM, and MO. Hypergeometric test with p-values adjusted by BH. (C) Heatmap of normalized expression of canonical NK/T-cell marker genes among clusters. (D) Heatmap illustrating expression of gene signatures across CD4+ T, CD8+ T, and NK clusters from IM, MO, and NAT. Abbreviations: IM, intrahepatic metastasis; MO, multicentric occurrence; NAT, normal tissues adjacent to the tumor; NK, natural killer; TCR, T-cell receptor.

The major compartments in tumor-infiltrating lymphocytes were 7 CD8+ exhausted T (Tex) cell populations including CD8T_C1_TNFRSF9, CD8T_C3_GZMK, CD8T_C6_GZMK, CD8_C7_CD38, CD8T_C9_CXCL13, ProT_C1_MKI67, and ProT_C2_MKI67, all of which were enriched in the CD8 exhaustion signatures and highly expressed markers related to exhaustion (eg, TOX, TIGIT, PDCD1, HAVCR2) but differed in gene expression, suggesting diverse transcriptional states of those Tex cells (Figure 4C, D). Tissue-resident memory (Trm) phenotypes are described as expressing significantly higher levels of Trm markers (eg, ENTPD1, ITGAE) and being enriched with signatures related to residing in tissues.²² Interestingly, they all shared an adhesion signature and were enriched in Trm features (Figure 4D). Further analysis found that both ProT_C1_MKI67 and ProT_C2_MKI67 showed a proliferative Trm gene signature as revealed by gene set enrichment analysis (GSEA) (Figure 5A, B and Supplemental Figure 8A, http://links.lww.com/HEP/J661).

FIGURE 5.

Functions of NK cells and T cells across IM and MO. (A) GSEA enrichment for Trm-mitotic gene set in Pro_C1_MKI67. (B) GSEA enrichment for Trm-mitotic gene set in Pro_C2_MKI67. NES, normalize enrichment score. (C) Quantification of the proportions of Pro_C1_MKI67 and Pro_C2_MKI67 between IM/MO and NAT, among IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (D) RNA velocity of exhausted CD8+ T cells with colored by cell clusters. (E) UMAP of T cells colored by terminally exhausted score (Tex-term, up) and precursor exhausted cell (Tpex, down). (F) Violin plot showing Tex-term score and Tpex score among all T-cell clusters. (G) Violin plot showing Tex-term score and Tpex score in T cells across IM, MO, and NAT. (H) Quantification of the proportions of CD8_C8_GZMA across IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (I) Quantification of the proportions of CD8_C5_ANXA1 across IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (J) the ORs of each T cluster occurring in each sample from the same patient (P3 and P6). Abbreviations: IM, intrahepatic metastasis; MO, multicentric occurrence; NAT, normal tissues adjacent to the tumor; NES, normalize enrichment score.

Notably, it is ProT_C1_MKI67 and ProT_C2_MKI67 that changed significantly across IM, MO, and NAT, while other Tex clusters remained subtly changed (Figure 5C and Supplemental Figure 8B, http://links.lww.com/HEP/J661). In addition, those 2 proliferative clusters above seemed to be more abundant in IM when compared with MO and NAT (fc = 2.33 and 1.84, p-value = 0.226 and 0.046; fc = 1.73 and 3.25, p-value = 0.133 and 0.02, respectively) (Figure 5C). RNA velocity revealed that ProT_C2_MKI67 bridged CD8_C1_TNFRSF9 and ProT_C1_MKI67, ProT_C1_MKI67 was located at the end of arrows where flows toward, suggesting those cells were terminally differentiated (Figure 5D). Moreover, gene expression scores characterizing precursor exhausted cells (Tpex) and terminally exhausted cells (Tex-term) were computed to evaluate T cells in IM and MO. The higher expression of two scores in certain cell clusters was accordant with unsupervised annotations for Tex cell populations (Figure 5E). Notably, the highest expression of Tpex scores was observed in CD4T_C6_CXCL13 while several proliferative cell clusters: ProT_C1_MKI67, ProT_C2_MKI67, and ProT_C3_MKI67, expressed Tex-term scores extremely (Figure 5F). These data support that ProT_C1_MKI67 and ProT_C2_MKI67 correspond to Tex cell expansion and differentiation. Most importantly, Tpex scores were higher in T cells derived from MO when compared with those in IM or NAT. Contrarily, T cells in IM harbored significantly higher Tex-term scores than MO or NAT (Figure 5G). That T cells were more terminally exhausted in IM than MO implied poorer effector function of T cells and more adverse immune suppressive circumstances during tumor metastasis.

CD8T_C2_GNLY, whose fractions were alike in IM, MO, and NAT, was characterized by expressing effector molecules (eg, FGFBP2) and cytotoxicity genes (eg, GNLY) alongside T-cell receptor (TCR) signaling enhanced (Figure 4D and Supplemental Figure S9A, http://links.lww.com/HEP/J661). Contrastively, CD8_C8_GZMA showed a moderate cytotoxicity signature and expressed several naive genes (eg, TCF7, LEF1), with TCR signaling absent (Figure 4D). Its proportions were higher in IM than MO or NAT (Figure 5H and Supplemental Figure 9A, http://links.lww.com/HEP/J661). Further check for its markers showed that it expressed genes like THEMIS and GIMAP4, the former could dampen the effector function of CD8+ T cells in acute viral infection, and the latter was reported to be critical in the regulation of T-cell apoptosis^23,24 (Supplemental Figure S7, http://links.lww.com/HEP/J661). CD8T_C5_ANXA1 was enriched in an effector memory signature. Its fractions increased in NAT relative to IM and MO, indicating effector memory function was dampened during tumorigenesis (Figure 5I and Supplemental Figure 9A, http://links.lww.com/HEP/J661).

Regarding CD4+ T cells, 4 major clusters were identified including 1 naive CD4+ T-cell cluster (CD4T_C5_CCR7), 3 Tregs (CD4_C3_TNFRSF18, CD4T_C2_IKZF2, and ProT_C3_MKI67), 1 follicular helper T (CD4T_C6_CXCL13), 3 memory T clusters (CD4T_C4_AREG, CD4T_C1_ANXA1, CD4T_C7_CCR6). As expected, a significant increase of fractions related to those Tregs clusters in IM or MO was observed while CD4T_C2_IKZF2 and CD4_C3_TNFRSF18 seemed to be higher in MO relative to IM (Supplemental Figure S9B, http://links.lww.com/HEP/J661).

Several unconventional T-cell subsets were also captured. CD8_C4_SLC4A10 mainly covered MAIT cells whose frequencies were lower in IM/MO than NAT (Supplemental Figure 9C, http://links.lww.com/HEP/J661), but indistinctive between IM and MO. NK_C1_FCER1G was FCER1G+, together with XCL1/2, CXCR6, and ITGAE expressed. As previously reported, FCER1G can sign innate-like T cells with high cytotoxic potential.²⁵ Both NK_C2_KLRC2 and NK_C4_KLRC2 showed high expression of activation markers, while TCR signaling signature and cytotoxicity signature were higher in NK_C2_KLRC2 and NK_C4_KLRC2, respectively. Those 3 NK clusters were reduced in IM or MO compared with NAT, but only NK_C1_FCER1G was increased in IM than MO (Supplemental Figure S9, http://links.lww.com/HEP/J661). NK_C3_GNLY cells were FGFBP2+, with high expression of cytotoxicity genes and IFN response markers, similar to CD8T_C2_GNLY. Regarding NK_C5_PPARG, it expressed a high level of TCF7, but lacked the expression of granzymes, indicating a naive-like phenotype. Pro_C4_MKI67 was described as a proliferative NK cell cluster. Interestingly, it was higher in IM than MO (Supplemental Figure S9D, http://links.lww.com/HEP/J661). Notably, T_stress can be distinguished from other clusters by heat shock gene expression (eg, HSPA1A, HSPA1B) (Supplemental Figure 9E, http://links.lww.com/HEP/J661). Those T cells with stress response state were highlighted and detectable in the TMEs across different cancer types, thus also linked to ICB treatment.¹⁵ The fractions of T_stress across IM, MO, and NAT did not meet any significant differences (Supplemental Figure S9E, http://links.lww.com/HEP/J661).

Above all, patient-wise odds ratios (ORs) were applied to evaluate the heterogeneity of T cells across different samples from the same patient. It showed that IM samples shared a similar distribution of each T-cell cluster, which was distinct from that in NAT, while those MO samples displayed a higher interregional heterogeneity of T cells with diverse distributions for each cluster (Figure 5J and Supplemental Figure S10, http://links.lww.com/HEP/J661). It implied that tumors from IM could share alike TMEs related to T cells.

MMP9+ macrophages were abundant in MO while cDC2 was absent in IM

In total, 28,532 myeloid cells were divided into 4 clusters covering dendritic cells (DCs), monocytes, macrophages, and neutrophils (Figure 6A, B and Supplemental Figure S11, http://links.lww.com/HEP/J661). Their fractions in IM, MO, and NAT were compared initially. When compared with NAT, the fractions of DCs decreased in IM and MO. Moreover, DCs were higher in MO relative to IM (fc = 1.36, p-value = 0.018) (Figure 6C). An ascending tendency for the fractions of monocytes through IM, MO, and NAT was also noticed (MO vs. IM: fc = 1.36, p-value = 0.177; NAT vs. MO: fc = 1.22, p-value = 0.351, respectively) (Figure 6C). To get an accurate probe into the myeloid components across IM, MO, and NAT, those cells also were grouped into fine clusters and annotated with marker genes (Figure 6D).

FIGURE 6.

Myeloid cell heterogeneity in IM and MO. (A) UMAP plots of myeloid cell clusters across IM, MO, and NAT colored by main cell types. (B) Dot plot showing the average expression levels of canonical marker genes of macrophage, DC, monocyte, and neutrophil. (C) Quantification of the proportions of macrophage, DC, monocyte, and neutrophil between IM/MO and NAT, among IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (D) UMAP plots of myeloid cell clusters across IM, MO, and NAT colored by fine cell types. (E) Dot plot showing the average expression levels of canonical marker genes related to cDC1s, cDC2, activation, and inhibitory function. (F) Quantification of the proportions of DC_C5_CLEC10A and DC_C17_CCL19 across IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (G) Representative images and quantification of immunohistochemistry staining of CD1C across IM and MO patients in the additional cohort (n=10). (H) Kaplan-Meier plot showing better clinical outcome in multifocal HCC with the higher expression of CD1C (n=35). (I) Dot plot showing the average expression levels of marker genes for macrophages from IM, MO, and NAT. (J) KEGG pathways enriched in MMP9+ macrophages. Hypergeometric test with p-values adjusted by BH. (K) RNA velocity of myeloid cells with colored by cell clusters. (L) Quantification of the proportions of Mac_C13_SPP1 and Mac_C1_MMP9 across IM, MO, and NAT with the median and IQR labeled. Two-sided Wilcoxon test with p-values adjusted by BH. (M) Scatter plot showing the correlation between Mac_C1_MMP9 and MAIT. Spearman rank correlation coefficient. (N) the ORs of each myeloid cluster occurring in each sample from the same patient (P3 and P6). Abbreviations: cDC2, conventional type II DC; IM, intrahepatic metastasis; MO, multicentric occurrence; NAT, normal tissues adjacent to the tumor.

The DCs were mainly divided into conventional type I DCs (cDC1s, DC_C12_IDO1 and DC_C19_CLEC9A), conventional type II DCs (cDC2s, DC_C5_CLEC10A), LAMP3+ DCs (DC_C17_CCL19) and CD274+ DCs (DC_C18_CXCL11) (Figure 6E). cDC1s and cDC2s were reported to trigger CD8+ T cells and CD4+ T cells, respectively.²⁶ LAMP3+ DCs were described as mature DCs enriched in immunoregulatory molecules.²⁷ CD274+ DCs were marked by inhibitory molecules including CD274 and PDCD1LG2 (Supplemental Figure S11, http://links.lww.com/HEP/J661). Remarkably, it is cDC2s (DC_C5_CLEC10A) that boosted significantly across IM, MO, and NAT with a gradient (MO vs. IM: fc = 1.24, p-value = 0.058; NAT vs. MO: fc = 1.93, p-value = 0.063, respectively) (Figure 6F and Supplemental Figure S12A, http://links.lww.com/HEP/J661). IHC staining of CD1C showed that IM patients harbored lower intensity of CD1C+ cDC2s than MO patients from the additional cohort (Figure 6G). It also confirmed that the higher expression of CD1C correlated with better survival in multifocal HCC cohort (n=35) (Figure 6H). The fractions of cDC1s (DC_C12_IDO1) were improved in NAT relative to IM or MO with no discrepancy observed between IM and MO (Supplemental Figure S12A, http://links.lww.com/HEP/J661). LAMP3+ DCs (DC_C17_CCL19) were increased across IM, MO and NAT (Figure 6F). Those evidence suggested that DCs were suppressed during tumorigenesis and tumor metastasis. Furthermore, the fractions of cDC2 and CD4T_C5_CCR7 were positively correlated, in line with its function on activating CD4+ T cells (Supplemental Figure S12B, http://links.lww.com/HEP/J661). Significantly negative correlation between cDC2 and Tregs/Tex clusters including CD4T_C2_IKZF2, CD4_C3_TNFRSF18, Pro_C1_MKI67, Pro_C2_MKI67, and ProT_C3_MKI67 were also noticed (Supplemental Figure S12C, http://links.lww.com/HEP/J661).

First, to characterize the heterogeneity of macrophages across IM, MO, and NAT, genes related to macrophages highly expressed in diverse groups were calculated. For macrophages in MO, CCL3 and CCL4 were highly expressed. SPP1 was exceedingly expressed in macrophages derived from IM when macrophages in NAT highly expressed CD5L and MARCO (Figure 6I). MMP9+ macrophages (Mac_C1_MMP9), the well-known SPP1+TREM2+ macrophages (Mac_C13_SPP1), the proliferative macrophages (Mac_C8_MKI67) and the high-metallothionein expressing macrophages (Mac_C16_MT1G) were all inclined to be enriched in IM or MO and identified as TAMs (Supplemental Figure S13A, http://links.lww.com/HEP/J661). Mac_C1_MMP9 were the most abundant and highly expressed molecules related to M2 or TAMs polarization, including MMP9, MMP19, and FGFR1.²⁸ Pathway analysis showed that several classical pathways, including the NF-kappa B signaling pathway, the Toll−like receptor signaling pathway, etc. were enriched in Mac_C1_MMP9 which implied its regulatory roles (Figure 6J). RNA velocity for myeloid cell clusters showed that clear RNA velocity flows toward MMP9+ macrophages which might be terminally differentiated (Figure 6K). Especially, the fractions of Mac_C1_MMP9 were higher in MO than IM, while Mac_C13_SPP1 seemed to be more abundant in IM than MO (fc = 1.79, p-value adjusted = 0.138; fc = 2.05, p-value adjusted = 0.327, respectively) (Figure 6L). HIF-1 signaling, PPAR, carbon, and cholesterol metabolism were enriched in Mac_C13_SPP1 (Supplemental Figure S13B, http://links.lww.com/HEP/J661). Notably, Mac_C13_SPP1 showed significantly positive correlations with several T cells populations including 3 Treg clusters (CD4T_C2_IKZF2, CD4T_C3_TNFRSF18, and ProT_C3_MKI67) and 2 proliferative/exhausted T clusters (ProT_C1_MKI67 and ProT_C2_MKI67) (Supplemental Figure S13C, http://links.lww.com/HEP/J661). Furthermore, the fractions of MAIT and MMP9+ macrophages in IM and MO were negatively correlated (Figure 6M). Accordingly, TAMs impair MAIT cell function at the border of tumors in HCC.²⁹ Assorted tissue preference patterns estimated by ORs were also noticed in MO, but it showed a similar trend for each myeloid cluster to be enriched or absent in IM (Figure 6N and Supplemental Figure S14, http://links.lww.com/HEP/J661).

Diverse communication networks for the TMEs between IM and MO

To investigate potential communication mechanisms in IM or MO, correlations between different cell cluster abundances were quantified primarily. Tregs clusters, MMP9+ macrophages, CXCL13+ CD4 T cells, proliferative Trm cell clusters, LAMP3+ or CD274+ DCs formed a parallel pattern related to abundances across IM, MO, and NAT (Figure 7A). Besides, those clusters seem to be more connected in IM than MO (Figure 7B).

FIGURE 7.

Potential cell communications beneath the tumor microenvironments of IM and MO. (A) A pairwise Spearman correlation analysis of the proportions of each T-cell cluster and each myeloid cell cluster across samples derived from IM, MO. (B) A pairwise Spearman correlation analysis of the proportions of each T-cell cluster and each myeloid cell cluster across samples derived from IM and MO, respectively. (C) Representative ligand-receptor pairs calculated by CellPhoneDB across IM and MO. (D) Ligand-receptor pairs showing the interaction between MMP9+ macrophages and Pro_C1_MKI67 colored by ligand activity across IM and MO calculated by NicheNet. (E) Representative regions depicting cellular niches, including Tregs, MMP9+macrophages, and proliferative and exhausted T cells across IM and MO by multiplex immunofluorescence (P2 and P3). Abbreviations: IM, intrahepatic metastasis; MO, multicentric occurrence.

Considering the immune suppressive functions of Tregs and the emerging role of CXCL13+ CD4 T cells,³⁰ the interactions between Tregs or CXCL13+ CD4 T cells and other T-cell clusters were estimated. Interestingly, those proliferative T-cell clusters were closely connected with Tregs and CXCL13+ CD4 T cells regardless of IM or MO (Supplemental Figure S15A, http://links.lww.com/HEP/J661). Sequentially, we focused on the MMP9+ macrophages (Mac_C1_MMP9) and cDC2s since the former was enriched in IM/MO and the latter gradually changed across IM, MO, and NAT. cDC2s did cooperate with CD4+ T-cell clusters intimately, which was consistent with its functions (Supplemental Figure S15B, http://links.lww.com/HEP/J661). Besides, MMP9+ macrophages and cDC2s interacted strongly with those proliferative T-cell clusters, together with CD4_C3_TNFRSF18 and CD4T_C4_AREG (Supplemental Figure 15B, http://links.lww.com/HEP/J661). Given the noticeable interactions between MMP9+ macrophages and ProT_C1_MKI67, special ligand-receptor pairs were further identified. Several pairs, including HLA-E_KLRK1, LGASL9_HAVCR2, TNFSF9_TNFRSF9, and CXCL16_CXCR6 were enriched in MO for MMP9+ macrophages and ProT_C1_MKI67, distinct from CXCL2_DPP4, CCL3_CCR5, and TGFB1_TGFBR3 in IM (Figure 7C). Besides, a complex related to SPP1 was evidently achieved in IM, consistent with higher SPP1 expression in macrophages from IM. Additional NicheNet analysis revealed ligand-receptor pairs related to MMP9, including MMP9_ITGB2, MMP9_RECK, MMP9_CD44, and MMP9_ ITGB2 were enriched between MMP9+ macrophages and ProT_c1_MKI67 (Figure 7D). Furthermore, multiplex immunofluorescence confirmed the presence of cellular niches comprised of MMP9+ macrophages, Tregs, Pro_C1_MKI67 among IM and MO (Figure 7E and Supplemental Figure S16, http://links.lww.com/HEP/J661). Those evidences above indicated the potential interactions between MMP9+ macrophages and proliferative Trm cells might be key in the TMEs of IM and MO.

DISCUSSION

When intertumor and intratumor heterogeneity fluctuate among patients with multifocal HCC, comprehensive genomic and immune silhouette indeed facilitate the understanding of tumorigenesis and tumor metastasis. Considering that single-cell omics have deciphered the malignant cells or immune cells profoundly,³¹ our study presents a landscape of the cellular ecosystems related to pathogenesis beneath IM and MO.

Tumors from IM and MO showed different genomic features and transcriptomic traits. Here, we distinguished IM from MO based on whether tumors shared certain mutations at the genomic level. Additionally, IM tumors exhibited higher ITH, stronger stemness, and more activated pathways, including glycolysis and fatty acid metabolism when compared with MO tumors, suggesting that IM tumors were more aggressive and pernicious. In total, those evidences indicated that IM and MO covered separate pathogenesis, which always led to independent treatment options.

scRNA-seq enables tumor biology to be explored at the single-cell level, which has been carried out to investigate arcane ecosystems in cancer.¹⁵ scRNA-seq to recognize tumors and the TMEs in IM and MO provided certain clues to separate them two. T-cell exhaustion, defined by absent effector function and active inhibitory receptors, is a common state of T cells associated with dysfunction during tumorigenesis and tumor metastasis.³² In this study, exhausted T cells in IM and MO shared several Trm signatures of which a proliferative signature was noticed. Indeed, proliferative CD8+ T cells related to T-cell exhaustion have been reported in certain tumors, including non–small cell lung cancer, colorectal cancer, and head and neck squamous cell carcinoma.^22,33,34 Interestingly, immune checkpoint inhibitor therapy in colorectal cancer could reduce proliferative CD8+ Trm-like cells with exhaustion features.²² Notably, the proportions of those proliferative and exhausted CD8+ T cells were increased significantly across NAT, MO, and IM. Those cells were also biased to a terminally exhausted cluster and more inclined to be enriched in IM than MO and NAT, indicating severe anergy and tolerance for T cells in IM. We proposed that those proliferative and exhausted CD8+ T cells were involved in a special transcriptional state of T cells during tumorigenesis, especially in tumor metastasis. However, the mechanism of their formation is still indeterminate, which needs to be further explored.

MMP9+ macrophages were identified as TAMs in IM and MO, consistent with a recent study.³⁵ We found that MMP9+ macrophages were involved in communications among diverse cell types. For example, its fractions were correlated with MAIT cells negatively in IM and MO. Actually, TAMs could spoil MAIT cell function at the HCC invasive margin.²⁹ The TMEs are usually shaped by a series of interactive cells. For example, CD4+ helper T cells could interact with DCs and influence ICB response in non-small-cell lung cancer.³⁶ Moreover, CD4+ helper T cells and DCs could form cell niches, emerging a cellular triad with progenitor CD8+ T cells following ICB in HCC.³⁷ Here, correlation analysis for each T-cell cluster and myeloid cell cluster across NAT, MO, and IM showed that the fractions of MMP9+ macrophages, Tregs, those proliferative and exhausted T cells, and LAMP3+ DCs were relevant intimately, which implied potential niches where they communicated with each other. Besides, the cell niches seemed to be more tightly in IM than MO, indicating a more durable and tougher milieu in IM. This could partly explain more aggressive tumors since more adverse communications in IM than MO. However, potential mechanisms covering cellular ecosystems and cell-to-cell communications (e.g., MMP9+ macrophages, Tregs, proliferative/ exhausted T cells, and LAMP3+ DCs) should be settled with more evidence from functional experiments based on cell lines or mouse models related to multifocal HCC which were absent in this study.

Our data showed that CD20+ B cells and cDC2s were decreased in IM relative to MO and NAT, suggesting that those effective cells were lacking during tumor metastasis related to multifocal HCC. This evidence demonstrated that the procedure of antigen presentation was damaged in IM. Clinically, patients with multifocal lesions within the liver are treated as advanced HCC, which may benefit from immune checkpoint inhibitor.³⁸ The exploration of effective treatment for multifocal HCC is still underway. Our data showed that 2 antigen antigen-presenting cells: B cells and cDC2s, were relatively absent in IM when compared with MO. Accordingly, B cells and tertiary lymphoid structures could facilitate immunotherapy response.¹¹ Besides. correcting DCs functional defects related to aging could induce CD4⁺ T cells with cytolytic activity and drive antitumor immunity in elderly mice whose tumors were hard to be eradicated by direct anti-PD-1 and anti-CTLA4 immune checkpoint inhibitors.³⁹ Beyond immune checkpoint blockade, recovering B cells and cDC2s in multifocal HCC, especially in IM, may provide a silver lining for multifocal HCC immunotherapies.

In this study, we primarily focused on scRNA-seq, which provided a high-resolution insight into multifocal HCC while some limitations need to be noticed. First, one limitation of this study is the lack of integration with other omics data (eg, proteomics, metabolomics), which can facilitate a more comprehensive view of cellular mechanisms beneath multifocal HCC. Second, a larger cohort with scRNA-seq/WES available will enhance the generalizability and robustness of our findings. Third, a cross-sectional study may restrict our understanding to assess changes over time and track temporal dynamics in multifocal HCC, where longitudinal studies can shed light on.

In summary, our study revealed genomic heterogeneity and transcriptional features of IM and MO by WES and scRNA-seq analysis. IM and MO do harbor distinct milieus with diverse abundances and transcriptional functions of certain cell types characterized. IM presents more energetic malignant cells and more repressive TMEs than MO. This study illustrates a comprehensive landscape of tumors and the TMEs across IM and MO, the understanding of which enable personalized immunotherapies to be suitable for IM and MO.

Supplementary Material

hep-82-0582-s001.pdf ^{(4.9MB, pdf)}