Cytotoxic T-Cell Trafficking Chemokine Profiles Correlate With Defined Mucosal Microbial Communities in Colorectal Cancer

The involvement of gut microbiota in T-cell trafficking into tumor tissue of colorectal cancer (CRC) remains to be further elucidated. The current study aimed to evaluate the expression of major cytotoxic T-cell trafficking chemokines (CTTCs) and chemokine-associated microbiota profiles in both tumor and adjacent normal tissues during CRC progression. We analyzed the expression of chemokine C-X-C motif ligands 9, 10, and 11 (CXCL9, CXCL10, and CXCL11), and C-C motif ligand 5 (CCL5), characterized gut mucosa-associated microbiota (MAM), and investigated their correlations in CRC patients. Our results showed that the expression of CXCL9, CXCL10, and CXCL11 was significantly higher in tumor than in adjacent normal tissues in 136 CRC patients. Notably, the high expression of CXCL9 in tumor tissues was associated with enhanced CD8+ T-cell infiltration and improved survival. Moreover, the MAM in tumor tissues showed reduction of microbial diversity and increase of oral bacteria. Microbial network analysis identified differences in microbial composition and structure between tumor and adjacent normal tissues. In addition, stronger associations between oral bacteria and other gut microbes were observed. Furthermore, the correlation analysis between the defined MAM and individual CTTCs showed that the CTTCs’ correlated operational taxonomic units (OTUs) in tumor and adjacent normal tissues rarely overlap with each other. Notably, all the enriched OTUs were positively correlated with the CTTCs in either tumor or adjacent normal tissues. Our findings demonstrated stronger interactions between oral bacteria and gut microbes, and a shifted correlation pattern between MAM and major CTTCs in tumor tissues, underlining possible mechanisms of gut microbiota–host interaction in CRC.


INTRODUCTION
Colorectal cancer (CRC) accounts for the third most commonly diagnosed cancer and the second leading cause of cancer-related death worldwide (1). High infiltration with cytotoxic T cells (CTCs) correlates with improved relapse-free and overall survival (OS) in patients with CRC (2)(3)(4). Therefore, recent therapeutic strategies for cancer such as immunotherapies focus on CTC trafficking to the tumor site (5)(6)(7)(8). Different from the remarkable treatment responses of adoptive immunotherapy in patients affected by advanced melanoma and hematologic malignancy, the effect of immunotherapy against CRCs has been more moderate (9). One of the major challenges is to effectively traffic CTCs to the tumor microenvironment (10,11).
The expression of chemokine receptors in CTCs, as well as the expression of their ligands in tumor tissues, is essential for localizing CTCs to tumor tissue (5,12,13). It has been reported that CTCs with higher CXCR3 expression can be recruited to the tumor site by ligands, including CXCL9, CXCL10, and CXCL11, which are known as IFNg-inducible chemokines (12,13). Additionally, studies have identified CCL5 and its receptor (C-C motif) receptor 5 (CCR5) as another critical component of T-cell chemotaxis that is closely associated with CTC infiltration and better survival (12). Therefore, expression of specific chemokines in tumors could be potentially correlated with the presence of CTCs, which might serve as useful targets for anti-cancer therapies.
The gut mucosa is a dynamic interface between host cells and microbiota (14). Progress of colorectal neoplasia has been linked to interactions between tumor microenvironment and mucosal microbiota barrier, whose process can be reversed by interfering the microbiota (15)(16)(17). In mouse models, mixture of microbes enhances anti-cancer immunity through inducing interferon-gproducing CTCs in the tumor tissues (18), pointing to plausible evidence for the use of gut microbiota as a therapeutic target. Additionally, it has recently been demonstrated that bacteria isolated from CRC tissues could upregulate expression of most types of CTTCs in CRC cell lines in vitro (19). However, it remains to be elucidated the change of microbiome profiles during the transition from normal mucosae to malignant lesions, and the correlation between CTTCs and defined microbial communities in the CRC tumor microenvironment. Although enrichment of Fusobacterium and its regulation of tumor microenvironment have been demonstrated in CRC (20,21), increasing evidence suggests that microbiota work as a community with nonnegligible contribution from various microbes (22)(23)(24), which remains to be clarified. Herein, we investigated the microbial transition in the tumor mucosae and adjacent normal mucosae, and the association between bacterial colonization and CTTCs in CRC patients.

Patient Recruitment and Sample Collection
A total of 136 CRC patients scheduled for a primary resection of their tumor at the Affiliated Hospital of Jiangnan University between 2016 and 2019 were recruited in the study. The participants did not receive chemo-radiotherapy before the resection and were not treated with antibiotics in the month prior to surgery but were administered antibiotics intravenously within a few hours of the resection. After surgery, there were 136 pairs of fresh tissues from colorectal tumor or as far away from the tumor as possible (adjacent tissue) collected. Biopsies were snap-frozen in a cryovial immediately with liquid nitrogen and then stored at −80°C until DNA extraction. Histopathological and clinical features were scored according to the International Union Against Cancer (UICC)-TNM staging system. This study was approved by the Ethics Committee of Jiangnan University and was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all of the participants after explanation of the nature of the study.

DNA Extraction and 16S rRNA Gene Sequencing
Paired mucosae samples from tumor and adjacent normal tissues were subjected to DNA extraction. Mucosal DNA was extracted using the AllPrep DNA/RNA extraction kit. Total DNA was purified from tumor and paired normal adjacent tissue samples. MAM was analyzed based on 16S rRNA gene sequencing. 16S rRNA gene amplicon sequencing was carried out employing the 16S rRNA gene Sequencing Library Preparation protocol developed by Illumina (San Diego, California, USA). Briefly, 200 ng of mucosal DNA was amplified from each sample using the primers 515F (5′ GTGCCAGCMGCCGCGGTAA 3′) with Titanium Adaptor B and 806R (5′ GGACTACHVGGGTWT CTAAT 3′) with Titanium Adaptor A and a sample-specific barcode sequence targeting the V4 hypervariable region of the 16S rRNA gene using FastStart Taq DNA Polymerase (Roche, USA).

Immunohistochemistry and Immunofluorescence Staining and Image Analysis
Specimens used for immunohistochemistry and immunofluorescence staining were obtained immediately after surgical resection and fixed in 10% neutral formalin, paraffin-embedded, and used for histological assays as previously described (25). Immunohistochemistry of paraffin sections was carried out using a two-step protocol. Briefly, 5-mm paraffin sections were first deparaffinized and hydrated, and endogenous peroxidase activity was blocked by incubating the slides in 0.3% H 2 O 2 . Antigen retrieval was performed by microwave treatment in citrate buffer, pH 6.0. Sections were blocked with normal sera from the same species from which the secondary antibodies were derived. After overnight incubation at 4°C with antibodies against human CD8 (1:300 dilution, Abcam, ab101500), CXCR3 (1:300 dilution, ABclonal, A2939), or control antibodies (Rabbit monoclonal IgG, Abcam, ab172730), we applied Envision System HRPlabeled polymer anti-rabbit (for CD8 and CXCR3) (Dako Cytomation) for 30 min and diaminobenzidine (5 min) and hematoxylin counterstain (1 min). Slides were scanned by an automated scanning microscope (Pannoramic Digital Slide Scanners, 3DHISTECH). The 3DHISTECH software (CaseViewer) was used to count the number of positive signals in each tissue core. We calculated the average density (cells/ mm 2 ) of each tumor-infiltrating CD8 + T-cell subset or CXCR3 + subset.
For multiple-color immunofluorescence staining, formalinfixed, paraffin-embedded sections were deparaffinized and rehydrated. Antigen retrieval was performed by microwave treatment in citrate buffer, pH 6.0, and blockage of nonspecific antibody binding was carried out with 5% BSA. Sections were then incubated with anti-human CXCR3 and CD3 overnight at 4°C, followed by specimen-paired immunofluorescence secondary antibodies. Negative controls were generated by replacing primary antibodies with isotypematched antibodies. Slides were analyzed on a fluorescent imaging microscope.

Bioinformatics Analysis
The Quantitative Insights into Microbial Ecology version 2 (QIIME2) software (subversion 2019.1) was used for quality filtering of DNA sequences, demultiplexing, taxonomic assignment, and calculating aand b-diversity. For details, selected sequences were clustered into OTUs with USEARCH (version 11, http://drive5.com/uparse/), with a threshold sequence identity of 99%. The reads were aligned to the Greengenes Core Set reference alignment using PyNAST. The Greengenes taxonomies were used to generate summaries of the taxonomic distributions of OTUs across different levels (phylum, order, family, genus, and species). A phylogenetic tree was built with FastTree and used for estimates of a-diversity (Rarefaction curves, Chao1, Shannon diversity) and b-diversity (using unweighted UniFrac). Metagenomic content of the microbiota samples was predicted from the 16S rRNA profiles, and KEGG pathway functions were categorized at Levels 1-3 using the phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt) tool (26).

Statistical Methods
All data were summarized as means ± SEM and analyzed with SPSS software (Version 22). Comparisons between CTTC groups were performed by Student's t-test. Correlations between continuous variables were determined by Pearson correlation analysis. Survival was estimated by the Kaplan-Meier method and compared by the log-rank test. Multivariate analysis of prognostic factors for OS was performed using the Cox proportional hazards model. The c 2 test was used to test for relationships between categorical variables. Values of p < 0.05 (two-tailed) were considered significant.
For microbiota, differential abundance analyses were performed using paired Wilcoxon signed-rank test to identify significant changed features between tumor and adjacent normal mucosae. Using the R implementation of Random Forests 10fold cross-validations with 100 iterations, we selected a minimum set of bacterial taxa that maximally discriminated against each Dukes stage, different patients' survival state, and lymph node metastasis state; the variable importance of a microbial taxon was determined by 100 iterations of the algorithm with 3,000 trees and the default mtry of p 1/2 , where p is the number of input phylotypes.

Expression of CTTCs in CRC Tissues and Their Association With Disease Progression
CTTC mRNA expression in tumor and adjacent normal tissues from 136 CRC patients was analyzed using quantitative PCR. The expression of CXCL9, CXCL10, and CXCL11 except for CCL5 was significantly higher in the tumor compared with the adjacent normal tissues (Figures 1A-D; p = 0.0707, p = 0.0001, p < 0.0001, and p = 0.4207). Correlation analysis showed that several chemokines were significantly associated with one another in adjacent normal tissues but not tumor tissues, indicating that different chemokines might be regulated simultaneously in adjacent normal tissues during tumor progression ( Figure 1E). In CRC, the expression level of CTTCs in tissues can significantly impact the distribution of CTCs and then the patients' clinical outcome (2)(3)(4). Therefore, we further examined the expression level of CTTCs from patients at different stages of CRC. In tumor, the expression level of chemokine CXCL9 and CXCL11 was significantly decreased in advanced stage CRC patients (stages IV, n = 11; p = 0.02 and p = 0.045; Figure 1F) compared to those in early stages (stages I, II and III, n = 18). Moreover, the expression level of chemokine CXCL11 and CCL5 in the adjacent normal tissues was significantly lower in advanced stage CRC (stages IV, n = 11; p = 0.027 and p = 0.0001; Figure 1G) than those in early stages (stages I, II, and III, n = 18). In addition, no significant difference was observed in the expression level of CXCL10 in the tumor or adjacent normal tissues between patients in advanced stage and those in early stages (p = 0.127 for tumor and p = 0.167 for adjacent normal; Figures 1F, G). Collectively, the results indicated that the three CTTCs (CXCL9, CXCL11, and CCL5) were selectively regulated in tumor or adjacent normal tissues and decreased with progressive stages in CRC patients.

Association Between CTTC Expression and Local CD8 + T-Cell Infiltration
The CRC patients were then categorized into two groups according to the expression level of CTTCs (CXCL9, CXCL10, CXCL11, and CCL5) in tumor. To evaluate the potential role of these chemokines in the localization of immune cells, immunofluorescence staining was applied to examine the in situ infiltration of CXCR3 + and CD3 + cells in CRC patients. CXCR3 + cells were enriched in tumor tissues from CXCL9 high patients, which also was the hotspot for CXCR3 + CD3 + cells ( Figure 2A). Additionally, the infiltration density of CD8 + T cells in tumor and adjacent normal tissues was assessed and compared between patients with high and low CTTC expression ( Figures 2B-D). We found that the density of CTCs was significantly higher in tumors in the CXCL9 high group than that in the CXCL9 low group ( Figure 2C). In adjacent normal tissues, the density of CTCs was higher in the CCL5 high group than in the CCL5 low group ( Figure 2D). Taken together, the results showed that the CXCL9 high group in tumor or CCL5 high group in adjacent normal tissues exhibited relatively higher CTC infiltration than the matched chemokine low group, implicating altered anti-tumor immune activity.

Prognostic Significance of CTTCs in CRC Patients
To address whether CTTC expression was associated with CRC progression, we analyzed the correlation between relevant   Multivariate Cox proportional hazards analysis was then performed, with variables associated with survival in univariate analysis adopted as covariates. In multivariate analysis shown in Table 2, the CXCL9 expression level in tumor could not emerge as an independent prognostic factor of both OS (HR, 0.205; 95% CI, 0.043-0.988; p = 0.062) and DFS (HR, 0.443; 95% CI, 0.265-0.742; p = 0.065). These results suggested that chemokine CXCL9 was significantly associated with CRC progression, but might not serve as a powerful predictor of CRC survival alone.

Difference of MAM Between Tumor and Adjacent Normal Mucosae
As gut MAM could serve as stimulation that impacted the chemokine expression, we next continued to identify MAM in tumors and paired adjacent normal mucosae from the CRC patients. To investigate how MAM changed in tumors compared with adjacent normal tissues, we compared paired tumor and adjacent normal tissues from 101 patients in our CRC cohort ( Table 3). Wilcoxon signed-rank tests showed that alpha-diversity indices, including the Shannon and Simpson indices and Pielou's evenness, were significantly decreased in tumors (p < 0.0001, p = 0.0041 and p = 0.0014, respectively, Figures 4A-C), while richness indices, such as ACE, Chao, and phylogenetic diversity (PD) whole tree, were also decreased in tumors (p < 0.0001, p < 0.0001, and p < 0.0001, Figures 4D-F) compared with paired adjacent normal tissues. As for beta-diversity, principal coordinate analysis (PCoA) could not separate the microbiomes from tumor and adjacent normal mucosae into different clusters, which could be due to significant inter-individual variation (data not shown).  The bold face indicates p-values of variables with significance in either univariate or multivariate analyses.

Large Centralities of Oral Bacteria in CRC MAM Network
The structure change of the mucosae microbiota is the result of dynamic interactions between community members. The SparCC algorithm was employed to construct microbial interaction networks. Here, we observed both co-occurrence and coexcluding interactions of significantly different OTUs between the tumor and adjacent normal mucosae ( Figures 6A, B and Supplementary Figures S3A, B). As shown in Figure 6A, the microbial network interactions in tumors mainly occurred among genera belonging to phyla Firmicutes and Proteobacteria. Notably, co-occurrence interactions among oral bacteria Fusobacterium, Peptostreptococcus, Parvimonas, Gemellaceae, and Campylobacter, and between Fusobacterium and Clostridium were observed in both tumor and adjacent normal network. In addition, cooccurrence interactions between oral bacteria Fusobacterium and Lachnoanaerobaculum, between Fusobacterium and Bulleidia, and between Lachnoanaerobaculum and Selenomonas were exclusively observed in tumors. The mucosae microbiota in adjacent normal tissues was also dominated by Firmicutes and Proteobacteria and consisted mainly of the six phyla observed in the tumor group ( Figure 6B). However, the richer interaction network incorporated more commensal bacteria, with interactions among the family S24-7, Streptophyta, and Acinetobacter presented in adjacent normal tissues. Weighted node connectivity (WNC) scores were calculated to determine hub microbes with significant roles in CRC microbial ecological network. To prioritize differentially abundant taxa, we next focused on OTUs with large WNC scores, which indicated large centralities and importance in the CRC MAM interaction network. These taxa with large centralities included genera Peptostreptococcus, Fusobacterium, Parvimonas, Bulleidia, Rhodoplanes, Candidatus Solibacter, Dialister, Lachnoanaerobaculum, and Selenomonas in tumors ( Figure 6A). Species-level identification of these OTUs included Clostridium aldenense, Lachnoanaerobaculum orale, Bacteroides fragilis, Ruminococcus bromii, Bacteroides uniformis, and Parabacteroides distasonis (Supplementary Figure S3A). Their centralities suggested that they can form a backbone of niche-specific relationships and might exhibit significant influence on the tumor microbial ecology.   Figure S4). Functions enriched in adjacent normal tissues compared to tumors at the L3 level such as pathogenic Escherichia coli infection, Fc gamma R-mediated phagocytosis, and p53 signaling pathway (Supplementary Figure S5). In contrast, 18 pathways at the L3 level were enriched in tumors compared to adjacent normal tissues, such as MAPK signaling pathway, Fc epsilon RI signaling pathway, Carbohydrate digestion and absorption, Lipopolysaccharide biosynthesis, Nucleotide metabolism, D-Alanine metabolism, Epithelial cell signaling in Helicobacter pylori infection, Bacterial toxins, Amino acid metabolism, and Antigen processing and presentation.

Distinctive CTTC Expression Profiles Were Associated With MAM
To assess the correlations between the relative abundance of defined MAM and the expression of individual CTTCs in CRC, correlation was analyzed between the expression of CTTCs and the abundance of differently enriched OTUs in tumor compared to adjacent normal mucosal tissues. CTTCs including CXCL9, CXCL10, CXCL11, and CCL5, which were differentially expressed between tumor and adjacent normal tissues ( Figures 1A-D), were significantly correlated with the abundance of several OTUs ( Figures 7A-H and Additional File 7). In particular, abundance of several OTUs, mainly including Acinetobacter lwoffii, species in Wautersiella genus, and Desulfobacteraceae family, was significantly correlated with expression of at least two kinds of CTTCs in tumor tissues. It has been reported that Methylobacteriaceae was associated with all prognostically favorable T-cell markers and most corresponding recruiting chemokines (19). In line with this, Methylobacterium adhaesivum was associated with the high expression of CXCL10 in tumor tissues. Moreover, Streptomyces mirabilis, Psychrobacter marincola, Acinetobacter johnsonii, Psychrobacter sanguinis, and species in Photobacterium and Acinetobacter genus and SHA-31 family were significantly correlated with expression of at least two kinds of CTTCs in adjacent normal tissues. Although for each of the CTTCs, their correlated OTUs in tumor and adjacent normal tissues rarely overlap with each other, we still found several OTUs that were correlated with CTTCs in both tumor and adjacent tissues such as Desulfosporosinus meridiei, Rothia mucilaginosa, and Haemophilus parasuis, and species in B-42 and Catonella genus, Neisseriaceae and ML1228J-1 family, and Spirobacillales and GN03 order ( Figures 7A-H). Notably, taxa that have been reported to be associated with CRC progression (22), such as species in Fusobacterium and Parvimonas genus, were only positively correlated with the CTTCs in tumor (Figures 7A-D).
Notably, taxa that have been reported to be associated with CRC progression (22), such as species in Fusobacterium and Parvimonas genus, were only positively correlated with the CTTCs in tumors (Figures 7A-D).
We then analyzed how the MAM was altered during CRC progression. Changes in the microbial communities were identified by comparing the patients at low Dukes' stages (stages I-II) with those at high Dukes' stages (stages III-IV) and linear discriminant analysis (LDA) effect size (LEfSe) to compare the taxonomic abundance. In addition, we added the chemokine levels into the risk index to generate the ROC curves ( Supplementary Figures S6A-D). Our results showed that the changes in abundances of specific taxa in tumor or adjacent normal mucosae can be used as a classifier that distinguishes between low/high Dukes' stage at a fixed specificity of 81.1% and 81.4% (Addition File 7). Besides, the addition of CTTC levels into the risk index would slightly improve the specificity at 83.0% and 88.3%.

DISCUSSION
In the present study, we aimed to explore the chemotactic factors associated with CRC prognosis and to profile MAM of CRC patients and to investigate the relationship between mucosae microbiota and chemokines involved in CTC recruitment during CRC progression. Based on existing reported clinical data and evidence, we explored major chemokines (including CCL5, CXCL9, CXCL10, and CXCL11) associated with recruitment of CTCs into CRC tissues (5,12,(27)(28)(29).
In our current cohort, significant positive correlations between expression of CXCL9, CXCL10, and CXCL11 were detected in adjacent normal tissues, suggesting that these chemokines, targeting the same chemokine receptor, may be concomitantly induced and regulated, pointing to their coexpression in normal tissues. Notably, such co-expression was dramatically weakened in tumors, indicating dysregulation of the CTTCs in the tumor microenvironment. Particularly, we found that in tumors, the expression of CXCL9 negatively correlated with tumor stage, nodal and distant metastases, and Dukes' stage, A B FIGURE 6 | Correlation network of differential MAM in tumor (A) and adjacent normal mucosae (B). The correlation coefficients were calculated with the Sparse Correlations for Compositional data algorithm (SparCC). A subset of significant correlations with strengths of at least 0.45 in the mucosae commensals at the genera level were selected for visualization using Cytoscape version 3.6. The size of the nodes corresponds to weighted node connectivity (WNC) scores. Red and green lines represent positive and negative correlations, respectively. and predictive of favorable clinical outcome. Due to the limitation of follow-up time, we could only evaluate the impact on identified chemokines on patients' 3-year survival instead of 5-year survival in our cohort. Therefore, more follow-up work needs to be completed in the future in order to assess the prognostic value of these CTTCs on the long-term survival and DFS of patients. Previous reports showed that the positive prognostic significance of these chemokines could rely on their capacity to attract the T-cell populations with receptors into tumor tissues (5,12,13). In consistent with previous findings, our data also showed that tumors with high CXCL9 expression level or adjacent normal tissues with high CCL5 expression level were infiltrated with a higher number of CTCs. It should be noted that a higher-level infiltration of CD8+ T cells does not directly confirm a higher immune activity before further functional validation. Nevertheless, patients with better prognosis outcome might benefit from higher levels of CTTCs within CRC tissues, and more CTC infiltration could possibly be   (30), indicating that certain bacterial species could be useful microenvironmental stimuli and dynamically regulate CD8 + Tcell function. These findings, taken together, underscore the possibility of CD8 + T cells' functional recovery by gut bacteria intervention, as well as the importance of potential mechanisms on how the microbiome may alter anti-tumor response by CD8 + T cells via the chemokine-chemokine receptor axis.
Our CRC patients showed heterogeneous CTTC expression and its dysregulation in tumors compared to adjacent normal tissues. Such changes might reflect distinct genetic backgrounds and/or exposure to different microorganisms in the tumor microenvironment. Recent studies reported that stimulation by gut commensal bacteria induced upregulation or de novo expression of multiple chemokines in tumor cell lines (18,19). However, specific microbes in MAM correlated to high chemokine expression, and immune cell infiltration in human CRC samples was not evaluated by far. In the current study, the gut MAM in tumor and adjacent normal tissues showed distinct taxonomic composition, with decreased diversity and richness in tumor tissues. Our findings in CRC are thus in line with a previous hypothesis that altered microbial diversity can be recognized as a feature of disease status, including inflammatory diseases and cancer (31)(32)(33).
In addition to confirmation of previously reported association between microbes and CRC, our MAM analysis also revealed novel taxonomic changes in the disease. At the phylum level, Fusobacteria was substantially enriched in tumor MAM in our CRC patients, in consistent with previous reports (34)(35)(36)(37). Notably, Synergistetes could be a novel phylum significantly enriched in tumors, although its abundance was much lower compared with that of Fusobacteria in the CRC patients. Synergistetes has not previously been reported as a tumorenriched phylum in gut microbiome studies with European and American CRC cohorts (22,22,35), suggesting that this phylum might be a characteristic in our local Chinese CRC cohort. At the genus level, Fusobacterium has been reported as the most abundant genus, which resides in the oral cavity as commensals, but can be an opportunistic pathogen for colon carcinogenesis via alterations in signaling pathways or impairment of antitumor immune functions (20). In addition, it has been demonstrated that Fusobacterium nucleatum plays a role in the development and progression of colon cancer (20,(38)(39)(40), and it has also been detected in patient samples with nodal and distant metastasis (41)(42)(43)(44). In our current study, we also confirmed a significantly increased abundance of Fusobacterium in tumor compared with adjacent normal tissues and in advanced stage CRC patients compared with early-stage CRC patients, suggesting it as an important and dominant candidate pathogen in both occurrence and development of CRC. Moreover, B. fragilis was shown in our data to be detected in 95% of CRC tumor tissues, and it could possibly stimulate an inflammatory status that can promote carcinogenesis via induction of proinflammatory toxins as reported (45).
It is noteworthy that our findings demonstrate substantial alteration of MAM microbial functions in CRC patients. Pathways related to genetic information processing and bacterial toxin biosynthesis were also found at higher abundance in tumor, which was in line with the increase of bacteria that could synthesize proinflammatory toxins, such as B. fragilis. In contrast, microbial pathways associated with metabolism were downregulated in tumor. Such findings, in line with a previous report in other cohorts (22), indicate that during CRC progression, altered microbiome could be involved in shifted host immune response and metabolism, which were key components of carcinogenesis and the maintenance of local microenvironment of CRC.
Previous reports showed that CTTCs might predict favorable outcome in CRC patients (3,4). However, our data showed that CTTCs could not serve as a powerful predictor of patient survival. Moreover, our results revealed that a set of defined microbes positively associated with the expression of CTTCs in CRC patients, suggesting that intervention of gut microbiota might be useful in targeting the balance of the tumor microenvironment of CRC. Notably, a group of defined microbes in tumor and adjacent normal mucosae were associated with expression of CTTC genes, possibly indicating their capacity to promote recruitment of CTC populations. Most of the correlations between OTUs and CTTCs in tumor and adjacent normal tissues were not found in the other tissue types, in line with the dysregulation of CTTCs in tumors. Interestingly, Methylobacteriaceae was positively correlated with CXCL10 in both tumor and adjacent normal mucosa in our CRC patients, and it has been reported to be associated with prognostically favorable T-cell markers and most corresponding recruiting chemokines (19). The results suggested that in view of the complex tumor microenvironment, both cancer-promoting and cancer-suppressing factors could co-exist in MAM. Considering that species in Fusobacterium genus was one of the mucosal bacteria that promoted tumor development, this could partially explain why the chemokine CXCL9 could not serve as a powerful independent predictor of our CRC patient survival. Importantly, among the microbes positively associated with CTTCs such as CXCL9, there were a set of gram-negative bacteria. The enriched gram-negative bacteria in tumors could possibly stimulate an inflammatory state, induce the increase of CTTCs in the tumor microenvironment, and affect the progression and prognosis of CRC patients. Such findings might improve our understanding of the microbiota dynamics along CRC progression and provide new insight into the development of treatment strategies with immune therapy. Further studies are thus warranted to clarify the species-level or strain-level impact of MAM microbes on chemokine secretion in tumor tissues and T-cell infiltration.
In conclusion, the current study reveals the significant correlation between cytotoxic trafficking chemokines and gut MAM in CRC patients and improved prognosis. This knowledge might eventually pave the way towards development of innovative treatments by modifying gut microbiota to promote cytotoxic T-cell infiltration for the favorable prognostic significance.

DATA AVAILABILITY STATEMENT
Our data has been successfully deposited to NCBI. The accession number is: PRJNA669258.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Medical Ethics Committee of Jiangnan University. The ethics committee waived the requirement of written informed consent for participation.

AUTHOR CONTRIBUTIONS
JZ and J-HC designed the study, analyzed the data, and wrote the paper. CQ and R-NG developed the clinical sample cohorts, and JT, C-YR, R-NG, and Q-CL managed collection of clinical samples and information, DNA extraction, and quantification. JZ analyzed the NGS data. JZ and K-WW performed the statistical analysis. GY, Y-SL, and J-HC, in addition to all other co-authors, reviewed the manuscript, provided feedback, and approved the manuscript in its final form. All authors contributed to the article and approved the submitted version.

FUNDING ACKNOWLEDGMENTS
We are grateful to the participants that have made this research possible.

SUPPLEMENTARY MATERIAL
The