Cancer-associated fibroblast-derived SEMA3C facilitates colorectal cancer liver metastasis via NRP2-mediated MAPK activation

Abstract

Liver metastasis remains the predominant cause of mortality in patients with colorectal cancer (CRC). Nevertheless, the mechanisms underlying the initiation of colorectal cancer liver metastasis remain poorly elucidated. During the metastatic process of CRC cells from the primary site to the liver, we performed time-resolved analyses and identified a subset of tumor cells spatially located in the primary tumor and temporally distributed in the early stages of liver metastasis. These cells were termed liver metastasis-initiating cells (LMICs). LMICs exhibit high stemness, low proliferation, active interaction with surrounding stromal components, and a close association with liver metastasis. Notably, we found significant interactions between cancer-associated fibroblasts (CAFs) and LMICs via the SEMA3C-NRP2 receptor-ligand pair. Further in vivo and in vitro experiments confirmed that CAF-secreted SEMA3C could bind to the NRP2 receptor, which activates the MAPK pathway and promotes colorectal cancer liver metastasis. Our findings suggest potential therapeutic strategies for the early prevention of colorectal cancer liver metastasis.

Keywords: colorectal cancer; liver metastasis; metastasis-initiating; single-cell.

Fig. 1.

Single-cell Analysis of CRC and Paired Liver Metastasis Reveals Tumor Heterogeneity. (A) Uniform manifold approximation and projection (UMAP) plot showing 11 cell clusters of primary CRC and liver metastatic patients. Cells from different clusters are marked by colors. (B) Dot plot showing the expression levels and percentages of classic marker genes across the 11 cell clusters. (C) Bar plot showing the percentages of cell types in primary CRC, liver metastasis, and peripheral blood mononuclear cells (PBMCs) samples. (D) UMAP plots showing the expression of marker genes (PTPRC, EPCAM, PECAM1, COL1A1, CD68, CD79A, CD3D, and NKG7), colored by gene expression. (E) Violin plots showing copy number variation (CNV) scores of six epithelial clusters (E1–E6). (F) UMAP plot of epithelial cells, colored by normal and malignant cells. The Right plot showing the distribution of CNV scores for normal and malignant epithelial cell cluster. (G) UMAP plot showing four malignant epithelial clusters (MC1–MC4) colored by their tissue of origin (primary CRC or liver metastasis). (H) UMAP plot showing four malignant epithelial clusters colored by CytoTRACE scores. The Right box plots compared CytoTRACE scores between the four malignant clusters, with TFF3+MC4 showing the highest differential potential. (I) Heatmap displaying the expression of stemness-related genes and AUCell scores for stem cell–related processes across the four malignant clusters. (J) Gene ontology (GO) enrichment analysis for each malignant cluster. (K) The semisupervised trajectory of malignant cells using Monocle, visualizingg pseudotime (Top), cell clusters (Middle), and tissue sources (Bottom). (L) Expression patterns of S100A2 and TFF3 along the pseudotime trajectory. (M) Heatmap showing the temporal expression of stemness and EMT-related genes along the pseudotime trajectory. (N) The spatial distribution of TFF3+MC4 cells in spatial transcriptomics data. (O) Heatmap showing the tissue distribution of TFF3+MC4 cells by Ro/e score. Ro/e >1, +++; 1 >or= Ro/e > 0.8, ++; 0.8 > or= Ro/e > 0.2, +; 0.2 >or= Ro/e > 0, ±; Ro/e = 0, −. (P) Kaplan–Meier survival curves showing overall survival in TCGA-CRC and GSE159216. (Q) Gene set enrichment analysis (GSEA) based on the genes ranked by the relationship with TFF3+MC4 score.

Fig. 2.

Identification of Liver Metastasis-Initialing Cells (LMICs) in Primary CRC Tissues. (A) UMAP plots showing the reclustering of TFF3+MC4 cells (Top) and VECTOR analysis results for the developmental sequence of TFF3+MC4 cells in CRC liver metastasis (Bottom). (B) Trajectory plot showing the timing sequence of tumor cell development colored by state (Top), cell types (middle top), sources (Middle down), and pseudotime (Bottom). (C) UMAP plot showing the LMICs, identified based on both VECTOR and Monocle algorithms. (D) UMAP plot showing the Slingshot pseudotime trajectory further validated the lineage relationship of the identified LMICs, positioning them at the beginning of the inferred developmental trajectory. (E) Heatmap of key genes across the LMICs. CDCA7 and LEFTY1 increasing and others like SEMA3F and SPHK1 decreasing as cells progress through the differentiation trajectory. (F) UMAP plots showing AUCell scores for EMT stem cell development and oxidative phosphorylation pathways. (G) GSEA results showing positive and negative correlations of LMIC score with pathway activities. (H) Scatter plots demonstrating correlations between LMIC score and stemness-related pathways. (I) Scatter plots showing the relationship between LMIC score and markers of epithelial (CDH1), mesenchymal (CDH2, VIM), and proliferation (MKI67). (J) Kaplan–Meier survival curves from the TCGA-CRC dataset showing that patients with higher LMIC score had worse overall survival compared to those with lower LMIC score. (K) COX analysis across multiple CRC datasets, showing that the LMIC score is associated with increased poor prognosis risk. (L) Heatmap of GO, KEGG, and hallmark pathway showing the biological characteristics of LMICs compared to other CRC cell populations. (M) Dot plot showing the expression of key genes (e.g., ECM1, FYN) in LMICs. (N) UMAP plots showing focal adhesion and ECM–receptor interaction pathway activities. (O) Bar plot showing ESTIMATE stromal and immune scores in TCGA-CRC. (P) Scatter plots showing significant correlations between LMIC score and focal adhesion (FAK), ECM, and cell adhesion molecule (CAM) pathway activities. (Q) Box plots showing increased expression of PTK2 (a key kinase in the FAK pathway) and VEGFA (a downstream target of FAK) in metastatic (M1) versus non-metastatic (M0) patients.

Fig. 3.

CAF–LMIC Interactions Promote Liver Metastasis via the SEMA3C–NRP2 Ligand–Receptor Axis. (A) Bar plot showing the number of cell–cell interactions between LMICs and various cells in TME via CellPhoneDB. LMIC exhibits the highest number of interactions with CAF. The heatmap on the Right highlights the interaction patterns between LMICs and different cell types, with LMIC–CAF interactions being the most prominent. (B) Spatial transcriptomics showing the spatial proximity of LMICs and CAFs. Spots colocalized with LMIC and CAFs, only CAFs, and only LMICs are shown in the different colors. (C) UMAP plot showing the expression of SEMA3C in CAFs. (D) Network plots illustrating cell–cell communication between CAFs and other cell types. SEMA3C+CAFs exhibits more interactions with other cell types than SEMA3C–CAFs and exhibits the strongest interactions with LMICs. (E) Function enrichment analysis showing pathways highly enriched in SEMA3C+ CAFs, including EMT, apoptosis, and cytokine–receptor interactions. (F) Scatter plot showing a positive correlation between SEMA3C+CAFs score and LMIC score in TCGA-CRC. (G) Kaplan–Meier survival curves showing that CRC patients with higher SEMA3C+CAFs score is associated with worse OS in TCGA-CRC. (H) Box plot showing the correlation between SEMA3C+CAF score and tumor stages in TCGA-CRC, indicating a higher score in patients with distant metastasis. (I) Box plot comparing SEMA3C+CAF score across different lymph node stages, indicating increased SEMA3C+CAFs activity in higher stages. (J) GSEA for genes highly correlated with SEMA3C+CAFs, showing enrichment for cytokine–receptor interaction, focal adhesion, and CAM pathways. (K) Box plots showing SEMA3C expression across CRC subtypes (CMS1-CMS4) and molecular subtypes (Desert, D; Fibrotic, F; Immune-Enriched, Non-Fibrotic, IE; Immune-Enriched, Fibrotic, IE/F). CMS4 and subtype F, both stromal-enriched subtypes, exhibit significantly higher SEMA3C expression. (L) Box plots comparing SEMA3C expression in metastatic and nonmetastatic samples from the GSE39084 and TCGA-CRC cohort, showing significantly higher expression in samples with distant metastasis. (M) Kaplan–Meier survival curve from GSE39084 showing that higher SEMA3C expression is associated with worse OS. (N) IHC images showing the SEMA3C expression in CRC tissue. Intense staining of SEMA3C is observed in CRC with liver metastasis (LM) compared to those without LM. The bar plot showing the quantitative analysis of IHC, revealing a significantly higher relative positive area of SEMA3C in CRC patients with LM compared to those without LM (P <0.05). Scale bars, black (100 μm) or red (50 μm). (O) mIHC assays showing the colocalization of α-SMA (CAF marker), SEMA3C, and PanCK (epithelial marker) in CRC samples with and without LM. (Scale bar, 100 μm.)

Fig. 4.

CAF-secreted SEMA3C Promotes Liver Metastasis and Poor Prognosis in CRC. (A) Transwell assays and wound healing assays in SW480 and CT26 cells only or treated with CAF–NC CM, and CAF–SEMA3C/CAF–Sema3c CM. The results showing that CAF–SEMA3C or CAF–Sema3c significantly promotes the invasion and migration ability of SW480 and CT26 cells compared with CAF–NC. Scale bars, white (100 μm) or black (200 μm). (B) Quantification of the invasion, migration, and wound healing assays in SW480 cells and CT26 cells only or treated with CM from CAF–NC or CAF–SEMA3C/CAF–Sema3c. (C) Western blotting assay of EMT-related proteins (ZEB1, N-cadherin, E-cadherin, and Vimentin) from SW480 only or treated with CAF–NC CM, CAF–SEMA3C CM or rhSEMA3C, and CT26 only or treated with CAF–NC CM, CAF–Sema3c CM or rmSEMA3C. (D) Tumor volume growth curves and tumor weight in subcutaneous tumor model treated with tumor cells only (n = 6), CAF–SEMA3C (n = 6), and CAF–NC (n = 6). Tumors in the CAF–SEMA3C group exhibit significantly larger volumes and weights than CAF–NC group. (E) Tumor volume growth curves and tumor weight in a patient-derived xenograft (PDX) model treated with rhSEMA3C (n = 4) versus control (n = 4). Mice treated with rhSEMA3C showing significantly larger tumor volumes and weights than controls. (F) Liver metastatic lesions and orthotopic primary tumors were collected from an orthotopic liver metastasis mouse model across three group: CT26 cells only (control, n = 6), CT26 coinjected with CAF–NC (n = 6), or CT26 coinjected with CAF–Sema3c (n = 6). Quantification of tumor weight showing a significantly larger tumor mass in the CAF–Sema3c group compared to CAF–NC, and CAF–NC compared to CT26 cells only. Quantification of the number of liver metastasis in three groups. (G) HE images of liver from mice injecting CT26 only or coinjecting CT26 cells with either CAF–NC or CAF–Sema3c. Scale bars, black (500 μm) or white (100 μm). (H) IHC staining for EMT-related protein (E-cadherin, N-cadherin, Vimentin, Snail, and Zeb1) in tumors from mice injecting CT26 cells only or coinjecting CT26 cells with either CAF–NC or CAF–Sema3c. (Scale bar, 100 μm.)

Fig. 5.

SEMA3C Mediates Liver Metastasis via Direct Interaction with NRP2. (A) UMAP plot showing NRP2 expression in LMICs. NRP2 is specifically expressed in LMICs, with overlapping distribution patterns between NRP2 and LMICs. (B) Correlation analysis showing strong positive correlations between NRP2 expression and LMIC score. (C) Box plot showing that higher NRP2 expression correlates with increased LMIC score in TCGA-CRC. (D) Spatial transcriptomics analysis depicting the colocalization of SEMA3C and NRP2. Ligand–receptor pairs are highlighted. (E) Scatter plot showing a positive correlation between SEMA3C and NRP2 expression in TCGA-CRC. (F) COIP assays confirming the physical interaction between SEMA3C and NRP2 in SW480 and CT26 cell lines. (G) Kaplan–Meier survival curves showing patients with higher NRP2 expression have significantly worse OS (Left) and disease-free survival (DFS) (Right). Box plots showing that higher NRP2 expression correlates with metastasis (Left), lymph node metastasis (Middle), and advanced tumor stages (Right). (H and I) GSEA showing that high SEMA3C expression is associated with activation of EMT and other metastasis-related pathways. (J) The scatter plot showing a positive correlation between NRP2 expression and EMT markers (CDH1, CDH2, VIM). NRP2 shows a strong positive correlation with mesenchymal markers (CDH2 and VIM), while the correlation with the epithelial marker (CDH1) is weak. (K) In vivo tumor growth assays in a subcutaneous mouse model. CAF–SEMA3C group developing larger tumors, and this effect is reversed by shNRP2. Tumor volume and weight are significantly reduced in mice injected with shNRP2 cells compared to controls. (L) Flowchart of orthotopic tumor model construction and subsequent processing. (M–O) In vivo orthotopic liver metastatic model using CT26 cells. Mice coinjecting with CT26 cells and CAF–Sema3c exhibiting increased tumor growth and liver metastasis, while shNRP2 reverses the enhanced tumor growth and liver metastasis. Images of orthotopic model and liver metastasis (M), quantality analysis (N), and HE (O) confirm that NRP2 knockout reduces liver metastasis. Scale bars, black (500 μm) or white (100 μm).

Fig. 6.

SEMA3C Promotes Liver Metastasis via NRP2-mediated MAPK activation. (A) Functional enrichment analysis of NRP2+ malignant cells showing significant activation of the MAPK signaling pathway and other tumor-related pathways, including regulation of actin cytoskeleton, focal adhesion, and ECM–receptor interactions. (B) UMAP plot showing the MAPK pathway activity score. Cells within the LMICs region exhibit significantly higher MAPK pathway activity. (C) Scatter plots showing positive correlations between MAPK pathway scores and NRP2 expression (Left) and MAPK pathway scores and EMT pathway activity (Right) in TCGA-CRC. (D) Schematic diagram of the experimental design. CT26 cells are coinjected with CAF–NC or CAF–Sema3c cells. After 6 wk of growth in an orthotopic mouse model, tumors are harvested for RNA sequencing. (E) KEGG pathway enrichment analysis of the transcriptomic data from CAF–Sema3c group, highlighting significant activation of tumor-related pathways, including MAPK, HIF-1, and fatty acid metabolism. (F) Heatmap showing the expression of key genes in the MAPK-p38 and MAPK-JNK pathways, with significant upregulation of MAP3K20 and other upstream regulators in CAF–SEMA3C group. (G) Scatter plots showing the correlation between SEMA3C expression and MAPK-p38 as well as MAPK-JNK pathway score. (H) IHC and mIHC analysis of p-p38 expression and colocalization with EPCAM in tumor tissues. mIHC of p-p38 (red) and EPCAM (green) in tumor tissues from CAF–Sema3c and CAF–NC groups. Colocalization of p-p38 and EPCAM indicates p38 activation specifically within epithelial tumor cells. Scale bars, black (100 μm) or white (50 μm). (I) Western blotting assays of SW480 and CT26 cells treated with recombinant SEMA3C protein and shNRP2. Recombinant SEMA3C protein increases phosphorylation of p38 and JNK, upregulates ZEB1, N-cadherin, and Vimentin, and downregulates E-cadherin. These effects are significantly reversed by shNRP2, indicating the role of the SEMA3C–NRP2 axis in MAPK-p38 and JNK activation. (J) Western blotting assays of SW480 and CT26 cells treated with recombinant SEMA3C protein and the p38/JNK inhibitor (LL-Z1640-4). Recombinant SEMA3C protein increases phosphorylation of p38 and JNK, upregulates ZEB1, N-cadherin, and Vimentin, and downregulates E-cadherin. Inhibition of p38/JNK signaling significantly reduced the phosphorylation levels of p38 and JNK. (K) Quantitative analysis of Western blotting assays. (L) Schematic model illustrating the proposed mechanism by which SEMA3C promotes CRC liver metastasis. SEMA3C secreted by CAFs interacts with NRP2 on tumor cells, activating the MAPK signaling pathway and promoting EMT, leading to CRC progression and liver metastasis.