Importance of lymph node immune responses in MSI-H/dMMR colorectal cancer

Patients with colorectal cancers (CRCs) generally exhibit improved survival through intensive lymph node (LN) dissection. However, recent progress in cancer immunotherapy revisits the potential importance of regional LNs, where T cells are primed to attack tumor cells. To elucidate the role of regional LN, we investigated the immunological status of nonmetastatic regional LN lymphocytes (LNLs) in comparison with those of the tumor microenvironment (tumor-infiltrating lymphocytes; TILs) using flow cytometry and next-generation sequencing. LNLs comprised an intermediate level of the effector T cell population between peripheral blood lymphocytes (PBLs) and TILs. Significant overlap of the T cell receptor (TCR) repertoire was observed in microsatellite instability–high/mismatch repair–deficient (MSI-H/dMMR) CRCs with high tumor mutation burden (TMB), although limited TCRs were shared between nonmetastatic LNs and primary tumors in microsatellite stable/MMR proficient (MSS/pMMR) CRC patients with low TMB. In line with the overlap of the TCR repertoire, an excessive LN dissection did not provide a positive impact on long-term prognosis in our MSI-H/dMMR CRC cohort (n = 130). We propose that regional LNs play an important role in antitumor immunity, particularly in MSI-H/dMMR CRCs with high TMB, requiring care to be taken regarding excessive nonmetastatic LN dissection in MSI-H/dMMR CRC patients.


Introduction
Colorectal cancers (CRCs) are the third most common cancer and the fourth most common cause of death from cancer in the world, accounting for approximately 1.1 million new cases and 550,000 deaths per year (1). Approximately 20% of CRC patients have distant metastases at the time of diagnosis. Despite recent progresses in treatment, including molecular-targeted therapy, patients suffering from metastatic CRCs have a poor 5-year survival rate of about 10% (2). More effective therapies therefore are urgently required.
Cancer immunotherapy, including immune checkpoint blockade (ICB), has been proven to be beneficial for patients with various types of cancer, including malignant melanoma and non-small cell lung cancer (3)(4)(5). Among gastrointestinal cancers, a recent phase III trial with an anti-PD-1 monoclonal antibody (mAb) for patients with advanced gastric cancers (GC) or esophageal cancers (EC) showed a survival benefit, resulting in the approval of anti-PD-1 mAb for treating GC or EC in Japan (6,7). However, the efficacy of anti-PD-1 mAb is not satisfactory, and ICB seems to be less effective against metastatic CRCs, especially microsatellite stable (MSS)/mismatch repair proficient (pMMR) CRCs (8,9). By contrast, PD-1 blockade was reported to be effective against microsatellite instability-high (MSI-H)/MMR deficient (dMMR) CRCs, with a response rate of 30%-70%, which can be explained by the high tumor mutation burden (TMB), potentially becoming tumor neo-antigens (8,10,11), although MSI-H/dMMR accounts for approximately 10% of CRCs (12). Therefore, studies to understand the detailed immunological features of CRCs, particularly in the tumor microenvironment (TME), are needed to develop better treatment strategies.
The existence of lymph node (LN) metastases is important in deciding the treatment strategy for CRCs, and patients who undergo an intensive LN dissection exhibit improved survival (13,14). One of the reasons for LN dissection is to accurately determine the disease stage. Accurate staging is a critical issue to consider whether further treatments are required or not. According to some guidelines, examination of a minimum of 12 LNs is recommended for the accurate staging (13,14). On the other hand, from an immunological view, regional LNs are considered to play an essential role in antitumor immunity because T cells are generally primed by antigen presenting cells (APCs) that capture tumor antigens in tumor tissues and infiltrate into regional LN (15,16). Therefore, an excessive LN dissection, especially of non-metastatic LNs, may lead to negative effects regarding antitumor immune responses. To reconcile this dilemma of cancer treatment, a broad T cell landscape in the TME, LNs and peripheral blood should be elucidated in humans, as it is difficult to recapitulate the landscape of immune responses with animal models, particularly with transplanted tumors. Here, we comprehensively investigated the immunological status of non-metastatic regional LNs (LN lymphocytes; LNLs) in comparison with those of the TME (tumor-infiltrating lymphocytes; TILs) and peripheral blood (peripheral blood lymphocytes; PBLs) in CRC patients to further understand the immunological features that lead to optimal CRC therapy.

LNLs exhibit an intermediate phenotype between PBLs and TILs especially for effector memory T cells.
We first addressed immunological phenotypes in CRC patients from whom sufficient amounts of PBLs, LNLs and TILs were available (NCCHE cohort). The clinical characteristics of 21 CRC patients enrolled in this study are summarized in Supplemental Table 1. One patient had double cancer of the transverse colon (C416_T) and sigmoid colon (C416_S), both of which were subjected to assessment. The location of tumors was 5 right and 6 left in pMMR CRCs, 8 right and 3 left in dMMR CRCs. All patients received LN dissection regardless of tumor location according to the Japanese guideline (17), and the number of LN dissection was not significantly different with location (Supplemental Figure 1A). Each dissected LN was halved with the maximum surface, all of which were pathologically examined. If there were pathological metastases including micrometastases in LNs, such LNs were excluded from the analyses for avoiding to include metastatic LNs in our assay. CD8 + T cell infiltration was higher in early stage or dMMR CRCs compared with late stage or pMMR CRCs as expected (Supplemental Figure 2).
The frequency of effector memory CD8 + T cell in LNLs tended to, though not significant, be higher in dMMR CRCs (Supplemental Figure 3). TCGA analyses have shown that a lot of genes related to cytotoxic activity (GZMA and PRF1) and T cell exhaustion (PDCD1, LAG3, HAVCR2 and etc.) were significantly highly expressed in MSI-H/dMMR CRCs and that MSI-H/dMMR CRCs contained abundant CD8 + T cell infiltration from CIBERSORTx (Supplementary Figure 4) (12, 18). In comparison with PBLs, LNLs and TILs, CD4 + T cell proportion was the highest in LNLs, followed by PBLs and TILs. By contrast, the CD8 + T cell proportion tended to be higher in TILs, followed by PBLs and LNLs (Supplemental Figure 5). T cells generally infiltrate into local tissues after priming at draining LNs, where APCs present cognate antigens (16). In accordance with this, TILs dominantly contained CCR7 -CD45RAcells (effector memory T cells), whereas limited CCR7 + CD45RA + T cells (naive T cells) were observed compared with PBLs and LNLs (Figure 1 and 2). LNLs comprised a similar level of the naive T cell population as PBLs and an intermediate level between PBLs and TILs for effector memory T cells in both CD4 + and CD8 + T cells. Distal LNLs were comparable with proximal LNLs in both CD4 + and CD8 + T cells (Figure 1 and 2). Surgical-resected tonsils by chronic tonsillitis were provided as another cohort of control, showing significant differences in many populations from LNs that we analyzed in our CRC cohort (Supplemental Figure 6).

Activation status of LNLs is in an intermediate range between PBLs and TILs.
We further examined the T cell activation status according to PD-1 expression, as PD-1 expression is induced upon T-cell receptor (TCR) stimulation and associated with clinical responses by PD-1 blockade in some cancer types (19)(20)(21). PD-1 was highly expressed by TILs, followed by LNLs and PBLs in both CD4 + and CD8 + T cells ( Figure 3A). PD-1 expression was the highest in CCR7 -CD45RAeffector memory T cells in PBLs, LNLs and TILs (Figure 3B and C). These findings suggest that PD-1 expression may reflect the activation status of tumor antigen-specific T cells.
To gain further insight into the high PD-1 expression in TILs, we explored the expression of T-bet and Eomes since these transcription factors are reportedly associated with PD-1 expression (22,23). We divided T cells into three fractions: T-bet -Eomes -, Tbet high Eomes low and T-bet low Eomes high ( Figure 4A). There were significant differences in each population among PBLs, LNLs and TILs in CD8 + T cells. T-bet low Eomes high CD8 + T cells were significantly higher in LNLs and TILs than those in PBLs, although more than half of PBLs were composed of T-bet high Eomes low CD8 + T cells ( Figure 4A). PD-1 expression was the highest in T-bet low Eomes high CD8 + T cells, followed by Tbet high Eomes low CD8 + T cells and T-bet -Eomes -CD8 + T cells in PBLs, LNLs and TILs, which was consistent with previous reports ( Figure 4B) (22,23). Interestingly, PD-1 was noticeably expressed by TILs even in T-bet -Eomes -CD8 + T cells, suggesting that other factors may be related to the high PD-1 expression in TILs in addition to T-bet and Eomes.
By contrast, the T-bet -Eomespopulation was comparably dominant in CD4 + T cells from PBLs, LNLs and TILs (Supplemental Figure 7). The activation statuses of TILs, LNLs and PBLs in humans reflect the hypothesis developed by animal models; T cells that are activated in draining LNs infiltrate into tumor tissues.

Immune suppressive FOXP3 + CD4 + T cells are abundant in LNLs.
We next interrogated immune suppressive cells, particularly CD4 + regulatory T (Treg) cells, as Treg cells inhibit antitumor immunity and contribute to unfavorable clinical courses in CRCs (24). Correctly identifying CD4 + Treg cells in humans is compromised due to the upregulation of FOXP3, the master transcription factor of Treg cells, upon TCR stimulation in conventional T cells (25). We therefore proposed a classification of human Treg cells based on the expression levels of a naive marker, CD45RA and FOXP3.  Figure  5A) (24,26,27). eTreg cells have higher expression of CTLA-4, which is an important immune checkpoint molecule and plays a crucial role in Treg cell-mediated immune suppression (28), than that of FOXP3 low non-Treg cells (Supplemental Figure 8). In addition, other Treg-related molecules, such as CD39, ICOS and GITR (27), were highly expressed by eTreg cells (Supplemental Figure 8). These data support the notion that eTreg cells are bona fide Treg cells with an immune suppressive function and FOXP3 low non-Treg cells are different from immune suppressive Treg cells (24,26,27).
The frequencies of eTreg cells and FOXP3 low non-Treg cells were significantly higher in TILs than those in PBLs, while naive Treg cells were more abundant in PBLs than in TILs ( Figure 5A). Similarly, the frequency of CD45RA -FOXP3 -CD4 + T cells was significantly higher in TILs than that in PBLs whereas the frequency of CD45RA + FOXP3 -CD4 + T cells was significantly higher in PBLs than in TILs Interestingly, the frequency of FOXP3 low non-Treg cells in LNLs was variable in type A (n = 11) and type B (n = 11), although there was a trend to be higher, particularly 4 patients, in type B than in type A ( Figure 5A and B). Furthermore, Fusobacteria was found in stools from two type B CRC patients with high FOXP3 low non-Treg cells in LNLs ( Figure   5C). These findings suggest that the frequency of eTreg cells in LNLs is detected at an intermediate level between that of PBLs and TILs and that FOXP3 low non-Treg cells may be induced by inflammation caused by Fusobacteria, not only in TILs but also in LNLs.

TCR repertoire is shared between TILs and LNLs in dMMR CRC patients.
As the frequencies and immunological phenotypes of effector T cells and immune suppressive cells were in an intermediate range in LNLs, we asked whether T cells activated in draining LNs infiltrated into tumor tissues via exploring shared TCRβ between LNs and primary tumors. TCR diversity, which was evaluated with Shannon's index, was significantly higher in PBLs and LNLs than that in TILs ( Figure 6A). While TCR repertoires of naive CD8 + T cells harbored high TCR diversity, TCR repertoires of activated CD8 + T cells (detected as effector memory CD8 + T cells and PD-1 + CD8 + T cells) in tumors was significantly skewed ( Figure 6B). Together with previous reports (19,21), we therefore propose that the skewing of T cell clones, particularly PD-1 + CD8 + T cells in the TME reflect the activation of tumor antigen-specific T cells. Some TCRs were shared between PBLs or LNLs and TILs, and shared TCRs with LNLs were frequently found in TILs compared to those with PBLs ( Figure 6C). These shared TCRs were expanded from PBLs or LNLs to TILs in many patients, especially from proximal LNs to primary tumors ( Figure 6D). Ten CRC samples harbored a considerable level of shared TCRs in primary tumors with proximal LNs (> 10%), 7 of which were observed in patients with dMMR CRCs. Remaining dMMR CRCs had intermediately shared TCRs with proximal LNs (5%-10%) ( Figure 6E). By sharp contrast, more than half of pMMR CRC samples had few shared TCRs with proximal LNs (< 5%) ( Figure 6E). There was no significant difference in TCR diversity of PBLs, LNLs or TILs between pMMR and dMMR CRCs despite a significant difference in shared TCRs (Supplemental Figure   10A). Although we analyzed non-metastatic LNs, these data were also examined according to pathological staging (pStage) because LN metastases might affect nonmetastatic LNs. There was no significant difference in TCR diversity or shared TCRs between pStage I or II (pN0) and III or IV (pN1-2) (Supplemental Figure 10B). The frequency of the shared TCRs in TILs with distal LNLs showed a trend to be lower than that with proximal LNLs, although the difference was not significant due to the small number ( Figure 6C). These findings suggest that a variety of T cell clones activated in regional, especially proximal, LNs infiltrate into the TME of MSI-H/dMMR CRC and attack tumor cells.
Additionally, we performed whole exon sequencing (WES) for CRC samples with sufficient volume and found that dMMR CRC samples had significantly higher TMB including both non-synonymous single nucleotide variations and insertion/deletion (Supplemental Figure 11A). There was no significant difference between TCR diversity and TMB in both proximal LN and primary tumor (Supplemental Figure 11B).
Abundant shared TCRs between proximal LNs and primary tumors, although not significant, were detected in higher TMB samples (Supplemental Figure 11C). The TCR repertoire was highly overlapped between TILs and LNLs in dMMR CRC patients with high TMB, but not in patients with pMMR CRCs with low TMB, which can be explained by the presence of antigen-specific T cells against neo-antigens derived from gene alterations.

Intensive LN dissection may induce negative impacts on prognosis in patients with MSI-H/dMMR CRCs.
Many patients with dMMR CRCs harbored substantially shared TCRs between LNLs and TILs, while most patients with pMMR CRCs showed a limited overlap. Thus, we hypothesized that regional LNs could play an important role in developing antitumor immunity, particularly in patients with MSI-H/dMMR CRCs compared to those with MSS/pMMR CRCs. Then, the prognostic significance of the number of dissected LNs was investigated in another MSI-H cohort (SCC cohort) (Supplemental Table 2). The characteristics of 130 patients with MSI-H/dMMR CRCs, such as female, right side, early stage, and poor differentiation, were similar to previous reports (29)(30)(31). No significant difference in LN dissection number was observed according to tumor location (right vs. left) as well as NCCHE CRC cohort from which we collected TILs and LNLs (Supplemental Figure 1B). Because some guidelines have recommended examination of a minimum of 12 LNs (13,14), only 21 CRC patients in the cohort received LN dissection with less than 12, and the number of LNs dissected in patients with pN1-2 was comparable with that in patients with pN0, supporting the accurate staging in our cohort ( Figure 7A). Indeed, the presence of LN metastases in patients with LN dissection number < 12 was comparable with that in patients with > 12 (3/21 vs. 28/109, P = 0.41).
This can also be demonstrated by the reduced incidence of LN metastases in MSI-H/dMMR CRCs, which is different from MSS/pMMR CRCs (32,33). An ROC curve determined the cut-off value of the number of excessive LN dissection as 38 (Supplemental Figure 12). The presence of LN metastases in patients with LN dissection number < 38 was also comparable with that in patients with > 38 (23/96 vs. 8/34, P > 0.99). In all stage, the small number of LN dissection (< 12) was not related to the 13 unfavorable prognosis ( Figure 7B). Interestingly, the high number of LN dissection (> 38) exhibited a significantly shorter recurrence-free survival (RFS) compared with the low number of LN dissection (< 38) (Figure 7B). We next focused on patients with each pStage: no patient experienced recurrence in pStage 0 or I. In pStage II, the number of dissected LNs was not correlated with RFS, although the number of LN dissection is reportedly a prognostic factor in pMMR CRCs ( Figure 7C) (13,14). Furthermore, in pStage III and pN0, the high number of dissected LNs (> 38) corresponded to a slightly shorter RFS, but the difference was not significant, compared with that of other groups ( Figure 7C). Taken together, excessive LN dissection could provide a negative impact on long-term prognosis in patients with MSI-H/dMMR CRCs, by removing T cells that are activated in draining LNs and infiltrate into the TME to attack tumors.

Discussion
While the importance of comprehensive analyses from tumor tissues to the periphery is appreciated based on the data from animal models, systemic evaluation has been limited in humans. Here, we extensively explored the immunological status of tumors, non- Accurate staging is critical for considering the proper treatment method. "More is better" has been well established including Japanese clinical settings for the accurate staging of CRCs. Accordingly, some guidelines have recommended examination of a minimum of 12 LNs (13,14). T cells are generally primed at LNs and infiltrate into tumor tissues (15) although T cells could reportedly be primed and activated in tumor tissues (42). Thus, regional LNs often play an important role in developing antitumor immunity.
Accordingly, if regional LNs are removed or lymphocyte migration from LNs is prevented, antitumor immunity fails to control tumor progression in mouse models (43,44). In fact, considerably shared TCRs have been observed between LNs and primary tumors from MSI-H/dMMR CRCs with high TMB, potentially becoming tumor neoantigens. One can therefore envision that intensive LN dissection may have negative effects on antitumor immunity especially in MSI-H/dMMR CRCs with high TMB. Indeed, patients who received a limited LN dissection (the number of LNs dissected < 12) was not correlated with a shorter RFS. Rather, those who received an intensive LN dissection (the number of LNs dissected > 38), who were expected to have more accurate staging compared with the others, exhibited a slightly shorter RFS regardless of pStage, while limited LN dissection is reportedly associated with a poor prognosis in MSS/pMMR CRCs (13,14). Thus, we need to be careful with excessive non-metastatic LN dissections in MSI-H/dMMR CRCs with high TMB based on the following reasons: 1) the reduced incidence of LN metastases (32,33), 2) substantial shared T cell clones between LNs and primary tumors, 3) no influence of LN dissection number on the outcome after surgery, and poor prognosis after recurrence (37) in MSI-H/dMMR CRCs. Furthermore, ICBs, which are reportedly effective against MSI-H/dMMR CRCs (8,10,11), could exhibit more efficacy in settings where regional LNs are remained, even after surgery or neoadjuvant settings, because ICBs promote a new infiltration of tumor antigen-specific T cells (45) and regional LNs may harbor such tumor antigen-specific T cells as observed in our study. Although the detailed role of PD-1 + T cells in LNs remains unclear, recent mouse studies have demonstrated that LNs contained enriched PD-1 + tumor-specific progenitor T cells and that those PD-1 + T cells play an important role in PD-1 blockademediated antitumor immunity (46,47). By contrast to MSI-H/dMMR CRCs, patients with CRCs, most of which consist of MSS/pMMR, reportedly show a better prognosis after intensive LN dissection, meaning that a larger LN dissection leads to accurate staging and reduced residual lesions (13,14). This is consistent with our present study showing that very few shared TCRs between non-metastatic LNs and primary tumors were observed.
It is important to correctly identify metastatic LNs in regional LNs to avoid remaining residual metastatic LNs with leaving non-metastatic LNs. In clinical settings, it should be useful for surgeons to employ PET/CT to find metastatic LNs before surgery, and intraoperative rapid diagnosis for regional LNs also helps surgeons for the proper discrimination. Furthermore, optical image-guided cancer surgery is a promising technique to adequately determine tumor margins by tumor-specific targeting, potentially leading to complete resection of tumor tissues (48)(49)(50). Combining these techniques can achieve accurate discrimination of metastatic LNs and avoid excessive LN dissection.
A recent study of breast cancer patients reported that shared TCRs between draining LNs and primary tumors were often observed (51), probably due to the assay of metastatic LNs. We analyzed LNs without metastasis, and very few shared TCRs between LNs and primary tumors were observed in MSS/pMMR CRCs. Whereas one concern is that the LNs tested may not be tumor-associated LNs, we collected LN samples of CRCs based on the anatomical location of lymphatic flow. Another important aspect of tumorassociated LNs should be the reflection of the TME: increased frequencies of dysfunctional effector T cells and immune suppressive cells. PD-1 + CD8 + T cells and terminally-differentiated CD8 + T cells in proximal LNs tended to be higher compared with tonsils. Additionally, Treg cells were also more abundant in proximal LNs than in tonsils. The frequencies of these dysfunctional effector T cells and immune suppressive cells were gradually decreased in distal LNs, supporting that the LNs in our study were draining LNs of CRCs. Similar results showing few shared TCRs in LNs without metastasis were also observed in another study of CRC (52). Yet, the mechanism(s) of considerable shared TCRs between LNLs and TILs, especially in MSI-H/dMMR CRCs, remains unclear. One plausible explanation is that MSI-H/dMMR CRCs generally harbor high TMB that are recognized by the immune system as neo-antigens, and T cells specific for these neo-antigens are elicited in regional LNs (53). Indeed, MSI-H/dMMR CRCs had high TMB, potential neo-antigens, in our present study. Another possibility is that micro-metastases are present in these regional LNs and shared T cell clones are highly present, as in breast cancers. If there were pathological metastases including micrometastases in LNs, such LNs were excluded from our analyses. While we are not completely able to exclude micro-metastases, we carefully avoided to include metastatic LNs in our non-metastasis regional LNs as much as possible. The detailed mechanism(s) should be elucidated by further basic and translational studies.
In conclusion, we systemically analyzed the immunological status of tumors, nonmetastatic LNs and peripheral blood in CRC patients, showing several differences in immunological status in each site and very few shared TCRs between LNs and primary tumors, especially those from MSS/pMMR CRC patients. By contrast, substantial shared

TCRs were detected between non-metastatic LNs and primary tumors in MSI-H/dMMR
CRCs with high TMB. Furthermore, in our MSI-H/dMMR cohort, intensive LN dissection could provide negative impacts on long-term prognosis, suggesting that regional LNs play an important role in developing antitumor immunity against MSI-H/dMMR CRCs and that careful consideration needs to be paid regarding non-metastatic LN dissection, particularly in MSI-H/dMMR CRC patients. However, since this was a hypothesis-generating study with small patient cohort for whom we had appropriate to optimal therapies against MSI-H/dMMR CRCs (54,55).

Patients and samples.
Immunological phenotypes of paired PBLs, LNLs and TILs from 21 CRC patients who underwent surgical resection at National Cancer Center Hospital East between 2017 and 2020 were examined (NCCHE cohort). All surgeries were performed according to the Japanese guideline (17).

Microsatellite instability status.
MSI analysis was performed using fluorescence-based PCR as described previously (57).
Briefly, MSI status was determined using five Bethesda markers (BAT25, BAT26, D5S346, D2S123, and D17S250) and classified as MSI-H (when two or more markers were demonstrated to be unstable), MSI-low (MSI-L; when only one marker was unstable) and MSS (when no markers were unstable). MSI-positive markers were reexamined at least twice to confirm the results.

WES.
DNA was extracted from available frozen tumor samples and paired blood samples using Amplified DNA fragments underwent enrichment of exonic fragments using a SureSelect Human All Exon Kit v5 (Agilent Technologies, Santa Clara, CA). Massively parallel sequencing of isolated fragments was performed with a Novaseq6000 (Illumina, Sandiego, CA). Paired-end WES reads were independently aligned to the human reference genome (hs37d5) using BWA-MEM (58). Picard MarkdDuplicates (http://broadinstitute.github.io/picard/) was used to remove PCR duplicates. The resultant BAM files were processed using GATK tools (59). Local realignments of insertions and deletion was performed using GATK IndelRealigner. Systematic errors in base quality scores were corrected using GATK BaseRecalibrator. Somatic mutations and short indels were called using GATK MuTect2 (http://www.broadinstitute.org/cancer/cga/mutect).
Variants calls from Mutect2 filtered with GATK FilterMutectCalls. Mutations were discarded manually if the mutant allele read depth was < 5, the variant allele frequency was < 0.1 or the variant base was observed in the normal tissue. Gene mutations were annotated by ANNOVAR (60). The data were deposited in the Japanese Genotypephenotype Archive (accession number: JGAD000385).

Immunological phenotype analyses.
Flow cytometry assays were performed as described (56). The antibodies used in the flow cytometry analyses are summarized in Supplemental Table 3  TCR usage was analyzed as previously reported (61). Briefly, total RNA was prepared from paired proximal LNs and primary tumors. Complementary DNA was synthesized from total RNA, and TCR β chains were amplified using adaptor ligation-mediated PCR.
Then, using the PCR products as templates, TCR sequences were analyzed using Miseq according to the manufacturer's protocol. Alignments among approximately 10,000 sequences/run were performed with IMGT/V-QUEST (http://www.imgt.org).

Public data analyses
Differentially expressed genes between MSS and MSI-H CRCs were extracted from TCGA datasets using fold changes with false discovery rate by edgeR, and CIBERSORTx was performed as previously reported (12,18,62).

Statistical analysis.
The relations of continuous variables between or among groups were compared with ttest or one-way ANOVA, respectively. For multiple testing, Bonferroni correction was employed. The univariate relationship between each independent variable was examined using the Fisher's exact test. RFS was defined as the time from surgery until the first observation of disease progression. An ROC curve of LN dissection number for recurrence within 2 years after surgery was constructed to determine a cut-off value. The RFS was investigated with the Kaplan-Meier method and was compared among groups using the logrank test. All tests were two-tailed, and P values less than 0.05 were considered statistically significant. All of the above-mentioned statistical analyses were performed using Prism version 7 software (GraphPad Software, Inc., La Jolla, CA).

Study approval.
This study was approved by the Institutional Review Board of the National Cancer Center  PBLs, LNLs and TILs from 21 CRC patients who received surgical resection were prepared, and immunological phenotypes were examined with flow cytometry. Representative flow cytometry staining (upper) and summaries (lower) for the frequency of CCR7 + CD45RA + CD4 + T cells (naive), CCR7 + CD45RA -CD4 + T cells (central memory) CCR7 -CD45RA -CD4 + T cells (effector memory) and CCR7 -CD45RA + CD4 + T cells (terminally differentiated effector memory) in conventional CD4 + T cells of PBLs, LNLs and TILs. Means and SDs are shown, and statistical analyses were performed using the one-way ANOVA tests with Bonferroni corrections. CM, central memory; EM, effector memory; TEMRA, terminally differentiated effector memory; PB, peripheral blood; dLN, distal LN; pLN, proximal LN; ns, not significant.

Figure 3. PD-1 expression by T cells in PBLs, LNLs and TILs.
PBLs, LNLs and TILs from 21 CRC patients who received surgical resection were examined with flow cytometry as in Figure 1. (A) Representative flow cytometry staining (left) and summaries (right) for the frequency of PD-1 expressing cells in CD4 + and CD8 + T cells in PBLs, LNLs and TILs. (B) Summaries for the frequency of PD-1 expressing cells in total CD4 + T cells, CCR7 + CD45RA + CD4 + T cells (naive), CCR7 + CD45RA -CD4 + T cells (central memory) CCR7 -CD45RA -CD4 + T cells (effector memory) and CCR7 -CD45RA + CD4 + T cells (terminally differentiated effector memory) in PBLs, LNLs and TILs. (C) Summaries for the frequency of PD-1 expressing cells in total CD8 + T cells, CCR7 + CD45RA + CD8 + T cells (naive), CCR7 + CD45RA -CD8 + T cells (central memory) CCR7 -CD45RA -CD8 + T cells (effector memory) and CCR7 -CD45RA + CD8 + T cells (terminally differentiated effector memory) in PBLs, LNLs and TILs. Means and SDs are shown, and statistical analyses were performed using the one-way ANOVA tests with Bonferroni corrections. CM, central memory; EM, effector memory; TEMRA, terminally differentiated effector memory; PB, peripheral blood; dLN, distal LN; pLN, proximal LN; ns, not significant.  PBLs, LNLs and TILs from 21 CRC patients who received surgical resection were examined with flow cytometry as in Figure 1. (A) Representative flow cytometry staining (upper) and summaries (lower) for the frequency of FOXP3 + CD4 + T cell populations in PBLs, LNLs and TILs. CRCs are classified into two types according to FOXP3 low non-Treg cell infiltration in TME: type A (FOXP3 low non-Treg low) and B (FOXP3 low non-Treg high). (B) Differences in the frequency of eTreg cells and FOXP3 low non-Treg cells according to CRC subtypes. (C) Fecal metagenome was analyzed by 16S rRNA sequencing. Fusobacteria was frequently found in stools from type B. Means and SDs are shown, and statistical analyses were performed using the one-way ANOVA tests with Bonferroni corrections in A and the t-tests in B, respectively. PB, peripheral blood; dLN, distal LN; pLN, proximal LN; ns, not significant. PBLs, LNLs and TILs from 21 CRC patients who received surgical resection were prepared, and the TCR sequencing was performed with next-generation sequencing. If there were remaining samples, the TCR sequencing was performed for PBLs and LNLs of distal LNs. (A) Diversity of the TCR repertoire in PBLs, LNLs and TILs evaluated with Shannon's index. The top 10 TCRs are colored (left), and the summary of Shannon's index is shown (right). (B) Correlation between TCR diversity and CCR7 + CD45RA + CD8 + T cell (naive), CCR7 -CD45RA -CD8 + T cell (effector memory) or PD-1 + CD8 + T cell proportion. Black, peripheral blood; blue, distal LN; green, proximal LN; red, primary tumor.