Pan-Cancer Analysis of Prognostic and Immune Infiltrates for CXCs

Simple Summary CXCs are important genes that regulate inflammation and tumor metastasis. While there are many studies with a focus on individual CXCs, few present a pan-cancer analysis of the whole CXC family. Our results indicate that CXCs are a potential therapeutic target in a variety of tumors and a potential prognostic marker that could improve the survival of cancer patients and the accuracy of prognosis. Meanwhile, we found that CXCs may be involved in diseases caused by intestinal flora. Abstract Background: CXCs are important genes that regulate inflammation and tumor metastasis. However, the expression level, prognosis value, and immune infiltration of CXCs in cancers are not clear. Methods: Multiple online datasets were used to analyze the expression, prognosis, and immune regulation of CXCs in this study. Network analysis of the Amadis database and GEO dataset was used to analyze the regulation of intestinal flora on the expression of CXCs. A mouse model was used to verify the fact that intestinal bacterial dysregulation can affect the expression of CXCs. Results: In the three cancers, multiple datasets verified the fact that the mRNA expression of this family was significantly different; the mRNA levels of CXCL3, 8, 9, 10, 14, and 17 were significantly correlated with the prognosis of three cancers. CXCs were correlated with six types of immuno-infiltrating cells in three cancers. Immunohistochemistry of clinical samples confirmed that the expression of CXCL8 and 10 was higher in three cancer tissues. Animal experiments have shown that intestinal flora dysregulation can affect CXCL8 and 10 expressions. Conclusion: Our results further elucidate the function of CXCs in cancers and provide new insights into the prognosis and immune infiltration of breast, colon, and pancreatic cancers, and they suggest that intestinal flora may influence disease progression through CXCs.


Introduction
Cancer is a major worldwide public health problem, causing an estimated total of 9 million deaths in 2016 (World Health Statistics, 2020). Although significant survival benefits have been achieved in recent years because of early detection, screening, and treatment methods, improving overall survival (OS) remains a challenge in the clinic. Tumor metastasis is an important factor that correlates with a poor prognosis of cancer. adjacent normal control samples were obtained from the Oncomine database. Difference in transcriptional expression was compared by Student's t-test. Cut-off of p value and fold change were as follows: p value: 0.05; fold change: 2.0; gene rank: 10%.

TISCH Analysis
Tumor Immune Single-cell Hub (TISCH, http://tisch.comp-genomics.org/, accessed on 21 February 2021), a database focusing on the tumor microenvironment (TME), provides single-cell level cell-type annotation [15]. In the present study, we evaluated the expression level of CXCs in each subgroup of cells in the three cancer datasets and analyzed the interrelationship between the level and tumor stage. GSEA enrichment analysis and visualization of inflammatory pathways were also performed.

GEPIA Analysis
Gene Expression Profiling Interactive Analysis (GEPIA, http://gepia.cancer-pku. cn/, accessed on 25 February 2021) is a web-based tool to deliver fast and customizable functionalities based on 9736 tumors and 8587 normal samples from GTEx database and The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov, accessed on 1 March 2021) [16]. In our study, tumor/normal differential expression analysis and pathological stages were obtained from them. Difference in transcriptional expression was compared by Student's t-test, and p < 0.05 was considered statically significant.

Kaplan-Meier Plotter
Kaplan-Meier plotter (https://www.kmplot.com, accessed on 5 March 2021) is a database that contains gene expression data and survival information of breast cancer patients. The prognostic value of mRNA expression was analyzed using this database [17]. To analyze the OS of patients with three aforementioned malignancies, samples of patients were segregated into two groups (high-expression group and low-expression group). These groups were assessed by a Kaplan-Meier survival plot, with the hazard ratio (HR) with 95% confidence intervals (CIs) and a log-rank p value. Only the JeSet best probe set was selected.

OncoLnc Dataset
OncoLnc (http://www.oncolnc.org/, accessed on 6 March 2021) is a tool for interactively exploring survival correlations, which contains survival data for 21 cancer studies performed by TCGA. The PDAC and COAD patients were divided into two groups; we assessed the OS of these groups by using Kaplan-Meier plots and log-rank p value; and the cut-off criterion was log-rank p value < 0.05.

TRRUST Dataset
TRRUST v2 (https://www.grnpedia.org/trrust/, accessed on 7 March 2021) is a dataset that provides the transcription factor (TF) of target genes and the regulatory network between them. It includes 8444 TF-target regulatory relationships of 800 human TFs and 6552 TF-target regulatory relationships of 828 mouse TFs. These data are derived from 11,237 PubMed articles that describe small-scale experimental studies of transcriptional regulations [18].

KnockTF Dataset
KnockTF (http://www.licpathway.net/KnockTF/index.html, accessed on 7 March 2021) is a database providing available resources of human gene expression profile datasets, which are associated with TF knockdown/knockout. The database annotates TFs and their targets in a tissue/cell-type-specific way [19].

MiRWalk Dataset
MiRWalk 2.0 (http://mirwalk.umm.uni-heidelberg.de, accessed on 8 March 2021), an open source platform, can predict and validate miRNA-binding sites of genes from humans, mice, rats, dogs, and cows. The core of miRWalk is the TarPmiR (random-forest-based approach) that can predict miRNA target sites of the transcript sequence.

CBioPortal Dataset
CBioPortal (https://www.cbioportal.org, accessed on 8 March 2021), a comprehensive database, provides analysis and visualization functions to process multi-tumor genomics data [20]. Based on data in TCGA, genetic alterations and co-expression of CXCs were obtained from cBioPortal. Protein expression z scores (RPPA) and mRNA expression z scores (RNA Seq V2 RSEM) were obtained using a z score threshold of 2.0.

STRING Dataset
STRING 11.0 Dataset (https://string-db.org/, accessed on 8 March 2021) collects and integrates PPI (protein-protein interaction) data from public sources and predicts potential functions [21]. A CXCs-PPI network analysis was used to inquire about the interactions. The visualization of those networks was achieved by Cytoscape v.3.6.

GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) Analysis
GO and KEGG analyses of 66 proteins interacting with CXCs found in the STRING database were performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID, https://david.ncifcrf.gov/summary.jsp, accessed on 9 March 2021) [22]. GO analysis can reveal the potential functional roles of CXCs, including biological processes (BP), cellular components (CC), and molecular functions (MF), while KEGG analysis can define the pathways related to CXCs.

TIMER Analysis
The Tumor Immune Estimation Resource (TIMER 1.0, https://cistrome.shinyapps. io/timer/, accessed on 10 March 2021) is a database that focuses on analyzing tumorinfiltrating immune cells throughout 32 kinds of malignancies from TCGA [23]. We used the gene module to inquiry correlations between CXCL expression and abundance of tumor-infiltrating immune cells by Spearman's correlation, which include CD8 + T cells, CD4 + T cells, macrophages, neutrophils, B cells, and dendritic cells.

Amadis
Amadis is a database that provides experimentally supported microbiota-disease associations [24]. With aid of Amadis's network analysis tools, we found that there could be an association between CXCL8, Fusobacterium nucleatum, and human diseases (including inflammatory bowel disease and colon cancer).

Bacterial Culture
Fusobacterium nucleatum (F. nucleatum) strain ATCC 25586, which was purchased from American Type Culture Collection (ATCC, Manassas, VA, USA), was cultured in brain heart infusion (BHI) broth at 37 • C under anaerobic conditions.

Mice
The animal experiments obtained permission through the Animal Ethics and Welfare Committee (AEWC) of the First Affiliated Hospital of Harbin Medical University. C57BL/6J wild-type (WT) mice were purchased from Beijing Vital River Laboratory Animal Technology Co. Ltd. (Beijiing, China). Six-to eight-week-old female C57BL/6J mice aged 6-8 weeks were housed in standard specific pathogen-free conditions. The mice were injected with a single intraperitoneal(i.p.) injection of the AOM (10 mg/kg). One week later, they were given three cycles of 2% DSS treatment (1 week per cycle). The mice were treated with F. nucleatum (1 × 10 9 CFU) by gavage from a fortnight before AOM injection until sacrifice. During the DSS intervention, F. nucleatum administration was suspended. The negative control mice were gavage-fed with PBS only. Intragastric gavage administration was carefully carried out with the animal immobilized, using a gavage needle appropriate for mice. Before bacterial intragastric administration, mice were fed with broad-spectrum antibiotics (BSA) in the drinking water for 5 days to ensure the consistency of regular microbiota and facilitate F. nucleatum colonization. The DAI (disease activity index) and body weight were observed daily.

Western Blot
Western blots were performed according to standard protocols. A 12% SDS-PAGE gel was used to separate total proteins extracted from mice colon tissue. Then, proteins were transferred onto polyvinylidene fluoride membranes. The membranes were incubated with primary antibodies for CXCL8 (Novus) and CXCL10 (Affinity) overnight at 4 • C after blocking with 5% non-fat dry milk in PBST. Anti-GAPDH (Beyotime, 1:1000) was used as the control. Each experiment has been repeated at least three times.

Enzyme-Linked Immunosorbent Assay (ELISA)
The mouse blood samples were centrifuged, serum was collected, and immediately cryopreserved in liquid nitrogen. According to the manufacturer's instructions, the quantification of serum cytokine was carried out using the Quantibody ® Mouse CXCL10 ELISA Kit (RayBiotech, Norcross, GA, USA).

Histology and Immunohistochemistry (IHC)
Cancer tissue samples and paracancerous tissue samples were collected from the First Affiliated Hospital of Harbin Medical University. The studies obtained permission through the Ethics Committee of the First Affiliated Hospital of Harbin Medical University. Written informed consents were signed by patients/participants to participate in this study.
For histologic evaluation, formalin-fixed colon tissue sections were embedded in paraffin and cut (5 µm) for H&E staining or immunohistochemistry (IHC). For IHC assays, we deparaffinized the paraffin sections, inactivated endogenous enzymes, and thermally repaired antigens. These sections were stained with CXCL8 (Novus) and 10 (Affinity) antibodies, followed by a corresponding secondary antibody and a Streptavidin Biotin Complex kit (Boster BioEngineering, Wuhan, China). Stained slides were scanned by KFBIO. SlideViewer and quantified by Image-pro-plus software.

Statistical Analysis
All data were analyzed using SPSS 22.0 software (Chicago, IL, USA) by ordinary one-way analysis of variances with Tukey's multiple comparisons. p < 0.05 was considered statistically significant.

Analysis Process and Data Processing
The analysis process is shown in Figure 1. The data used by this study are from TCGA datasets and Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo, accessed on 15 March 2021) datasets. We conducted a comprehensive analysis of CXCs in eight steps (Figure 1).

Analysis Process and Data Processing.
The analysis process is shown in Figure 1. The data used by this study are from TCGA datasets and Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo, accessed on 15 March 2021) datasets. We conducted a comprehensive analysis of CXCs in eight steps (Figure 1).

Transcriptional Levels of CXCs in Various Cancers
First, we used the Oncomine database to analyze the differential expression levels of CXC transcripts in 20 types of cancer tissues versus the corresponding normal tissues. We found that each of the 16 genes of this family had approximately 400 unique analyses, except CXCL16 and 17. For these 16 genes, we identified cancer types with significant differences in the expression confirmed by multiple unique analyses ( Figure 2).

Transcriptional Levels of CXCs in Various Cancers
First, we used the Oncomine database to analyze the differential expression levels of CXC transcripts in 20 types of cancer tissues versus the corresponding normal tissues. We found that each of the 16 genes of this family had approximately 400 unique analyses, except CXCL16 and 17. For these 16 genes, we identified cancer types with significant differences in the expression confirmed by multiple unique analyses ( Figure 2). CXCL1 was significantly expressed at high levels in 21 unique analyses in colon cancer and significantly expressed at low levels in 15 unique analyses in breast cancer. CXCL2 was significantly expressed at high levels in 14 unique analyses and significantly expressed at low levels in 27 unique analyses in breast cancer. CXCL3 was significantly expressed at high levels in 23 unique analyses in colon cancer. CXCL8 was significantly expressed at high levels in 19 unique analyses in colon cancer and 3 unique analyses in pancreatic cancer. CXCL9 was significantly expressed at high levels in 15 unique analyses in breast cancer and 21 unique analyses in lymphoma. CXCL10 was significantly expressed at high levels in 16 unique analyses in breast cancer and 16 unique analyses in lymphoma. There were 12 unique analyses with significantly high expression of CXCL11 in breast cancer and 12 unique analyses in colon cancer, and 16 unique analyses with significantly low expression of CXCL12 in breast cancer, and 3 unique analyses with significantly low expression in pancreatic cancer. According to these analyses, we found significant differences in the expression of the CXCs in BRCA, COAD, and PDAC; thus, we selected these three types of cancer for follow-up analysis.
In the TCGA datasets with more than 100 samples, invasive ductal breast cancer samples showed low expression of CXCL1, 2, 3, 12, and 14 (fold change > 2). The Curtis dataset of invasive ductal breast cancer showed significant differences in the expression of CXCL2, 8,9,10,12, and 14 (fold change > 2) [25]. Colon cancer samples from the TCGA showed high expression of CXCL1, 3, 5, 6, and 11 and low expression of CXCL12 (fold change > 2); the Bittner poly-cancerous dataset confirmed the differential expression of CXCs in colon cancer and breast cancer. There were fewer samples of pancreatic cancer. In the Barretina dataset with 44 samples, the levels of CXCL2, 3, 5, and 16 (fold change > 2) were significantly higher [26], and the levels of CXCL3, 5, 8, 10, and 16 (fold change > 2) were significantly higher in the Badea and Pei datasets (Table 1) [27]. CXCL1 was significantly expressed at high levels in 21 unique analyses in colon cancer and significantly expressed at low levels in 15 unique analyses in breast cancer. CXCL2 was significantly expressed at high levels in 14 unique analyses and significantly expressed at low levels in 27 unique analyses in breast cancer. CXCL3 was significantly expressed at high levels in 23 unique analyses in colon cancer. CXCL8 was significantly expressed at high levels in 19 unique analyses in colon cancer and 3 unique analyses in pancreatic cancer. CXCL9 was significantly expressed at high levels in 15 unique analyses in breast cancer and 21 unique analyses in lymphoma. CXCL10 was significantly expressed at high levels in 16 unique analyses in breast cancer and 16 unique analyses in lymphoma. There were 12 unique analyses with significantly high expression of CXCL11 in breast cancer and 12 unique analyses in colon cancer, and 16 unique analyses with significantly low expression of CXCL12 in breast cancer, and 3 unique analyses with significantly low expression in pancreatic cancer. According to these analyses, we found significant differences in the expression of the CXCs in BRCA, COAD, and PDAC; thus, we selected these three types of cancer for follow-up analysis.
In the TCGA datasets with more than 100 samples, invasive ductal breast cancer samples showed low expression of CXCL1, 2, 3, 12, and 14 (fold change > 2). The Curtis dataset of invasive ductal breast cancer showed significant differences in the expression of CXCL2, 8,9,10,12, and 14 (fold change > 2) [25]. Colon cancer samples from the TCGA showed high expression of CXCL1, 3, 5, 6, and 11 and low expression of CXCL12 (fold change > 2); the Bittner poly-cancerous dataset confirmed the differential expression of CXCs in colon cancer and breast cancer. There were fewer samples of pancreatic cancer. In the Barretina dataset with 44 samples, the levels of CXCL2, 3, 5, and 16 (fold change > 2) were significantly higher [26], and the levels of CXCL3, 5, 8, 10, and 16 (fold change > 2) were significantly higher in the Badea and Pei datasets (Table 1) [27]. We used the TISCH database to analyze subpopulation distribution (Material S1) of 16 genes in single-cell sequencing datasets of breast, colon, and pancreatic cancers. Among them, CXCL10 and 16 were significantly increased in mononuclear/macrophage cells of the three cancers ( Figure 3).
According to the number of cells detected in the dataset and the expression of CXC in each dataset, the breast cancer dataset BRCA_GSE114727_inDrop, colon cancer dataset CRC_GSE146771_10X, and pancreatic cancer dataset PAAD_CRA001160 were selected for further analysis.

mRNA and Protein Expression of CXCs in Three Kinds of Cancer
Using the GEPIA dataset, we compared the expression of mRNAs in three types of cancer tissues versus normal tissues. The results showed that the expression of CXCL1, 2, 3, 12, and 14 in BRCA was lower in tumor tissues, and the expression of CXCL9, 10, 11, and 13 was higher than its expression in normal tissues; in COAD, the expression of

mRNA and Protein Expression of CXCs in Three Kinds of Cancer
Using the GEPIA dataset, we compared the expression of mRNAs in three types of cancer tissues versus normal tissues. The results showed that the expression of CXCL1, 2, 3, 12, and 14 in BRCA was lower in tumor tissues, and the expression of CXCL9, 10, 11, and 13 was higher than its expression in normal tissues; in COAD, the expression of CXCL12, 13 and 14 was lower than that in normal tissues, and the expression of CXCL1, 2, 3, 5, 8, 9, 10 and 11 was higher than that in normal tissues; in PDAC, the expression of CXCs was significantly higher than that in normal tissues, except CXCL2, 7, 11 and 12 ( Figure 4).
In addition, we detected expression differences of CXCRs (CXC receptor) in three cancers. The results showed that there was no significant difference between normal tissue and cancer tissue, except for the fact that CXCR4 and 6 were less expressed in pancreatic cancer tissue (Material S2).
Cancers 2021, 13, x FOR PEER REVIEW 10 of 34 CXCL12, 13 and 14 was lower than that in normal tissues, and the expression of CXCL1, 2, 3, 5, 8, 9, 10 and 11 was higher than that in normal tissues; in PDAC, the expression of CXCs was significantly higher than that in normal tissues, except CXCL2, 7, 11 and 12. (Figure 4) In addition, we detected expression differences of CXCRs (CXC receptor) in three cancers. The results showed that there was no significant difference between normal tissue and cancer tissue, except for the fact that CXCR4 and 6 were less expressed in pancreatic cancer tissue (Material S2).  We also analyzed the expression of CXCs in three types of cancers at various stages. The TISCH database was used to analyze the relationship between CXC expression and staging in different subsets of cells. In BRCA, the staging differences in CXCL1, 2, 5, 8, 12, and 14 were statistically significant (Figure5A). Among them, CXCL2, 8, and 12 were generally significantly correlated with staging in each cell subgroup, and six genes were significantly correlated with staging in the mononuclear/macrophage subgroup ( Figure 5B). In COAD, only the staging differences in CXCL9, 10, and 11 were statistically significant ( Figure 5C) and were significantly associated with staging in the mononuclear/macrophage subpopulation ( Figure 5D). In PDAC, the staging differences in CXCL3, 5, and 8 were statistically significant ( Figure 5E) and were generally significantly associated with staging in all cell subsets ( Figure 5F). We also analyzed the expression of CXCs in three types of cancers at various stages. The TISCH database was used to analyze the relationship between CXC expression and staging in different subsets of cells. In BRCA, the staging differences in CXCL1, 2, 5, 8, 12, and 14 were statistically significant ( Figure 5A). Among them, CXCL2, 8, and 12 were generally significantly correlated with staging in each cell subgroup, and six genes were significantly correlated with staging in the mononuclear/macrophage subgroup ( Figure 5B). In COAD, only the staging differences in CXCL9, 10, and 11 were statistically significant ( Figure 5C) and were significantly associated with staging in the mononuclear/macrophage subpopulation ( Figure 5D). In PDAC, the staging differences in CXCL3, 5, and 8 were statistically significant ( Figure 5E) and were generally significantly associated with staging in all cell subsets ( Figure 5F).  We examined CXCL protein levels through IHC and found that the protein expressions of CXCL8 and 10 were statistically significantly up-regulated in human breast cancer, pancreatic cancer, and colon cancer tissues versus the corresponding normal samples ( Figure 6). We examined CXCL protein levels through IHC and found that the protein expressions of CXCL8 and 10 were statistically significantly up-regulated in human breast cancer, pancreatic cancer, and colon cancer tissues versus the corresponding normal samples ( Figure 6).

Prognostic Value of CXCs in Three Kinds of Cancer
We further made the survival analysis of CXCs in three cancers. A public dataset was used to analyze the associations between CXC mRNA levels and the survival time of breast cancer patients using the Kaplan-Meier mapping tool. The public dataset OncoLnc was used to analyze the associations between CXC mRNA levels and the survival of patients with colon and pancreatic cancer.

Prognostic Value of CXCs in Three Kinds of Cancer
We further made the survival analysis of CXCs in three cancers. A public dataset was used to analyze the associations between CXC mRNA levels and the survival time of breast cancer patients using the Kaplan-Meier mapping tool. The public dataset OncoLnc was used to analyze the associations between CXC mRNA levels and the survival of patients with colon and pancreatic cancer.

Prediction of Transcription Factors (TFs) Regulating CXCs
Because of the significant differences in the expression of CXCs in the cancer tissues versus normal tissues, we used the TRRUST database and KnockTF database to identify possible TFs and regulatory relationships between CXCs and TF.
We determined that the key TFs of the CXC family include RELA, NFKB1, and SP1 (Table 2) (predicted). Additionally, we evaluated all TFs of the nine CXCs, including possible regulation modes ( Figure 8A, Table 3) (experimentally validated). Interestingly, the same TF may induce different regulations in different studies, such as ELF4, NFKB1, and RELA, which are present in the lists of transcriptional activators and transcriptional suppressors of CXCL8. Meanwhile, we extracted a transcriptional regulatory subnetwork between CXCs and TFs using the KnockTF database. TF-target relationships supported by

Prediction of Transcription Factors (TFs) Regulating CXCs
Because of the significant differences in the expression of CXCs in the cancer tissues versus normal tissues, we used the TRRUST database and KnockTF database to identify possible TFs and regulatory relationships between CXCs and TF.

Regulation of CXCL8 and 10 by F. nucleatum in CAC
The important contribution of the gut microbiota to human health and disease is widely recognized. Until now, more and more online databases have been developed to manage signatures of microbiota genomes, disease-related genes and proteins, as well as providing some analysis. Amadis is a database that provides microbiota-disease associa-tions supported by experiments and interaction networks between them. By constructing an interaction network of CXCs, CXCRs, intestinal flora, and human diseases, we found that there could be a possible association between CXCL8, Fusobacterium nucleatum, and human diseases (including inflammatory bowel disease (IBD) and colon cancer) ( Figure 9A). Additionally, we analyzed the gene expression profile of the RNA-seq dataset (GSE90944) in HT-29 cell lines treated with or without F. nucleatum. Differential expression of CXCL8 and significant differential expression of TFs (predicted as shown in Table  3) could have activated CXCL8 transcription and activated CXCL10 transcription ( Figure  9B). Correlation analysis showed that CXCL8 was significantly correlated with CEBPB, FOSB, JUN, NFE2L2, HDAC2, and SFPQ. CXCL10 was significantly correlated with IRF1 and IRF7 (Figure 10).  Additionally, we analyzed the gene expression profile of the RNA-seq dataset (GSE90944) in HT-29 cell lines treated with or without F. nucleatum. Differential expression of CXCL8 and significant differential expression of TFs (predicted as shown in Table 3) could have activated CXCL8 transcription and activated CXCL10 transcription ( Figure 9B). Correlation analysis showed that CXCL8 was significantly correlated with CEBPB, FOSB, JUN, NFE2L2, HDAC2, and SFPQ. CXCL10 was significantly correlated with IRF1 and IRF7 (Figure 10). Subsequent analysis of the miRNA data presented in the Supplementary Materials identified a total of 64 differentially expressed miRNAs (p < 0.05, log2FC > 2) ( Figure  11A,B). Analysis of the miRWalk database identified 804 miRNAs that can bind to the 3'-UTR of CXCL8 and 1015 miRNAs that can bind to the 3'-UTR of CXCL10 (Material S3) (predicted using bioinformatics tools). The intersection of differentially expressed miRNA with possibly bound miRNA identified seven downregulated miRNAs in the case of CXCL8 and 13 downregulated miRNAs in the case of CXCL10 ( Figure 11C; Table 5). Subsequent analysis of the miRNA data presented in the Supplementary Materials identified a total of 64 differentially expressed miRNAs (p < 0.05, log2FC > 2) ( Figure 11A,B). Analysis of the miRWalk database identified 804 miRNAs that can bind to the 3'-UTR of CXCL8 and 1015 miRNAs that can bind to the 3'-UTR of CXCL10 (Material S3) (predicted using bioinformatics tools). The intersection of differentially expressed miRNA with possibly bound miRNA identified seven downregulated miRNAs in the case of CXCL8 and 13 downregulated miRNAs in the case of CXCL10 ( Figure 11C; Table 5).   The result of the above analysis is in accordance with the in vivo experiment. By using the AOM/DSS-induced, colitis-associated cancer mouse model, we verified F. nucleatum's regulatory role in the expression of CXCs ( Figure 12A). Oral gavage with F. nucleatum aggravates the loss of body weight in CAC mice ( Figure 12C,D). At the time of sacrifice, colons were removed, and colon length and tumor number were measured. Treatment with F. nucleatum significantly shortened the colon length and promoted tumorigenesis ( Figure 12B,E). Inflammation of the intestine was histologically analyzed. Compared with the control group, treatment with F. nucleatum significantly increased the mucosal breaks of the oral administration group (Figure 12F,G). Blood was collected and assayed by ELISA. CXCL10 levels in the blood of mice with F. nucleatum gavage were significantly upregulated ( Figure 12H). WB analyses of colon tissue from CAC mice after their F. nucleatum administration revealed significant up-regulation of CXCL8 and 10 ( Figure 12I-L). Generally, these results proved that treatment with F. nucleatum could aggravate inflammation of the intestine, promote tumorigenesis, and increase CXCL8 and 10 gene expression in AOM/DSS-induced CAC mice. These results suggest that in the presence of F. nucleatum, the expression of TFs and miRNAs is different and thus regulates the expression of CXCL8 and 10 to influence the occurrence and development of colon cancer. Data are presented as means ± SD. * p < 0.05; ** p < 0.01; *** p < 0.001; Student's t-test (two-tailed). The original Western blot images of (I&K) was shown in Material S8.
We have explored the co-expression relationships of CXCs. In BRCA samples, there were significant positive correlations between the expression of CXCL1 and that of CXCL2, 3, 5, 6, and 8. These correlations were also found in the expression of CXCL2 and CXCL3, 5 and 6. and in the expression of CXCL3 and CXCL5, 6 and 8. It is also found that CXCL5, 6, and 9 were positively correlated with CXCL10, 11, and 13, and so is the expression of CXCL10 with the expression of CXCL11 and 13; similarly, the expression of CXCL11 was found to be positively correlated with the expression of CXCL13 (p < 0.05, R 2 > 0.5).
In COAD samples, there are similarities in the correlations and also distinct differences. The expression of CXCL1 was found to be highly correlated with the expression CXCL2 and 9; however, CXCL4 was negatively correlated with CXCL12 and 13 (p < 0.05, R 2 < −0.5). Positive correlations were also found between CXCL6 and CXCL8; CXCL8 and CXCL9, 10; and so were CXCL9 and CXCL10, 11, 12, and 13. It is also found that CXCL10 was positively correlated with CXCL12 and 13; and that CXCL12 was positively correlated with CXCL13 (p < 0.05, R 2 > 0.5).
In PDAC samples, CXCL1 was highly correlated with CXCL2, 3, 6, and 8; CXCL2 was positively correlated with CXCL3 and 8. These correlations were also found in CXCL3 and CXCL5 and 8 and in CXCL9 and CXCL10 and 11. Similarly, the expression of CXCL10 was positively correlated with the expression of CXCL11 (p < 0.05, R 2 > 0.5) (Figure 13C-E).
With the above co-expression analysis results, we found that the co-expression of CXCs may be related to the chromosomal localization of genes and transcription factors. CXCL1, 2, 3, 5, 6, 7, and 8 are located at 4q12-13. In BRCA samples, the expression of CXCL1 was highly correlated with CXCL2, 3, 5, 6, and 8. In PDAC samples, the expression of CXCL1 was positively correlated with CXCL2, 3, 6, and 8. Meanwhile, according to the analysis of transcription factors, the co-expressed genes may be regulated by the same one or more transcription factors, as NFκB and RELA may be responsible for multiple CXCs, including CXCL1, 2, 5, 8, 10, and 12.

Prediction of CXC-Interacting Proteins and Their Functions and Pathways
The CXC family performed functions by binding to receptors, so it is important to analyze the relation between CXCs and proteins interacting with CXCs. We analyzed 50 proteins interacting with CXCs using the String database. As a result, 66 nodes and 1498 edges were obtained in the PPI network, and a network map was constructed using Cytoscape ( Figure 14A).   The TISCH database was used to analyze and visualize the enrichment scores of inflammatory response signaling pathways in each cell subgroup. Inflammatory response signaling pathways were found to be enriched in mononuclear/macrophage subsets in all three cancer datasets ( Figure 15). Additionally, the functions of CXCs and their 50 interacting proteins were analyzed using the DAVID database by GO and KEGG enrichment analysis. The results presented the top 10 highly enriched biological processes pathways include chemokine-mediated signaling pathways, inflammatory responses, chemotaxis, immune responses, G proteincoupled receptor signaling pathways, cell chemotaxis, and other biological processes, suggesting that CXCs in cancer are involved in chemotaxis and function in the inflammatory response ( Figure 14B). The extracellular space, extracellular region, outer plasma membrane, cell surface, plasma membrane, and cell area were the main enrichment terms of CXCs ( Figure 14C). In the molecular function categories, CXCs and CXCs-interacting proteins were enriched in chemokine activity and CXCR-chemokine-receptor-binding activity ( Figure 14D).
It is known that CXCL1, 2, 3, 5, 6, 7, and 8 are bound to CXCR1 and 2; CXCL9, 10, and 11 are bound to CXCR3. Correlation analysis was performed for CXCs sharing the same receptor and their targets. All correlation coefficients between CXCR and CXCs were not significant (R < 0.8), suggesting that there was no strong correlation between CXCs and CXCR expressed in colon cancer (Material S5).
The TISCH database was used to analyze and visualize the enrichment scores of inflammatory response signaling pathways in each cell subgroup. Inflammatory response signaling pathways were found to be enriched in mononuclear/macrophage subsets in all three cancer datasets ( Figure 15).

Immune Cell Infiltration and CXCs in Three Types of Cancer
At present, the function of CXCs is still controversial. Some studies have found that tumor cells secrete CXCs to act on their own surface receptors [28], while other studies have revealed that CXCs can act as a signal to recruit immune cells [29]. The results of the

Immune Cell Infiltration and CXCs in Three Types of Cancer
At present, the function of CXCs is still controversial. Some studies have found that tumor cells secrete CXCs to act on their own surface receptors [28], while other studies have revealed that CXCs can act as a signal to recruit immune cells [29]. The results of the functional enrichment and pathway analyses suggest that CXCs may influence the clinical outcome of cancer patients through regulating inflammatory response and immune cell infiltration. Therefore, we used the TIMER database to explore specific features of CXCs.
We analyzed the correlations between each CXC and tumor purity, B cells, CD8 + T cells, CD4 + T cells, macrophages, neutrophils, and dendritic cells in three types of cancer. A total of 244 pairs with significant correlation were detected, including 24 pairs with a partial correlation coefficient (Partial.cor) > 0.5; all pairs were positively correlated. As shown, for these 24 pairs of data, we mainly focused on the association between CXCL9, 10, and 13 and infiltrating immune cells ( Figure 16A,B,C). Other related data are shown in Material S6.
We used the TISCH database to analyze the distribution of CXCL9, 10, and 13 cells in each subgroup of three types of cancer. It was found that CXCL9 and 10 were essentially enriched in mononuclear/macrophage subsets among the three cancers. CXCL13 is enriched in fibroblasts and CD8 + T cells in breast cancer, CD8 + Tex and CD4 + Tconv cells in colon cancer, and plasma in pancreatic cancer ( Figure 16D,E,F).
We also analyzed the distribution of CXCR in colon cancer. The results showed that CXCR1 mainly expresses in NK cells. CXCR2 mainly expresses in neutrophils and monocytes/macrophages. CXCR3 is widely distributed in Treg, Tprolif, CD8T, CD8Tex, and CD4Tconv cells. CXCR4 is widely distributed in T cells, such as Treg, Tprolif, and CD8T, as well as in NK cells and B cells. CXCR5 is mainly distributed in B cells. CXCR6 is mainly distributed in NK cells and T cells, such as Treg, Tprolif, and CD8T (Material S7). These results may indicate that CXCs play a role in recruiting immune cells by binding to receptors on the surface of immune cells.
shown, for these 24 pairs of data, we mainly focused on the association between CX 10, and 13 and infiltrating immune cells ( Figure 16A,B,C). Other related data are sh in Material S6.

Discussion
The imbalance of CXC expression has a considerable impact on tumorigenesis, tumoral cell proliferation, apoptosis, and tumor metastasis. Intercellular communications between stromal cells and tumor cells affect the expression of CXCs in various types of cells, thus regulating tumor metastasis and invasion. Some studies have already shown correlations between CXCs and the tumor microenvironment, suggesting that CXCs can regulate tumor progression and immunotherapy. Our previous studies have shown that the protection of colorectal cancer cells from radiotherapy by CXCL12/CXCR4 is mediated by survivin in colorectal cancer [12]. CXCL10 is considered a potential therapeutic target for melanoma [30]. The application of CXCL8 for the diagnosis of CRC is more practical than the use of the classical tumor marker CEA. Serum CXCL8 may be a potential biomarker of colorectal cancer progression [31]. Some studies have demonstrated unique weak binding between CXCL8 and CXCR2 and interaction between CXCR2 and G proteins [32]. However, there is a lack of a bioinformatics analysis that demonstrates the prognostic values and biological functions of CXCs in multiple tumors. In this study, we demonstrated abnormal expression of CXCs in 20 types of cancer and significant differences in the mRNA expression of the CXC family members that have significant prognostic value in breast cancer, colon -cancer, and pancreatic cancer. This study is the first to suggest that the CXCL family may be involved in the interactions between intestinal flora and colonic epithelium of the host. We hope that our findings will help to improve our understanding of the roles of the CXCL family members and improve treatment design and the accuracy of prognosis in patients with these tumors.
We initially investigated the expression of CXC chemokines and their relationships with pathological stages of the tumors. We found that nine genes were differentially expressed in breast cancer versus normal tissues (CXCL9, 10, 11, and 13 were up-regulated, and CXCL1, 2, 3, 12, and 14 were downregulated). Additionally, we demonstrated that the expression of CXCL1, 2, 5, 8, 12, 13, and 14 was closely associated with the stage of breast cancer. Similarly, 11 genes were differentially expressed in colon cancer (CXCL1, 2, 3, 5, 8, 9, 10, and 11 were up-regulated, and CXCL11, 12, 13, and 14 were downregulated). The development of tumors was associated with an increase in the expression of CXCL9, 10, and 11. The results on pancreatic cancer data showed that 12 genes were up-regulated (CXCL1, 2, 3, 5, 6, 8, 9, 10, 13, 14, 16, and 17). The expression of CXCL1, 3, 5, and 8 were associated with the stages of pancreatic cancer. These data suggest that differentially expressed CXC chemokines may play important roles in these three types of tumors.
Analysis using large groups of patients with breast, colon, or pancreatic cancer in the K-M plotter database indicated that a number of CXC family members were significantly associated with survival and had specific associations. In patients with colon cancer, the survival time of patients with higher levels of expression of CXCL1, 3, 8, 10, and 14 was longer than that of patients with lower expression. In pancreatic cancer patients, the survival time of patients with higher levels of expression of CXCL5, 8,9,10,11, and 17 was remarkably shortened. In breast cancer, the groups with higher expression of mRNAs of CXCL2, 6, 9, 10, 12, 13, and 14 and the groups with lower expression of CXCL3, 8, and 17 had significantly better overall survival (OS).
There are contradictory evidences in the role of CXCL8 in the development and progression of colon cancer. High amounts of serum CXCL8 prevent liver metastasis of CRC and are correlated with good favorable prognostic outcomes [33]. In contrast, elevated CXCL8 levels promote carcinogenesis and are associated with poor prognosis [34]. The analysis showed that the expression of CXCL8 in colon cancer was higher than in normal controls, and patients with high expression of CXCL8 in colon cancer had a longer survival time. In breast cancer and pancreatic cancer patients, the expression of CXCL8 was higher in the tumor tissues; however, the OS time of patients who had a higher expression of CXCL8 was significantly shorter. This contradictory phenomenon reflects the complex role of CXCL8 in the occurrence and development of colon cancer. The intestinal microflora is closely linked to colonic disease. Colonic tissue directly interacts with intestinal flora, and multiple studies noted that intestinal microorganisms play a significant role in the development and progression of colon cancer [35], inflammationrelated colon cancer [36], the colon cancer microenvironment [37], and colon cancer drug resistance [38]. Network analyses were carried out using the Amadis database analysis tool, and a possible association between CXCL8, Fusobacterium nucleatum, and human diseases (including inflammatory bowel disease and colon cancer) was found. We investigated the results of sequencing obtained after coculture of Fusobacterium nucleatum with colon cancer cell lines and determined that the expression of CXCL8 was significantly increased, and the expression of CXCL10 was decreased in the HT29 cell line cocultured with Fusobacterium nucleatum. Fusobacterium nucleatum may change tumor proliferation, invasion, metastasis, and drug resistance by increasing the expression of CXCL8 and reducing the expression of CXCL10, thus affecting the prognosis of patients. Combining these data with the data on differential expression of miRNAs in SW480 cells cocultured with Fusobacterium nucleatum indicated changes in the expression levels of transcription factors related to CXCL8 and 10, transcriptional suppressors, and miRNAs acting on the corresponding 3'-UTR of mRNAs. At present, there are no reports on the influence of the intestinal flora on the prognosis of colon cancer patients and the progression of colon cancer mediated by the expression of chemokines. Our analysis demonstrated that Fusobacterium nucleatum might influence the changes in the chemokine family members at the transcriptional and posttranscriptional levels and, thus, influence the development and progression of colon cancer and the prognosis. Although the expression of CXCL8 and 10 increased significantly, their role in the process of F. nucleatum aggravating CAC remains unclear. Future experimental verification of the mechanism may identify new pathogenic pathways and therapeutic targets.
Co-expression analysis revealed that the co-expression of CXCs might be related to their chromosomal locations, and this co-expression might be regulated by transcription factors. Interestingly, in the analysis of the expression of CXCRs, we found that there was no significant difference in the expression of CXCRs in tumor tissue and normal tissue, and there was no diagnostic or prognostic value of CXCRs. Through the single-cell sequencing database data, we found that CXCRs are mostly distributed on the surface of immune cells, which may indicate that CXCs play a role in recruiting immune cells by binding to receptors on the surface of immune cells.
Our study has certain limitations. The results at the transcriptional level can reflect the immune status; however, this analysis cannot reflect the overall changes. Independent cohort studies should be performed to verify our results. The CXC family members may play a dual role in disease progression. Increased expression of CXCs in tumor tissues may promote carcinogenesis and regulate the tumor microenvironment; however, in some tumors, high expression of the CXC family members may suggest a better overall survival time. Most of the results were predicted by bioinformatics analysis, and, as a result, further experiments in vitro or in vivo are needed to demonstrate the associations between these factors.

Conclusions
In this study, we systematically analyzed the expression and prognostic value of CXCs in a variety of tumors and provided a thorough evaluation of the heterogeneity and complexity of the molecular and biological characteristics of the tumors. High expression of certain CXCs can be used as a molecular marker to identify tumor patients in high-risk groups. This is the first study to propose the theory that intestinal flora may influence disease by influencing the transcriptional changes in CXCs, thus providing a direction for further research. Our results indicate that CXCs are a potential therapeutic target in a variety of tumors and a potential prognostic marker to improve the survival of cancer patients and accuracy of prognosis, and they may be involved in diseases caused by intestinal flora.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/10 .3390/cancers13164153/s1, Material S1: Subpopulation distribution of CXCs in single-cell sequencing datasets of the three cancers. Material S2: Expression differences of CXCRs in three cancers. Material S3: MiRNAs predicted binding to CXCL8 and 10 by miRWalk database. Material S4: CXC gene expression and mutation analysis in pancreatic cancer and colon cancer. Material S5: Correlation analysis of CXCRs and CXCs in colon cancer. Material S6: Correlation of other differentially expressed CXCs and immune cell infiltration in BRCA, COAD, and PDAC. Material S7: Distribution of CXCRs in colon cancer; Material S8: The original Western blot images of (I&K).