Abstract

Background & Aims

   Cancer stem cells (CSCs) are well-established drivers of tumorigenesis,
   but their role in regulating tumor metastasis remains poorly
   understood. Here, we report the identification and characterization of
   a cluster of metastasis-promoting CSC-like cells in hepatocellular
   carcinoma (HCC).

Methods

   CSC-like cells in HCC were identified through the analysis of single
   cell RNA-sequencing data from 19 HCC samples. The stemness and invasive
   characteristics of these cells were evaluated using bioinformatical
   analyses of nine clinical cohorts and experimental validations. Spatial
   transcriptomics sequencing of 12 HCC samples revealed the cellular
   interactions between the CSC-like cells and tumor microenvironments,
   which were validated through gene co-expression analyses and
   immunohistochemistry. Finally, signaling pathway blockade was used to
   assess the potential clinical application of CSC-like cells.

Results

   Through comprehensive analyses of single cell RNA-sequencing data from
   19 patients with HCC and spatial transcriptomics data from 12 patients
   with HCC, a metastasis-promoting CSC-like subpopulation was identified.
   These CSC-like cells expressed high levels of epithelial–mesenchymal
   transition genes and were associated with poor prognosis of HCC.
   Histologically, CSC-like cells were enriched in highly aggressive
   tumors, especially in intrahepatic disseminated foci, where they
   interacted with immune cells. Functionally, CSC-like cells induced
   macrophage M2 polarization and T cell exhaustion through the ICAM1
   signaling pathway, forming immunosuppressive microenvironments.
   Downregulation of ICAM1 expression in CSC-like cells suppressed
   macrophage M2-polarization and T cell exhaustion, thereby reversing
   antitumor immune effects.

Conclusions

   Our study identified a metastasis-promoting CSC subpopulation,
   providing a potential perspective for CSC-targeted therapies in HCC.

Impact and implications

   The heterogeneity of CSCs in HCC has been identified, yet the
   identification and characterization of metastasis-promoting CSC
   subpopulations remain unexplored. Here, we identified a CSC-like tumor
   cell subpopulation that promotes HCC metastasis by increasing cell
   invasiveness and suppressing antitumor immune responses via the ICAM1
   signaling pathway. Our study uncovers novel mechanisms of HCC
   metastasis from the perspective of CSCs, and proposes potential tumor
   therapeutic strategies by inhibiting cellular interactions between
   CSC-like cells and immune cells.

   Keywords: Hepatocellular carcinoma, Single-cell RNA sequencing, Spatial
   transcriptomics, Cancer stem cell, Metastasis, Tumor microenvironment

Graphical abstract

   [55]Image 1
   [56]Open in a new tab

Highlights

     * •
       Comprehensive analysis of scRNA-seq and spatial transcriptomic data
       was used to investigate the characteristics of HCC CSCs.
     * •
       CSC-like cells, a distinct subpopulation of CSCs, exhibited higher
       invasiveness compared with conventional CSCs.
     * •
       CSC-like cells were characterized by high expression levels of
       CD24, ICAM1, ACSL4, BAG3, C17orf67, GOLGA8B, and RBM26.
     * •
       Functional interactions between CSC-like cells and immune cells
       promoted an immunosuppressive microenvironment.
     * •
       Targeting ICAM1 signaling disrupted CSC-like cell-mediated
       immunosuppression, enhancing antitumor immune responses.

Introduction

   Primary liver cancer has as the third highest mortality rate of all
   human malignancies, with ∼75–85% of liver cancers classified as
   hepatocellular carcinoma (HCC).[57]^1^,[58]^2 Tumor metastasis and
   recurrence are the leading cause of HCC-related death,[59]^3 with
   short-term tumor recurrence closely linked to tumor metastasis.[60]^4
   Therefore, tumor metastasis represents a crucial focus in HCC treatment
   investigations. Unfortunately, recently developed HCC treatment
   strategies, including surgical resection, radiofrequency ablation,
   tumor embolization, and tumorigenesis-based chemotherapy drugs, have
   limited therapeutic effects on HCC metastasis.[61]^5 Hence, there is an
   urgent need for tumor metastasis-targeted therapeutic strategies for
   HCC treatment.

   Tumor heterogeneity, driven by cancer stem cells (CSCs), is the leading
   cause of treatment failure, metastasis, and recurrence.[62]^6 CSCs are
   a rare population of tumor cells with self-renewing capacity that drive
   tumor initiation and are resistant to standard radiation and
   chemotherapeutic interventions.[63]^7 Increasing evidence suggests the
   heterogeneity of the CSC population, with CSC subtypes identified in
   gastric and colorectal cancer.[64]^8^,[65]^9 Remarkably, different CSC
   subtypes have been shown to have different roles in tumorigenesis and
   metastasis. In colorectal cancer, the tumorigenesis of the primary
   tumor is initiated by leucine-rich repeat-containing G-protein-coupled
   receptor 5^+ (LGR5^+) CSCs, whereas metastases are driven by LGR5^–
   CSCs.[66]^9^,[67]^10 However, the exact CSC subpopulation that
   initiates tumor metastasis, and the characteristics of these cells,
   remain elusive.

   Single-cell RNA-sequencing (scRNA-seq) is a high-throughput sequencing
   method used to uncover gene expression profiles at a single cell
   resolution. However, histological information is lost during the
   experimental process. Spatial transcriptomics (ST) maps transcriptomics
   in situ, thereby complementing scRNA-seq.[68]^11 The combination of
   scRNA-seq and ST has elucidated the landscape of tumor
   microenvironments (TMEs) in multiple tumor types.[69][12], [70][13],
   [71][14] Nevertheless, the integration of scRNA-seq and ST has not yet
   been applied to the investigation of CSC. Thus, in this study, we used
   scRNA-seq and ST in HCC and identified a CSC-like tumor subpopulation.
   Unlike conventional tumorigenic CSCs, this CSC-like population exhibits
   distinct characteristics. Comprehensive bioinformatic analyses were
   used to reveal gene expression profiles, histological distribution
   features, and in situ cellular interaction networks of CSC-like cells.
   Experimental validation and multiple clinical cohorts were also used to
   confirm the functions of this cell type ([72]Fig. 1A). Based on these
   strategies, we identified CSC-like cells as a metastasis-promoting
   subpopulation of CSCs in HCC, and uncovered the underlying mechanisms
   involved.

Fig. 1.

   [73]Fig. 1
   [74]Open in a new tab

   Identification of CSC-like subclusters in HCC.

   (A) Schematic overview of the study design. (B) UMAP plot of integrated
   HCC scRNA-seq datasets. (C) Feature plots showing expression levels and
   distribution characteristics of marker genes for each partition. (D)
   UMAP plot displaying the functional clusters of HCC tumor cells. (E)
   Dot plot showing the marker gene expression levels of functional
   clusters in tumor cells. (F) UMAP plot visualizing CSC-con and CSC-like
   subclusters in HCC tumor cells. (G, H) Top 20 enriched signaling
   pathways of highly expressed genes in CSC-con (G) and CSC-like (H)
   subclusters. Stemness-related pathways are highlighted with red boxes.
   CSC, cancer stem cell; CSC-con, CSC-conventional; HCC, hepatocellular
   carcinoma; scRNA-seq, single cell RNA-sequencing; UMAP, uniform
   manifold approximation and projection.

Materials and methods

Patient samples

   Our study was approved by the Ethics Committee of Peking University
   Third Hospital (Beijing, China). All research was conducted in
   accordance with both the Declarations of Helsinki and Istanbul.
   Patients with liver cancer who underwent surgery at the Department of
   General Surgery of the Peking University Third Hospital were enrolled
   according to the following criteria: (1) pathologically confirmed HCC;
   (2) no history of other malignancies; and (3) no anti-cancer therapy
   before surgery. All patients were informed and kept anonymous.

   All other information regarding the materials and methods is provided
   in the supplementary data online.

Results

Identification of CSC subclusters in HCC

   To investigate the transcriptomes of CSCs in HCC, we performed
   scRNA-seq analysis on tumors and peritumoral specimens from five
   patients with HCC (cohort 1, [75]Table S1). The clinicopathological
   information about these patients is summarized in [76]Table S2. In
   addition, we integrated our scRNA-seq data with a published HCC dataset
   (cohort 2, [77]Table S1),[78]^15 which contained tumor tissues from 14
   patients. Collectively, the integrated dataset comprised 116,858 single
   cells collected from 19 patients with HCC. We then clustered all cells
   into five major partitions: T/natural killer (NK) cells, stromal cells,
   hepatocytes, myeloid cells, and B cells ([79]Fig. 1B). To validate the
   identities of these cell types, we used lineage-specific markers,
   including CD3D and GNLY for T/NK cells, CD68 and CD14 for myeloid
   cells, CD79A for B cells, ALB for hepatocytes, and PLVAP and ACTA2 for
   stromal cells ([80]Fig. 1C).[81]^16^,[82]^17 Cellular distribution in a
   uniform manifold approximation and projection (UMAP) plot indicated low
   batch effects between patients from the two original datasets
   ([83]Fig. S1A). Significant differences in the transcriptomic profiles
   of hepatocytes, lymphocytes, and myeloid cells were observed between
   tumors and peritumoral regions, indicating their significant roles in
   HCC ([84]Figs. S1B and C).

   Next, we selected the hepatocyte partition from the tumor samples to
   analyze CSC. Reclustering of the hepatocytes identified eight
   functional clusters ([85]Fig. 1D): HCC_CSC; HCC_ECM (extracellular
   matrix); HCC_Immune; HCC_Metabolism; HCC_Oncogene; HCC_Replication;
   HCC_Secretion; and HCC_Stress. The identities of these functional
   subclusters were confirmed through the analysis of differentially
   expressed genes (DEGs) ([86]Fig. 1E). Considering the significance of
   CSCs in tumorigenesis and progression,[87]^6 we focused on HCC_CSC for
   further investigations. Apart from the selective expression of CSC
   markers ([88]Fig. 1E),[89]^17^,[90]^18 the HCC_CSC subpopulation also
   exhibited selective high expression of CSC signature epigenetic
   regulators ([91]Fig. S1D),[92]^19 transcription factors (TFs)
   ([93]Fig. S1E),[94]^19 and activation of stemness-related signaling
   pathways ([95]Fig. S1F).[96]^19 Taken together, these results validated
   the stemness of the HCC_CSC cluster.

   Notably, although CSCs were functionally identified as a whole, the
   HCC_CSC cluster in fact comprised two transcriptionally distinct
   subclusters ([97]Fig. 1F). To further characterize these two
   subclusters, we independently screened these subclusters for DEGs, and
   performed signaling pathway enrichment analysis. The results revealed
   activation of the Hippo signaling pathway ([98]Fig. 1G, red box) and a
   signaling pathway regulating stem cell pluripotency ([99]Fig. 1H, red
   box) in the DEGs from these two subclusters. This finding further
   confirms the stemness of these two subclusters.

Transcriptional characteristics and prognostic significance of CSC-like cells

   To illustrate the transcriptomes of the CSC subclusters, we compared
   their gene expression profiles, identifying substantial DEGs between
   them ([100]Fig. 2A). These subclusters were distinguished by different
   marker genes. One subcluster showed selective expression of EPCAM,
   PROM1, TACSTD2, KRT19, and CD24, whereas the other showed selective
   expression of CD24 and ICAM1 ([101]Fig. 2B). Therefore, we designated
   these subclusters as CSC-conventional (CSC-con) and CSC-like cells,
   respectively.

Fig. 2.

   [102]Fig. 2
   [103]Open in a new tab

   Marker gene identification and prognosis evaluation of CSC-like cells.

   (A) DEG analysis comparing CSC-con (upper right) and CSC-like cells
   (below left). Dots and labels colored red, blue, and gray show the
   preferential gene expression in CSC-con, CSC-like, and no significant
   difference, respectively, between the two CSC subclusters. (B) Violin
   plots showing relative gene expression levels of CSC marker genes in
   CSC-con, CSC-like, and non-CSC cells. (C) Multiplex immunofluorescence
   staining of HCC samples highlighting CSC-like cells (indicated by white
   arrows for CK8/18^+SMA^–ICAM1^+ cells). CK8/18 marks HCC tumor cells
   and SMA marks fibroblasts. (D) Violin plots showing relative gene
   expression levels of CSC-like marker genes in CSC-con, CSC-like, and
   non-CSC cells. (E) Flow cytometry analysis confirming the existence of
   CSC-like cells in an HCC sample. (F) qRT-PCR detecting expression
   levels of ACSL4, BAG3, C17orf67, and ICAM1 in CSC-con, CSC-like, and
   non-CSC cells from an HCC sample. Unpaired Student t test: ∗p <0.05,
   ∗∗p <0.01. (G) Relative expression levels of CSC-con markers (EPCAM,
   KRT19, CD44 and CD24) and CSC-like cell markers (CD24, ICAM1, C17orf67,
   ACSL4, and GOLGA8B) in HCC tumor cell subclusters in cohort 3. (H)
   Survival analysis of patients with HCC (cohort 7) showing survival
   correlation between prognosis risk score and RNA expression levels of
   CSC-like cell markers, Log-rank (Mantel-Cox) test. (I) Survival
   analysis of patients with HCC (cohort 8) showing correlation between
   prognosis risk score and protein expression levels of CSC-like cell
   markers, Log-rank (Mantel-Cox) test. CSC, cancer stem cell; CSC-con,
   CSC-conventional; DEG, differential gene expression; HCC,
   hepatocellular carcinoma.

   The presence of CSC-like cells in tumors was further confirmed through
   multiplex immunofluorescence staining ([104]Fig. 2C). Compared with the
   well-documented CSC-con, the CSC-like population remained obscure.
   Subsequently, we screened the DEGs of CSC-like cells for potential
   marker genes, identifying five additional genes: ACSL4, GOLGA8B,
   C17orf67, BAG3, and RBM26 ([105]Fig. 2D). Notably, these genes have
   been reported to be involved in cell stemness,[106][20], [107][21],
   [108][22], [109][23], [110][24] further supporting the stemness of
   CSC-like cells. Using HCC specimens, we confirmed the high levels of
   these genes in CSC-like cells ([111]Fig. 2E and [112]S2A). Compared
   with the CSC-con (CD24^+ICAM1^–) and non-CSC (CD24^–) subpopulations,
   the CSC-like (CD24^+ICAM1^+) subpopulation in both samples showed
   higher levels of ACSL4, C17orf67, and BAG3 ([113]Fig. 2F and [114]S2B).

   To validate the existence of CSC-like cells in a large number of
   patients with HCC, we performed bioinformatic analyses on three
   additional HCC cohorts, designated as cohort 3,[115]^25 cohort
   4,[116]^26 and cohort 5[117]^27 ([118]Table S1). Following similar
   analysis procedures as used above, we identified CSC-like
   subpopulations in these cohorts and observed specific expression of
   CSC-like cell markers ([119]Fig. 2G and [120]S2C,D). Moreover, ACSL4
   and BAG3 were also validated as CSC-like cell markers at the protein
   level ([121]Fig. S2E) (cohort 6, [122]Table S1).

   To investigate the clinical significance of CSC-like cells, we detected
   the correlation between CSC-like cell markers and the prognosis of
   patients with HCC. We first analyzed the relationship between the
   expression levels of CSC-like cell-specific genes and HCC prognosis
   using clinical data from The Cancer Genome Atlas (TCGA) Program (cohort
   7, [123]Table S1). The results showed that CSC-like cell signature
   genes, including CD24, ICAM1, ACSL4, BAG3, C17orf67, GOLGA8B, and
   RBM26, were effective independent predictors of poor HCC prognosis
   ([124]Fig. S2F).

   To comprehensively evaluate HCC prognosis based on CSC-like cell
   abundance, we applied the least absolute shrinkage and selection
   operator (LASSO) regression algorithm to establish a prognostic
   evaluation model: HCC prognosis risk score = 0.059 × Expr[CD24] +
   0.035 × Expr[ICAM1] + 0.083 × Expr[BAG3] + 0.030 × Expr[C17orf67] +
   0.133 × Expr[RBM26]. The prognosis evaluation model outperformed the
   single signature gene model ([125]Fig. S2G). Patients with a higher
   risk score had significantly worse prognosis ([126]Fig. 2H). We
   verified the prognostic prediction model using data from HCC cohort 8
   ([127]Table S1).[128]^28 This resulted in validation of the prognostic
   prediction model at the protein level, with patients showing higher
   risk scores exhibiting significantly shorter survival periods compared
   with those with lower risk scores ([129]Fig. 2I). These findings
   suggest that CSC-like cells are robust predictors of poor HCC prognosis
   at both the RNA and protein levels.

Cell invasiveness phenotype of CSC-like cells

   To explore the functions of CSC-like cells in HCC development, we
   compared the enrichment of DEGs between CSC-con and CSC-like cells in
   an epithelial cell proliferation-related gene set, and observed
   significant enrichment of CSC-con DEGs in this gene set
   ([130]Fig. S3A). Compared with the highly proliferative CSC-con,
   CSC-like cells exhibited a quiescent phenotype ([131]Fig. S3B),
   suggesting distinct roles for CSC-like cells beyond cell proliferation.
   Next, we analyzed the stemness-related signaling pathways between
   CSC-con and CSC-like cells, and observed greater activation of JNK/MAPK
   signaling in CSC-like cells compared with CSC-con cells ([132]Fig. 3A).
   In addition to its stemness-related roles, the JNK/MAPK signaling
   pathway has also been implicated in regulating cell metastasis.[133]^29
   Accordingly, we found enrichment of CSC-like DEGs in multiple
   metastasis-associated signaling pathways ([134]Fig. 3B). Moreover,
   compared with CSC-con, CSC-like cells showed higher expression levels
   of mesenchymal-associated signature genes, including ZEB1, FN1, GPC6,
   HTRA1, and CALD1 ([135]Fig. 3C). Higher expression of these mesenchymal
   signature genes in CSC-like cells compared with CSC-con cells was also
   observed in HCC cohorts 3 ([136]Fig. 3D) and 4 ([137]Fig. 3E),
   indicating the enhanced metastatic potential of CSC-like cells.

Fig. 3.

   [138]Fig. 3
   [139]Open in a new tab

   Cell invasiveness of CSC-like cells.

   (A) Heatmap comparing the activity of typical CSC signaling pathways
   between CSC-con and CSC-like cells. (B) Gene set enrichment analysis
   plots showing the enrichment of CSC-like DEGs in cell metastasis
   signaling pathways. p values and normalized enrichment score are
   indicated. (C–E) Heatmaps showing differential expression levels of
   epithelial and mesenchymal genes in CSC-con and CSC-like cells in (C)
   the scRNA-seq data, (D) cohort 3, (E) and cohort 4. (F) Violin plots
   showing the relative expression levels of CSC marker genes in CTC from
   cohort 9 compared with CSC-con and CSC-like cells from our data. (G)
   Stacked histogram showing the CSC-like cell proportion in patients with
   HCC with (M) or without (NM) tumor metastasis in cohort 7. (H) qRT-PCR
   analysis of HTRA1 expression levels in non-CSC versus CSC-like cell
   sorted from the HCC sample described in [140]Fig. 2E: unpaired Student
   t test: ∗∗p <0.01. (I) In vivo bioluminescence imaging of nude mice
   injected intrahepatically with luciferase-labeled CSC-like cells or
   non-CSC cells at 7 weeks after cell injection. Red arrows indicate the
   intrahepatic disseminated foci. CSC, cancer stem cell; CSC-con,
   CSC-conventional; CTC, circulation tumor cell; DEG, differential gene
   expression; HCC, hepatocellular carcinoma; scRNA-seq, single-cell RNA
   sequencing.

   To further verify the invasiveness of CSC-like cells, we detected the
   marker gene expression levels of CSC-con and CSC-like cells in
   circulation tumor cells (CTCs) of HCC (cohort 9,
   [141]Table S1).[142]^30 These CTCs are considered to be one of the most
   frequent origins of HCC metastasis. Therefore, the higher expression
   levels of CSC-like markers, rather than of CSC-con markers, indicated
   the enrichment of CSC-like cells in CTCs and their critical roles in
   HCC metastasis ([143]Fig. 3F). Moreover, we observed a higher
   proportion of CSC-like cells in primary tumors with metastasis compared
   with those without metastasis in cohort 7 ([144]Fig. 3G), further
   supporting the metastatic-promoting functions of CSC-like cells.

   The high invasiveness of CSC-like cells was further validated using HCC
   samples and a mouse model. CSC-like cells sorted from HCC samples
   exhibited elevated HTRA1 expression levels compared with non-CSC cells
   ([145]Fig. 3H and [146]S3C). Similarly, CSC-like cells sorted from the
   HCC cell line PLC/PRF/5 ([147]Fig. S3D) showed greater in vivo
   metastatic potential compared with non-CSC cells ([148]Fig. 3I and
   [149]S3E,F). Collectively, these findings suggest the high invasiveness
   of CSC-like cells.

   To elucidate the mechanisms underlying the invasiveness of CSC-like
   cells, we used SCENIC to explore the TFs. CEBPG, NR3C1, CEBPA, DDIT3,
   ATF3, BCLAF1, JUN, NR2F6, and CHD2 were selectively activated in
   CSC-like cells ([150]Fig. S3G). Among these TFs, CEBPA was predicted to
   be a TF of CALD1 (a representative mesenchymal gene),[151]^31
   supporting the metastasis-promoting function of CSC-like cells. In
   conclusion, CSC-like cells represent a highly invasive subpopulation of
   CSCs in HCC.

Spatial distribution characteristics of CSC-like cells

   To further investigate the histological characteristics of CSC-like
   cells, we analyzed the HCC microenvironments using ST. ST was performed
   on 12 fresh-frozen tumor specimens (cohort 10, [152]Table S1) from
   patients with HCC ([153]Table S3), and these tumor samples were
   classified based on their invasiveness. According to the completeness
   of the tumor capsules and the severity of intrahepatic spread, all 12
   samples were classified into one of three types ([154]Fig. 4A and
   [155]S4). Tumors with an intact capsule were classified as type 1,
   comprising samples H4, H8, H9, H11, and H14 ([156]Fig. 4A, yellow box).
   Type-2 tumors were characterized by incomplete capsules and the absence
   of intrahepatic spread, including samples H1, H2, H3, H12, and H13
   ([157]Fig. 4A, orange box). Type-3 tumors lacked complete capsules and
   exhibited intrahepatic disseminated foci, represented by samples H7 and
   H10 ([158]Fig. 4A, red box). Spot clustering results from ST data
   revealed a perfect correspondence between the H&E staining
   ([159]Fig. 4A [first lane] and [160]S4A) and the spot annotation
   outcomes ([161]Fig. 4A [second lane] and [162]S4B). Signature gene
   analyses were used to confirm the annotations ([163]Fig. 4A [third
   lane] and [164]S4C). Furthermore, copy number variation (CNV) analysis
   of the ST subclusters identified significant CNV in the tumor regions
   ([165]Fig. 4A [last lane] and [166]S4D), further supporting the ST
   annotation results.

Fig. 4.

   [167]Fig. 4
   [168]Open in a new tab

   Histopathological characteristics of CSC-like cells by ST data
   analysis.

   (A) Histopathological subtypes of HCC samples revealed by ST analysis.
   H&E staining (first lane), spot clustering results (second lane),
   marker gene expression levels (third lane), and CNV analysis (last
   lane) of ST data from HCC samples of H4 (type 1, yellow box), H1 (type
   2, orange box), and H7 (type 3, red box). ‘T’ represents tumor tissue
   (indicated by APOC2); ‘S’ represents stroma (indicated by RGS5 and
   COL3A1); ‘N’ represents peritumoral normal tissue (indicated by HP);
   and ‘I’ represents immune components (indicated by ITGB2, SLC40A1,
   IL7R, and MZB1). (B) Comparison of the relative proportion of CSC-like
   cells in tumor regions of type-1 (circles), type-2 (triangles), and
   type-3 (squares) HCC samples; unpaired Student t test: ∗∗∗∗p <0.0001.
   (C, D) Histopathological distribution (left) and statistical analysis
   (right) of CSC-like cell abundance in H7 (C) and H10 (D). Primary
   tumors and metastasis foci are indicated by black and red circles,
   respectively; unpaired Student’s t test: ∗∗∗∗p <0.0001. (E) IHC
   staining showing the histopathological distribution of ICAM1 in primary
   and metastasis regions. Red arrows indicate the enrichment of ICAM1
   protein in the tumor invasive frontline. (F, G) RNA velocity of H7 (F)
   and H10 (G). Arrows indicate directional cell state transition, with
   red circles highlighting intrahepatic disseminated foci. CSC, cancer
   stem cell; HCC, hepatocellular carcinoma; M, metastasis; P, primary;
   ST, spatial transcriptomics.

   We next used SPOTlight analysis to map the spatial distribution of CSCs
   by integrating scRNA-seq and ST data. When analyzed according to tumor
   sample type, both CSC-con ([169]Figs. S5A and B) and CSC-like cells
   ([170]Fig. 4B and [171]S5C) were more abundant in type-2 and type-3 HCC
   samples compared with type 1, suggesting a higher proportion of CSCs in
   aggressive tumors. Furthermore, type-3 HCC samples exhibited an
   abundance of CSC-like cells compared with type-2 HCC samples
   ([172]Fig. 4B). In particular, in situ analysis revealed that CSC-like
   cells were more localized to metastasis sites rather than to primary
   tumor sites in both H7 ([173]Fig. 4C) and H10 ([174]Fig. 4D),
   indicating the involvement of CSC-like cell in tumor metastasis.
   Immunohistochemical (IHC) staining further validated the distribution
   feature of CSC-like cells, revealing remarkable enrichment of ICAM1
   expression at the tumor invasive frontline ([175]Fig. 4E, red arrows).

   To further confirm the localization of CSC-like cell in metastases, we
   analyzed RNA velocity in situ. In both H7 ([176]Fig. 4F) and H10
   ([177]Fig. 4G) samples from our ST data, the evolution trajectory was
   directed from tumor disseminated foci, suggesting the enrichment of
   CSC-like cells in metastasis sites. Collectively, CSC-like cells
   exhibited preferential distribution in tumor metastasis sites.

Cellular interaction between CSC-like cells and tumor immune
microenvironments

   To explore the in situ functional mechanisms of CSC-like cells, we
   analyzed the cellular interaction network using CellChat. The
   clustering analysis of CSC-related signaling pathways identified four
   functional subgroups (Groups 1–4) ([178]Fig. 5A). Signaling pathways in
   groups 1, 3, and 4 were functionally enriched in immunomodulatory
   pathways, including T cells, NK cells, dendritic cells (DCs), and
   macrophage activation pathways, indicating frequent interactions
   between CSCs and the tumor immune microenvironments (TIMEs)
   ([179]Fig. 5B). Notably, CSC-like cells showed a distinct cellular
   interaction network compared with CSC-con ([180]Fig. 5C). These
   signaling pathways were enriched in immune recruitment and
   chemokine-mediated cellular responses ([181]Fig. 5D), further
   supporting the interactions between CSC-like cells and TIMEs. To
   visualize these interactions in situ, we resolved the cellular
   composition within CSC-like-positive spots in ST chips. The results
   showed a high abundance of immune cells, including myeloid cells and
   T/NK cells, within the CSC-like niche ([182]Fig. 5E), indicating
   colocalization of CSC-like cells with immune cells and their cellular
   interactions.

Fig. 5.

   [183]Fig. 5
   [184]Open in a new tab

   Cellular interaction between CSC-like cell and tumor immune
   microenvironments.

   (A) Functional similarity analysis of activated signaling pathways in
   HCC_CSC cells. (B) Gene ontology analysis of pathways in the four
   groups identified in (A). (C) Heatmap comparing the relative signaling
   pathway strength between CSC-con and CSC-like cells. Interaction
   strength ranges from high (dark green) to low (white). Contribution to
   total outgoing or incoming signaling of each pathway is shown on the
   right. (D) Top 20 enriched signaling pathways of activated cellular
   interactions in CSC-like cells. (E) In situ distribution and cellular
   composition of CSC-like-positive spots in HCC samples of H3, H7, and
   H10. CSC, cancer stem cell; CSC-con, CSC-conventional; HCC,
   hepatocellular carcinoma.

   To elucidate the exact cellular subtypes interacting with CSC-like
   cells, we reclustered the cell partitions in TME. Reclustering of
   myeloid cells resulted in 16 subclusters ([185]Figs. S6A and B).
   Reclustering of stromal cells resulted in 21 subclusters
   ([186]Figs. S6C and D), and that of lymphocytes resulted in 12
   subclusters ([187]Figs. S6E and F).

Immunosuppressive microenvironment induced by CSC-like cells

   Next, we identified the distinct cell interactions of CSC-like cells by
   comparing the CellChat analysis results of CSC-con and CSC-like cells.
   In contrast to CSC-con cells, CSC-like cells exhibited selective
   activation of the ICAM signaling pathway ([188]Fig. 6A). Components of
   the ICAM pathway comprised the ligand ICAM1 and various receptors,
   including SPN, ITGAL, ITGAX/ITGB2, ITGAM/ITGB2, and ITGAL/ITGB2
   ([189]Fig. S7A). Given the abundance of receptor pairs in macrophages
   and T cells ([190]Fig. S7A), we hypothesized that CSC-like cells
   regulate the TME via macrophages and T cells. Among the macrophage
   subclusters, Macro_1, and Macro_4 exhibited the strongest interaction
   with CSC-like cells in the ICAM signaling pathway ([191]Fig. 6B). M2
   polarization of macrophages has been reported following ICAM1 signaling
   pathway activation.[192]^32^,[193]^33 Consistently, we observed
   M2-polarization phenotypes in these macrophage subclusters, evidenced
   by high expression levels of MRC1, IL10, and TGFB1 ([194]Fig. 6C). In
   addition, elevated expression levels of SLC40A1, TREM2, and
   APOE[195]^34 were observed in Macro_1 and Macro_4 ([196]Fig. S7B),
   further confirming the M2 polarization of these subclusters. High
   expression levels of these M2-polarization markers and ICAM1 receptors
   were specifically found in tumor regions rather than in peritumoral
   normal regions at the protein level (cohort 6, [197]Table S1 and
   [198]Fig. S7C), indicating the tumor-promoting function of this
   signaling pathway. Interestingly, CSC-like cells also displayed
   selective interaction with subclusters T cell_2, T cell_5, and T cell_6
   ([199]Fig. 6B). These T cell subclusters were characterized by high
   expression of CTLA4, PDCD1, HAVCR2, LAG3, and TIGIT ([200]Fig. 6D).
   Hence, the enrichment of these immune checkpoint molecules[201]^35 in
   CSC-like cell-interacting T cell subclusters offered direct evidence
   supporting the immunosuppressive states in the CSC-like niche.

Fig. 6.

   [202]Fig. 6
   [203]Open in a new tab

   Immunosuppressive microenvironment induced by CSC-like cells through
   the ICAM1 signaling pathway.

   (A) Chord diagram showing ICAM signaling pathway networks between the
   CSC-con, CSC-like cells, and HCC TME. (B) Bubble plot showing the
   communication probability of the ICAM pathway ligand–receptor pairs
   between subclusters of T cell, macrophage, and CSC-like cells. (C, D)
   Violin plots displaying the relative expression levels of (C)
   M2-polarization markers (MRC1, IL10, and TGFB1) in macrophage
   subclusters and (D) exhaustion markers (CTLA4, PDCD1, HAVCR2, LAG3, and
   TIGIT) in T cell subclusters. (E) Heatmap showing co-expression
   probabilities of ICAM1/ITGB2 in the same or adjacent spots from HCC
   samples. Spearman’s correlation analysis: p values of co-expression
   significance are colored from red (p <0.05) to white (p = 0.05) and to
   blue (p >0.05). (F) Distribution of ICAM1 and ITGB2 in the ST slide of
   HCC sample from H7. Zoomed-in portions of the HCC section (middle) show
   multiple areas of colocalization. (G, H) IHC double staining showing
   adjacent distribution of (G) CSC-like cells (ICAM1-positive) and
   M2-polarized macrophages (CD163-positive) and (H) CSC-like cells
   (ICAM1-positive) and exhausted T cells (PD-1-positive) in HCC. (I)
   Correlation analysis of the expression of ICAM1 and immunosuppressive
   markers, including CD163, TIGIT, and HAVCR2, in cohort 7, Pearson’s
   correlation analysis. CSC, cancer stem cell; CSC-con, CSC-conventional;
   HCC, hepatocellular carcinoma; IHC, immunohistochemistry; TME, tumor
   microenvironment.

   We next investigated ICAM1 signaling histologically by reanalyzing the
   ST data. In specimens containing CSC-like cells (H1, H3, H7, H10, H12,
   and H13), we observed co-expression of ICAM1 with ITGB2, ITGAL, and
   ITGAM in the same or adjacent spots on the ST chips ([204]Fig. 6E,F and
   [205]S7D), indicating activated ICAM1 signaling in the TME. The
   adjacent histological distribution of CSC-like cells (ICAM1-positive
   cell) with M2-polarized macrophages (CD163-positive cells) and CSC-like
   cells (ICAM1-positive cell) with exhausted T cells (PD-1-positive
   cells) was further confirmed through IHC staining ([206]Fig. 6G,H).
   Furthermore, the correlation between the expression of ICAM1 and
   macrophage M2-polarization marker genes, such as CD163 and CD200R, as
   well as T cell exhaustion marker genes, including TIGIT, HAVCR2, and
   CTLA4, was further confirmed in HCC cohort 7 ([207]Fig. 6I and [208]S7E
   and [209]Table S1). Taken together, these findings show that CSC-like
   cells induce M2 polarization of macrophages and promote T cell
   exhaustion.

Modulation of TIME through regulation of ICAM1 signaling

   To verify the cellular interaction between CSC-like cells and the TIME,
   we sorted CSC-like cells (CD24^+ICAM1^+) and non-CSCs (CD24^–) from the
   HCC cell line PLC/PRF/5, and co-cultured these cells with THP-1 cells
   ([210]Fig. 7A). THP-1 cells co-cultured with CSC-like cells showed
   higher expression levels of M2-polarization marker genes, including
   CD163 and CD200R, compared with those co-cultured with non-CSC
   ([211]Fig. 7B), suggesting that CSC-like cells induce M2 polarization
   of macrophages. Similarly, co-culture of CSC-like cells or non-CSCs
   with Jurkat ([212]Fig. 7C) revealed consistent results. Jurkat cells
   co-cultured with CSC-like cells showed higher expression levels of
   exhaustion marker genes, including PDCD1 and TIGIT, compared with those
   co-cultured with non-CSCs ([213]Fig. 7D), indicating that CSC-like
   cells promoted T cell exhaustion.

Fig. 7.

   [214]Fig. 7
   [215]Open in a new tab

   Knockdown of ICAM1 reverses the immunosuppression state of TME.

   (A) Schematic of the co-culture system of THP-1 cells with CSC-like
   cells and non-CSC cells. (B) qRT-PCR analysis of CD163 and CD200R
   expression levels in THP1 cells co-cultured with CSC-like or non-CSC
   cells sorted from PLC/PRF/5; (C) Schematic of the co-culture system of
   Jurkat cells with CSC-like cells and non-CSC cells. (D) qRT-PCR
   analysis of PDCD1 and TIGIT expression levels in Jurkat cells
   co-cultured with CSC-like or non-CSC cells sorted from PLC/PRF/5. (E)
   Flow cytometry confirming the downregulation of ICAM1 expression in
   PLC/PRF/5 shICAM1 cells. (F) qRT-PCR detection of MRC1 and CD200R
   expression levels in THP1 cells co-cultured with sh-scramble (scr) or
   shICAM1 PLC/PRF/5 cells. (G) qRT-PCR detection of PDCD1 and TIGIT
   expression levels in Jurkat cells co-cultured with sh-scramble (scr) or
   shICAM1 PLC/PRF/5 cells. (H) Schematic showing the functional
   mechanisms of CSC-like cells in tumor metastasis. CSC-like cells
   promote tumor metastasis through high invasiveness and
   immunosuppression roles via cellular interaction with macrophages and T
   cells via ICAM1 signaling. Knockdown of ICAM1 in CSC-like cells
   reverses the suppressive TME and inhibits tumor metastasis. Unpaired
   Student t test: ∗p <0.05, ∗∗∗p <0.001, ∗∗∗∗p <0.0001. CSC, cancer stem
   cell; IHC, immunohistochemistry; TME, tumor microenvironment.

   Given the immunoinhibitory function of CSC-like cells, we next explored
   whether this phenomenon could be reversed by repressing ICAM1
   signaling. We suppressed ICAM1 expression in tumor cells using short
   hairpin (sh)RNA ([216]Fig. 7E), and co-cultured these cells with THP-1
   or Jurkat cells. As a result, the immunoregulatory role of CSC-like
   cells was attenuated following ICAM1 knockdown ([217]Fig. 7F,G).

   Collectively, our data identified a CSC-like subpopulation in HCC that
   promotes tumor metastasis through high invasiveness and regulation of
   TIME, driving it into immunosuppressive states. Furthermore, ICAM1
   knockdown inhibited M2 polarization of macrophages and T cell
   exhaustion, thereby strengthening antitumor immune responses and
   repressing tumor metastasis ([218]Fig. 7H).

Discussion

   Although the heterogeneity of CSCs in HCC has been previously
   characterized using scRNA-seq and ST,[219][36], [220][37], [221][38]
   the functions of CSC subpopulations remain poorly understood. Our study
   integrated scRNA-seq and ST data and identified CSC-like cells as the
   metastasis-promoting CSC subpopulation in HCC. CSC-like cells function
   in tumor metastasis through the following mechanisms. First, the
   CSC-like subpopulation exhibits high invasiveness by expressing high
   levels of epithelial–mesenchymal transition (EMT) signature genes.
   Second, CSC-like cells induce macrophage M2 polarization and T cell
   exhaustion via the ICAM signaling pathway, reshaping the immune
   microenvironment into immunosuppressive states and permitting tumor
   metastasis. Intriguingly, CSC-like cell markers, including ICAM1,
   ACSL4, GOLGA8B, and BAG3, may provide additional potential mechanisms
   for the metastasis-promoting behavior of CSC-like cells. In HCC, high
   levels of ICAM1 in serum and tumor tissue have been shown to correlate
   with tumor metastasis[222]^39^,[223]^40: ACSL4 reprograms fatty acid
   metabolism and induces tumor progression[224]^41; GOLGA8B regulates
   cell invasion and metastasis through the STAT3 signaling
   pathway[225]^42; and BAG3 locates at the leading edge of migrating
   cells and regulates EMT.[226]^43 Collectively, these findings not only
   support the metastasis-promoting function of CSC-like cells, but also
   imply the involvement of more underlying mechanisms.

   CSCs are highly involved in tumorigenesis and recurrence by
   definition,[227]^44^,[228]^45 making them ideal therapeutic targets.
   Numerous clinical trials are targeting CSC surface antigens.[229]^46
   However, the results of many CSC-targeted clinical trials have been
   disappointing because of severe side effects. Approximately one-third
   of patients participating in anti-CD44 or anti-ALK-1 (TGF-β receptor)
   clinical trials suffered severe adverse events.[230]^47^,[231]^48 These
   results are likely the result of the extensive similarities of normal
   stem cells and CSCs,[232]^49 especially the CSC-con population.
   Therefore, targeting metastasis-specific CSC subpopulations, referred
   to as the CSC-like population, could offer novel insights into
   suppressing HCC progression. Hence, inhibitors targeting CSC-like cell
   markers, including ICAM1, C17orf67, ACSL4, GOLGA8B, BAG3, and RBM26,
   have the potential to suppress HCC progression without affecting normal
   stem cells, thereby significantly reducing side effects.
   Transformations between CSC-con and CSC-like cells should be considered
   when designing CSC-targeting drugs. Therefore, future investigations
   could focus on evaluating the plasticity of CSC subpopulations.
   Simultaneously targeting both CSC subpopulations could improve
   prognosis, while identifying key transformation regulators between
   CSC-con and CSC-like cells could provide new treatment strategies for
   HCC.

   By comprehensive use of scRNA-seq and ST data, our study not only
   elucidates the gene expression profile and in situ cellular
   interactions of the CSC subpopulation, but also provides the first
   spatial maps of its distribution. Both CSC-con and CSC-like cells were
   more abundant in invasive tumors lacking intact capsules. Previous work
   highlighted the associations between tumor capsule and patient
   prognosis, including tumor metastasis and recurrence.[233]^50 The
   widely accepted explanation for this phenotype is that the fibrotic
   structure forms a physical barrier against tumor migration, while our
   findings introduce a novel theory based on CSCs. Tumors can be
   classified into highly aggressive and less aggressive categories
   depending on their invasiveness. CSCs, especially CSC-like cells, are
   enriched in highly aggressive tumors and infiltrate stroma regions
   before the formation of thick capsules. These CSCs highjack immune
   cells and invade microvessels, ultimately driving tumor metastasis.

   In conclusion, by integrating bioinformatics analyses of scRNA-seq and
   ST data, our study identifies a metastasis-promoting CSC subtype,
   referred to as CSC-like cells. This subpopulation is characterized by a
   high expression of CD24, ICAM1, ACSL4, GOLGA8B, C17orf67, BAG3, and
   RBM26, and shows strong metastatic potential. Spatially, CSC-like cells
   are enriched in invasive tumors, especially in metastatic sites.
   Functionally, CSC-like cells drive macrophage M2 polarization and T
   cell exhaustion, contributing to the establishment of an
   immuno-suppressive microenvironment. Thus, this metastasis-initiating
   CSC-like subpopulation represents a promising therapeutic target for
   HCC treatment.

Abbreviations

   CNV, copy number variation; CSC-con, CSC-conventional; CSC, cancer stem
   cell; CTC, circulation tumor cell; DC, dendritic cells; DEG,
   differentially expressed gene; ECM, extracellular matrix; EMT,
   epithelial–mesenchymal transition; HCC, hepatocellular carcinoma; IHC,
   immunohistochemistry; LASSO, least absolute shrinkage and selection
   operator; LGR5, leucine-rich repeat-containing G-protein-coupled
   receptor 5; M, metastasis; NK, natural killer; P, primary; scr,
   sh-scramble; scRNA-seq, single-cell RNA sequencing; shRNA, short
   hairpin RNA; ST, spatial transcriptomics; TCGA, The Cancer Genome
   Atlas; TF, transcription factor; TIME, tumor immune microenvironment;
   TME, tumor microenvironment; UMAP, uniform manifold approximation and
   projection.

Financial support

   This work was supported by grants including the National Key Research
   and Development Program of China (2021YFA1300601 to YY), National
   Natural Science Foundation of China (82030081 to YY, 30700349, 30440012
   to LG, and 32101047 to YL), the Beijing Municipal Science and
   Technology Commission (Z131100004013036 to LG), the Shu Fan Education
   and Research Foundation, and the Lam Chung Nin Foundation for Systems
   Biomedicine.

Authors’ contributions

   Conceived the study and designed major experiments: YY, LG, CY.
   Performed sample preparation: CY, YL, ZW, GW, YB, GL. Performed DNA
   library construction of scRNA-seq and ST sequencing: XM. Performed
   sample collection, clinical diagnosis, and evaluation of the validation
   study: LG, YB. Performed IHC: YL, LG. Performed the animal experiments:
   ZH, CY. Analyzed scRNA-seq data: CY, GZ, AL, YJ. Analyzed ST data: CY,
   GZ, XZhang. Performed flow cytometry and quantitative real-time PCR:
   CY, YL. Performed MS and analyzed the data: HS, XZhao. Wrote the
   manuscript: CY, JL, LG, YY. The first-author position is determined by
   the relative degree of contribution to this study.

Data availability statement

   The raw data of scRNA-seq and ST results have been added to the Genome
   Sequence Archive for Human ([234]https://ngdc.cncb.ac.cn/gsa-human/)
   under accession number [235]HRA001399.

Conflicts of interest

   The authors declare no conflicts of interest that pertain to this work.

   Please refer to the accompanying ICMJE disclosure forms for further
   details.

Acknowledgements