The study population comprised 99 treatment-naïve patients with CRC who underwent surgical resection between January 2006 and December 2012 ( Table 1 ). All authors followed good clinical practice, and the study was conducted according to the principles of the Declaration of Helsinki. The study protocol was approved by the institutional review board of Severance Hospital (IRB no: 4-2018-0291).

The bacterial strain F. nucleatum (ATCC 25586) was obtained from the American Type Culture Collection (ATCC). The strain was cultured on Brucella agar plates with 5% sheep blood and incubated in a Coy chamber (Coy Laboratory Product, USA) for 3 days at 37°C under anaerobic conditions (atmosphere: 20% CO₂, 5% H₂, and 75% N₂). A single colony was inoculated into a Gifu anaerobic medium (GAM) broth (KisanBio, Korea) and incubated at 37°C for 18 hr. Bacterial cells were centrifuged at 2,000 х g for 5 min at 4°C and washed twice with PBS. For calculating colony-forming units per millimeter (CFU/mL) of a fluid, the pellet was serially diluted 10-fold in fresh GAM broth and spread onto the Brucella agar plate. The mean colony counts were 1.0 x 10⁹ CFU/mL.

Genomic DNA (gDNA) was isolated from 99 clinical tumor tissues using a QIAamp DNA Mini kit (Qiagen, UK) according to the manufacturer’s protocol. Briefly, the samples were suspended with protease K in ATL buffer and incubated at 55°C for 2 hr. Both AL buffer and absolute ethanol were added to the samples before applying the QIAamp spin column. Each sample was centrifuged and washed. DNA was eluted from the column with 50 μL of the supplied AE buffer. The quality and quantity of isolated DNA were determined using a NanoDrop spectrophotometer (ND-1000; Thermo Scientific, USA). All samples were stored at -20°C until used.

FN detection from 99 tumor samples was evaluated by quantitative real-time PCR using the TaqMan assay system, as previously described (Lee et al., 2021). Each reaction was conducted in a final volume of 20 μL reaction containing 1× TaqMan Universal PCR Master Mix (Applied Biosystems, USA), 300 nM of each primer, 200 nM TaqMan probe, and 30 ng of gDNA in a 96-well optical PCR plate. Amplification, detection, and data analysis were performed using the StepOnePlus Real-Time PCR System (Applied Biosystems, USA) as follows: 10 min pre-incubation at 95°C, amplification cycles of 95°C for 15 s, and 60°C for 1 min were repeated 50 times. The cycle threshold (Ct) values for FN were normalized to the amount of gDNA in each reaction, using prostaglandin transporter (PGT) as a human reference gene. The primer and probe sequences were as follows: 16S ribosomal RNA (16s rRNA) gene DNA sequence of FN forward primer, 5′-GGATTTATTGGGCGTAAAGC-3′; FN reverse primer, 5′-GGCATTCCTACAAATATCTACGAA-3′; FN FAM probe, 5′-CTCTACACTTGTAGTTCCG-3′; PGT forward primer, 5′- ATCCCCAAAGCACCTGGTTT-3′; PGT reverse primer, 5′- AGAGGCCAAGATAGTCCTGGTAA-3′; PGT FAM probe, 5′- CCATCCATGTCCTCATCTC-3′.

Total RNA was prepared from 99 clinical tumor tissues using the RNeasy Mini kit (Qiagen, UK) according to the manufacturer’s protocol. Briefly, clinical samples were lysed with a reagent in a microcentrifuge tube, and 70% ethanol was added to the lysate. After vortexing for 10 s, the lysate was transferred to a RNeasy spin column and centrifuged at 8,000 x g for 15 s. The spin columns were then washed with RW1 and RPE buffer. Total RNA was eluted in 50 μL RNase-free water and stored at -80°C until use. The concentrations of total RNA were determined using a NanoDrop spectrophotometer (ND-1000; Thermo Scientific, USA) and RNA integrity number (RIN) values were assessed using Agilent 2100 bioanalyzer (Agilent Technologies, Inc. USA).

For Illumina microarray analysis, 1.5 μg of total RNA was amplified and labeled with biotinylated nucleotides using Illumina Total Prep RNA amplification kit (Ambion, USA) according to previous studies (Kwon et al., 2017). Samples were purified using the RNeasy kit (Qiagen, UK). Hybridization, washing and scanning with Sentrix HumanHT-12 v4 Expression Bead Chip (Illumina, USA) were performed according to the Illumina BeadStation 500× System Manual.

Microarray data was extracted, normalized, and analyzed using Illumina BeadStudio provided by the manufacturer. Expression values were then log₁₀-transformed. The categorization was carried out depending on whether the gene was upregulated (log₁₀ [fold change] ≥ 0.3 and P<0.05), downregulated (log₁₀ [fold change] ≤ −0.3 and P<0.05), or unchanged in FN-positive tumor tissues compared with those of FN-negative tumors.

The identified DEGs were subjected to GO enrichment analysis using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) (http://david.abcc.ncifcrf.gov/). GO terms (“biological processes”) with a P-value <0.05 were considered significantly enriched by the DEGs. Only the GO terms with false discovery rate (FDR) <0.10 were considered as significant and ranked by FDR in ascending order.

Consensus molecular subtype (CMS) classification was performed based on the gene level according to a previous study (Guinney et al., 2015). CMS subtypes were classified using the random forest (RF) predictor implemented in the R package “CMSclassifier.” The similarity of the expression profile to the four subtypes (CMS1 to CMS4) was calculated for each case.

Surgical specimens were fixed in 10% buffered formalin for 48 hr and dehydrated with a series of ethanol before being embedded in paraffin. Immunohistochemistry was carried out on a fully automated VENTANA Benchmark XT Stainer (VENTANA Medical Systems; Roche Group, USA), according to a previous study (Kim et al., 2019). The paraffin-embedded tissue blocks were cut into 4 µm sections, mounted on glass slides, deparaffinized, and rehydrated. The sections were incubated with the following primary antibodies: anti-CD-3 (1:200; Cat# IS503, Dako, Denmark) and anti-CD-8 (1:200; Cat# IR623, Dako, Denmark). The staining was developed using the ultraView Universal DAB Detection Kit (Ventana Medical Systems, USA), followed by hematoxylin II counterstain according to the manufacturer’s instructions (Ventana, USA). To measure CD3⁺ or CD8⁺ T lymphocyte density, all stained slides were scanned (magnification ×200) using a VENTANA iScan HT slide scanner (Ventana Medical Systems, USA).

Lymphocytes were isolated from the tumor tissue by enzymatic dissociation with the human tumor tissue dissociation kit (Miltenyi Biotec, USA) according to the manufacturer’s instructions. The single cell suspension was then filtered through a 70 μm cell strainer, washed with PBS, and centrifuged at 300 x g for 5 min. The supernatant was discarded, then stained with the LIVE/DEAD fixable dead cell staining kit (Invitrogen, USA) to remove dead cells. The surface proteins were stained with fluorophore-conjugated antibodies. Intracellular staining was performed with fluorophore-conjugated antibodies and FoxP3 staining buffer kit (Thermo Fisher Scientific, USA). After washing, the cells were analyzed with an LSR II instrument and FlowJo software. The antibodies used for the flow cytometry analysis are listed in Table 2 . Gating strategies are shown in Supplementary Figures 3A, B . First, single cells were discriminated by forward-light scattering signal height (FSC-H) versus forward-light scattering signal area (FSC-A) gating. A lymphocyte gate was gated using a side scattering signal area (SSC-A) versus an FSC-A. We then gated for CD3⁺ T cells on live lymphocytes to exclude dead cells. In the CD3⁺ gating, CD4⁺ or CD8⁺ T cells were further gated on the CD8⁺ versus CD4⁺ plot.

T cells from FN-negative tumors were stimulated for 48 hr in RPMI1640 (2 mM L-glutamine) containing 10% fetal bovine serum (FBS) in the 24-well plates coated with anti-CD3 antibody. During stimulation, cells were infected with FN at a multiplicity of infection (MOI) of 100 for 48 hr at 37°C in a humidified air atmosphere containing 5% CO₂. After incubation, cells were harvested and stained with anti-CD3, anti-CD4, anti-CD8, anti-FOXP3, anti-PD-1, anti-TIM-3, and anti-TIGIT antibodies. The stained cells were analyzed by flow cytometry.

Data analyses were performed using GraphPad Prism 8.0 and R statistical software. Paired or unpaired t-tests were used to compare continuous variables. The chi-squared test was used to compare qualitative variables. Pearson or Spearman correlation analyses were used to evaluate correlations between parameters. The one-tailed Fisher’s exact test was used to assess the significance of overlapping genes between different groups and the log-rank test to compare survival outcomes. Significance was set at P<0.05.

We analyzed tumor tissues of 99 patients with stage III colorectal carcinoma cases ( Table 1 ). Our cohort was predominantly comprised of left-sided colon cancers (64.6%) and microsatellite stable cancers (91.9%). KRAS mutation and BRAF mutations were found in 40 (40.4%) patients and 1 (1.0%) patient, respectively. To investigate the classification of CMS subtypes in tumor tissues, we analyzed the mRNA expression patterns using microarray. Patients were classified as CMS1 (14.1%), CMS2 (24.2%), CMS3 (15.2%), CMS4 (26.3%), and mixed (20.2%) ( Figure 1A ). Real-time quantitative PCR was used to detect FN infection in tumor tissues. Of the 99 patients with CRC, 53 (53.5%) were positive for FN infection and 46 (46.5%) were negative. The proportion of CMS subtypes in 99 patients was examined according to the FN infection ( Figure 1B ). Interestingly, mesenchymal subtype (CMS4) was the most prevalent in FN-positive patients (32.1%). Next, we assessed the relationship between FN infection and patient outcomes in disease-free survival (DFS) and overall survival (OS) using the Kaplan–Meier method ( Figures 1C, D ). Both DFS and OS were lower in patients with FN-positive compared to those FN-negative (P=0.0019 for DFS and P=0.0304 for OS). CMS groups were not associated with survival outcomes in our cohort ( Supplementary Figures S1A, B ).

We performed microarray analysis of tumor tissues and a total of 563 differentially expressed genes (DEGs) were identified by comparing FN-positive tumor tissues and those of FN-negative tumors (corrected P-value <0.05). We found 436 down-regulated genes and T-cell transcript genes, including CD3E, CD4 and CD8A ( Figure 2A ). Among the 127 upregulated DEGs, we identified FCGR2B, FCGR3A, ADAR, WASL, SDC4, ADCY9, ZBTB7B, PAG1, CEBPB, VIPR1, and POMC involved in the negative regulation of lymphocyte proliferation and activation. We hypothesized that FN infection would be related to the depletion of T cells in the tumor microenvironment.

To determine this possibility, we performed a GO term analysis of biological processes using DEGs identified by FN infection. Downregulated DEGs were involved in immune responses, including T cell receptor signaling pathway, T cell co-stimulation, T cell activation, and interferon-gamma-mediated signaling pathway ( Figure 2B ). In contrast, upregulated DEGs were enriched in the pathways related to transcription and differentiation ( Supplementary Figure S2 ).

Unsupervised principal component analysis (PCA) was used to cluster the samples in terms of gene expression profiling ( Figure 2C ). PCA plots show that the colon tumor tissues are not distinct in terms of FN-positive or FN-negative classes whereas the samples are largely segregated in accordance to the CMS classification.

Next, we examined 21 immune cell types in 99 tumor tissues by using the CIBERSORT analytical deconvolution tool ( Figure 2D ). Interestingly, the proportions of CD8 T cells in FN-positive tumor tissues were lower than those in FN-negative tumors (P<0.001). The T follicular helper cells, M1 macrophages, activated dendritic cells, and neutrophils exhibited a decreasing tendency according to FN infection, but there was no significance. On the contrary, the proportions of mast cells resting (P=0.043) were higher than those in FN-negative tumors.

To determine whether CD8⁺ T cells in tumors exhibit any exhausted phenotype in response to FN infection, we examined the relative expression of CD8⁺ T cells among total tumor T cells ( Figure 2E ). The proportion of CD8⁺ T cells among total T cells was lower in FN-positive tumors compared with FN-negative samples (P=0.0039). We also assessed differences in the frequency of regulatory T cells and CD8⁺ T cells in total T cells ( Figure 2F ). The differences in the frequency of regulatory T cells and CD8⁺ T cells were diminished by FN infection (P=0.0051).

Then, immunohistochemical staining for CD3 and CD8 was performed to assess T-cell infiltration in tumor tissue. We observed that the CD3⁺ and CD8⁺ cells were reduced in FN-positive tumor tissues compared to FN-negative tumors ( Figure 2G ).

Our findings suggest that poorer prognosis in CRC patients is probably derived from T cell-mediated immune responses by FN infection.

To investigate whether FN infection contributes to T cell composition, we performed flow cytometry analysis for tumor-infiltrating lymphocytes (TILs) derived from tumor tissues in patients with mismatch repair-proficient CRC. In the analysis of CD3⁺ T cells, we found that the relative frequency of FoxP3⁺CD4⁺ regulatory T cells among CD3⁺ T cells were higher in FN-positive tumors, whereas the relative frequency of CD8⁺ T cells among CD3⁺ T cells were lower in this group ( Figures 3A–D ). We also found that the relative frequency of regulatory T cells (FoxP3⁺CD4⁺ T cells) was higher in FN-positive tumor tissues than in those FN-negative tumors in the analysis of CD4⁺ T cells ( Figure 3E , P=0.0014). Next, we examined the frequencies of FoxP3⁺ T cell subpopulations (Fr. II, FoxP3^hiCD45RA^- and Fr. III, FoxP3^lowCD45RA^-) in tumor tissue according to FN infection. Among the CD4⁺T cell population of TILs, the frequencies of both Fr. II ( Figure 3F , P=0.0428) and Fr. III ( Figure 3G , P=0.0111) regulatory T cells were significantly increased in FN-positive tumor tissues compared to those FN-negative tumors, respectively. The association of FN infection with Fr. II and Fr. III regulatory T cells in FoxP3⁺ regulatory T cells in tumor tissues was not statistically significant ( Figures 3H, I ; P=0.343 and P=0.322, respectively).

We further tested whether FN infection affects immune checkpoint inhibitory receptors including PD-1, TIM-3, and TIGIT in CD8⁺ TILs. Interestingly, we observed that the frequencies of PD-1^hi CD8⁺ T cells ( Figure 3J , P<0.0001) and TIM-3⁺ CD8⁺ T cells ( Figure 3K , P=0.001) were higher in FN-positive tissues. Figure 3L shows that the frequencies of TIGIT⁺ CD8⁺ T cells tended to increase in FN-positive tumor tissues, but there was no significant difference between FN-positive and FN-negative tumor tissues ( Figure 3L , P=0.0898). To investigate whether the FN burden is associated with immune phenotypes, FN-positive samples (n=32) were categorized into low- and high-burden FN (n=16 for each group). Our results showed that patients with a high burden of FN exhibited a higher frequency of FoxP3⁺ regulatory T cells among CD4⁺ T cells, compared to negative controls ( Supplementary Figure 4A , P=0.0013). We also observed that FN burden affects immune checkpoint inhibitory receptors including PD-1, TIM-3, and TIGIT in CD8⁺ TILs. The frequencies of PD-1^hi CD8⁺ T cells ( Supplementary Figure 4B , P<0.0001), TIM-3⁺ CD8⁺ T cells ( Supplementary Figure 4C , P=0.0005), and TIGIT⁺ CD8⁺ T cells ( Supplementary Figure 4D , P=0.0203) were relatively increased in high-burden FN samples compared to FN-negative samples. There was no significant difference between low burden FN samples and negative controls.

Our results suggest that FN contributes to the generation of an immunosuppressive tumor microenvironment (TME) by inducing a disproportionate T cell composition in colorectal carcinoma tissues.

To assess whether FN infection directly induces T cell-mediated immune responses in TILs, ex vivo stimulation experiment was performed. We isolated TILs from FN-negative tumor tissues. Single cell suspension of tumor tissues was prepared and then stimulated by anti-CD3 antibody. During stimulation, single cell suspension was infected with FN at a MOI of 100:1 (bacteria: cells) for 48 hr. The control group was treated with PBS instead of FN. Similar results to those shown in Figure 3 were observed from the T cell stimulation system. In FN-treated group, the frequencies of FOXP3⁺CD4⁺ T cells were significantly increased compared to control group ( Figure 4A , P<0.0001). We next examined the expression of immune checkpoint inhibitory receptors (PD-1, TIM-3, and TIGIT) in CD8⁺ T cells according to FN infection. Interestingly, PD-1 (P=0.0053), TIM-3 (P=0.0063), and TIGIT (P=0.0045) were highly expressed in CD8⁺ T cells after treatment with FN ( Figures 4B–D ). Representative histograms presenting these data are presented in Figure 4E , showing increased expression of CD4⁺ FoxP3⁺ T cells, CD8⁺ PD-1⁺, CD8⁺ TIM-3⁺, and CD8⁺ TIGIT⁺ T cells in FN-positive tumors. These findings provide evidence that FN infection is capable of inducing an immunosuppressive tumor microenvironment in patients with colorectal cancer.