结直肠癌微转移的空间与分子特征

Figure 1. Overview of study cohort, pathological analysis, and spatial mapping of CRC metastases (A) Overview of the study design integrates Visium spatial transcriptomics (ST), Visium HD ST, laser-capture microdissection with whole-genome sequencing (LCM-WGS), and PhenoCycler-Fusion protein imaging. A total of 49 tumor specimens from 19 patients, including colorectal liver metastases (CLiMs), lung metastases (CLuMs), primary CRC, and adjacent normal tissues, were analyzed. (B) Composition of datasets used in this study, including spatial transcriptomics, LCM-WGS, protein imaging, single-cell RNA-seq, and bulk transcriptomic cohorts. Sample numbers for each modality and cohort are indicated. (C) Representative H&E-stained sections show a CRC macrometastasis and two micrometastatic foci in the liver (1, scale bar, 1 mm; 2–3, scale bars, 200 μm), lung metastases (4, scale bar, 1 mm; 5, scale bar, 200 μm), and primary CRC with multilayer invasion and perineural invasion (6, scale bar, 1 mm). Insets show magnified views of micrometastatic regions (2–3). (D and E) Pathology annotation of Visium ST spots (n = 341,328) (D) and Visium HD 8-μm bins (n = 3,843,553) (E) across CLiM, CLuM, and primary CRC samples. Sankey plots show relationships across annotation layers. Bar plots indicate counts per annotation, and boxplots show gene expression and detected gene numbers per annotation. Box middle lines, median; box limits, upper and lower quartiles; box whiskers, 1.5× the interquartile range. (F and G) Spatial pathology annotations for representative Visium ST (CLiM, CLuM, primary CRC) and Visium HD ST (CLiM, primary CRC) samples. See also Figure S1 and Table S1.

Figure 2. Spatial clonal architecture in CRC and multiple metastatic sites (A) Phylogenetic trees, CNV distributions, and clonal compositions across primary and metastatic samples for representative patient 1 (PAT1). Spatial maps show H&E-stained images, pathology annotations, clonal distributions, and signature scores (CytoTRACE, quiescence, differentiation, and cell cycle). Patient timelines indicate therapy and sampling. Scale bars, 1 mm. (B) Experimental design of LCM–WGS based on spatially inferred clonal regions. (C) Comparison of large-scale CNVs inferred from Visium ST and low-depth WGS in matched regions. (D–H) CNV levels (D) and signature scores for CytoTRACE (E), quiescence (F), differentiation (G), and cell cycle (H) across patients (n = 11). p values were calculated using paired two-sided Student’s t-tests. (I) Phylogenetic tree shows relationships between clones from CLiMi and CLiMa. (J) Representative multiplex immunofluorescence of CLiMi and CLiMa showing Ki67 and SOX2. Scales bars, 50 μm. (K) Quantification of SOX2+ and Ki67+ tumor cells in CLiMi and CLiMa (n = 18). p values from two-sided Wilcoxon tests. For boxplots in panels (D–H, K), center lines indicate the median; box limits denote the upper and lower quartiles; and whiskers extend to 1.5 × the inter­ quartile range. See also Figures S2–S6 and Table S2.

Figure 3. Identification and characterization of spatial MPs (A) Identification of spatial MPs in CLiM by NMF. Similarity matrix based on Jaccard index with sample and MP annotations. (B) Spatial distribution of MPs in a representative sample (M-ST-36), showing their expression levels and spatial distribution. (C) PhenoCycler-Fusion multiplex imaging design using adjacent sections matched to Visium ST. (D) Spatial maps of MPs (left) and corresponding multiplex immunofluorescence images (right) for M-ST-36. Scale bar, 1 mm. ll OPEN ACCESS Article (E) Zoomed-in views show correspondence between MPs and marker proteins (EPCAM, CD20, CD3e, MPO, CD68, SMA, Collagen IV, CD31, and iNOS) in the regions shown in (D). Scale bars, 50 μm. (F) UMAP of cell types identified from Visium HD ST (n = 801,392). (G) Heatmap of canonical marker gene expression across cell types in CLiM. (H) Spatial distribution of annotated cell types and zoomed views shows spatial organization at single-cell resolution. Scale bar, 1 mm. (I) Jaccard similarity between MP gene sets (Visium ST) and cell-type marker genes (Visium HD ST). See also Figures S7 and S8 and Table S2.

Figure 4. Molecular and pathway characteristics in primary CRC and CLiM (A) Proportions and enrichment (Ro/e, ratio of observed to expected; *: 1–2; **: 2–3) of MPs across primary CRC regions. (B) Pathway signature scores across primary CRC regions. (C) Spatial mapping of tumor and metabolic pathways in a representative primary CRC sample. ll OPEN ACCESS Article (D) Proportions and enrichment (Ro/e) of MPs across CLiM regions. (E) Pathway signature scores across CLiM regions. (F and G) Multiplex immunofluorescence of CLiMa (F) and CLiMi (G) shows EPCAM, iNOS, CD3e, PD-1, and GZMB. Scale bars, 50 μm and 20 μm. (H) Proportions of PD-1+ and GZMB+ T cells among total T cells in CLiMa and CLiMi (n = 18). (I) Immunosuppressive signature expression across CLiM and CLuM regions. (J) Spatial mapping of pathology annotations and immunosuppressive signature in CLiMi and CLiMa. Scale bar, 1 mm. Two-sided Wilcoxon tests were used in (B), (E), (H), and (I), with significance indicated as *p < 0.05 and ****p < 0.0001. For box plots in (B), (E), (H), and (I), center lines indicate the median; box limits denote the upper and lower quartiles; box whiskers, 1.5× the interquartile range. See also Figure S9 and Table S2.

Figure 5. Spatial network of MPs in CLiM, CLuM, and primary CRC (A) Schematic of spatial neighborhood and proximity for Visium ST spots. (B) Spatial network of MPs across CLiM, CLuM, and primary CRC. (C and D) Spatial neighborhood (C) and proximity (D) scores between TU_CEACAM5 and TME-related MPs across CLiM, CLuM, and primary CRC. Box middle lines, median; box limits, upper and lower quartiles; box whiskers, 1.5× the interquartile range. (E–G) Spatial mapping and colocalization of TU_CEACAM5 with MPs across tumor types: (E) Mac_CD68 and myCAF in CLiM, (F) T_cell in CLuM, and (G) Mac_CD68 in primary CRC. Colors denote distinct MPs. (H–J) Multiplex immunofluorescence in liver metastases (H: EPCAM, CD3e, CD20, CD68, SMA, Collagen IV, iNOS), lung metastases (I: EPCAM, CD3e, CD68, SMA, Collagen IV), and primary CRC (J: EPCAM, CD3e, CD20, CD68, SMA, Collagen IV). Zoomed views highlight spatial relationships between EPCAM and Collagen IV, CD68, or CD3e. Scale bars, 50 μm and 1 mm. (K–M) Zoomed regions show co-expression patterns: (K) collagen IV with SMA or vimentin and CD68 with CD163 or iNOS; (L) CD3e with PD-1 or TOX; (M) CD68 with CD163 or iNOS. Scale bars, 20 μm. Two-sided Wilcoxon tests were used in (C) and (D). See also Figure S10.

Figure 6. Spatial network of MPs in CLiMa and CLiMi (A) Spatial distribution of macro- and micrometastases and surrounding liver tissue. (B) Spatial neighborhood and proximity scores between TU_CEACAM5 and T_B, and between TU_CEACAM5 and myCAF MPs in CLiMa and CLiMi. Box middle lines, median; box limits, upper and lower quartiles; box whiskers, 1.5× the interquartile range. (C–D) Spatial colocalization of MPs in representative CLiM samples. (E–F) Multiplex immunofluorescence images show EPCAM, CD3e, CD20, Collagen IV, and iNOS. Zoomed views highlight EPCAM association with collagen IV in macrometastases and with CD3e/CD20 in micrometastases. Scale bars, 50 μm and 1 mm. (G–H) Zoomed regions from (E) show co-expression patterns: Collagen IV with SMA or Vimentin (top), and CD3e with PD-1 or TOX (bottom). M-ST-05 corre­ sponds to (G) and M-ST-36 to (H). Scale bars, 20 μm. (I) Ligand-receptor interactions between TU_CEACAM5 and myCAF or T cells in CLiMa and CLiMi. (J) Spatial distribution of macro- and micrometastatic regions (macro: purple, extended: light purple; micro: red, extended: pink). Scale bar, 1 mm. (K) Spatial density maps of LAMA4, ITGA1 + ITGB1, NECTIN3, and TIGIT (yellow to dark purple, high to low). Two-sided Wilcoxon tests were used in (B). One-sided permutation tests were used in (I). See also Figure S11.

Figure 7. Identification and validation of the CLiMi six-gene signature and its association with clinical outcomes (A) Workflow for defining the CLiMi six-gene signature. (B) Six-gene signature expression across tumor, stroma, and normal liver regions. (C and D) Spatial mapping of six-gene signature expression in Visium ST (C) and Visium HD ST (D). Scale bars, 1 mm and 20 μm. (E) UMAP of tumor cells shows five clusters (top) and six-gene signature expression (bottom). (F) UMAP feature plots of individual signature genes. (G) Six-gene signature scores across clusters before and after cell-cycle adjustment. (H) Pathway enrichment and stemness marker expression in tumor cells. (I) Kaplan-Meier analysis of DFS and MRD-FS in the MDACC cohort, comparing patients with high (60–100%) versus low (0–40%) CLiMi six-gene signature expression. (J) Six-gene signature scores in CLiM samples with or without relapse (left; two-sided Wilcoxon test; n = 117). Proportions of relapse in high- and low-signature groups (right; chi-square test; n = 94). (K) Six-gene signature scores in baseline and recurrent CLiM samples (n = 12). (L and M) Spatial mapping of six-gene signature expression (L) and MicroMetSig^hi tumor cell fractions (M). (N–P) Validation across independent cohorts: (N) scRNA-seq (GEO: GSE285991; n = 33,357) with chemotherapy response (pre-treatment, PR, and PD); (O) liver metastasis microarray (GEO: GSE131418; n = 170) with pre- and post-chemotherapy/radiotherapy samples; (P) primary CRC microarray (GEO: GSE21510; n = 146) with metastatic status. Signature scores are stratified by condition. (Q) Kaplan-Meier analysis of DFS in the TCGA and GEO: GSE17538 cohorts using the high (60–100%) versus low (0–40%) stratification. p values were calculated using two-sided Wilcoxon tests in (B), (G), (K), and (N–P), and log rank tests in (I) and (Q). For box plots in (B), (J), (K), (O) and (P), center lines indicate the median; box limits denote the upper and lower quartiles; and whisker extend to 1.5 × the interquartile range. See also Figure S12 and Table S1.