Patient characteristics

In this study, 62,447 stage III CRC patients were assigned to train cohort, 41,631 patients were assigned to test cohort from SEER database, 2,622 patients from FAH-SYSU database and 143 from TCGA database were analyzed in this study. We compared baseline characteristics of CRC patients from these databases (Table 1). The cumulative 1-, 3, 5 years OS were 86.2%, 68.8% and 57.2% in train cohort, and were 86.1%, 68.7% and 57.1% in test cohort, were 95.5%, 77.9% and 67.4% in FAH-SYSU cohort and 90.9%, 73.5% and 50.2% in TCGA cohort.

Prognostic impact of the ELNs count for stage III CRC patients

In conducting clinicopathological analyses, both univariate and multivariate Cox proportional hazards regression models were utilized to forecast OS across the examined cohorts (Supplementary Tables 1–3). The results suggested that tumor location, size, pathological type, differentiation status, pathological TNM stage, administration of adjuvant therapy, and ELNs count were positively associated with OS in stage III CRC patients in the training cohort, test cohort, FAH-SYSU cohort and TCGA cohort (P < .05 for all). Aligning with these observations across the overall cohort, an increased ELNs count was distinctly associated with improved OS outcomes within both the N1 and N2 subgroups (P < .05), as detailed in Supplementary Table 4.

Innovative subtype designation for patients with III CRC based on the ELNs counts

For a more thorough evaluation, a Cox regression model with RCS was developed to identify an innovative subtyping strategy and show the link between OS risk and the ELNs count in patients with stage III CRC. The results of RCS analyses supported the established cut-off values for ELNs (Fig. 1A-B). The results of the RCS model adjusted for age at CRC diagnosis, sex, tumor location, tumor size, pathological type, differentiation status, and pathological T stage revealed a significant linear association between the ELNs count and OS in stage III CRC patients (Poverall<0.001, Pnonlinear<0.001) (Fig. 1B and Supplementary Fig. 1). The 50th quintile of the ELNs count was 16. Consistent with the results observed in the overall cohort, this ELNs cut-off was noted in the right colon, left colon and rectal cancer subgroups (Fig. 1C). The AJCC have established a standard yield of at least 12 ELNs for CRC. Figure 1B shows different cut-off values for the ELNs yield in predicting OS. To ensure survival and achieve accurate representation and generalizability, we suggest using 16 ELNs as the ideal yield cut-off, as determined by the RCS curve (Fig. 1C). The stage III CRC patients were then divided into three groups: the small number of ELNs (SN-ELN; ELNs count ≤ 11), medium number of ELNs (MN-ELN; 12 ≤ ELNs count < 16), and large number of ELNs (LN-ELN; ELNs count ≥ 17) groups. In both univariate and multivariate Cox regression analyses, the ELNs-related subtype was associated with OS in all cohorts (P < .05) (Table 2, Supplementary Tables 5–8).

Fig. 1

Innovative subtype for III CRC based on number of examined lymph nodes (ELNs). (A) Distribution of stage III CRC patient population across varyingELNs count categories. (B) A restricted cubic spline graph depicting the relationship between ELNs count and overall survival (OS). The graph illustrates the adjusted hazard ratio (HR, represented by the solid line) and the 95% confidence interval (CI, shaded in blue), demonstrating the association between the number of ELNs and OS in CRC patients. Lower HR correspond to improved survival outcomes. (C) derived from univariable Cox regression analysis for OS and cancer-specific survival (CSS) across distinct ELNs count thresholds, and the ELNs counts ≤ 12 serve as the reference group for comparison

Table 2 Multivariable Cox regression analysis in stage III colorectal cancer patients based on OS in different cohortsPrognostic impact of the ELNs-related subtype in stage III CRC patients

Survival analyses for stage III CRC patients were conducted according to ELNs-related subtypes. Kaplan-Meier (K-M) curves indicated superior OS and CSS for the LN-ELN group compared to the MN-ELN and SN-ELN groups within the training cohort, with the following HRs and CIs: MN-ELN vs. SN-ELN: HR = 0.83, 95% CI = 0.80–0.86, P < .001; LN-ELN vs. SN-ELN: HR = 0.71, 95% CI = 0.68–0.72, P < .001 for OS (Fig. 2A) ; and MN-ELN vs. SN-ELN: HR = 0.82, 95% CI = 0.78–0.85, P < .001; LN-ELN vs. SN-ELN: HR = 0.68, 95% CI = 0.65–0.71, P < .001 for CSS (Fig. 2B). To corroborate the prognostic significance of ELNs-related subtyping, comparisons of OS, CSS, and progression-free survival (PFS) were extended to the LN, MN, and SN-ELN groups across the test, FAH-SYSU, and TCGA cohorts. These K-M analyses reaffirmed that patients in the LN-ELN group consistently exhibited better OS, CSS, and PFS than those in the MN-ELN and SN-ELN groups, with statistical significance observed across the board (MN-ELN vs. SN-ELN and LN-ELN vs. SN-ELN, P < .05), as detailed for OS (Fig. 2C) and CSS (Fig. 2D) in test cohort, OS (Fig. 2E) in FAH-SYSU cohort, OS (Fig. 2F), CSS (Fig. 2G) and PFS (Fig. 2H) in TCGA cohort.

Fig. 2

The prognostic impact of ELNs-related subtypes on stage III CRC patients. KM curves of the (A) a small number ELNs (SN-ELN) ≤ 11,12 ≤ a medium number of ELNs (MN-ELN) ≤ 16 and 17 ≤ a large number of ELNs (LN-ELN) for overall survival (OS) in train cohort; (B) cancer-specific survival (CSS) in train cohort; (C) OS in test cohort; (D) CSS in test cohort; (E) OS in FAH-SYSU cohort; (F) OS in TCGA cohort; (G) CSS in TCGA cohort; (H) progression free survival (PFS) in TCGA cohort. And then, the survival analysis was used to evaluated the OS for patients with both N1 stage and LN-ELN (N1-LN-ELN), N1-MN-ELN, N1-SN-ELN, both N2 stage and LN-ELN (N2-LN-ELN), N2-MN-ELN and N2-SN-ELN in (I) train cohort, (J) test cohort, (K) FAH-SYSU cohort and (L) TCGA cohort. The survival analysis was used to evaluate the OS for patients with N2-LN-ELN and N1-SN-ELN in (M) train cohort, (N) test cohort, (O) FAH-SYSU cohort and (P) TCGA cohort

To ascertain the significance of ELNs-related subtyping in patients with pathological N1 and N2 staging, we assessed the OS across the LN-, MN-, and LN-ELN groups within various cohorts. Patients with a higher ELNs count had better OS in both the N1 and N2 stage subgroups. Across the four cohorts, patients in the both N1 stage and LN-ELN (N1-LN-ELN) subgroup had the best OS, while those in the both N2 stage and LN-ELN (N2-LN-ELN) subgroup had the least favorable OS in train cohort (Fig. 2I), in test cohort (Fig. 2J), in FAH-SYSU cohort (Fig. 2K), and TCGA cohorts (Fig. 2L) (P < .05). Comparative analysis revealed no significant OS differences between patients in the N2-LN-ELN and N1-SN-ELN subgroups across the training (Fig. 2M), testing (Fig. 2N), FAH-SYSU (Fig. 2O) and in TCGA cohort (Fig. 2P) (P < .05), despite the association of pathological N stage with OS in both univariate and multivariate Cox regression analyses (Table 2). These results suggest that an elevated ELNs count might mitigate the adverse effects of LN metastasis on survival.

Additionally, subgroup survival analyses were performed based on the three ELNs-related subtype groups. Consistent with the results observed in the overall cohort, the LN-ELN status was significantly related to better OS for different age, sex, tumor differentiation and TNM stage subgroups in the SEER and FAH-SYSU cohorts (P < .05) (Fig. 3A-B).

Fig. 3

Subgroup analyses of ELNs-related subtypes of stage III CRC patients in SEER and FAH-SYSU cohort. Forest plot showing the factors associated with overall survival (OS) of stage III CRC patients with ELNs-related subtypes in (A) SEER and (B) FAH-SYSU cohort

To delve deeper into the impact of the ELNs count on the efficacy of AC in stage III CRC patients, we examined the prognosis following AC treatment versus no AC treatment within the LN, MN and SN-ELN groups in the FAH-SYSU cohort (Fig. 4A-F). Comparisons of the baseline characteristics of stage III CRC patients in the FAH-SYSU cohort were performed between the Yes-AC group and the No-AC group before and after PSM (Supplementary Tables 9–11). Notably, in the LN-ELN group, a significant difference in OS was identified between the Yes-AC and No-AC groups both pre- and post-PSM (HR = 0.58, 95% CI = 0.43–0.78, P < .001 before PSM (Fig. 4C); HR = 0.52, 95% CI = 0.33–0.83, P = .005 after PSM (Fig. 4F)). A similar trend was observed in the MN-ELN group, where significant OS differences were noted between the Yes-AC and No-AC groups before and after PSM (HR = 0.60, 95% CI = 0.41–0.89, P = .012 before PSM (Fig. 4B); HR = 0.53, 95% CI = 0.29–0.95, P = .034 after PSM (Fig. 4E)). Conversely, within the SN-ELN group, no significant OS disparities were found between the Yes-AC and No-AC groups, neither before nor after PSM (HR = 0.98, 95% CI = 0.71–1.34, P = .879 before PSM (Fig. 4A); HR = 1.14, 95% CI = 0.72–1.80, P = .578 after PSM (Fig. 4D)), suggesting that AC may not enhance prognosis for stage III CRC patients with a low ELNs count.

Fig. 4

The survival analysis between receiving postoperative adjuvant chemotherapy (Yes-AC) and not receiving postoperative adjuvant chemotherapy (No-AC) group within different ELNs related subtypes group before and after propensity score matching (PSM) based on the FAH-SYSU cohort. To lessen selection bias, PSM matched the baseline parameters of the two groups in a ratio of 1:1, including age at diagnosis, gender, tumor location, tumor grade, histology type, tumor size, pathological T and N stage. (A) Overall survival (OS) comparison between YES-AC and No-AC group within SN-ELN before PSM. (B) OS comparison between YES-AC and No-AC group within MN-ELN before PSM. (C) OS comparison between YES-AC and No-AC group within LN-ELN before PSM. (D) OS comparison between YES-AC and No-AC group within SN-ELN after PSM. (E) OS comparison between YES-AC and No-AC group within MN-ELN after PSM. (F) OS comparison between YES-AC and No-AC group within LN-ELN after PSM

Evaluation of tumor microenvironment (TME) characteristics among the three ELNs-related subtypes

To discern the transcriptomic variations among stage III CRC patients categorized into the LN-, MN- and SN-ELN groups, RNA sequencing data from tumor samples within the TCGA cohort were analyzed. This analysis entailed comparing gene expression in tumor tissues between the LN-ELN group and the combined MN- and SN-ELN groups, as well as between the SN-ELN group and the combined MN- and LN-ELN groups. The comparison indicated that in the LN-ELN group, there were 2243 genes identified as upregulated and 228 as downregulated (Fig. 5A). Advanced comparisons to elucidate the biological distinctions between the LN-ELN group and the others (MN-ELN and SN-ELN) involved Gene Ontology (GO), hallmark gene set, and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses. These analyses highlighted significantly upregulated genes within the LN-ELN group (Fig. 5B-D), showcasing enriched terms and pathways related to immune response-activating signal transduction, B-cell-mediated immunity, T-cell receptor signaling pathway, and differentiation of Th1 and Th2 cells. Conversely, the comparison involving the SN-ELN group against the MN- and LN-ELN groups revealed 989 upregulated and 3100 downregulated genes in the SN-ELN group (Fig. 5E), with GO, hallmark gene set, and KEGG enrichment analyses pointing to enrichment in terms related to cell development and differentiation within the SN-ELN group (Fig. 5F-G).

Fig. 5

The analysis on potential biological mechanisms among different ELNs-related subtypes. (A) The volcano plot showed the differentially expressed genes (DEGs) between LN-ELN and MN- and SN-ELN groups in TCGA cohort. Red points: up-expression DEGs in LN-ELN with log2-fold change > 0.5 and P < .05; Green points: down-expression DEGs in LN-ELN with log2-fold change < -0.5 and P < .05; Black point: gene expression with |log2-fold change| < 0.5 or P < .05. The (B) Hallmarker, (C) GO and (D) KEGG analyzed presented the enrichment biological pathways between LN-ELN and MN- and SN-ELN groups. (E) The volcano plot showed the DEGs between SN-ELN and MN- and LN-ELN groups. The (F) Hallmarker and (G) GO analyzed presented the enrichment biological pathways between SN-ELN and MN- and LN-ELN groups. (H) The level of tumor microenvironment cell infiltration among LN-, MN- and SN-ELN groups by using the RNA-seq sequencing in TCGA cohort. (I) The characters of gene mutation among LN-, MN- and SN-ELN groups by using the DNA-seq sequencing in TCGA cohort. (J) The level of CD4 + T cell and CD8 + T cell compared among LN-, MN- and SN-ELN groups by using multiplex immunofluorescence method in FAH-SYSU cohort. The level of CD4 + T cell and CD8 + T cell on serial tissue microarrays

To investigate the variances in immune cell composition among the SN-, MN- and LN-ELN groups, we employed a deconvolution approach using CIBERSORT and Gene Set Variation Analysis (GSVA) tools (Fig. 5H). The results indicated pronounced differences in TME. Specifically, the quantities of T cells, follicular helper T cells, activated memory CD4 + T cells, M1 macrophages, and the neoantigen load were notably higher in the LN-ELN group compared to the SN- and MN-ELN groups (Fig. 5H). Additionally, the gene mutations were observed within these three groups (Fig. 5I).

To refine our understanding of immune cell infiltration in stage III CRC patients across the ELNs-related subtypes, MIF was used to quantify CD3+, CD4 + and CD8 + T lymphocytes within tumor tissues (Fig. 5J). The analysis indicated notable increases in the numbers of CD3+, CD4 + and CD8 + T cells in the LN-ELN subgroup compared with the SN-ELN subgroup.