Comprehensive analysis of CXCL14 uncovers its role during liver metastasis in colon cancer

The flowchart of whole study was shown in Figure S1. The GSE41258 provided the transcription profile data of primary focus and liver metastasis tissue of colon cancer. The data preprocessing process was shown in Figure S2. Then, the limma package was utilized to conduct the DEG analysis between primary focus and liver metastasis colon cancer tissue. Totally, 84 genes were upregulated and 83 genes were downregulated were identified in the liver metastasis colon cancer tissue compared with the control tissue (Figure S3). For the upregulated genes, biological enrichment analysis showed that these molecules were mainly involved in the negative regulation of blood coagulation, negative regulation of hemostasis, regulation of blood coagulation, regulation of hemostasis and negative regulation of response to wounding (Fig. 1A, GO-BP); for GO-CC, the top five enriched terms were blood microparticle, endoplasmic reticulum lumen, vesicle lumen, collagen-containing extracellular matrix, secretory granule lumen (Fig. 1B, GO-CC); for GO-MF, the top five enriched terms were lipoprotein particle receptor binding, endopeptidase inhibitor activity, peptidase inhibitor activity, endopeptidase regulator activity and enzyme inhibitor activity (Fig. 1C, GO-MF); for KEGG, the top five enriched terms were complement and coagulation cascades, cholesterol metabolism, drug metabolism-cytochrome P450, retinol metabolism and metabolism of xenobiotics by cytochrome P450 (Fig. 1D, KEGG). The downregulated genes were mainly enriched in the extracellular matrix disassembly, extracellular matrix organization, extracellular structure organization, external encapsulating structure organization and humoral immune response (Fig. 1E, GO-BP); for GO-CC, the top five enriched terms were immunoglobulin complex-circulating, sarcolemma, immunoglobulin complex, collagen-containing extracellular matrix and endoplasmic reticulum lumen (Fig. 1F, GO-CC); for GO-MF, the top five enriched terms were CXCR chemokine receptor binding, collagen binding, metalloendopeptidase activity, serine-type endopeptidase activity and serine-type peptidase activity (Fig. 1G, GO-MF); for KEGG, the top five enriched terms were rheumatoid arthritis, IL-17 signaling pathway, viral protein interaction with cytokine and cytokine receptor, ECM-receptor interaction and pancreatic secretion (Fig. 1H, KEGG). Moreover, the PPI network was constructed to illustrate the underlying protein interaction. The network of upregulated genes was illustrated in Fig. 1I and the network of downregulated genes was shown in Fig. 1J.

Based on the genes identified above, we tried to identify the molecules involved in the process of colon cancer liver metastasis, as well as remarkably affecting patients prognosis. Univariate Cox regression analysis was conducted and the genes with P < 0.1 were regarded as the molecules significantly affecting patients survival (Fig. 2A). LASSO regression analysis was used to optimize data dimension (Fig. 2B-C). Multivariate Cox regression analysis indicated that the molecules SERPINA3, SERPINA1, MMP3, ALDH1A3, PBK and CXCL14 were the independent factors for patients survival (Fig. 2D). Clinical correlation showed no significant difference in these genes between female and male patients (Fig. 2E); SERPINA1 and PBK were downregulated, while ALDH1A3 was upregulated in stage III-IV patients compared with stage I-II patients (Fig. 2F); SERPINA1 and CXCL14 were downregulated in the T3-4 patients compared with the T1-2 patients (Fig. 2G); SERPINA1 was downregulated, while ALDH3A1 was upregulated in the M1 patients compared to the M0 patients (Fig. 2H); SERPINA1 and PBK were downregulated, while ALDH1A3 was upregulated in N1-3 patients compared with N0 patients (Fig. 2I).

CXCL14 has the most significant P value in multivariate Cox regression analysis and therefore was selected for further analysis. Based on the open-accessed data of CRC_EMTAB8107 and CRC_GSE166655, we investigated the gene expression pattern of CXCL14 in the colon cancer microenvironment. The results showed that in both CRC_EMTAB8107 and CRC_GSE166655 cohorts, CXCL14 was mainly expressed in the fibroblasts (Fig. 3A-D). Next, we explored the role of fibroblasts in the colon cancer microenvironment. GSEA analysis based on the Hallmark gene set indicated that fibroblasts are associated with a higher activity of coagulation, epithelial-mesenchymal transition (EMT), myogenesis and UV response DN, yet a lower activity of allograft rejection (Fig. 3E-F). Meanwhile, GSEA analysis based on the KEGG gene set showed that the fibroblasts are associated with a higher activity of angiogenesis, coagulation, complement, EMT, myogenesis, UV response DN, while a lower activity of estrogen response early and estrogen response late (Fig. 3G-H). In the CRC_EMTAB8107 cohort, the fibroblasts interacted closely with the endothelial and malignant cells (Fig. 3I-J). In the CRC_GSE166655 cohort, the fibroblasts interacted closely with the endothelial, epithelial and malignant cells (Fig. 3K-L).

Pan-cancer analysis indicated that the CXCL14 was differentially expressed in most cancers and corresponding normal tissue, indicating its widespread effect on cancer development (Fig. 4A). Based on the data from TCGA-COAD and GTEx database, we noticed a higher RNA level of CXCL14 in colon cancer compared to the normal tissue (Fig. 4A). Nevertheless, the protein level of CXCL14 between colon cancer and normal tissue seemed to have no significant difference based on the data from the HPA database, indicating the underlying epigenetic regulation (Fig. 4B). KM survival curves indicated that the patients with higher CXCL14 expression levels might have worse overall survival (OS), disease-free survival (DSS) and progression-free survival (PFI) (Fig. 4C-E, OS, HR = 0.63, P = 0.018; DSS, HR = 0.62, P = 0.06; PFI, HR = 0.67, P = 0.025). Univariate and multivariate analysis revealed that the CXCL14 is a prognosis-protective marker independent of multiple clinical parameters like age, gender, and clinical and TNM classification (Fig. 4F-G).

Following, we tried to explore the biological effect of CXCL14 exerting in colon cancer. GSEA analysis based on the Hallmark gene set showed that CXCL14 is predominantly involved in the activating of the WNT/β-catenin pathway, pancreas beta cells, peroxisome and bile acid metabolism (Fig. 5A). The GSEA analysis based GO gene set indicated that the CXCL14 can significantly weaken the activity of chromatin assembly or disassembly, DNA conformation change, NDA packaging, protein DNA complex, protein DNA complex subunit organization and protein heterodimerization activity (Fig. 5B-G).

Different cells exert a diverse role in the tumor microenvironment and influence cancer progression in multiple manners. We next explored the correlation between CXCL14 and different cells in the tumor microenvironment. Results showed for the patients with high CXCL14 levels, the plasma B cells, CD8 + T cells, neutrophil and NK cells might infiltrate more, in contrast to B cells, monocyte and macrophages (Fig. 6A). Additionally, we observed lower activity of APC_co_inhibiton, APC_co_stimulation, check-point, cytolytic activity, inflammation-promoting, T cell co_inhibition, type I IFN response in patients with high CXCL14 expression (Fig. 6B). Drug sensitivity analysis indicated that the patients with low CXCL14 expression might be more sensitive to etoposide, rapamycin and sunitinib (Fig. 7A-L).