Abstract

   Intrahepatic cholangiocarcinoma (ICC) tumor cells and their
   interactions with the immune microenvironment, particularly at the
   leading‐edge area, have been underexplored. This study employs
   single‐cell RNA sequencing (scRNA‐seq) and spatial transcriptome (ST)
   analysis on samples from the tumor core, adjacent non‐tumorous tissue,
   and the leading‐edge area of nine ICC patients. These findings indicate
   that tumor cells at the leading‐edge area demonstrate enhanced
   proliferation and are tightly associated with the stroma, including
   endothelial cells and POSTN+ FAP+ fibroblasts. Notably, CD8+ T cells in
   this region exhibit a naive phenotype with low cytotoxicity and signs
   of exhaustion, likely due to compromised antigen presentation by
   antigen‐presenting cells (APCs). The predominant CD8+ T cell subset,
   mucosal‐associated invariant T (MAIT) cells, recruits SPP1+ macrophages
   within the stroma. This interaction, along with the presence of POSTN+
   cancer‐associated fibroblasts (CAFs) and endothelial cells, forms a
   unique “triad structure” that fosters tumor growth and ICC progression.
   The research highlights the intricate characteristics and interactions
   of ICC tumor cells in the leading‐edge area, offering insights into
   potential therapeutic targets for intervention.

   Keywords: immune landscape, intrahepatic cholangiocarcinoma,
   leading‐edge area, spatial transcriptomics
     __________________________________________________________________

   This study employs single‐cell RNA sequencing and spatial
   transcriptomics to delineate the microstructure and immune landscape of
   Intrahepatic cholangiocarcinoma (ICC), particularly in the leading‐edge
   area. It reveals enhanced tumor cell proliferation, a unique immune
   environment, and a new defined microstructure called “triad structure,”
   that occurred in the ICC tumors’ leading edge.

   graphic file with name ADVS-12-2412740-g005.jpg

1. Introduction

   Intrahepatic cholangiocarcinoma (ICC) ranks as the second leading type
   of primary liver malignancy, noted for its exceptionally aggressive
   essence and a generally poor prognosis.^[ [44]^1 ^, [45]^2 ^] Most
   patients have limited treatment chances for complete surgical resection
   when they are often first diagnosed at an advanced stage, and there is
   still a high risk of recurrence after surgery.^[ [46]^3 ^] At present,
   new treatment methods, such as targeted therapy (inhibitors or
   mutation‐targeting agents) and immunotherapy, become the more promising
   approaches for late‐stage patients.^[ [47]^4 ^] Targeted therapy has
   achieved promising results through targeting specific genetic mutations
   in patients. Targeted drug therapy, such as FGFR2/IDH‐1 inhibitors, is
   used in subgroups of patients with specific mutational alterations and
   has achieved good clinical outcomes.^[ [48]^5 , [49]^6 , [50]^7 ,
   [51]^8 ^] Patient responses to identical treatment modalities can
   diverge significantly, a phenomenon particularly pronounced in the
   realms of targeted and immunotherapies. This underscores the imperative
   to explore the intricacies of the tumor microenvironment (TME), aiming
   to achieve a more nuanced understanding of its characteristics.

   In the quest to enhance our grasp of the TME in ICC, the advent of
   scRNA‐seq and ST technologies has bestowed upon researchers a suite of
   potent and exacting investigative instruments. The combination of two
   technologies enables researchers to explore the complex cellular
   interactions and the spatial relationships that define the intricate
   network within the tumor's ecosystem, which also enables us to focus on
   a long‐overlooked area, the leading‐edge area of ICC tumor, also known
   as the tumor‐normal interface or invasive tumor front. We and others
   have delved into the heterogeneity of ICC tumors, uncovering that the
   immune microenvironment exhibits significant variation across different
   regions within the same tumor.^[ [52]^9 ^] This discovery prompted our
   subsequent inquiry into the unique phenomena occurring in a
   particularly critical area—the leading‐edge zone of the ICC tumor.

   In prior studies, Zhou and colleagues pinpointed a 500‐µm‐wide band
   encircling the periphery of liver cancer tumors, naming it “the
   invasive zone.” Within this region, the emission of Serum Amyloid A
   proteins (SAAs) by compromised hepatocytes was found to summon
   macrophages and incite their M2 polarization. This process is believed
   to foster a localized state of immunosuppression, which could, in turn,
   fuel the advancement of the tumor.^[ [53]^10 ^] Meanwhile, Li's
   research shed light on the significant aggregation of renal cell
   carcinoma (RCC) cells at the invasive tumor front, characterized by a
   pronounced epithelial‐mesenchymal transition (EMT).^[ [54]^11 ^]
   Bouchard et al. found the crucial interplay between tumor and stromal
   cells, showing that their communication within the tumor
   microenvironment activates distinct biological pathways based on
   location—edge or core. They emphasized phenotypic distinctions in tumor
   cells when cultivated with fibroblasts sourced from the tumor's
   invasive edge versus its core.^[ [55]^12 ^]

   The collective evidence points to the tumor's leading edge, or the
   tumor‐normal interface, as a region that is intimately associated with
   the tumor's specific behaviors and progression. Given the heightened
   malignancy observed in ICC, our study undertook an in‐depth single‐cell
   and spatial transcriptome analysis on a cohort of nine ICC patients.
   This research was designed to reveal the characteristics of the TME in
   tumor's leading edge, encompassing the distribution of stromal and
   immune cells, the tumor niche's surrounding ecosystem, and the complex
   interplay among diverse cell types. The findings from this study
   provide novel and profound insights into the intricate dynamics of the
   ICC ecosystem, suggesting potential therapeutic targets that may be
   instrumental in impeding tumor advancement.

2. Results

2.1. The General Outline of the Leading‐Edge Area Depicted by Single‐Cell and
Spatial Transcriptome Analysis

   To fully elucidate the features of the leading‐edge area in ICC, we
   amassed tissue samples from nine ICC patients. This collection included
   core tumor tissues (T, with seven samples), samples from the
   leading‐edge area (L, with nine samples), and corresponding
   non‐neoplastic adjacent tissues (N, with nine samples), all of which
   were subjected to single‐cell RNA sequencing (scRNA‐seq). Furthermore,
   well‐preserved leading‐edge area samples (n = 5) underwent spatial
   transcriptomics (ST) sequencing on the 10x Genomics Visium platform
   (Figure [56] 1A; Figure [57]S1A, Supporting Information). Comprehensive
   clinical profiles and pathological evidence for the ICC patients
   enrolled in this study are detailed in Table [58]S1 and Figure [59]S1B
   (Supporting Information), respectively.

Figure 1.

   Figure 1
   [60]Open in a new tab

   The general outline of the leading‐edge area depicted by single‐cell
   and spatial transcriptome analysis. A) Schematic diagram of the entire
   project, including patient‐matched sequencing information. T: tumorous
   core tissues, N: matched non‐tumor tissue (>3 cm from the border), L:
   leading‐edge area (expand ≈0.5 cm to both sides from the borderline).
   B) The UMAP visualization of the distribution of predominant cell
   populations, each distinguished by its specific color coding. C) UMAP
   plots illustrate the expression levels of key genes across diverse cell
   types. D) The distribution of major cell types across the indicated
   groups. E) Proportion of all cell types within the indicated samples.
   F) The tree plot represents the transcriptomic similarity of cell
   subtypes of four major classical cell types. G) Spatial transcriptomic
   results of the P6 patient, including H&E staining of tumor tissue,
   tumor‐side marker (KRT19), and non‐tumor side marker (ALB), and the
   distribution of fibroblast (fibro) and endothelial (endo) signature and
   corresponding markers (Fibroblast: COL1A1, ACTA2, TAGLN, and MYL9;
   endothelial: PECAM1, VWF, and CLDN5). *p < 0.05, **p < 0.01, ***p <
   0.001. No significant difference (n.s.).

   In summary, ≈230 000 high‐quality single‐cell transcriptomes were
   retained after quality control assessments. Across the patient cohort,
   we discerned six predominant cell types, including myeloid cells (LYZ),
   epithelial (including malignant tumor cells, hepatocytes, and
   cholangiocytes) (KRT19), fibroblasts (ACTA2), endothelial cells (CDH5),
   T/NK cell (CD3E, GNLY), and B cells (BANK1, JCHAIN), based on their
   canonical markers (Figure [61]1B,C).

   Diverse primary subpopulations are present across various tumor locales
   (Figure [62]1D) and within each sample (Figure [63]1E). However, these
   subpopulations exhibit varying proportions, thus highlighting the
   regional, as well as the inter‐ and intra‐patient heterogeneity within
   the tumor immune microenvironment in ICCs. In agreement with other
   studies,^[ [64]^13 ^] our study revealed a marked enrichment of
   fibroblasts within the tumor tissues (Figure [65]1D). Apart from tumor
   cells, other major immune cells may not show much difference in
   distribution tendency, implying the difference in different regions of
   ICC tumors may need more subtle investigation and the refined cluster
   of major cell types. Therefore, we here have summarized all cell
   subpopulations identified and depicted in this study, and the “hclust”
   algorithm was used to demonstrate the transcriptomic similarity among
   cell subpopulations (Figure [66]1F).

   To depict the characteristics of ICC spatial organization in the
   leading‐edge (L) area, we used ALB and KRT19 to define the tumor side
   and non‐tumor side on the L‐area slice (Figure [67]1G, left, and Figure
   [68]S1C, Supporting Information). Meanwhile, we used the endothelial
   and fibroblast signature derived from matched scRNA data of the L area
   to outline the stromal area in each slice (Figure [69]1G, right, and
   Figure [70]S1C, Supporting Information). The corresponding canonical
   markers of endothelial cells (PECAM1, VWF, and CLDN5) and fibroblasts
   (COL1A1, ACTA2, TAGLN, and MYL9) also prove the feasibility of this
   strategy. In line with the previous study,^[ [71]^14 ^] the stromal
   region acts as a barrier at the tumor‐normal interface and also extends
   into the tumor region, dispersing or encircling the tumor cells. This
   observation underscores the complex interplay and spatial distribution
   of stromal elements with tumor cells.

2.2. ScRNA Sequencing and Spatial Transcriptomics Reveal ICC Tumor Cell
Heterogeneity at the Invasive Tumor Front

   To evaluate the spatially characteristic disparities of tumor cells
   within ICC, we first applied the inferCNV algorithm which used immune
   cells as a reference, together with a marker‐based strategy, to
   identify tumor cells from epithelial (Figure [72] 2A,B). All patients
   contain cholangiocytes and ICC tumor cells, but a small number of
   patients (P5, P6, P7) lack hepatocytes based on our single‐cell
   dissociation strategy (Figure [73]S2A,B, Supporting Information). We
   found that tumor cell similarity is more obvious in intra‐patient
   samples than in inter‐patient samples (Figure [74]S2C, Supporting
   Information). And those previously reported genes such as CFTR, TM4SF4,
   and FXYD2, which characterize bile duct cells or cholangiocytes, have
   also been well reproduced in our single‐cell data (Figure [75]S2D,
   Supporting Information). We extracted and re‐clustered ICC tumor cells
   and named those overexpressing MKI67, TOP2A, and UBE2C as proliferating
   tumor cells (Prolif) (Figure [76]2C). We found that proliferating tumor
   cells were more enriched in the leading‐edge area (L) compared with the
   tumor‐core area (T) (Figure [77]2D; Figure [78]S2E–H, Supporting
   Information), which is further proved by immunohistochemistry (IHC)
   evidence (Figure [79]2E). Further, we investigated the potential driver
   transcription factors (TFs) using SCENIC and found that E2F1 and CEBPB
   (Figure [80]S2I, Supporting Information), associated with proliferation
   and stemness, were significantly upregulated in L‐side tumor cells,
   which further proved our above finding. We then collected a set of
   differentially expressed genes (DEGs) resulting from the comparative
   analysis of tumor cells from the L side and T side, conducting a KEGG
   pathway enrichment analysis. We found that in addition to the
   above‐mentioned represented proliferating characteristic (e.g., DNA
   replication), tumor cells from the leading‐edge area are also enriched
   in pathways like ribosome, ECM receptor interaction, and cell adhesion
   molecules (CAMs), etc (Figure [81]2F). We used the “addmodulescore”
   algorithm to evaluate other common behaviors of tumor cells and found
   that the tumor cells derived from the tumor‐core area exhibit obvious
   hypoxia, while the tumor cells in the leading‐edge region have
   significant advantages in stemness and EMT behaviors (Figure [82]2G).
   We also found that stemness‐related genes, e.g., CD44 and CD133, were
   overexpressed in L‐side tumor cells, and overexpression of CD44 brought
   a worse prognosis in the FDU‐ICC cohort (Figure [83]S2J,K, Supporting
   Information).

Figure 2.

   Figure 2
   [84]Open in a new tab

   ScRNA sequencing and spatial transcriptomics reveal ICC tumor cell
   heterogeneity at the leading‐edge area. A) Schematic diagram of the
   strategy to identify tumor cells. B) UMAP plots of three epithelial
   subtypes (Cholangiocyte, hepatocyte, and tumor cell) were identified
   using inferCNV and gene marker‐based strategy. C) UMAP plots of tumor
   cell subtypes (top) and their corresponding markers as demonstrated by
   violin plot (bottom). D) The proportion of proliferating tumor cells in
   the lead‐edge and core regions of ICC tumor. E) Immunohistochemistry
   (IHC) staining images (bottom) and statistic differences (top) of
   Ki‐67+ tumor cells in the L‐ and T‐side regions within ICC samples. F)
   KEGG pathway enrichment analysis was applied to examine the
   differentially expressed genes (DEGs) between tumor cells from the L
   and T sides. G) The “addmodulescore” algorithm was used to score the
   signature of nine biological processes in tumor cells from L‐side and
   T‐side ICC samples. H) The spatial dot plots showed the distribution of
   proliferating tumor cells and stromal (fibroblast and endothelial) and
   the tumor niches within were sliced by layers and calculated the score
   of proliferating signature. I) The mIHC results of ICC tumors in the
   leading‐edge zone show the distribution relationship between
   proliferative tumor cells (Pan‐ck+, Ki‐67+) and the stromal area
   [composed of endothelial cells (CD31+) and fibroblasts (collagen I+,
   periostin+)]. J) The invasive tumor front was sliced by layers (left)
   and the IHC scores (right) were calculated. *p < 0.05, **p < 0.01, ***p
   < 0.001. No significant difference (n.s.).

   To observe the distribution of proliferating tumor cells in ICC
   tissues, we constructed a signature of proliferative markers and
   enriched them in the ST data. We found that proliferating tumor cells
   were mainly distributed around the tumor niche (or cancer cell
   aggregates), close to the stromal area which is composed of endothelial
   and fibroblast (Figure [85]2H; Figure [86]S2L, Supporting Information).
   We circled out the entire tumor niche and performed onion‐like
   stratification. The proliferative signal of the inner‐layer tumor cells
   is weaker than that of the outer‐layer tumor cells (Figure [87]2H).
   Using multiplex immunohistochemistry (mIHC) technology, we confirmed
   that the location of proliferating tumor cells (Ki‐67+, Pan‐CK+) is
   close to where endothelial cells (CD31+) and fibroblasts (Collagen I+)
   are found (Figure [88]2I). However, in the leading‐edge area, as tumor
   cells approach the normal‐tumor interface line, it seems that they do
   not necessarily proliferate faster (Figure [89]2J).

2.3. Naive CD8+ T Cells Dominate and MAIT Cells Infiltrate the Leading‐Edge
Area

   T/NK cells play a pivotal role in the TME. To delve deeper into the
   characteristics of T/NK cells in the leading‐edge zone, we conducted a
   refined clustering analysis, categorizing the T/NK cells—comprising
   CD4+ T cells, CD8+ T cells, and NK cells—into 20 distinct clusters.
   (Figure [90] 3A,B; Figure [91]S3A, Supporting Information). CD8+ T cell
   subsets include the effector T cells (c0_Effector_GZMK), two types of
   mucosal‐associated invariant T (MAIT) cells (c2_MAIT_SLC4A10 and
   c9_MAIT_CCL20), one naïve subset (c4_Naive_NELL2) and two natural
   killer cell‐like subsets(c2_NKlike_XCL1 and c1_NKlike_GNLY), one
   interferon responded (IFNR) T cells(c8_IFNR_IFIT1), and two types of
   exhausted T cells (c3 and c6) and a proliferating subset
   (c7_Prolif_MKI67). CD4+ T cell (six clusters) and NK cell (four
   clusters) were also subclustered according to their canonical markers
   (Figure [92]S3A, Supporting Information). Our findings indicate that
   both CD8+ and CD4+ naïve T cells are notably enriched in the
   leading‐edge area of the tumor. In contrast, Treg cells and exhausted
   CD8+ T cells exhibit a propensity for localization within the
   tumor‐core region (Figure [93]3C). Upon scrutinizing the signature
   genes as previously characterized,^[ [94]^15 ^] we discerned that CD8+
   T cells predominantly exhibit a naïve phenotype, with diminished
   cytotoxic potential and a low level of exhaustion in the L‐side area
   (Figure [95]3D). These evaluated CD8+ T cells did not include CD8+ MAIT
   cells due to the different anti‐tumor modes of these two types of CD8+
   T cells. Pseudotime analysis constructed the trajectory of CD8+ T cell
   maturation from a naive state (Figure [96]S3B–D, Supporting
   Information). We observed that the majority of CD8+ T cells from the
   leading‐edge area were positioned at the onset of pseudo‐time, whereas
   those from the tumor‐core area were predominantly found at the terminus
   of pseudo‐time. This distribution suggests distinct states of CD8+ T
   cells across different tumor regions (Figure [97]3E). The results of
   mIHC also observed that there were a few naïve CD8+ T cells i) in the
   tumor side of the leading‐edge area, and CD8+ T cells were scarce ii),
   with most of them located on the non‐tumor side iii, iv)
   (Figure [98]3F). Unlike other classic CD8+ T cells, which rely on the
   cytotoxic molecule Granzyme K (GZMK), CD8+ MAIT cells primarily exert
   their anti‐tumor effects through the non‐specific molecules, tumor
   necrosis factor (TNF) (Figure [99]3G). We also noticed a tendency for
   CD8+ MAIT cells to be enriched in the tumor side of the leading‐edge
   area, although the difference was not statistically significant
   (Figure [100]3H). ST and mIHC results proved that CD8+T cells, CD4+T
   cells, and NK cells are mostly located on the non‐tumor side of the
   frontier region, while MAIT infiltrates into the tumor side
   (Figure [101]3I,J), suggesting the initial anti‐tumor role of MAIT in
   the leading‐edge area. The expression of MAIT effector molecules (TNF)
   and GZMB, the marker of cytotoxic CD8+T cells (Figure [102]S3E,
   Supporting Information), as well as the signature of naïve, cytotoxic,
   and exhausted activities (Figure [103]3K), also prove the above
   findings. However, for the entire CD8+ cell subpopulation, the
   cytotoxic effect of MAIT is not outstanding (Figure [104]S3F,
   Supporting Information), unlike those effector (c0), exhausted (c3), or
   NK‐like CD8+ T cells (c1 and c5), and the occurrence of MAIT cells
   suggest they serve as the vanguard of anti‐tumor CD8+ cells in this
   region. Meanwhile, we noticed that CD8+ MAIT cells exhibit stronger
   expression of anti‐tumor molecules (IFNG and TNF) in the normal liver
   tissue and exhibit a gradual decrease in anti‐tumor ability from normal
   liver tissue to the invasive tumor front, and to core tissue (Figure
   [105]S3G, Supporting Information), which might be associated with the
   metabolic change in different regions, noticing the elevated levels of
   hypoxia and glycolysis activities (Figure [106]S3H, Supporting
   Information). While in tumor‐core tissues, canonical CD8+T cells with
   GZMK/B/A and exhausted molecules (TOX and PDCD1) dominate (Figure
   [107]S3I, Supporting Information), which is completely different from
   the naïve and low‐cytotoxic/exhausted states of CD8+T cells in the
   leading‐edge area.

Figure 3.

   Figure 3
   [108]Open in a new tab

   Naive CD8+ T cells dominate and MAIT cells infiltrate in the
   leading‐edge area. A,B) UMAP plots show the major T/NK cluster (A) and
   their cell subsets (B). C) The distribution preference of CD8+ T cell
   subpopulations at different regions. D) The naïve, cytotoxic, and
   exhausted score of CD8+ T cell (CD8+ MAIT excluded) at different
   positions. E) Pseudo‐time trajectory analysis examines the evolution of
   CD8+ T cells at various pseudo‐time points and positions (top),
   alongside their distribution across the pseudo‐time spectrum (bottom).
   F) The multicolor immunofluorescence images show that naive CD8+ T
   cells (CD8+, TCF+) were detected in the tumor side of the leading‐edge
   zone, while the majority of CD8+ T cells are located in the non‐tumor
   area. G) Dot plot shows the expression of effector markers across CD8+
   cell subsets. H) The distribution of MAIT cells in different regions.
   I) Spatial dot plots depict the arrangement of CD8+ T cells, MAIT
   cells, CD4+ T cells, and NK cells within the tumor's leading‐edge
   region. J) The mIHC images confirmed that only a small number of MAIT
   cells and CD4+ T cells existed in the tumor side of the leading‐edge
   area, while the majority of CD4+ T cells, NK cells, and cytotoxic CD8+
   T cells were on the non‐tumor side. K) The spatial dot plots show the
   distribution of naïve‐, cytotoxic‐, and exhausted‐enriched spots
   calculated by the “addmodulescore” algorithm. *p < 0.05, **p < 0.01,
   ***p < 0.001. No significant difference (n.s.).

   It is worth noting that this phenomenon of naive cells accumulating in
   the tumor‐normal interface not only occurs in CD4+ and CD8+T cells but
   also in B cells (Figure [109]S4A,B, Supporting Information). We found
   that naïve (B_c03_naive) and memory B cells (B_c02_memory) enriched in
   the leading‐edge region, while tumor‐core tissue is more abundant with
   B cells that produce antibodies (B_c01_IGHG1) (Figure [110]S4C,
   Supporting Information). Surprisingly, Breg‐related genes were found to
   be enriched in the non‐tumoral tissue (N), with a small portion
   enriched in the L area (Figure [111]S4D, Supporting Information). This
   may be associated with the distribution of a greater number of tertiary
   lymphoid structures (TLS) in the non‐tumoral region (data not shown).
   And B cells infiltrated into tumor‐core tissues acquired other
   phenotypes like angiogenesis, or colonization ability due to
   environment shift (Figure [112]S4E, Supporting Information).

   Together, these above findings suggest the immature state of
   lymphocytes in the leading‐edge area and the behavior of lymphocytes
   against tumor cells may be more intense in the tumor‐core area, whether
   using antibodies or cytotoxic molecules.

2.4. The Antigen‐Presenting Capacity of APCs in the Leading‐Edge Area is
Impaired

   To delve into the function of myeloid cells at the leading‐edge area,
   we applied dimensionality reduction and unsupervised clustering to all
   myeloid cells (Figure [113] 4A). This chapter mainly discussed two main
   subclusters, including dendritic cells (DCs) and neutrophils. Among the
   DCs, we distinguished cDC1 (XCR1 and CLEC9A), cDC2 (CLEC10A, and
   FCER1A), LAMP3 DC (LAMP3 and CCR7), and pDC (IRF7 and IL3RA)
   (Figure [114]4B). We found that cDC2 was mostly enriched in the
   leading‐edge and tumor‐core area, and cDC1, however, shows the opposite
   condition (Figure [115]4C). DC is one of the classic antigen‐presenting
   cells (APCs). However, we found that the antigen presentation ability
   and co‐simulation activity of DCs were significantly reduced in the
   leading‐edge (L) area compared to the tumor‐core (T) region
   (Figure [116]4D,E). Our above results indicate that CD8+ T cells in the
   leading‐edge area exhibit a naïve, low‐cytotoxic phenotype. It is worth
   noting that naïve CD8+ T cell maturation requires antigen presentation
   with cDC1 and their proximity in location to occur a co‐stimulation
   event.^[ [117]^16 ^] However, in our ICC samples, cDC1 rarely
   infiltrates the tumor side of the leading‐edge area and mostly
   aggregates in the tertiary lymphoid structures (TLS) of the non‐tumor
   area, so as CD8+ naïve T cells (Figure [118]4F; Figure [119]S5A,
   Supporting Information). The results of mIHC also proved that there
   were only a few CD8+ T cells on the tumor side of the L area, and
   almost no cDC1 cells were found ii), compared to the tumor‐core area i)
   and the non‐tumor side iii) of the L area (Figure [120]4G). The above
   finding suggesting the immaturity of CD8+ T cells might be associated
   with the incompetent antigen presentation of DCs.

Figure 4.

   Figure 4
   [121]Open in a new tab

   The antigen‐presenting capacity of APCs in the leading‐edge area is
   impaired. A) The UMAP visualization of myeloid cells subsets (left) and
   subsets of DC and neutrophils (right). B) The circular plot delineates
   the expression profiles of signature genes across different DC subsets.
   C) Distribution of DC subsets at different locations. D) Heatmap shows
   the relative levels of antigen‐presentation‐related genes at different
   regions. E) The comparison of three different biological processes of
   DC at different locations. F) The spatial dot plots show the relative
   distribution of cDC1, CD8+ naive T cell, and TLS and the levels of
   their colocalization signal. G) The mIHC images revealed the quantity
   and positional relationship of CD8+ T cells and cDC1 in the leading
   edge and core area of the ICC tumor. H) The results of GSEA enrichment
   of gene sets in all neutrophil subsets. I) The circular plot
   illustrates the gene expression levels associated with antigen
   presentation processes, categorized by location. *p < 0.05, **p < 0.01,
   ***p < 0.001. No significant difference (n.s.).

   Despite DCs, our data also identified 4 clusters of neutrophil subsets
   (Figure [122]4A; Figure [123]S5B, Supporting Information). Neutrophils
   of cluster 3 (Neu_C3_VEGFA, CD74+/HLA‐DRA+) displayed a mixed phenotype
   with angiogenesis, lipid metabolism‐related, and antigen‐presentation
   characteristics and they enriched in both the leading‐edge and
   tumor‐core region (Figure [124]S5B–D, Supporting Information;
   Figure [125]4H). However, whether in the major neutrophil group or the
   subgroup (Neu_C3_VEGFA), their antigen‐presenting function in the
   leading‐edge area seems to be deficient (Figure [126]4I; Figure
   [127]S5E, Supporting Information). This is consistent with the
   phenomenon observed on DCs. It is worth noting that this subpopulation
   (Neu_C3_VEGFA) may represent a potentially harmful subgroup, similar to
   the SPP1+ macrophages that we will discuss next. Enrichment analysis of
   its genesets in bulk RNA‐seq data of FDU‐ICC cohort suggests that it is
   a high‐risk factor for patient mortality (Figure [128]S5F, Supporting
   Information). We also noticed that the interferon‐responded and cell
   stress‐related cluster (Neu_c2_IFIT1) (Figure [129]4H) is mostly
   enriched in the leading‐edge area (Figure [130]S5C, Supporting
   Information). Unlike cluster 3 (Neu_c3_VEGFA), transcription factor
   regulons enriched in c2_IFIT1 neutrophils were mostly cell
   stress‐related (IRF9, IRF7, and STAT2) (Figure [131]S5G, Supporting
   Information), this may be related to the stress state generated by this
   subgroup after receiving interferon signals.

2.5. Pro‐Angiogenic SPP1+ Macrophages are the Main Macrophage Subsets that
Infiltrate into the Tumor Side of the Leading‐Edge Area

   Macrophages are also an important component of anti‐tumor immunity, and
   their functions in the immune microenvironment are more versatile.
   Antigen presentation ability is also one of their abilities. However,
   in our ICC cohort, like many other myeloid cells, their antigen
   presentation and processing ability are also weaker in the leading edge
   compared to the core area (Figure [132]S6A, Supporting Information). To
   further investigate the regional heterogeneity of macrophages, we
   further refined macrophage subgroups (Figure [133] 5A,B), including the
   previously reported FOLR2+ macrophages (c4),^[ [134]^17 , [135]^18 ^]
   the transitional state IL1B+ subsets (c3), the cell stress‐related JUN+
   subsets (c0), and the previously considered harmful SPP1+ subgroup
   (c6).^[ [136]^19 ^] In terms of distribution, SPP1+ cluster (c6),
   APOC1+ cluster (c2), and JUN+ cluster (c0) show gradual enrichment from
   the invasive tumor front to the core region, while those previously
   deemed “good” subsets, like FOLR2+ (c4) and IL1B+ (c3) cluster,
   preferentially accumulated in the non‐tumor area (Figure [137]5C). We
   found that overall, macrophages have the worst phagocytic ability in
   the frontier area of invasion, while macrophages in the core area of
   the ICC tumor exhibit a hypoxic phenotype, which enriched in pathways
   such as hypoxia and glycolysis, consistent with the phenomenon observed
   in tumor cells (Figure [138]5D). Among these subsets of macrophages,
   Mac_c06_SPP1 (SPP1+, PLIN2+) exhibits a distinct malignant phenotype,
   manifested in its high expression of the degradation of the
   extracellular matrix (ECM)‐related genes (MMP14, MMP12, and MMP9),
   enhanced ability to promote angiogenesis, and weakened ability to
   phagocytose tumor cells (Figure [139]5E,F). We inferred developmental
   trajectories of macrophages in these subsets (Figure [140]5G; Figure
   [141]S6B, Supporting Information) and found that FOLR2+ macrophages(c4)
   gradually developed toward SPP1+ macrophages (c6) and finally to MKI67+
   subsets (c5) (Figure [142]5H), accompanying the faded ability of
   phagocytosis and enhanced fatty acid metabolism, hypoxia‐ and
   glycolysis‐related process (Figure [143]5I). A diverse array of TFs was
   observed to be activated across the varied macrophage subpopulations.
   Notably, TFs associated with fatty acid metabolism, particularly PPARG,
   were selectively activated in the SPP1+ macrophages (Figure [144]S6C,
   Supporting Information). This activation pattern implies a connection
   between the emergence of the SPP1+ subset and the metabolic pathways
   centered on fatty acid processing.

Figure 5.

   Figure 5
   [145]Open in a new tab

   Pro‐angiogenic SPP1+ macrophages are the main macrophage subsets that
   infiltrate into the tumor side of the leading‐edge area. A) The UMAP
   plot delineates distinct subsets of macrophages, with each cell type
   demarcated by a unique color. B) The circular plot reveals the marker
   gene expression of each macrophage subset. C) The preferential
   distribution patterns of macrophage subpopulations across distinct
   anatomical regions. D) The comparison of nine different biological
   processes of macrophages among different regions. E) The expression of
   specific functional genes among different macrophage subsets. F) The
   comparison of phagocytic and angiogenic functions among macrophage
   subpopulations. G,H) Pseudo‐time analysis of macrophages (G) and
   distribution of various macrophage subpopulations (H, left) and
   macrophages at different locations (H, right) over pseudo‐time. I) The
   changes in five different cellular processes of macrophages over
   pseudo‐time. J) Spatial dot plot shows the distribution of
   SPP1+macrophage, fibroblast, and endothelial and corresponding
   tricellular colocalization. K) The mIHC results prove the presence of
   pro‐angiogenic SPP1+ macrophages (CD68+ VEGF+) within the Triad
   structure. *p < 0.05, **p < 0.01, ***p < 0.001. No significant
   difference (n.s.).

   To better depict the distribution of each macrophage subset, we
   performed spatial transcriptome projection on the markers of each
   subset (Figure [146]S6D, Supporting Information), and we noticed an
   irony phenomenon that the FOLR2+ subgroup with the intact phagocytic
   function (Figure [147]5F) was mainly located in the non‐tumor side of
   the invasive tumor front (Figure [148]S6E, Supporting Information),
   while the SPP1+ subgroup, which was the worst of that function, was
   mainly distributed in the tumor side, more specifically in the stromal
   region, composed of endothelial cells and fibroblasts
   (Figure [149]5J,K; Figure [150]S6F, Supporting Information), which we
   termed “triad structure.” As previously described, this triad structure
   forms an immune barrier that hinders T cells from infiltrating into the
   tumor side in HCC and CRC,^[ [151]^14 ^] and this phenomenon was also
   seen in ICC tumors (Figure [152]3I,J; Figure [153]S6G, Supporting
   Information). In addition to the distribution of this structure at the
   tumor‐normal interface, we also found that this structure extends to
   the tumor side, scattering around the tumor niches (Figure [154]5K).
   Considering the proximity relationship between the SPP1+ subgroup and
   endothelial cells, through ligand‐receptor interaction analysis, which
   was based on the “Cellchat” algorithm, we found that it has a
   significant pro‐angiogenic function through VEGFA‐VEGFR1 interaction,
   and this phenomenon is more evident in the leading‐edge area (Figure
   [155]S6H,I, Supporting Information).

2.6. POSTN+ FAP+ CAFs were the Major Fibroblast Subset that Composed the
Triad Structure

   Our above results indicate that SPP1+ macrophage subsets (Mac_c6_SPP1),
   endothelial cells, and fibroblasts have a co‐localization relationship
   and form a triad structure. We have not yet elucidated the functional
   characteristics of fibroblasts, since it was an important component of
   this structure. We then explored the fibroblasts of three
   aforementioned regions in our ICC cohort and identified 4 subclusters
   (Figure [156] 6A). The c03_MYH11 cluster was enriched with smooth
   muscle‐related genes, including MYH11 and TAGLN, suggesting it is a
   myofibroblast identity. The c02_RGS5 subset demonstrated heightened
   expression of pericyte markers, such as MCAM and RGS5. The c04_CXCL12
   subgroup displayed characteristics of both antigen‐presenting
   cancer‐associated fibroblasts (apCAFs), marked by CD74 and HLA‐DRA, and
   immune‐related cancer‐associated fibroblasts (iCAFs), indicated by
   CXCL14 and CXCL12. This duality suggests that the c04_CXCL12 subgroup
   could be an immune‐modulatory subset (Figure [157]S7A, Supporting
   Information). Meanwhile, the c01_CST1 subset was distinguished by
   elevated levels of FAP and POSTN expression, indicative of their
   identity as quintessential CAFs (Figure [158]6B). Notably, this FAP+
   POSTN+ CAF(c01_CST1) is significantly enriched in the tumor side, while
   myofibroblasts (c03_MYH11) show less proportion in this area
   (Figure [159]6C). The results of ST also indicate that the marker genes
   of c01_CST1 subset, such as POSTN and FAP, are mainly expressed in the
   tumor side of the leading‐edge area, and so did the signature of this
   subset, which is composed of highly expressed genes calculated by
   FindAllMarkers algorithm, while the myofibroblast (c03_MYH11) subsets
   are mainly distributed in the non‐tumor side (Figure [160]6D). Our mIHC
   results also demonstrated that the fibroblast marker, Collagen I, is
   distributed on both sides of the L area, while POSTN+ CAFs are found
   only on the tumor side, corroborating our findings from single‐cell and
   spatial transcriptomics studies (Figure [161]6E). Among these four
   fibroblast subsets, a notable enrichment of gene sets linked to
   oxidative phosphorylation was identified within the myofibroblast
   subset (c03_MYH11), which is understandable, considering this subset
   mainly in the perivascular region. While the POSTN+FAP+ CAFs(c01_CST1)
   show enrichment in pathways like glycolysis, unfolded protein response,
   TGF‐beta signaling, and angiogenesis, suggesting that the environment
   in which it exists is related to hypoxia, and resulting corresponding
   countermeasures accordingly (Figure [162]6F). Together, the above
   findings show that POSTN+FAP+ CAFs were the major fibroblast subsets
   that composed that triad structure, not the myofibroblast or
   immune‐related subsets.

Figure 6.

   Figure 6
   [163]Open in a new tab

   POSTN+ FAP+ CAFs were the major fibroblast subset that composed that
   triad structure. A) The UMAP visualization on the left delineates the
   distinct fibroblast subpopulations, while the corresponding plot on the
   right maps their distribution. B) The markers of CAFs in the c01
   subgroup are highly expressed. C) The distribution of fibroblast
   subsets in different regions. D) The signature score of the c01 cluster
   and its markers enriched in the tumor side of the leading‐edge area,
   with the c03 cluster mainly located at the non‐tumor side. E) Collagen
   I‐producing fibroblasts are present on both sides of the leading‐edge
   area, while POSTN+ fibroblasts appear only on the tumor side. F)
   Analysis of pathway activity, quantified through the Gene Set Variation
   Analysis (GSVA), reveals comparative patterns among the diverse
   subtypes of fibroblasts, highlighting distinctive functional
   signatures. *p < 0.05, **p < 0.01, ***p < 0.001. No significant
   difference (n.s.).

2.7. Cellular Crosstalk Between Tumor Cells and Immune Cells Shapes the
Unique TME of the Invasive Tumor Front

   To understand the cellular interactions that occurred in the invasive
   frontier region of ICC tumors. We first focused on tumor cells. Using
   the “cellchat” algorithm, we calculated the cellular interactions using
   tumor cells as receptors. We found that most of the receptors that
   interact with tumor cells are CD44 and ITGA families (Figure [164] 7A),
   which are related to the processes of tumor cell proliferation and
   invasive movement. Given the high proliferation of tumor cells in this
   area, we analyzed the cells that provide this ligand and found that
   SPP1+ macrophages (Mac_c6_SPP1) interact with tumor cells via the
   SPP1‐CD44 axis (Figure [165]7B), and this interaction was more intense
   at the leading edge (Figure [166]S8A, Supporting Information). Prior
   research has established that the engagement of the SPP1‐CD44 axis is
   instrumental in fostering the stem‐like properties of tumor cells,
   which in turn enhances their proliferative capacity and induces
   epithelial‐mesenchymal transition (EMT) phenotypes (Figure [167]7C,
   top).^[ [168]^20 ^] This is consistent with the aforementioned
   characteristics of ICC tumor cells we have found in this region.
   Furthermore, our above investigations have identified that
   proliferative tumor cells predominantly reside in close proximity to
   the triad structure, which comprises SPP1‐expressing macrophages,
   endothelial cells, and POSTN+FAP+ fibroblasts (Figure [169]7C, middle).
   To elucidate the spatial relationship between SPP1+ macrophages and
   proliferating tumor cells (MKI67+), we screened for SPP1+
   macrophage‐enriched spots on the ST slide, then calculated the
   surrounding signals of proliferative tumor cells, and finally
   determined the correlation between the two. The results indicate that
   the enriched spots of SPP1+ macrophage on the spatial transcriptome are
   indeed accompanied by enhanced proliferative tumor signals, and their
   distribution is spatially positively correlated, although weak
   (Figure [170]7C, bottom). We also found from the GEO database that
   similar experiments have been conducted, where co‐culturing tumor cells
   alone with monocyte‐derived macrophages (MDM, the initial state of
   macrophage, like FOLR2+ macrophage in our study) does not promote tumor
   growth. However, if exogenous human SPP1 protein is added, the
   expression of proliferation‐related genes (UBE2C, E2F1) in the
   co‐cultured tumor cells increases (Figure [171]S8B, Supporting
   Information). The bulk RNA‐seq data from the ICC cohort at Fudan
   University corroborate the positive correlation between the enrichment
   of the SPP1+ macrophage signature and the elevated levels of both the
   tumor stemness signature and CD44 expression (Figure [172]S8C,D,
   Supporting Information). The proximity relationship between SPP1+
   macrophages and proliferating tumor cells has also been confirmed by
   mIHC technology (Figure [173]S8E, Supporting Information). Notably, in
   ICC tumors, higher infiltrations of SPP1+ macrophage would incur a poor
   prognosis (Figure [174]S8F, Supporting Information). Meanwhile, our
   above findings underscore the pivotal role of hypoxia in the genesis of
   SPP1+ macrophages (Figure [175]5I). Considering the distribution of
   relevant cells near SPP1+ macrophages, this hypoxia might be mostly
   caused by proliferating tumor cells due to their high proliferation and
   high oxygen‐consumption behaviors in the leading‐edge area
   (Figure [176]7D). Consequently, originally a good subgroup (FOLR2+
   macrophage) was educated by the environment cellularly or metabolically
   and transformed to SPP1+ subsets, accompanied by fade‐away phagocytic
   ability, enhanced angiogenesis and the occurrence of bypass energy
   metabolism‐related processes (fatty acid oxidation and glycolysis)
   (Figure [177]7D). To confirm the scenario inferred by the results from
   single‐cell and spatial transcriptomics, we also performed in vitro
   experiment using the ICC cell line RBE. We observed that ICC tumor
   cells exhibited enhanced proliferation ability when stimulated with
   SPP1 protein and in the presence of fetal serum (mimicking the
   condition of tumor cells near blood vessels) (Figure [178]S8G–J,
   Supporting Information). Additionally, after co‐culturing tumor cells
   with macrophages under limited oxygen conditions, we did observe an
   increase in VEGF expression in the macrophages (Figure [179]S8K–M,
   Supporting Information).

Figure 7.

   Figure 7
   [180]Open in a new tab

   Cellular crosstalk between tumor cells and immune cells shapes the
   unique TME of the invasive tumor front. A) The L‐R contribution of
   tumor cells in the leading‐edge area. B) The CD44 receptor on tumor
   cells receives ligand sources from other cell subtypes. Circle
   dimensions correspond to the P values, and the likelihood of
   interaction is represented by color intensity. C) Schematic diagram of
   SPP1‐CD44 interaction promoting tumor cell stemness and altering tumor
   biological behavior (top). The relative positional relationship between
   proliferative tumor cells and triad signals (indicates the special
   stromal area, composed of endothelial, fibroblast, and
   SPP1+macrophages) (middle), as well as the correlation between the
   enriched SPP1+macrophage spots and the signal intensity of surrounding
   proliferating tumor cells (bottom). D) An assessment of differential
   pathway activities across cell subtypes within the leading edge, as
   determined by Gene Set Variation Analysis (GSVA). The scores of
   pathways are scaled across cell types. E) The L‐R contribution of SPP1+
   macrophage in the leading‐edge region. F) The distribution of MAIT's
   signature on spatial slice, alongside the levels and spatial
   arrangement of ligand‐receptor interaction‐related genes (CSF1‐CSF1R,
   CCL5‐CCR1), as well as the co‐localization signaling intensity and
   distribution of MAIT with SPP1+macrophage. G) The specified
   interactions between SPP1+ macrophages and diverse cellular subtypes,
   mediated by L‐R pairs. H) The association between SPP1+ macrophage
   signature and MAIT cell signature within the bulk RNA‐sequencing
   dataset from Fudan University's ICC cohort. I) The mIHC images
   demonstrate the proximity of CD8+ MAIT cells to SPP1+ macrophages
   within the triad structure. J) Schematic diagram of intercellular
   interactions in the leading‐edge area.

   Then we continued to consider how SPP1+ macrophages were recruited into
   the TME. Hence, we used SPP1+ macrophages as receptor cells to identify
   the top‐ranked interacting molecules (Figure [181]7E). Observing that
   CD8+ MAITs, rather than other CD8+ T cells or NK cells, infiltrate the
   tumor's leading edge, we identified a potential mechanism wherein CD8+
   MAIT cells recruit SPP1+ macrophages into the TME via the CCL5‐CCR1 or
   CSF1‐CSF1R axis. Additionally, a co‐localization phenomenon with SPP1+
   macrophages was observed (Figure [182]7F,G). In the ICC bulk RNA‐seq
   cohort of Fudan University, there exists a positive correlation between
   two signatures, SPP1+ macrophages and CD8+ MAIT cells (Figure [183]7H).
   The results of mIHC also confirmed the presence of SPP1+ macrophages
   around CD8+ MAIT cells in the triad structure of the leading‐edge area,
   as well as other types of macrophages (Figure [184]7I). This
   recruitment is notably enhanced at the leading edge (Figure [185]S8N,
   Supporting Information). In addition, considering that SPP1+
   macrophages were later educated and then formed, we also found that
   CD8+ MAIT also has a strong recruitment effect on the initial good
   subgroup (FOLR2+ subgroup) (Figure [186]S8O, Supporting Information).

   Therefore, here we have discovered and depicted an immune environment
   soil favorable for ICC tumor growth that appears in the leading‐edge
   zone. Here, the classic MHC‐I molecule‐dependent CD8+T cells scarcely
   infiltrate the tumor side, and show a naïve, low‐cytotoxicity
   phenotype, while the CD8+ MAITs with weak cytotoxic ability infiltrate
   it. CD8+ MAIT cells recruit a good subset of macrophages (FOLR2+) to
   infiltrate into the tumor microenvironment but are transformed to
   low‐phagocytosis SPP1+ subsets due to the oxygen competition with
   proliferating tumor cells nearby. Thus, a unique triad structure is
   formed, composed of SPP1+ macrophages, POSTN+ fibroblasts, and
   endothelial cells. Within this structure, POSTN+ fibroblasts and SPP1+
   macrophages form a barrier that shields tumor cells from immune cell
   attacks. Endothelial cells provide nutrients to the tumor cells. The
   potential angiogenic effect of SPP1+ macrophage and its potential
   interaction with tumor cells (SPP1‐CD44) in the triad‐structure area
   leads to the formation of an environment favorable for ICC tumor growth
   in leading edge, thus forming a positive feedback loop
   (Figure [187]7J). It is worth noting that this phenomenon may not be
   limited to ICC tumors. In our subsequent studies, we observed a similar
   phenomenon in HCC (Figure [188]S9A, arrow spotted, Supporting
   Information) as well, suggesting that this triad structure may be
   conserved across different tumors. This structure could potentially
   support tumor growth and serve as a potential therapeutic target.

3. Discussion

   Our study, leveraging scRNA‐seq and ST technologies, has provided a
   comprehensive view of the characteristics of ICC tumor cells at the
   leading‐edge area and their orchestrated interactions with the triad
   structure that shape the immune microenvironment. Our findings
   underscore the enhanced proliferative capacity of tumor cells in the
   leading‐edge area, juxtaposed with a distribution that interfaces
   closely with stromal elements such as endothelial cells and POSTN+
   fibroblasts.

   Notably, the predominance of naïve CD8+ T cells and the infiltration of
   MAIT cells in this region suggest a unique immunological landscape that
   may influence tumor progression. It is noteworthy that in
   hepatocellular carcinoma (HCC), research by Wang and colleagues has
   identified an enrichment of double‐positive T cells (expressing both
   CD4+ and CD8+) in the leading‐edge zone. Specifically, this T cell
   subset displays hallmarks of activation and exhaustion, indicative of
   the robust anti‐tumor immune response occurring at the leading‐edge
   region of HCC.^[ [189]^21 ^] This observation contrasts with the
   scenario in ICC, where the leading‐edge zone appears to be
   characterized by an immune‐incompetent phenotype in our study. The
   naïve state of CD8+ T cells in the ICC leading‐edge area, characterized
   by low cytotoxicity and exhaustion, may be due to an impaired
   antigen‐presenting capacity of APCs, which is critical for T cell
   activation and differentiation. Our results also show the scarcity of
   cDC1 in the tumor side of the leading‐edge area. This observation is
   significant, as cDC1 is known to be crucial for initiating effective
   anti‐tumor immune responses.

   Furthermore, the presence of SPP1+ macrophages in the triad structure,
   alongside endothelial and POSTN+ FAP+ CAFs, suggests a role in
   promoting tumor development through pro‐angiogenic properties and
   activation of CD44 in tumor cells. The spatial distribution of these
   macrophages and their correlation with proliferating tumor cells
   indicate a spatially coordinated interaction that may facilitate a
   supportive microenvironment for tumor growth. Our study also highlights
   the heterogeneity of macrophage subsets within the tumor
   microenvironment, with SPP1+ macrophages exhibiting a distinct
   phenotype associated with poor phagocytic ability and enhanced
   angiogenic potential. The transition from FOLR2+ macrophages to SPP1+
   subsets may represent an adaptation to the microenvironmental
   stressors, such as hypoxia, which is prevalent in the leading‐edge area
   due to the high proliferative rate of tumor cells. It is worth noting
   that in liver cancer, researchers have proposed an onco‐fetal niche,
   which is similar to the triad structure we observed in ICC. The
   difference is that they have FOLR2+ tumor‐associated macrophage (TAM),
   PLVAP+ endothelial, and POSTN+CAF, and promote tumor EMT behavior in
   liver cancer.^[ [190]^22 ^] This may be related to the difference in
   blood supply richness between HCC and ICC tumors. Given that in
   clinical practice, although both occurred in the liver, we have clearly
   found that HCC has an out better blood supply than ICC, it is
   speculated that FOLR2+ TAM lacks this oxygen competition phenomenon
   with HCC tumor cells, without the phenotype transformation to SPP1+
   macrophages. It also suggests that the triad structure composed of
   macrophages, fibroblasts, and endothelial cells exhibits different
   forms and functions in different types of cancer.

   Despite the novel insights provided by our study, some limitations
   warrant consideration. The sample size is modest, which may constrain
   the generalizability of our findings. Additionally, while our study has
   concentrated on the interplay between tumor cells and finite immune
   cells, other cellular roles of TME, such as B cells and dendritic
   cells, may also play significant roles that require further
   investigation. Mast cells in our cohort may also have a potential
   recruitment effect on SPP1+ macrophages (Figure [191]7G; Figure
   [192]S8N, Supporting Information); however, due to their abundance, we
   cannot determine whether they play a major role, and it is possible
   that we have underestimated their function in tumor immunity. Future
   research directions should encompass the validation of our findings
   within a more extensive patient cohort to ascertain the universality of
   the observed patterns in ICC. Elucidating the intricate mechanisms
   underlying the crosstalk between tumor cells and immune cells,
   particularly the specific role of triad structure in tumor progression,
   will be critical. Finally, based on our discoveries, developing novel
   therapeutic strategies targeting SPP1+ macrophages or CD8+ MAIT cells
   could offer more effective treatment options for ICC patients.

   In conclusion, our study offers new perspectives on the characteristics
   of ICC tumor cells in the leading‐edge area and the complex interplay
   with the immune microenvironment, providing potential targets for the
   interception of interaction‐based therapies within ICC.

4. Experimental Section

Data Resources and Descriptions—Tissue Acquisition, Storage, and Sequencing

   Treatment‐naive ICC specimens, including adjacent normal liver tissue,
   the leading‐edge area (previously defined by Wang^[ [193]^21 ^]), and
   tumor core tissues, were collected from ICC patients with voluntary
   written consent, and with the sanction of the Peking University
   People's Hospital's medical ethics committee. These specimens were
   stored at Tissue Storage Solution (MACS, # 130‐100‐008), and then
   divided into three parts, which were used for single‐cell sequencing,
   spatial transcriptome sequencing, and paraffin embedding, respectively.
   A comprehensive compilation of patient details is presented in Table
   [194]S1 (Supporting Information).

Data Resources and Descriptions—scRNA Sequencing

   Fresh samples of adjacent normal liver, the tumor's leading edge, and
   core tissues were rinsed with 4 °C PBS and dissected into fragments.
   These fragments were immersed in a 10 mL EDTA‐enriched solution and
   agitated for an hour at a temperature of 37 °C. Subsequently, the
   tissues were treated with a 10 mL DTT‐enriched PBS solution (65 mm DTT,
   with 10% FBS added) for a duration of 15 min at the same temperature,
   with continuous agitation. The EDTA and DTT were eliminated by rinsing
   the tissues with PBS twice. The cells were then strained through a
   100‐micron filter, centrifuged, and rinsed twice with PBS. The
   resulting cell suspensions were prepared for scRNA‐seq, performed by
   LC‐Bio Technology (Hangzhou, China), following the standard operating
   procedure on the 10X Genomics Chromium platform.

Data Resources and Descriptions—Samples Processed for Spatial Transcriptomics

   Spatial transcriptomic libraries were crafted utilizing the 10x
   Genomics Visium Spatial Gene Expression Reagent Kit. The leading‐edge
   tissue samples were bisected, with the well‐preserved half being
   encased in OCT and preserved at −80 °C for subsequent analysis. The
   counterpart half was allocated for scRNA sequencing. Prior to executing
   the complete procedure, a tissue optimization test was conducted using
   the 10x Genomics Visium Spatial Tissue Optimization with cryosections
   from ICC patient tumors, accompanied by fluorescence signal
   visualization, which determined 24 min as the ideal permeabilization
   duration. Subsequently, tissue sections from a cohort of 5 ICC patients
   were prepared for spatial transcriptomic sequencing in accordance with
   the protocol provided by LC‐Bio Technology (Hangzhou, China).

Data Resources and Descriptions—Spatial Genomics Data Preparation and
Computational Processing

   The FASTQ files were aligned to the GRCh38 human genome using the Space
   Ranger (v2.1.0) software from 10x Genomics, yielding a count matrix as
   per genome annotations. Spatial transcriptomics samples underwent
   quality control through the Loupe Browser, with the selection of
   undamaged spots devoid of tissue folds. Further quality checks,
   including gene count, mitochondrial read percentage, and ribosomal
   fraction, were unnecessary as all spots met the criteria.

   Processed gene‐spot matrices were subsequently analyzed using the
   Seurat package in R (version 4.2.1). The “SCTtransform” function was
   applied for normalization across spots. Dimensionality reduction and
   clustering were conducted using principal components analysis (PCA),
   selecting the top 9–11 principal components. Signature scoring was
   executed with the AddModuleScore function in Seurat. The gene lists of
   particular signatures were generated from scRNA‐seq data and pathways
   from the “msigdbr” library, with details provided in Table [195]S2
   (Supporting Information). Spatial feature expression was visualized
   with Seurat's SpatialFeaturePlot function.

   For a tumor niche circled by stroma, the Loupe Browser 8 was used to
   select this area and then use the Concaveman algorithm to stratify this
   area into the onion‐like layer. Then map the intensity of the ICC
   proliferating signature of each layer onto it.

   To determine the spatial relationship between proliferating tumor cells
   and SPP1+ macrophages, single‐cell‐derived SPP1+ macrophage signatures
   were utilized and projected onto spatial transcriptional data. The
   average proliferative signal from tumor cells within the hexagonal
   zones surrounding areas of elevated SPP1+ macrophage signatures was
   then computed and analyzed each spot on the tumor side of the spatial
   transcriptome. By assessing the correlation between the signals of
   SPP1+ macrophages and those of proliferating tumor cells, the spatial
   interplay between these two cell types was deduced. This streamlined
   method efficiently elucidates the potential co‐distribution and
   interaction of SPP1+ macrophages with proliferating tumor cells.

Data Resources and Descriptions—Co‐Localization Signal Calculation

   For the assessment of co‐localization signals among two or three cell
   types, which were represented by genes, signatures, or distinct signal
   types, a tiered system was employed that categorizes signal intensities
   into quartiles (0–3). This methodological approach allows us to
   quantify the co‐localization by multiplying the signal intensities at
   identical spatial locations. Consequently, the co‐localization signal
   was minimal when only a single signal was detected. Conversely, the
   co‐localization signal reaches its peak when both signals were
   concurrently present and exhibit high intensities, indicative of a
   robust co‐occurrence. This analytical framework was equally applicable
   for scenarios involving tricellular co‐localization, enhancing the
   ability to discern complex cellular interactions within the tissue
   microenvironment.

Data Resources and Descriptions—scRNA‐Seq Data Processing

   The raw data was executed with the “Seurat” package in R (version 4.2.1
   and 4.3.1). Cells exceeding 25% mitochondrial gene content were
   filtered out, along with genes detected in less than 20 cells and cells
   expressing less than 500 genes. Seurat's “FindVariableFeatures”
   function pinpointed the genes with the highest variability.
   Dimensionality reduction was achieved through Principal Component
   Analysis (PCA), and batch effects were mitigated using Seurat's CCA
   (canonical correlation analysis) method. UMAP clustering revealed six
   primary clusters, with cell type annotations determined by integrating
   known and differentially expressed marker genes.

   Signature scores were derived by averaging the scaled and centered
   expression values of genes within each signature, as calculated by
   Seurat's “AddModuleScore” function. The cell type signature scores
   utilized are detailed in Table [196]S2 (Supporting Information).

Hematoxylin and Eosin (H&E) Staining

   Tissues were preserved in 10% neutral‐buffered formalin and subjected
   to a series of ethanol treatments for dehydration, followed by paraffin
   embedding and sectioning into 5‐µm slices using standard histological
   techniques. H&E staining was applied to the thin sections for
   microscopic examination.

Multiplex Immunohistochemistry (mIHC), Immunohistochemistry (IHC), and
Quantitative Analysis

   The process of mIHC was carried out following the Opal Kit guidelines
   from PerkinElmer. The list of primary antibodies and their dilutions is
   provided in Table [197]S3 (Supporting Information). The procedure began
   with deparaffinization of the sections in xylene and subsequent
   rehydration through a graded ethanol series. Microwave antigen
   retrieval was performed in a heated citric acid buffer at pH 6.0 for 10
   min. To quench endogenous peroxidase activity, 3% hydrogen peroxide was
   applied for 30 min, followed by a 30‐min blockade of nonspecific
   binding sites with goat serum. Incubation with primary antibodies took
   place for 1 h in a humidified box at room temperature, succeeded by the
   application of secondary antibodies conjugated to horseradish
   peroxidase. Fluorescein TSA Plus was used for target visualization at a
   dilution of 1:200. Between staining steps, slides were subjected to a
   second round of microwave antigen retrieval in a heated citric acid
   buffer to eliminate excess antibodies. Nuclear staining was completed
   with DAPI, and the sections were mounted using a fluorescent mounting
   medium. The slices were then scanned and imaged utilizing the
   PerkinElmer Vectra3 imaging system.

   Immunohistochemistry (IHC) was conducted in accordance with the methods
   detailed in previous research.^[ [198]^23 ^] Pathologists, who were
   unaware of the patient outcomes, evaluated the immunoreactivity of
   proliferative markers (Ki‐67) through image analysis, focusing on both
   staining density and intensity. The scoring system was defined with
   specific criteria: a score of 1 was given for staining that was either
   0–25% dense or showed negative intensity, 2 for 26–50% density or weak
   intensity, 3 for 51–75% density or moderate intensity, and 4 for
   76–100% density or strong intensity. The overall semi‐quantitative
   score was derived by multiplying the density score with the intensity
   score, yielding a possible score range between 1 and 16.

Developmental Trajectory Inference

   Developmental trajectory analysis was performed using the Monocle2
   (version 2.12.0) algorithm to explore the potential lineage progression
   within CD8+ T cells, excluding the CD8+ MAIT subset, and macrophages.
   For this analysis, the 2000 highly variable genes, as determined by the
   differentialGeneTest algorithm, were chosen to build the developmental
   trajectory map.

Transcription Factor Regulon Analysis

   To expedite the computational process, the examination of regulatory
   networks and regulon activities was facilitated by the pySCENIC tool.^[
   [199]^24 ^] The expression matrix with cell type information was first
   transformed into “*.loom” profile. And then use Python (3.7.10) to run
   “pyscenic grn,” “pyscenic ctx,” and “pyscenic aucell” sequentially. The
   differential transcription factors among cell types were visualized in
   the R platform.

Cancer Hallmarks and GSEA Analyses

   Gene sets for analysis were sourced from the “msigdbr” library
   ([200]http://software.broadinstitute.org/gsea/msigdb/). The scoring
   methodology for gene signatures adhered to a previously reported
   approach. Essentially, each cell's gene signature score was derived
   from averaging the scaled expressions of the genes within the
   signature. To detect significant shifts in gene expression between
   leading‐edge and core tumor cells, the “FindAllMarkers” algorithm
   (Seurat) was applied.

Survival Analysis

   Survival analysis was conducted utilizing the “survfit” function in R.
   The Cox proportional hazards model was employed to determine the Hazard
   Ratio (HR), with the 95% confidence interval (CI) being calculated. The
   Kaplan–Meier (K–M) method was applied to estimate survival functions
   using the survfit function. The ICC mRNA cohort from Fudan University
   was utilized to elucidate the correlation between SPP1+ macrophages and
   the characteristic genes or gene sets of related cells. This dataset
   can be accessed in the supplementary materials of the publication.^[
   [201]^25 ^] Patients were stratified into two cohorts based on the
   median expression levels of specific genes—such as CD44—or based on the
   calculated scores of certain gene signatures, like the SPP1+ macrophage
   or stemness signatures. The two‐sided log‐rank test was then applied to
   statistically compare the K–M survival curves between these groups.

InferCNV

   The R package inferCNV (version 1.14.2), accessible at its GitHub
   repository ([202]https://github.com/broadinstitute/inferCNV), was
   employed to distinguish malignant cells by deducing chromosomal copy
   number variations (CNVs) from gene expression profiles. Leveraging T/NK
   cells, B cells, and myeloid cells as representative comparators of
   normal cell populations, The epithelial with high CNVs and high
   epithelial markers were estimated to be malignant. A gene ordering file
   was crafted, grounded in the human GRCh38 genome assembly, which
   outlined the precise chromosomal coordinates—both the start and end
   positions—for each gene. This file was integral to the
   “gene_order_file” parameter required by the inferCNV algorithm. By
   inputting the raw count matrix and annotation file into inferCNV, a
   rigorous cutoff threshold of 0.1 was implemented to identify
   significant CNVs, thereby refining the precision of the analysis in the
   detection of tumor cells.

Cell–Cell Interactions Analysis

   Interactions between cells, as indicated by the expression patterns of
   established Ligand‐Receptor (L‐R) pairs across various cell types, were
   quantified using the CellChat tool (version 1.6.1), available on its
   GitHub repository ([203]https://github.com/sqjin/CellChat).^[ [204]^26
   ^] Briefly, CellChat was utilized with gene expression data to map
   overexpressed ligands and receptors onto a protein‐protein interaction
   network, identifying significant ligand‐receptor interactions.
   Probability assessment and permutation testing within CellChat were
   applied to deduce significant cell–cell communications. The resulting
   interaction networks were visualized using bubble plots for clarity. In
   this study, SPP1+ macrophages and tumor cells were set as receptors to
   calculate likely interacting ligands with the
   “netAnalysis_contribution” function, followed by “netVisual_circle” to
   trace the origin of these ligands, succinctly delineating the cellular
   communication landscape.

Cell Culture and Reagents

   The THP‐1, RBE cells, were purchased from Meisen Cell (Zhejiang,
   China). Cells were cultured in RPMI 1640 medium (Gibco, USA) with 10%
   fetal bovine serum (FBS) (Procell, Wuhan, China) at 37 °C in a
   humidified atmosphere with 5% CO2. For macrophage generation, 3 × 10^5
   THP‐1 cells were seeded in six‐well dish treated with 100 ng PMA
   (MedChemExpress, USA) for 72 h and polarized into macrophages
   (THP1‐M0).

   Macrophages (THP1‐M0) and RBE cell lines co‐cultivation was conducted
   using the non‐contact co‐culture transwell system (0.3 µm, Corning,
   USA). Place the pre‐cultured RBE cells in the upper chamber and combine
   it in a six‐well plate containing THP1‐M0. After that, seal the
   six‐well plate to create a limited oxygen environment. After 12 h of
   co‐culture, macrophages (THP1‐M0) were harvested for further analyses.

   For tumor cell stimulation assays, human recombinant SPP1 protein (Cat:
   [205]P02393, Solarbio, China) was added to RBE cells and cultured for
   24 h in the presence or absence of fetal bovine serum. After that,
   cells were harvested for further analyses.

Statistical Analysis

   All data underwent suitable statistical evaluations. For variables
   exhibiting normal distribution, differences between two groups were
   assessed using an unpaired two‐tailed t‐test, while non‐normally
   distributed variables were compared with the Mann–Whitney U test.
   Multiple group comparisons, involving three or more groups, were
   conducted using the Tukey post‐hoc test. Survival differences were
   examined with the Kaplan–Meier method complemented by the log‐rank
   test. GraphPad Prism version 8.4.2 was utilized for scoring
   immunohistochemical results and for visualizing data. A p‐value of less
   than 0.05 was set as the threshold for statistical significance (*p <
   0.05, **p < 0.01, ***p < 0.001).

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   L.Z., L.Z., and C.Q. contributed equally to the work. Z.J. and L.Z.
   conceived and designed the study. L.Z. analyzed the acquired data.
   Z.J., G.J., L.Z., and C.Q. supervised the integral analysis. Z.C.,
   M.D., H.J., C.Z., Z.J., and L.Y. contributed to data collection,
   curation, and interpretation. All authors contributed to the manuscript
   and reviewed it carefully before submission.

Code Availability

   Codes used in this study are available at GitHub
   ([206]https://github.com/TsuyinLee/Leading‐edge‐area‐in‐ICC).

Supporting information

   Supporting Information
   [207]ADVS-12-2412740-s001.docx^ (4.9MB, docx)

   Supporting Information
   [208]ADVS-12-2412740-s002.xlsx^ (23.9KB, xlsx)

Acknowledgements