Main

The expression of FOXP3 is highly specific for naturally occurring Treg cells1,11. Yet there is ample evidence that FOXP3+CD4+ T cells in humans are functionally and phenotypically heterogeneous, and include suppressive and nonsuppressive T cells10,12,13. For example, FOXP3+CD4+ T cells in peripheral blood mononuclear cells (PBMCs) can be dissected into three subpopulations on the basis of FOXP3 and CD45RA expression levels. Fraction I (Fr-I; FOXP3loCD45RA+; referred to as naive Treg (nTreg)) cells, after antigenic stimulation, differentiate into Fr-II (FOXP3hiCD45RA−; referred to as effector Treg (eTreg)) cells, which are terminally differentiated, highly suppressive and functionally stable. In contrast, Fr-III (FOXP3loCD45RA−) cells do not possess suppressive activity and can secrete pro-inflammatory cytokines10. Here we have examined how these FOXP3+ T cell subpopulations in tumor tissues contribute to the prognosis of CRCs.

Among the CD4+ T cell population of tumor-infiltrating lymphocytes (TILs) in individuals with CRC, there were twofold and fourfold greater numbers of FOXP3+ T cells in the tumor than in the normal colonic mucosa or PBMCs, respectively (Fig. 1a,b and Supplementary Table 1; CRC TILs, 30.9 ± 10.0%; normal colon mucosa, 14.4 ± 2.7%; PBMCs, 6.9 ± 2.5%). The number of Fr-II cells notably increased, with a prominent reduction in the number of Fr-I cells and naive CD45RA+FOXP3−CD4+ conventional T cells (Fr-V) (Fig. 1a,b and Supplementary Fig. 1a). We also noted markedly higher numbers of Fr-III cells in some CRCs but not in other types of cancers, such as malignant melanoma (Fig. 1a,b and Supplementary Fig. 1b,c). These findings prompted us to classify CRCs into two types, A or B, by the frequencies of Fr-III TILs, i.e., type A as less than and type B as greater than 9.8%, which was the upper limit of the mean + (2 × s.d.) of the Fr-III cell frequencies in normal colonic mucosa (n = 7). Both types contained similarly high percentages of Fr-II cells, with significantly higher numbers of total FOXP3+ T cells in type B than type A tumors (Fig. 1a,b). Functionally, Fr-II cells from either type of CRC tissue showed strong in vitro suppressive activity, whereas Fr-III cells from type B CRCs did not (Fig. 1c). The lack of suppressive activity in Fr-III cells correlated with markedly lower expression of suppression-related molecules, such as T cell immunoreceptor with Ig and ITIM domains (TIGIT) and cytotoxic T lymphocyte–associated protein 4 (CTLA-4) (Supplementary Fig. 2a,b)14,15. A significantly higher frequency of IL-17-secreting Fr-III cells were present in type B CRCs than in PBMCs, whereas there were equally high frequencies of interferon (IFN)-γ-secreting Fr-III cells in type B CRCs and PBMCs (Fig. 1d,e and Supplementary Fig. 2c)16. Both PBMCs and TILs contained very few IL-10-producing CD4+ T cells, indicating that only a few immunosuppressive type 1 regulatory T cells infiltrated into tumor tissues (Supplementary Fig. 2d). Moreover, assessment of the Treg cell–specific CpG methylation status of FOXP3 revealed partial hypomethylation in the Fr-III cells from TILs and PBMCs, in contrast to the profound hypomethylation observed in Fr-II cells and the complete methylation seen in memory FOXP3−CD45RA− (Fr-IV) cells (Fig. 1f)17. Thus, FOXP3+ T cells in CRC TILs are heterogeneous in function and include FOXP3hi suppression-competent Treg cells and FOXP3lo nonsuppressive T cells. In addition, whereas the phenotype, function and epigenetic status of the two populations are equivalent to those in PBMCs, the populations differ in frequency. Fr-I nTreg cells are scarce, and Fr-II eTreg cells are abundant, in both types of CRCs, whereas cytokine-secreting, nonsuppressive Fr-III T cells constitute a sizable fraction of FOXP3+ TILs in type B CRCs.

Figure 1: Classification of CRCs based on phenotypic and functional heterogeneity of FOXP3+ TILs.
figure 1

(a,b) Representative flow cytometry plots for CD45RA and FOXP3 staining of CD4+ T cells (a) and quantification of the frequency of total FOXP3+ cells and that for cells of each indicated fraction from CD4+ T cells from PBMCs (n = 33), colonic mucosa (n = 7), type A CRCs (n = 13) and type B CRCs (n = 22) (b). (c) Representative flow cytometry plots (of three independent experiments) showing suppressive activity of FOXP3+CD25+CD4+ T cells from type A (top) and type B (bottom) CRCs. Carboxyfluorescein diacetate succinimidyl ester (CFSE)-labeled CD25−CD4+ T cells from PBMCs were cocultured with an equal number of cells from the indicated fractions of TILs for 5 d with a CD3-specific mAb and irradiated APCs. CD45RA−CD25hi (Fr-II) cells from PBMCs from the same donor were used as a control. Corresponding subsets—defined by FOXP3 and CD45RA staining and by CD25 and CD45RA staining—are shown using the same colors. (d,e) IFN-γ and IL-17 production by phorbol 12-myristate 13-acetate (PMA)- and ionomycin-stimulated Fr-II, Fr-III and Fr-IV CD4+ T cells. Representative flow cytometry plots for IFN-γ and IL-17 expression (d) and quantification of IFN-γ-expressing (top) and IL-17-expressing (bottom) cells (e) in Fr-II, Fr-III and Fr-IV cells from PBMCs (n = 20), type A CRCs (n = 6) and type B CRCs (n = 14). (f) Representative analysis (of three independent experiments) for DNA methylation status of Treg-specific DNA demethylated regions (in FOXP3 intron 1) in Fr-II, Fr-III and Fr-IV CD4+ T cells from TILs (left) and PBMCs (right) of individuals with CRC. White and black dots indicate individual demethylated and methylated CpG residues, respectively. In b,e, horizontal lines indicate medians. *P < 0.05, **P < 0.01, ***P < 0.001; by one-way analysis of variance (ANOVA) with post hoc Tukey test. n.d., not determined.

Source data

Full size image

We performed microarray analysis of type A and type B CRCs and found that genes involved in immune responses and inflammation were significantly upregulated in type B CRCs (Fig. 2a,b and Supplementary Fig. 3). For example, transcription of IL12A, IL12B, TGFB1 and TNF, which encode the cytokines IL-12, TGF-β and tumor necrosis factor (TNF)-α, respectively, was higher in type B CRCs, as confirmed by quantitative RT–PCR (Fig. 2c), but not in the paired normal colonic mucosa (Supplementary Fig. 4). Normal mucosal tissue from type B CRCs showed increased transcription of interleukin 6 (IL6), as previously reported18. RNA in situ hybridization analysis revealed that stromal cells, such as macrophages and fibroblasts, produced TGF-β, TNF-α and IL-12 in type B tumor tissues (Supplementary Fig. 5). These findings indicated possible roles of these cytokines in inducing FOXP3+ T cells. Indeed, addition of TGF-β to an in vitro culture of CD45RA+CD25− naive CD4+ T cells stimulated with monoclonal antibodies (mAbs) specific for CD3 and CD28 induced a high frequency of FOXP3hi T cells, whereas addition of IL-12 or TNF-α did not. IL-12 but not TNF-α inhibited this TGF-β-dependent induction of FOXP3+ T cells, particularly of FOXP3hi cells, whereas treatment with IL-12 and TGF-β together induced FOXP3lo T cells more efficiently than treatment with TGF-β alone (Fig. 2d,e and Supplementary Fig. 6a). Such FOXP3lo cells produced larger amounts of IFN-γ after in vitro stimulation (Fig. 2f) and lacked suppressive activity, even in the presence of an IFN-γ-specific mAb (Supplementary Fig. 6b,c). Treatment with IL-12 or TGF-β neither inhibited differentiation of Fr-I cells into FOXP3hi cells nor converted Fr-II cells into FOXP3lo cells (Supplementary Fig. 7). In addition, treatment with IL-10 failed to induce FOXP3hi or FOXP3lo cells or to inhibit TGF-β-dependent induction of FOXP3+ cells (Supplementary Fig. 8). Collectively, type A and type B CRCs possess distinct gene expression profiles, with type B CRCs showing high expression of inflammation- or immune-response-related genes, especially IL12A, IL12B, TGFB1 and TNF. Furthermore, we suggest that Fr-III nonsuppressive FOXP3lo T cells, which are abundant in type B CRCs, are probably derived from non-Treg cells following their activation with cytokines, particularly IL-12 and TGF-β.

Figure 2: Distinct cytokine dependency of two CRC types.
figure 2

(a,b) Volcano plot for the expression of total genes (15,180 genes) and genes classified into immune-response-related (blue dots, 959 genes) or inflammatory-response-related (red dots, 469 genes) genes by gene ontology (GO) terms (a) and cumulative distribution function for the t-statistics between mRNA expression profiles of type A and type B CRCs (black line, total genes; blue line, immune-response-related genes; red line, inflammatory-response-related genes) (b) (n = 2 each). P values are based on the two-sided Kolmogorov–Smirnov test (b). (c) mRNA expression levels of indicated genes, as measured by quantitative RT–PCR, in type A (n = 12) and type B (n = 22) CRC tumors. Horizontal lines indicate medians. (d,e) Schematic for generation of FOXP3hiCD45RA−CD4+ and FOXP3loCD45RA−CD4+ T cells from CD45RA+CD25− naive CD4+ T cells from PBMCs of healthy individuals (d, top), representative flow cytometry plots for the analysis of FOXP3 expression after treatment of CD45RA+CD25− naive CD4+ T cells (d, bottom) and quantification (n = 3 independent experiments) of FOXP3hi (e, left) and FOXP3lo (e, right) cells, demarcated as shown in Supplementary Figure 7a. (f) Representative flow cytometry plots (left) and quantification of three independent experiments (right) for IFN-γ production by CD4+ T cells induced in d,e, as assessed by intracellular cytokine staining after stimulation with PMA and ionomycin. In e,f, error bars indicate mean ± s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001; by Mann–Whitney U-test (c) or one-way ANOVA with post hoc Tukey test (e,f).

Source data

Full size image

Our findings indicate that assessing the transcription levels of IL12A, TGFB1 and TNF in CRC tissues can be a surrogate for enumerating Fr-III cells for the purposes of classifying type A and type B CRCs. Indeed, the combination of IL12A and TGFB1 expression levels most substantially distinguished type A and type B CRCs, as compared to other combinations of cytokine gene transcription (Fig. 3a and Supplementary Fig. 9a,b); type A CRCs were mostly IL12AloTGFB1lo, whereas type B CRCs were IL12Ahi and/or TGFB1hi. Notably, the IL12AhiTGFB1hi group showed significantly longer disease-free survival (DFS) times than the IL12AloTGFB1lo group (P = 0.020) in a second cohort of 109 individuals with CRC, despite other clinical parameters being comparable between the two groups (summarized in Supplementary Table 2) (Fig. 3b). Moreover, whereas FOXP3 expression was not a prognostic factor in the entire cohort of subjects with CRC (P = 0.75), high FOXP3 expression (defined as greater than the median of the values of FOXP3 transcripts) (Fig. 3c) was clearly associated with poor prognosis (P = 0.038) in type A CRCs that were defined as IL12AloTGFB1lo (Fig. 3d). In contrast, high FOXP3 expression indicated a better prognosis, although not significantly so (P = 0.34), in type B CRCs that were defined as IL12AhiTGFB1hi, despite their much higher expression levels of FOXP3 than those in type A CRCs. In addition, two clusters (1 and 2) that were found by unsupervised hierarchical clustering with the correlation matrix according to the expression levels of IL12A and TGFB1 mostly corresponded to type A and type B CRCs, respectively (Fig. 3e)19. Also, FOXP3hi CRCs (i.e., CRCs with FOXP3 expression levels above the median in each cluster (Fig. 3f)) showed poor prognosis in cluster 1 (P = 0.017) but not cluster 2 (Fig. 3g). Thus, in type A CRCs, in which Fr-II eTreg cells were the vast majority of FOXP3-expressing cells, high FOXP3 expression was associated with poor prognosis. In contrast, in type B CRCs, in which Fr-III non-Treg cells significantly contributed to the total FOXP3 expression levels, high FOXP3 expression was linked to better prognosis.

Figure 3: Dependency of CRC prognosis on distinct subsets of FOXP3+CD4+ TILs.
figure 3

(a) mRNA expression of IL12A and TGFB1 in type A (n = 12) and type B (n = 22) CRCs, as defined by flow cytometry (see Fig. 1a). The cut-off value for the expression of each mRNA (0.6 for IL12A and 0.6 for TGFB1) was defined by Youden's index (J) = sensitivity + specificity − 1. (b) Kaplan–Meier curves for DFS of the indicated patient groups, as classified by IL12A and TGFB1 expression levels as in a. The hazard ratio (HR) for DFS in the IL12AloTGFB1lo group, as compared to that of the IL12AhiTGFB1hi group, was calculated with Cox regression analysis, and the 95% confidence interval is shown in parentheses. (c) FOXP3 expression, as assessed by quantitative RT–PCR, in the patient groups in b. Each circle indicates an individual patient. (d) Kaplan–Meier curves for DFS of individuals with CRC, classified on the basis of FOXP3 mRNA levels. DFS of the total number of patients with CRC (n = 109) (top), patients with IL12AloTGFB1lo expression (n = 35) (middle) or patients with IL12AhiTGFB1hi expression (n = 27) (bottom) was compared by classifying each group into FOXP3hi or FOXP3lo for FOXP3 mRNA expression levels that were above or below the median, respectively, as in c. (e) Heat map for IL12A and TGFB1 expression in 109 patients with CRC. Red and blue represent high and low levels of expression for the indicated genes, respectively. Two clusters (1 and 2) were found by unsupervised hierarchical clustering with the correlation matrix on the basis of IL12A and TGFB1 expression levels. (f) Relative FOXP3 mRNA levels in CRC tumor samples from each cluster. (g) Kaplan–Meier curves for DFS of individuals with CRC, classified on the basis of FOXP3 mRNA levels. DFS of patients in cluster 1 (top) or cluster 2 (bottom) was compared by classifying each group into FOXP3hi or FOXP3lo for FOXP3 mRNA expression levels that were above or below the median, respectively, as in f. In b,d,g, P values were calculated by using the log-rank test. In c,f, FOXP3 mRNA expression levels were normalized to GAPDH expression and are shown relative to FOXP3 levels of a reference tumor sample; horizontal lines indicate medians. *P < 0.05, ***P < 0.001; by one-way ANOVA with post hoc Tukey test (c) or the Mann–Whitney U-test (f).

Source data

Full size image

Tumor-infiltrating CD8+ T cells and NK cells are associated with better prognosis in CRCs19,20,21,22; however, the expression levels of CD8A, KLRG1 or B3GAT1, which encode CD8a, killer cell lectin-like receptor G1 (KLRG1) or CD57, respectively, by themselves were not a significant prognostic factor in our study (Supplementary Fig. 10). Yet, high ratios of IFNG/FOXP3 were significantly associated with better prognosis in type A CRCs (P = 0.041), suggesting that Treg cells suppress IFN-γ producing cells (including CD8+ T cells, NK cells and other cells) in type A CRCs (Supplementary Fig. 11).

We next attempted to determine the factors responsible for such distinct patterns of FOXP3+ T cell infiltration into CRCs. Correlations between bacterial infiltration into tumors and CRC development have been reported23,24, and by using fluorescent in situ hybridization (FISH) analysis, we found that intestinal bacteria were present in tumor tissues at significant frequencies in IL12AhiTGFB1hi type B but not IL12AloTGFB1lo type A CRC tissues (Fig. 4a,b and Supplementary Fig. 12). Sequencing of bacterial 16S ribosomal DNA from CRC tissue or the stool of the same individual detected F. nucleatum in IL12AhiTGFB1hi but not IL12AloTGFB1lo CRC tissues (as confirmed by FISH analysis), and it did not detect F. nucleatum in the stool from individuals with either type of CRC (Fig. 4b,c)25. CCND1 and NFKB2, which encode cyclin D1 and NF-κB, respectively, were also expressed at significantly higher levels in type B than in type A CRCs (Fig. 4d), suggesting an association of F. nucleatum with oncogenic and inflammatory responses, as there is no causal evidence here.23,24. Thus, F. nucleatum, and possibly other intestinal bacteria, might invade tumor tissues26,27 and have a role in the production of inflammatory cytokines (such as IL-12, TGF-β and TNF-α) by the tissues, thereby contributing to the expansion of FOXP3lo non-Treg cell population in type B CRCs, although the presence of the bacterium has been reported to inhibit T cell and NK cell effector activities and to correspond to poor CRC prognosis28,29,30. The relevance of the presence of F. nucleatum in the tumor requires further study.

Figure 4: Contribution of colonic bacteria to determining CRC types.
figure 4

(a) The frequency of bacterial infiltration into IL12AloTGFB1lo (n = 8) and IL12AhiTGFB1hi (n = 7) CRC tissues, as assessed by FISH using the EUB338 probe for bacteria. (b) Representative images of the samples in a showing H&E staining and FISH analysis for whole bacteria with the EUB338 probe (green) or Fusobacterium spp. with the FUSO227 probe (red) in IL12AloTGFB1lo (type A) and IL12AhiTGFB1hi (type B) CRC tissues. Pt., patient. Scale bars, 100 μm. (c) Meta-sequencing analysis of bacterial 16S ribosomal DNA that was amplified from stools (left) and the surfaces of tumor tissues (right), using 454 sequencing. The relative abundance of operational taxonomic units (OTUs) is shown for each sample. Each color indicates bacterial species that belong to the same phylum. Of 23 CRC samples collected for FISH and meta-sequencing analysis, analysis was limited to the indicated number of samples in which the quality of DNA was adequate for analysis. (d) Relative expression of CDH1 (which encodes E-cadherin), MYC, CCND1, NFKB1 and NFKB2, as measured by quantitative RT–PCR, in type A (n = 12) and type B (n = 22) CRCs. Horizontal lines indicate medians. *P < 0.05, **P < 0.01; by the Mann–Whitney U-test (d).

Source data

Full size image

In conclusion, the high expression of FOXP3 can be a marker of poor prognosis in type A CRCs, in which tumor-infiltrating FOXP3+CD4+ T cells are predominantly Fr-II eTreg cells, but not type-B CRCs, in which mostly inflammatory Fr-III non-Treg FOXP3+ cells are present. The difficulty of distinguishing Fr-III cells from Fr-II cells in tumor tissues by immunohistochemistry (Supplementary Fig. 13) would have been a major confounding factor in previous studies evaluating the clinical significance of FOXP3+CD4+ T cells in CRCs by using immunohistochemistry7,8,9. Furthermore, the significant correlation between the frequency of Fr-III cells and the transcription levels of IL12A and TGFB1 in CRC tissues indicates that the latter is a useful prognostic marker of CRCs, by enabling the distinction of type A and type B tumors. Our results also suggest that Treg cell depletion may augment antitumor immunity and provide clinical benefits in individuals with type A CRCs or other cancers that are characterized by large amounts of infiltrating FOXP3hi Treg cells3,13,31. In addition, clinical strategies to locally increase FOXP3lo non-Treg cells—for example, by modulating colonic microbiota (although further studies are required) or intratumoral IL-12 injection32—could be beneficial in suppressing the development and growth of CRCs and other cancers.

Methods

Patients and samples.

Peripheral blood, tumor tissue, adjacent normal colonic tissue and stool samples were obtained from healthy individuals and those with CRC or malignant melanoma. All healthy donors were subjects with no history of autoimmune diseases and malignant tumors. Patients with CRC (summarized in Supplementary Table 1) who underwent surgery at Osaka University Hospital between 2011 and 2015, and whose tumor sizes were sufficient for collection of tumor-infiltrating lymphocytes, were included in this study. Samples from another cohort of patients with CRC (summarized in Supplementary Table 2) (used for survival analysis) who received curative resection was collected between 2003 and 2005 at Osaka University Hospital. PBMCs were isolated by density gradient centrifugation with Ficoll-Paque (GE Healthcare). To collect TILs, tumor tissues were minced and treated with gentleMACS Dissociator (Miltenyi Biotec), as previously described13. All donors provided written informed consent before sampling, according to the Declaration of Helsinki. This study was performed in a nonblinded and nonrandomized manner, and was approved by the Osaka University Research Ethics Committee (http://www.osaka-u.ac.jp/en/research/iinkai/moral/index.html) (Osaka, Japan).

Antibodies and reagents.

Violet 450–conjugated anti-CD8 (RPA-T8) mAb, fluorescein isothiocyanate (FITC)-conjugated anti-CD45RA (HI100) mAb, allophycocyanin (APC)- and Cy7-conjugated anti-CD4 (RPA-T4) mAb, Brilliant Violet 711 (BV711)-conjugated anti-CD25 (2A3) mAb, Alexa Fluor 700–conjugated anti-CD3 (UCTH1) mAb, V450-conjugated anti-IL-10 (JES3-9D7) mAb and APC-conjugated anti-CTLA-4 (BNI3) mAb were purchased from BD Biosciences. APC-conjugated anti-CD3 (UCTH1) mAb, PE-conjugated anti-Foxp3 (Forkhead Box P3, 236A/E7) mAb, peridinin chlorophyll protein complex (PerCP)- and Cy5.5-conjugated anti-IFN-γ (4S.B3) mAb, phycoerythrin (PE)- and Cy7-conjugated anti-TNF-α (Mab11) mAb, PE–Cy7-conjugated anti-TIGIT (MBSA43) mAb and eFluor 506–conjugated fixable viability dye were obtained from eBioscience. BV421-conjugated anti-CD25 (BC96) mAb and BV421-conjugated anti-IL-17 (BL168) mAb were purchased from BioLegend. Recombinant IL-12, TGF-β, TNF-α and IL-10 were purchased from PeproTech (Rocky Hill, NJ).

Surface marker and FOXP3 staining.

Cells, washed using PBS with 2% fetal calf serum (FCS), were stained with mAbs specific for CD3, CD4, CD8, CD25, CD45RA, TIGIT or CTLA-4 and with fixable viability dye (Invitrogen). Intracellular staining of FOXP3 was performed with anti-Foxp3 mAb and Foxp3 Staining Buffer Set (eBioscience) according to the manufacturer's instructions. After washing, cells were analyzed with an LSR Fortessa instrument (BD Biosciences) and FlowJo software (Treestar, Ashland, OR). The dilution of the staining antibodies was done according to the manufacturer's instructions.

Intracellular cytokine staining.

Cells were stimulated for 6 h with 50 ng/ml phorbol 12-myristate 13-acetate (PMA; Sigma) and 1 μg/ml ionomycin (Sigma), and GolgiStop reagent (BD Biosciences) was added for the last 5 h of culture. Cells were stained for cell surface markers (mAbs specific for CD3, CD4 and CD45RA and fixable viability dye) and then for intracellular cytokines and FOXP3. After washing, the cells were analyzed with an LSR Fortessa instrument and FlowJo software.

Suppression assay.

CD45RA−CD25hiCD4+ (Fr-II) T cells or CD45RA−CD25loCD4+ (Fr-III) T cells were sorted from TILs or PBMCs of patients with CRC, using FACS Aria II (BD Biosciences). 1 × 104 CFSE-labeled (1 μM) responder CD25−CD4+ T cells from PBMCs were cocultured with 1 × 104 unlabeled Fr-II or Fr-III cells in the presence of 1 × 105 irradiated antigen-presenting cells (APCs) while being stimulated with 0.5 mg/ml anti-CD3 (OKT3) mAb. In some cultures, neutralizing anti-IFN-γ (10 μg/ml) was added. Proliferation of CFSE-labeled cells was assessed by an LSR Fortessa instrument 5 d later.

Methylation analysis.

FOXP3hiCD45RA−CD4+ (Fr-II) T cells, FOXP3loCD45RA−CD4+ (Fr-III) T cells and FOXP3−CD45RA−CD4+ (Fr-IV) T cells from TILs and PBMCs of patients with CRC were sorted, and genomic DNA was prepared with the NucleoSpin Tissue XS kit (Macherey Nagel). After sodium bisulfite treatment (MethylEasy Xceed, Human Genetic Signatures), modified DNA was amplified by PCR and subcloned into the PCR2.1-TOPO vector (Invitrogen). PCR primers specific for FOXP3 conserved noncoding sequence-2 region were previously described17. The colonies (16–48 colonies/region) were directly amplified with the Illustra TempliPhi Amplification Kit (GE Healthcare) and sequenced with Hiseq2000 (Illumina).

Microarray analysis.

Total RNA from CRC tumor tissues was isolated with the RNeasy Mini Kit (Qiagen) and subjected to microarray analysis (Human Gene 2.0 ST Array; Affymetrix). Obtained raw data was normalized by the robust multi-array average algorithm (RMA) (R 2.15). Gene ontology (GO) biology processes are described with 15,180 genes. The t-statistic was calculated by comparing type A and type B CRCs.

Quantitative real-time PCR.

cDNA (n = 34) was synthesized from 0.1 μg of total RNA using SuperScript III reverse transcriptase kit (Invitrogen) and the oligo(dT) primer in a total volume of 20 μl. cDNAs were amplified in a final volume (20 μl) containing 10 μM of each Taqman probe (Taqman Gene Expression Arrays, Life Technologies) and 10 μl of Taqman Gene Expression Master Mix (Life Technologies) according to the manufacturer's instructions. Primers for TGFB1, TNF, IFNG, IL1A, IL6, IL8, IL10, IL12A, IL12B, IL17, FOXP3, CDH1, MYC, CCDN1, NFKB1 and NFKB2 were purchased from Applied Biosystems. Relative mRNA expression was evaluated after normalization with GAPDH expression. IL10 and IL12B mRNA were not assessed in two samples due to limited amounts of samples.

T cell culture.

CD4+ T cells that were isolated by negative selection with human CD4+ Isolation Kit (Miltenyi Biotec) were further separated into CD45RA+CD25−CD4+ naive T cells, CD45RA+CD25+CD4+ nTreg cells and CD45RA−CD25+CD4+ eTreg cells, using a FACS ARIA II instrument. Sorted 5 × 104 CD4+ T cells were cultured in the presence of anti-CD3- and anti-CD28-coated Dynabeads (0.1 bead per cell) (Invitrogen) according to the manufacturer's instructions. In some cultures, IL-12 (5 ng/ml), TGF-β (5 ng/ml), TNF-α (50 ng/ml) and IL-10 (10 ng/ml) were added. Cells were analyzed with an LSR Fortessa instrument 7 d later.

Meta 16S sequencing by 454.

Freshly collected human stool and tumor samples were suspended in 4× volume (wt/vol) of a 20% glycerol solution in PBS, frozen using liquid nitrogen and stored at −80 °C. Genomic DNA was prepared as previously reported33,34. DNA was dissolved in Tris–EDTA (TE) buffer and stored at 4 °C until use. The 16S rRNA gene hypervariable regions (V1–2) were PCR-amplified using barcoded 27Fmod and 338R primers33. PCR was conducted by using the following reaction mixture: 1× Ex Taq PCR buffer composed of 10 mM Tris-HCl (pH 8.3), 50 mM KCl, and 1.5 mM MgCl2 in the presence of 250 mM dNTPs, 1 Unit of Ex Taq polymerase (Takara Bio), 0.2 mM of both forward and reverse primers and 20 ng of template DNA (40 ng in case of fecal samples). Thermal cycling was performed under the following conditions: initial denaturation at 96 °C for 2 min, followed by 30 cycles (20 cycles in case of fecal samples) of denaturation at 96 °C for 30 s, annealing at 55 °C for 45 s and extension at 72 °C for 1 min, and a final extension step at 72 °C for 10 min on a 9700 PCR system (Life Technologies). Multiplexed amplicon pyrosequencing was carried out using a 454 GS FLX Titanium or 454 GS JUNIOR instrument (Roche Applied Science) according to manufacturer's instructions. The sequences generated were processed as previously described33. Briefly, reads were assigned to samples on the basis of the barcode sequence. Reads with an average quality value <25 and those that did not have the sequences exactly matched to both PCR primer sequences were filtered off. After removing possible chimeric reads, 1,000 high-quality reads were randomly selected per sample, sorted by their quality value and clustered into operational taxonomic units (OTUs) using a 96% pairwise-identity cut-off value with the UCLUST program. A representative sequence in each OTU was used for taxonomic assignment by using a homology search against the 16S (RDP ver. 10.27. and CORE update 2-9-12) and NCBI genome databases using the GLSEARCH program.

FISH (fluorescence in situ hybridization).

FISH was performed using formalin-fixed paraffin-embedded CRC specimens. 4-μm-thick sections were prepared and hybridized with the 5′ Alexa Fluor 647–labeled universal bacterial probe EUB338 and the 5′ Alexa Fluor 546–labeled Fusobacterium targeted probe FUSO. The sequences of the FISH probes were obtained from probeBase (http://probebase.csb.univie.ac.at/): pB-00159 for EUB338 and pB-00782 for FUSO. Slides were deparaffinized, dried, and treated with 0.2 M HCl for 20 min, and they were then hybridized overnight with the indicated FISH probes at a concentration of 10 ng/μl at 50 °C in hybridization buffer (0.9 M NaCl, 20 mM Tris-HCl, pH 7.4, 0.05% sodium dodecyl sulfate). Slides were washed for 10 min at 50 °C in wash buffer (0.09 M NaCl, 20 mM Tris-HCl, pH 7.4, 0.01% sodium dodecyl sulfate) and rinsed in water. Tissue sections were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) and mounted with coverslips using Fluoromount/Plus (Diagnostic BioSystems). Slides were imaged using a Leica SP5 confocal microscope and analyzed using an open-source software, Fiji.

RNA in situ hybridization.

RNA in situ hybridization for TGFB1, TNF, and IL12 was performed using formalin-fixed paraffin-embedded tissue specimens of representative cases. RNA scope system (Advanced Cell Diagnostics, Hayward, CA) was adopted and performed according to the manufacturer's instruction.

Immunohistochemistry.

Tissue specimens were deparaffinized in xylene and a series of graded concentrations of alcohol and then immersed into preheated antigen-retrieval solution (DAKO high-pH solution), incubated at 95 °C for 20 min, and allowed to cool to room temperature. The specimens were incubated with 2.5 μg/ml anti-Foxp3 mAb (236A/E7) and anti-CD8 (C8/144B) mAb overnight at 4 °C. A horseradish peroxidase (HRP)-conjugated dextran polymer system (Histofine Max-PO (M), Nichirei Biosciences, Tokyo, Japan) was used for secondary detection. Endogenous peroxidase activity was blocked by a 20-min incubation in 0.3% hydrogen peroxide and 0.1% sodium azide solution in PBS. 3,3′-diamino-benzidine (Nichirei Biosciences) was used as a chromogen, and hematoxylin counterstain was performed. For quantification of FOXP3+ or CD8+ T cells in tumor tissue, tissue sections were scanned at NanoZoomer 2.0-HT (Hamamatsu Photonics, Shizuoka, Japan) to ascertain areas with high numbers of FOXP3+ or CD8+ T cells. The number of FOXP3+ or CD8+ T cells per high-power field in these areas was automatically scored with Tissue Studio (Definiens, Munich, Germany), and the average values of three high-power fields were calculated.

Statistical analyses.

Comparisons between subjects were evaluated using the nonparametric Mann–Whitney U-test or the one-way ANOVA with post hoc Tukey test, and P < 0.05 was considered significant. Survival curves were estimated using the Kaplan–Meier method and compared by the log-rank test. Hazard ratios were calculated by Cox regression analysis. Comparisons of patients' background were performed by Fisher's exact test. All statistical analyses were performed using the SPSS software version 21.0 (SPSS Inc., Chicago, IL) or Prism version 6 software (GraphPad Software, Inc., La Jolla, CA).

Accession codes.

Gene Expression Omnibus: all raw CEL files for samples used in the microarray analysis of this study can be accessed with accession number GSE79038. The 16S V1–V2 sequences analyzed in the present study were deposited in the DDBJ database with accession number DRA004536.