Tertiary lymphoid structure-related genes drive tumor microenvironment heterogeneity and prognostic disparities in left- vs. right-sided colon cancer

Introduction

Colorectal cancer (CRC) ranks as the third most common malignancy and the second leading cause of cancer-related mortality worldwide, with over 1.9 million new cases and approximately 900,000 deaths annually (1). Despite advances in screening, diagnosis, and targeted therapies, CRC exhibits profound heterogeneity, particularly between left-sided colorectal cancer (LCC) and right-sided colorectal cancer (RCC) (2). LCC, encompassing tumors in the descending colon, sigmoid colon, splenic flexure, and rectosigmoid junction, often presents with distinct clinical features such as rectal bleeding and obstruction, while RCC, including the cecum, ascending colon, and hepatic flexure, is associated with iron-deficiency anemia and advanced stages at diagnosis (3). These anatomical differences translate into divergent molecular profiles: RCC is characterized by higher rates of microsatellite instability, BRAF mutations, and CpG island methylator phenotype, whereas LCC shows more frequent chromosomal instability, KRAS mutations, and APC alterations (4). Consequently, RCC patients generally experience poorer prognosis, reduced response to conventional chemotherapy, and limited benefits from anti-EGFR therapies compared to LCC counterparts (5). This sidedness-specific heterogeneity underscores the need for tailored therapeutic strategies, yet, the underlying mechanisms remain incompletely elucidated.

Emerging evidence highlights the role of the tumor immune microenvironment in CRC progression and sidedness disparities. Tertiary lymphoid structures (TLS), ectopic lymphoid aggregates resembling secondary lymphoid organs, form within tumors and facilitate adaptive immune responses, including B-cell maturation, T-cell priming, and antigen presentation (6). TLS presence correlates with improved prognosis in various cancers, including CRC, by enhancing anti-tumor immunity through chemokine recruitment (e.g., CXCL9, CXCL11) and lymphocyte activation (7). TLS-related genes (TLSRGs), such as chemokines (CXCL13, CCL21), immune checkpoints (PDCD1, ICOS), and B/T-cell markers (MS4A1, CD38), orchestrate leukocyte recruitment and functional reprogramming within the tumor microenvironment (TME). These genes drive site-specific immune landscapes and modulate therapeutic vulnerabilities through the activation of divergent signaling pathways in LCC versus RCC (8,9). RCC often displays enriched inflammatory and interferon pathways, while LCC shows hypoxia and complement activation dominance (10). However, comprehensive analyses integrating multi-omics data—encompassing bulk transcriptomics, somatic mutations, single-cell RNA sequencing, and cell communication—are lacking, particularly in elucidating TLSRGs’ contributions to CRC sidedness heterogeneity and prognostic modeling (11).

While the prognostic significance of TLS in CRC is increasingly recognized, and recent multicenter studies have successfully developed histopathology-based nomograms to predict outcomes in LCC and RCC [e.g., Mao et al., 2023 (12)], the molecular mechanisms driving these sidedness-specific immune disparities remain under-explored. Specifically, a systematic integration of somatic mutations, single-cell transcriptomics, and intercellular communication networks to delineate the ‘TLS-permissive’ microenvironment is lacking. Unlike histological models, our study aims to construct a transcriptome-based signature that reflects the underlying biological machinery—ranging from metabolic stress to cell cycle dysregulation—that governs TLS efficacy (Figure 1). We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2367/rc).

Figure 1 The workflow diagram of this study. AUC, area under the curve; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis; KM, Kaplan-Meier; LCC, left-sided colorectal cancer; RCC, right-sided colorectal cancer; ROC, receiver operating characteristic; ssGSEA, single-sample gene set enrichment analysis; TCGA-COAD, The Cancer Genome Atlas Colon Adenocarcinoma; TLSRG, tertiary lymphoid structures-related gene.

Methods

Data acquisition and preprocessing

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. CRC transcriptomic data were integrated from The Cancer Genome Atlas Colon Adenocarcinoma (TCGA-COAD) and Gene Expression Omnibus (GEO) databases. For TCGA-COAD, RNA-Seq raw counts were downloaded from the Genomic Data Commons (GDC) portal, comprising 424 patient samples (50 normal and 374 tumor tissues). Somatic mutation data were also retrieved from TCGA. Low-expression genes (expressed as zero in >50% of samples) were filtered out. Tumors were classified into LCC and RCC based on anatomical sites: RCC included cecum, ascending colon, and hepatic flexure; LCC included descending colon, sigmoid colon, splenic flexure, and rectosigmoid junction, yielding 189 RCC and 135 LCC samples. For validation, the GSE103479 dataset was downloaded from GEO, containing 149 CRC samples reclassified as 62 RCC (ascending colon, hepatic flexure, cecum, transverse colon) and 87 LCC (descending colon, sigmoid colon, rectum). Batch effects were removed using the sva package (v3.50.0) in R. A total of 39 TLSRGs were included in this study (Table S1).

Consensus clustering

Consensus clustering was performed to identify molecular subtypes in CRC samples based on the expression matrix of TLSRGs. The ConsensusClusterPlus package (v1.64.0) was employed with a maximum cluster number (K) of 6, 50 resamplings (each subsampling 80% of samples and features), partitioning around medoids (PAM) algorithm, and Pearson correlation as the distance metric. K=2 was selected for subgrouping into cluster1 (C1) and cluster2 (C2). Principal component analysis (PCA) was conducted to validate subtype separation. Proportions of LCC/RCC in each subtype and vice versa were calculated.

Enrichment analysis

Gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) were conducted to elucidate pathway differences between LCC and RCC using hallmark gene sets from the Molecular Signatures Database (MSigDB; h.all.v2025.1.Hs.symbols.gmt). For GSEA, a ranked gene list was generated from t-statistics derived via limma differential analysis (v3.58.0) of RCC versus LCC. Parameters included minGSSize =10, maxGSSize =500, pvalueCutoff =0.25 with Benjamini-Hochberg adjustment, and 1,000 permutations for significance estimation using clusterProfiler (v4.10.0). Results were visualized as ridge plots (top 30 pathways) and enrichment curves. GSVA computed pathway scores non-parametrically with a Gaussian kernel (kcdf = “Gaussian”, minSize =5, maxSize =500) via the GSVA package (v1.50.0). Analyses encompassed Wilcoxon tests for TLSRG score differences by site, Spearman correlations between TLSRG and hallmark pathways [false discovery rate (FDR) <0.0001 threshold], and gene-pathway network construction via clustered heatmaps (Euclidean distance, complete linkage) with significance markers (P<0.05).

Somatic mutation analysis

TCGA-COAD somatic mutation data were integrated with clinical information, incorporating hemicolon grouping (LCC/RCC) into mutation annotation format (MAF) objects. TLSRG-based mutation subsets were extracted for LCC and RCC groups. Mutation frequency differences were visualized using coBarplot for side-by-side bar plots.

Drug sensitivity and immune infiltration analysis

Drug sensitivity in TCGA-COAD was assessed using the pRRophetic package, stratified by hemicolon. Half-maximal inhibitory concentration (IC₅₀) values for LCC and RCC were compared via Wilcoxon rank-sum tests to identify differentially sensitive drugs. Spearman correlations were computed between IC₅₀ values, TLSRG scores, and gene expressions, visualized as heatmaps. Immune cell proportions were estimated using CIBERSORT, with Wilcoxon tests identifying significant differences between LCC and RCC. Spearman correlations linked these proportions to TLSRG scores and expressions, supplemented by scatter plots with regression lines and coefficients for significant cells versus TLSRG scores.

Construction and evaluation of prognostic risk model

A candidate gene pool was formed by merging TLSRG-correlated pathway genes (FDR <0.0001) with TLSRGs (2,662 genes total). Prognostic genes were screened via univariate Cox regression for OS association in TCGA-COAD and GSE103479, retaining intersections (8 genes: HSPB1, NCOR2, STIL, ASNS, GCLM, SLC4A4, NPTXR, YKT6). To construct a robust prognostic model, we employed an integrative machine learning pipeline using the Mime package (https://github.com/l-magnificence/Mime) (13). The Mime package integrates 10 established machine learning algorithms: Random Survival Forest (RSF), CoxBoost, Stepwise Cox (StepCox), Lasso, Ridge, Elastic Net (Enet), Partial Least Squares Cox (plsRcox), Gradient Boosting Machine (GBM), Survival Support Vector Machine (survivalSVM), and Supervised Principal Components (SuperPC). Using TCGA-COAD as the training set, we evaluated these algorithms and their combinations (101 combinations in total) using a 10-fold cross-validation approach. The random seed was fixed (seed =123) to ensure reproducibility. The performance of each model was quantified using the concordance index (C-index) in both the training and validation (GSE103479) cohorts. The SuperPC algorithm (implemented internally via the standard SuperPC methodology within the Mime framework) exhibited the highest and most consistent C-index and was selected to construct the final risk model. Performance was evaluated via C-index distribution plots. Survival analysis used Kaplan-Meier (KM) curves for high/low-risk groups (median risk score cutoff), log-rank tests, and confidence intervals. Time-dependent receiver operating characteristic (ROC) curves assessed 1-/3-/5-year predictive accuracy. Subgroup analyses computed ROC and KM curves for LCC/RCC separately.

Single-cell analysis

Single-cell RNA-seq data from GSE200997 (CRC tumors) were loaded and processed using Seurat (CreateSeuratObject), filtering for tumor samples and quality control via mitochondrial gene percentage. Normalization (NormalizeData), variable feature identification (FindVariableFeatures, n=3,000), and PCA were performed. Batch effects from hemicolon were corrected using Harmony. Dimensionality reduction (t-SNE/UMAP) and clustering yielded 14 clusters annotated into seven cell types (e.g., EPCAM for epithelial cells, CD3D for T cells, LYZ for myeloid cells) based on marker genes. Cell proportions were compared between LCC/RCC via stacked bar plots and Wilcoxon tests. TLSRG scores were calculated using ssGSEA, visualized spatially by median grouping. Cell communication was analyzed with CellChat (v1.6.0) using the “Secreted Signaling” ligand-receptor database (CellChatDB.human). Separate CellChat objects were created for LCC/RCC, identifying overexpressed interactions and aggregating pathway-level probabilities. Group comparisons merged objects for differential network visualization (netVisual_diffInteraction) and heatmaps (netAnalysis_signalingRole_heatmap).

Statistical analysis

All analyses were conducted in R (v4.5.0). Group comparisons used Wilcoxon rank-sum tests; correlations employed Spearman tests; enrichment via GSEA/GSVA (clusterProfiler/GSVA); prognostic modeling integrated univariate Cox with machine learning (Mime), evaluated by C-index and KM curves (log-rank); cell communication via CellChat. Statistical significance was defined as P<0.05.

Results

Differential expression, mutation, and prognostic significance of TLSRGs in LCC vs. RCC

To systematically analyze the distinctions in TLSRGs between LCC and RCC, we first performed somatic mutation analysis. As depicted in Figure 2A, the somatic mutation landscape of TLSRGs exhibited significant differences between LCC and RCC. In LCC, the highest mutation frequencies were observed in FBLN7 (25%), CD4 (20%), and IL1R1 (20%). Conversely, in RCC, the highest mutation frequencies were found in MS4A1 (16%), CCR5 (14%), and CD4 (14%). Expression analysis revealed significant differential expression of 27 TLSRGs between LCC and RCC (Figure 2B). Among these, TRAF6 and CXCR3 showed significantly higher expression in LCC, while the remaining 25 genes exhibited lower expression in LCC compared to RCC. Regarding prognosis, several TLSRGs demonstrated significant associations with clinical outcomes in the overall cohort: TRAF6, IL2RA, CXCL11, and CD38 were significantly associated with disease-free interval (DFI); PDCD1 and FBLN7 with disease-specific survival (DSS); ICOS and CXCL11 with overall survival (OS); and CXCL11, FBLN7, and CXCL9 with progression-free interval (PFI) (Figure 2C). Importantly, the prognostic significance of TLSRGs differed markedly between LCC and RCC subtypes (Figure 2D). In LCC: SGPP2 [hazard ratio (HR) =0.52, P<0.05] was associated with OS; FBLN7 (HR = 7.09, P<0.05) with DSS; MS4A1 (HR =7.79, P<0.05) and TRAF6 (HR =3.55, P<0.05) with DFI; and FBLN7 (HR =6.87, P<0.05), CXCL11 (HR =0.71, P<0.05), IL1R1 (HR =1.56, P<0.05), and CCL21 (HR =1.26, P<0.05) with PFI. In RCC: ICOS (HR =0.42, P<0.05) was associated with OS; PDCD1 (HR =1.59, P<0.05) with DSS; TRAF6 (HR =12.05, P<0.05), IL2RA (HR =0.18, P<0.05), and IGSF6 (HR =0.29, P<0.05) with DFI; and CXCL11 (HR =0.82, P<0.05) and GFI1 (HR =0.64, P<0.05) with PFI.

Figure 2 Comparative analysis of TLSRG mutations, expression, and prognostic associations in LCC and RCC. (A) Co-barplot illustrating differential somatic mutation frequencies of TLSRGs in LCC and RCC. (B) Differential expression of TLSRGs between LCC and RCC assessed by Wilcoxon test. (C) Forest plot depicting HRs for associations between TLSRGs and OS, DSS, DFI, and PFI in the TCGA-COAD cohort. (D) Forest plot depicting HRs for associations between TLSRGs and OS, DSS, DFI, and PFI stratified by LCC and RCC within the TCGA-COAD cohort. Significance levels: ns, not significant; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. DFI, disease-free interval; DSS, disease-specific survival; HR, hazard ratio; LCC, left-sided colorectal cancer; OS, overall survival; PFI, progression-free interval; RCC, right-sided colorectal cancer; TCGA-COAD, The Cancer Genome Atlas Colon Adenocarcinoma; TLSRG, tertiary lymphoid structures-related gene.

Consensus clustering and GSVA enrichment based on TLSRGs

To further investigate the relationship between TLSRGs and the molecular heterogeneity of LCC and RCC, we performed consensus clustering using TLSRGs on the TCGA-COAD cohort, identifying two distinct molecular subtypes (C1 and C2). PCA demonstrated discernible separation between the two subtypes (Figure 3A). The proportion of C1 was higher than C2 in both LCC and RCC, and the C1/C2 ratio was significantly higher in LCC compared to RCC (Figure 3B). Furthermore, the proportion of RCC cases was significantly higher within the C1 subtype compared to the C2 subtype (Figure 3C). GSVA enrichment scores for the TLSRG gene set were significantly elevated in RCC compared to LCC (Figure 3D). Subgroup analysis revealed that within both LCC and RCC, the GSVA score was significantly higher in the C1 subtype than in C2 (Figure 3E). Within the C2 subtype, the GSVA score was significantly higher in RCC than in LCC; however, no significant difference was observed between LCC and RCC within the C1 subtype (Figure 3F).

Figure 3 Consensus clustering and GSVA enrichment assessment based on TLSRGs in colon cancer. (A) PCA plot showing separation of the two clusters (C1 and C2) identified by consensus clustering. (B) Stacked bar chart showing the distribution of C1 and C2 subtypes within LCC and RCC groups. (C) Stacked bar chart showing the distribution of LCC and RCC cases within C1 and C2 subtypes. (D) Comparison of TLSRG GSVA enrichment scores between LCC and RCC. (E) Comparison of TLSRG GSVA scores between C1 and C2 subtypes within LCC and RCC groups. (F) Comparison of TLSRG GSVA scores between LCC and RCC within C1 and C2 subtypes. Significance levels: ns, not significant; ****, P<0.0001. GSVA, gene set variation analysis; LCC, left-sided colorectal cancer; PCA, principal component analysis; RCC, right-sided colorectal cancer; TLSRG, tertiary lymphoid structures-related gene.

Single-cell level expression analysis of TLSRGs

We further assessed TLSRG expression differences at single-cell resolution. Clustering analysis of the GSE200997 dataset identified 14 clusters, which were annotated into 7 major cell types (Figure 4A). The cellular composition differed between LCC and RCC (Figure 4B): LCC exhibited a higher proportion of T cells, while RCC showed higher proportions of epithelial cells, B cells, myeloid cells, and fibroblasts. Using single-sample gene set enrichment analysis (ssGSEA), we calculated TLS enrichment scores at the single-cell level. The overall TLS score was significantly lower in RCC compared to LCC (Figure 4C). Analysis at the cellular level revealed that T cells, epithelial cells, B cells, and myeloid cells in LCC had significantly higher TLS scores than their counterparts in RCC. Conversely, endothelial cells and dendritic cells in LCC had significantly lower TLS scores than those in RCC (Figure 4D). We also observed cell-type-specific expression patterns of individual TLSRGs: MS4A1 was prominently expressed in B cells, CD38 in dendritic cells, and CCL5 in T cells (Figure 4E).

Figure 4 Single-cell resolution analysis of TLSRG expression. (A) UMAP plot showing clustering, cell type annotation, and ssGSEA-derived TLS scores for cells in the GSE200997 dataset. (B) Stacked bar chart showing cellular proportions in LCC and RCC samples within the GSE200997 dataset. (C) Comparison of overall TLS ssGSEA scores between LCC and RCC. (D) Comparison of TLS ssGSEA scores for matched cell types between LCC and RCC. (E) Expression levels of representative TLSRGs across different cell types. Significance levels: ns, not significant; *, P<0.05; **, P<0.01; ****, P<0.0001. LCC, left-sided colorectal cancer; RCC, right-sided colorectal cancer; ssGSEA, single-sample gene set enrichment analysis; TLS, tertiary lymphoid structures; TLSRG, tertiary lymphoid structures-related gene; UMAP, Uniform Manifold Approximation and Projection.

Differential cell-cell communication in LCC vs. RCC

We next evaluated differences in intercellular communication networks between LCC and RCC. The total number of inferred interactions and the overall interaction strength were significantly lower in RCC compared to LCC (Figure 5A). While the major communication patterns occurred between similar cell types in both groups (Figure 5B,5C), the interaction strength was generally higher in LCC. Notably, epithelial cell interactions with T cells and B cells were stronger in LCC. Analysis revealed 8 signaling pathways shared between LCC and RCC (Figure 5D). Among these, the MIF and MK pathways were the most active. Within these shared pathways, epithelial cells were the dominant signal senders in RCC, whereas T cells were the dominant signal senders in LCC.

Figure 5 Analysis of cell-cell communication networks in LCC and RCC. (A) Comparison of the number of inferred interactions and overall interaction strength between LCC and RCC. Circle plots depicting the major intercellular communication networks in (B) LCC and (C) RCC. Line thickness represents interaction strength. (D) Signaling role heatmap illustrating the relative activity of shared signaling pathways and the dominant signaling roles (sender cells) in LCC and RCC. LCC, left-sided colorectal cancer; RCC, right-sided colorectal cancer.

Validation of TLSRG expression and prognostic associations in an independent cohort

We validated TLSRG expression and prognostic associations using the independent GSE103479 cohort. Differential expression analysis identified 5 TLSRGs with significantly altered levels: CD38, CXCL13, and MS4A1 were upregulated in LCC, while CCL5 and IL1R1 were downregulated (Figure 6A). Prognostic analysis within the entire GSE103479 cohort showed associations between specific TLSRGs and outcomes: CCL5 (HR =1.17, P<0.05) with OS; and CSF2 (HR =2.38, P<0.05), SDC1 (HR =0.49, P<0.05), IRF4 (HR =0.49, P<0.05), MS4A1 (HR =0.21, P<0.05), CD5 (HR =0.35, P<0.05), TRAF6 (HR =0.51, P<0.05), and CD200 (HR =1.72, P<0.05) with PFI (Figure 6B). Subtype-specific analysis revealed distinct prognostic associations (Figure 6C): in LCC, CCL5 (HR =1.26, P<0.05), CD4 (HR =0.23, P<0.05), and CD200 (HR =0.50, P<0.05) were associated with OS, while IGSF6 (HR =0.06, P<0.05), MS4A1 (HR =0.16, P<0.05), CSF2 (HR =2.39, P<0.05), and CD5 (HR =0.21, P<0.05) were associated with PFI. In RCC, CD200 (HR =2.72, P<0.05) was associated with OS, and CD200 (HR =2.90, P<0.05), IRF4 (HR =0.34, P<0.05), CSF2 (HR =2.55, P<0.05), and SDC1 (HR =0.38, P<0.05) were associated with PFI.

Figure 6 Validation of TLSRG expression and prognostic significance in the GSE103479 cohort. (A) Box plot highlighting 5 TLSRGs significantly differentially expressed between LCC and RCC in the GSE103479 cohort. (B) Forest plot showing associations between TLSRGs and OS/PFI in the entire GSE103479 cohort. (C) Forest plot showing associations between TLSRGs and OS/PFI stratified by LCC and RCC within the GSE103479 cohort. Significance levels: *, P<0.05; ***, P<0.001. HR, hazard ratio; LCC, left-sided colorectal cancer; OS, overall survival; PFI, progression-free interval; RCC, right-sided colorectal cancer; TLSRG, tertiary lymphoid structures-related gene.

Association of TLSRGs with drug sensitivity and immune cell infiltration

We further analyzed the relationship between TLSRGs and drug sensitivity, as well as immune cell infiltration, in the TCGA-COAD cohort. Analysis of drug sensitivity (using public pharmacogenomic data) revealed significant differences in sensitivity to 22 compounds between LCC and RCC (e.g., erlotinib, salubrinal, sorafenib). The expression levels of individual TLSRGs and the overall TLSRG GSVA score showed significant correlations with sensitivity to these drugs, with most genes exhibiting consistent correlation patterns (Figure 7A). Furthermore, significant differences in the infiltration levels of four immune cell types were observed between LCC and RCC: CD8⁺ T cells, follicular helper T (Tfh) cells, M0 macrophages, and M1 macrophages. The expression levels of TLSRGs correlated significantly with the abundance of these cell types. Correlations with M0 macrophages were predominantly negative, while correlations with CD8⁺ T cells, Tfh cells, and M1 macrophages were predominantly positive (Figure 7B). Importantly, the TLSRG GSVA score itself showed significant positive correlations with CD8⁺ T cell, Tfh cell, and M1 macrophage infiltration, and a significant negative correlation with M0 macrophage infiltration (Figure 7C).

Figure 7 Assessment of associations between TLSRGs, TLSRG GSVA score, drug sensitivity, and immune cell infiltration. (A) Heatmap depicting Spearman correlation coefficients between TLSRG gene expression, TLSRG GSVA score, and drug sensitivity (IC₅₀ values) for compounds showing differential sensitivity between LCC and RCC. (B) Heatmap depicting Spearman correlation coefficients between TLSRG gene expression, TLSRG GSVA score, and immune cell infiltration (estimated abundance). (C) Scatter plots illustrating significant correlations between the TLSRG GSVA score and infiltration levels of specific immune cell types (CD8⁺ T cells, Tfh cells, M0 macrophages, M1 macrophages). Correlation coefficients (r) and P values are shown. *, P<0.05; **, P<0.01; ***, P<0.001. GSVA, gene set variation analysis; IC₅₀, half-maximal inhibitory concentration; LCC, left-sided colorectal cancer; RCC, right-sided colorectal cancer; TLSRG, tertiary lymphoid structures-related gene.

Relationship between TLSRGs and hallmark signaling pathways

Using GSEA, we identified hallmark pathways dysregulated between LCC and RCC, including interferon gamma response, inflammatory response, apoptosis, and others (Figure 8A). We subsequently calculated GSVA enrichment scores for hallmark gene sets and analyzed their correlation with the TLSRG GSVA score. The TLSRG score showed significant positive correlations with hallmark pathways exhibiting differential activity, particularly allograft rejection, inflammatory response, complement, and hypoxia (Figure 8B). Figure 8C presents a comprehensive heatmap illustrating the correlation between individual TLSRG expression, the TLSRG GSVA score, and hallmark pathway GSVA scores.

Figure 8 Association of TLSRGs with hallmark signaling pathways. (A) GSEA enrichment plot highlighting hallmark pathways significantly dysregulated between RCC and LCC (ranked by NES). (B) Heatmap depicting Spearman correlation coefficients between individual TLSRG gene expression, the TLSRG GSVA score, and hallmark pathway GSVA scores. (C) Barplots illustrating significant correlations between the TLSRG GSVA score and GSVA scores of hallmark pathways. GSEA, gene set enrichment analysis; GSVA, gene set variation analysis; LCC, left-sided colorectal cancer; NES, Normalized Enrichment Score; RCC, right-sided colorectal cancer; TLSRG, tertiary lymphoid structures-related gene.

Construction and validation of a prognostic risk signature based on TLSRGs and associated pathways

To develop a robust prognostic signature reflecting TLSRGs and their related biology, we identified hallmark pathways significantly correlated with the TLSRG GSVA score (FDR <0.0001). The union of genes from these pathways and the TLSRGs yielded 2,662 genes. We analyzed the association of these genes with OS in both the TCGA-COAD and GSE103479 cohorts. Genes significantly associated with OS (P<0.05) in both cohorts were selected, resulting in an 8-gene signature: HSPB1, NCOR2, STIL, ASNS, GCLM, SLC4A4, NPTXR, YKT6. Subsequently, we utilized the Mime package to systematically construct and evaluate 101 machine learning models based on ten different algorithms. As shown in the C-index distribution plot (Figure 9A), the SuperPC-based model outperformed other algorithms (such as RSF and Lasso) regarding predictive accuracy and stability across both cohorts. Consequently, the SuperPC model was selected as the optimal prognostic signature. Kaplan-Meier survival analysis revealed that patients in the high-risk group had significantly worse OS compared to the low-risk group in TCGA-COAD (HR =2.06, P=0.001; Figure 9B). Time-dependent ROC analysis showed area under the curve (AUC) values for predicting 1-, 3-, and 5-year OS of 0.646, 0.658, and 0.647 in TCGA-COAD (Figure 9C). Similarly, in the GSE103479 cohort, high-risk patients exhibited significantly worse OS (HR =1.86, P=0.02; Figure 9D), with corresponding AUC values of 0.702, 0.627, and 0.630 for 1-, 3-, and 5-year OS prediction (Figure 9E). Furthermore, the model demonstrated comparable predictive performance for OS when applied separately to LCC and RCC patients within both cohorts (Figure 9F-9I).

Figure 9 Construction and validation of a risk prognostic model based on TLSRGs and associated pathways. (A) Distribution of C-index for the 117 evaluated machine learning models across the TCGA-COAD and GSE103479 cohorts. (B) Kaplan-Meier survival curves comparing OS between high- and low-risk groups defined by the SuperPC model in the TCGA-COAD cohort. (C) Time-dependent ROC curves showing the performance of the SuperPC model in predicting 1-, 3-, and 5-year OS in the TCGA-COAD cohort. (D) Kaplan-Meier survival curves for OS in the GSE103479 cohort stratified by the SuperPC model risk groups. (E) ROC curves for predicting 1-, 3-, and 5-year OS in the GSE103479 cohort using the SuperPC model. ROC curve analysis demonstrating the performance of the SuperPC model in predicting OS within LCC and RCC subtypes of the (F,G) TCGA-COAD cohort and (H,I) GSE103479 cohort. AUC values are indicated. AUC, area under the curve; C-index, concordance index; OS, overall survival; ROC, receiver operating characteristic; TCGA-COAD, The Cancer Genome Atlas Colon Adenocarcinoma; TLSRG, tertiary lymphoid structures-related gene.

Discussion

Our integrated analysis revealed significant heterogeneity in somatic mutation landscapes of TLSRGs between left- and right-sided colon cancers. The high mutation burden of IL1R1 in LCC is particularly noteworthy, as this gene encodes a key pro-inflammatory receptor in the interleukin-1 signaling pathway. Prior studies demonstrate that IL1R1 polymorphisms promote carcinogenesis through sustained NF-κB activation and stromal remodeling in multiple malignancies, including thyroid and head/neck cancers (14,15). Its mutation in LCC likely disrupts immunoregulatory balance, potentially explaining the aggressive phenotypes associated with this gene in our survival analyses. Conversely, MS4A1 mutations in RCC may impair B-cell mediated anti-tumor immunity. Experimental evidence indicates that MS4A1 loss diminishes antibody-dependent cellular cytotoxicity (ADCC)—a key mechanism of anti-CD20 therapies like rituximab (16). This biological context underscores why MS4A1 mutations predicted poor DFI specifically in LCC, where intact B-cell function may be more critical for immune surveillance. The prominence of CCR5 mutations in RCC (14%) aligns with its established role in recruiting immunosuppressive cells to tumor microenvironments. The CCR5/CCL5 axis facilitates metastasis by promoting cancer-associated fibroblast interactions and T-reg recruitment (17,18), mechanisms that may contribute to RCC’s characteristically immunosuppressive niche. Pharmacological CCR5 inhibition (e.g., maraviroc) has shown preclinical efficacy in disrupting these protumorigenic networks (19).

Consensus clustering highlighted the potential of TLSRGs in capturing laterality-specific molecular features of CRC. The elevated GSVA scores in the C1 subtype suggest heightened activity of immune-related pathways (e.g., inflammatory response, complement activation) within RCC (20,21). This may be linked to RCC’s richer microbiome and chronic inflammatory background, potentially leading to more frequent but functionally impaired TLS formation (22,23). Conversely, the C2 subtype, more prevalent in LCC, showed lower pathway activity, implying that LCC’s immune microenvironment relies more on alternative mechanisms, such as angiogenesis or metabolic reprogramming (24,25). GSEA and GSVA analyses further delineated pathway differences: RCC-enriched hallmark pathways included MYC targets and WNT beta-catenin signaling, consistent with its recognized aggressiveness and poorer prognosis. LCC, on the other hand, showed stronger activation of apoptosis and P53 pathways, supporting its better documented response to conventional chemotherapy (24,26). These observations align with established classifications like the consensus molecular subtypes (CMS), where RCC is often CMS1 (microsatellite-immune) and LCC is frequently CMS2/3 (CIN/metabolic), but our study innovatively integrates TLSRGs into the clustering framework, revealing a novel immune-related dimension to molecular subtyping.

Single-cell RNA-seq analysis revealed significant differences in the cellular microenvironment between LCC and RCC. LCC exhibited a higher proportion of T cells, whereas RCC showed elevated proportions of epithelial cells, B cells, myeloid cells, and fibroblasts. This cellular composition reflects RCC’s tendency towards a more immunosuppressive and fibrotic tumor microenvironment, conducive to tumor progression (10,27). The overall TLS ssGSEA score was significantly lower in RCC. Cell-type-specific analysis revealed significantly reduced TLS scores in RCC within T cells, epithelial cells, B cells, and myeloid cells, while endothelial cells and dendritic cells showed higher scores. This cell-type-specific TLS score disparity implies impaired immune cell recruitment and activation in RCC (12), possibly due to immune evasion mechanisms driven by higher mutational burden or other factors (28). CellChat analysis further demonstrated weakened intercellular communication strength in RCC, particularly in epithelial-T cell and epithelial-B cell interactions. This attenuation could undermine the formation of effective anti-tumor immune networks, contributing to poorer prognosis (29). These communication differences likely arise from altered expression of ligand-receptor pairs, evidenced by reduced aggregated probabilities for secreted signaling pathways in RCC, visually highlighted in differential network heatmaps. In contrast, the stronger communication observed in LCC supports the potential for more effective TLS formation and localized immune responses. Consistent with existing literature, our single-cell data confirms the impact of CRC laterality on immune infiltration (e.g., RCC often harbors more regulatory T cells), but this study uniquely emphasizes the role of TLSRGs in mediating these differences.

In the risk signature, HSPB1 functions as a molecular chaperone implicated in tumor progression. Its low expression in CRC tissues correlates with deeper tumor invasion (30), and it exhibits anti-angiogenic properties by interacting with VEGF (31,32), aligning with its prognostic value in our model. STIL, a centrosomal protein regulating the cell cycle, demonstrates context-dependent roles. While its overexpression shortened lifespan and reduced tumor formation in mice (33), contrasting studies in triple-negative breast cancer show it promotes progression via the Fanconi anemia pathway (34). This duality may reflect tissue-specific functions relevant to CRC heterogeneity. ASNS is crucial for amino acid metabolism and chemoresistance. Knockdown of ASNS suppresses proliferation and tumor growth in gastric cancer (35), and its role in mediating resistance to platinum-based therapy in CRC models (36) underscores its significance in our prognostic signature. GCLM is a rate-limiting enzyme in glutathione synthesis. Beyond its metabolic function, recent findings reveal a moonlighting role where nucleus-translocated GCLM, phosphorylated by P38 MAPK, promotes chemoresistance in CRC by enhancing NF-κB activity (37). This non-canonical mechanism provides a compelling explanation for its association with poor prognosis in our high-risk group. SLC4A4, a Na⁺/HCO₃⁻ cotransporter, is consistently downregulated in CRC tissues (38,39). Its low expression promotes proliferation, migration, and correlates with altered immune infiltration (e.g., reduced CD8⁺ T cells) (39,40), directly linking it to the immunosuppressive features often observed in RCC and high-risk phenotypes. NPTXR is primarily studied in neurology as a CSF biomarker for Alzheimer’s disease progression (41), but emerging evidence identifies it as a prognostic biomarker in esophageal squamous cell carcinoma (42), suggesting broader roles in epithelial malignancies. YKT6, a SNARE protein involved in vesicle trafficking and autophagy, is upregulated in multiple cancers where it promotes proliferation, migration, invasion, and correlates with poor prognosis (43,44). Its inclusion in our signature highlights the contribution of membrane dynamics and potential autophagy regulation to CRC aggressiveness.

It is important to note that the 8-gene signature consists not of the structural components of TLS itself, but of downstream or collateral pathway genes significantly correlated with the TLS-rich microenvironment (45-48). For instance, STIL is a core component of centriole duplication and a key regulator of the cell cycle (49); its inclusion is consistent with the established crosstalk between tumor cell-cycle programs (E2F-related) and anti-tumor immunity (50). Furthermore, the presence of metabolic genes like ASNS and GCLM suggests that the prognostic value of our model stems from capturing the metabolic stress and redox balance of the TME, which are key determinants of lymphocyte activation/dysfunction (51,52). Finally, pH regulation and extracellular acidosis are increasingly recognized as major immunomodulatory forces in the TME; bicarbonate transporters (including SLC4A4) can reshape acidity and anti-tumor immunity (53).

Specifically, recent multicenter studies have established the prognostic value of TLS in CRC through histopathological assessment (54). For instance, Mao et al. [2023] developed a robust nomogram integrating TLS density from H&E slides to predict outcomes in LCC and RCC (12). Likewise, a multicenter whole-slide-image analysis further suggested that TLS density along the invasive margin is a robust prognostic index across independent cohorts (54), and a large registry-based study proposed a pathological TLS score combining density and maturation to stratify survival and recurrence risk in CRC (55). While such pathology-based models provide direct visual evidence of immune aggregates, they do not, by design, resolve the molecular programs and signaling networks that enable TLS neogenesis, organization (e.g., B/T zoning, FDC/HEV development), and functional maturation—topics that have been extensively discussed in mechanistic TLS literature (45). Moreover, standardized TLS assessment remains challenging because histologic definitions vary across studies and conventional staining [e.g., hematoxylin and eosin (H&E)] has recognized limitations for consistently classifying maturation states in some settings (56).

In contrast, our study offers a distinct and complementary perspective. By leveraging a transcriptomic approach, our 8-gene signature is positioned to capture a “molecular ecosystem” that may support TLS formation and function, including metabolic regulation (ASNS, GCLM) and cell-cycle-linked programs (STIL). Mechanistically, asparagine availability and ASNS induction are implicated in optimizing CD8⁺ T-cell activation and metabolic reprogramming (57), while glutathione synthesis capacity—whose rate-limiting step is catalyzed by glutamate-cysteine ligase (GCLC/GCLM)—is central to cellular redox buffering and stress adaptation (58) and has been linked to dendritic-cell differentiation/cross-presentation capacity (a key cell type in TLS biology) (59). In parallel, STIL is a cell-cycle-regulated centriole-duplication factor required for faithful cell division (49), consistent with the broader concept that TLS maturation (especially germinal-center-like reactions) is coupled to proliferative immune programs rather than being a purely structural phenomenon (60).

Unlike histological counting, which primarily evaluates the structure, a molecular signature evaluates the functional soil. This distinction is crucial, because transcriptomic TLS-associated signatures (e.g., chemokine-based programs) have been shown to associate with TLS-related biology and clinical outcomes in CRC cohorts (61) and multiple groups have developed CRC TLS gene-signature models for prognosis and/or therapy-response stratification (62,63). Therefore, our model may be particularly useful for identifying patients with TLS-permissive immune programs when mature TLS architecture is not readily appreciable on routine histology (e.g., limited tissue, heterogeneous sampling, or suboptimal architectural preservation), and in scenarios where small or microdissected FFPE sections can still yield RNA suitable for sequencing-based assays (56,64).

Despite providing comprehensive insights into CRC laterality heterogeneity, there are several limitations in this study. Firstly, data were primarily sourced from public databases, potentially introducing selection bias. Future multi-center, prospective studies are needed for validation. Secondly, although Harmony batch correction is effective, it cannot completely eliminate technical variation, possibly exaggerating LCC/RCC differences. Thirdly, while mutation and expression analyses focused on TLSRGs, they lacked validation at the protein level (e.g., immunohistochemistry), overlooking potential post-transcriptional regulation. The prognostic model, although optimized by machine learning, was not adjusted for key clinical variables (e.g., stage, age) in multivariate analysis, potentially overestimating the independent contribution of TLSRGs. Fourthly, drug sensitivity predictions using pRRophetic based on in vitro GDSC data require cautious interpretation regarding clinical relevance, as they are not derived from actual in vitro/in vivo experiments within this specific cohort context. Future research should integrate multi-omics approaches (e.g., proteomics, metabolomics) and functional experiments (e.g., CRISPR knockout models, organoid co-cultures) to strengthen the mechanistic evidence. Despite these constraints, this study establishes a crucial foundation for understanding the role of TLSRGs in CRC laterality and provides clear directions for subsequent research.

Conclusions

This study demonstrates that TLSRGs drive profound heterogeneity between left- and right-sided colon cancer through distinct mutation profiles, differential expression patterns, and altered immune microenvironment organization. We established an 8-gene prognostic signature that effectively stratifies patient risk across anatomical subtypes. The robust performance of our machine learning model highlights its clinical utility for prognostication while revealing novel therapeutic targets for side-specific precision oncology approaches in CRC.

Acknowledgments

We gratefully acknowledge The Cancer Genome Atlas (TCGA) program and the Gene Expression Omnibus (GEO) database for providing open-access high-quality data.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2367/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2367/prf

Funding: This study was supported by the Natural Science Foundation of Ningbo (No. 2023J220), the Zhejiang Provincial Medical and Health Science and Technology Plan (No. 2025KY211), and the Beijing Science and Technology Innovation Medical Development Foundation.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2367/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Global cancer burden growing, amidst mounting need for services. Saudi Med J 2024;45:326-7.
2. Siegel RL, Kratzer TB, Wagle NS, et al. Cancer statistics, 2026. CA Cancer J Clin 2026;76:e70043. [Crossref] [PubMed]
3. Quinones-Olvera N, Owen SV, McCully LM, et al. Diverse and abundant phages exploit conjugative plasmids. Nat Commun 2024;15:3197. [Crossref] [PubMed]
4. Nunes L, Li F, Wu M, et al. Prognostic genome and transcriptome signatures in colorectal cancers. Nature 2024;633:137-46. [Crossref] [PubMed]
5. Abdel Hamid M, Pammer LM, Oberparleiter S, et al. Multidimensional differences of right- and left-sided colorectal cancer and their impact on targeted therapies. NPJ Precis Oncol 2025;9:116. [Crossref] [PubMed]
6. Lv J, Zhang X, Zhou M, et al. Tertiary lymphoid structures in colorectal cancer. Ann Med 2024;56:2400314. [Crossref] [PubMed]
7. Dai NN, Hu MY, Wang JP, et al. Tertiary lymphoid structures in the microenvironment of colorectal cancer: exploring new therapeutic targets. Cancer Immunol Immunother 2025;74:245. [Crossref] [PubMed]
8. Yi T, Liao Q, Jiang Z. Distinct regulation mechanism of immune-related genes in left-sided and right-sided colon cancer. Nucleosides Nucleotides Nucleic Acids 2025; Epub ahead of print. [Crossref]
9. Yang J, Zhang Y, Cheng S, et al. Identification of a tertiary lymphoid structure (TLS)-related signature for ovarian cancer prognosis suggests a potential role of STAT5A in TLS maturation. Genes Dis 2025;12:101514. [Crossref] [PubMed]
10. Guo W, Zhang C, Wang X, et al. Resolving the difference between left-sided and right-sided colorectal cancer by single-cell sequencing. JCI Insight 2022;7:e152616. [Crossref] [PubMed]
11. Li Y, Li H, Sun G, et al. Integrative analyses of multi-omics data constructing tumor microenvironment and immune-related molecular prognosis model in human colorectal cancer. Heliyon 2024;10:e32744. [Crossref] [PubMed]
12. Mao Y, Wang X, Xi L, et al. Prediction values of tertiary lymphoid structures in the prognosis of patients with left- and right-sided colon cancer: a multicenter propensity score-matched study. Int J Surg 2023;109:2344-58. [Crossref] [PubMed]
13. Liu H, Zhang W, Zhang Y, et al. Mime: A flexible machine-learning framework to construct and visualize models for clinical characteristics prediction and feature selection. Comput Struct Biotechnol J 2024;23:2798-810. [Crossref] [PubMed]
14. Tian Y, Sun Y, Wu J, et al. Impact of IL1R1 polymorphisms on the risk of head and neck cancer in Chinese Han population. Gene 2020;757:144927. [Crossref] [PubMed]
15. Xiong Z, Sun Y, Wu J, et al. Genetic polymorphisms in IL1R1 and IL1R2 are associated with susceptibility to thyroid cancer in the Chinese Han population. J Gene Med 2019;21:e3093. [Crossref] [PubMed]
16. Kawabata KC, Ehata S, Komuro A, et al. TGF-β-induced apoptosis of B-cell lymphoma Ramos cells through reduction of MS4A1/CD20. Oncogene 2013;32:2096-106. [Crossref] [PubMed]
17. Nishikawa G, Kawada K, Nakagawa J, et al. Bone marrow-derived mesenchymal stem cells promote colorectal cancer progression via CCR5. Cell Death Dis 2019;10:264. [Crossref] [PubMed]
18. Schlechter BL, Stebbing J. CCR5 and CCL5 in metastatic colorectal cancer. J Immunother Cancer 2024;12:e008722. [Crossref] [PubMed]
19. Tanabe Y, Sasaki S, Mukaida N, et al. Blockade of the chemokine receptor, CCR5, reduces the growth of orthotopically injected colon cancer cells via limiting cancer-associated fibroblast accumulation. Oncotarget 2016;7:48335-45. [Crossref] [PubMed]
20. Mao Y, Xu Y, Chang J, et al. The immune phenotypes and different immune escape mechanisms in colorectal cancer. Front Immunol 2022;13:968089. [Crossref] [PubMed]
21. Li D, Liu X, Gao W, et al. An Immune Subtype Classification System Enables the Development of Strategies to Predict and Enhance Immunotherapy Responses in Colorectal Cancer. Cancer Res 2025;85:1441-58. [Crossref] [PubMed]
22. Han S, Cui L, Wang B, et al. USP15 Facilitates Colorectal Cancer Immune Evasion through SMYD3/CCL2-Dependent Myeloid-Derived Suppressor Cell Recruitment. Cancer Immunol Res 2025;13:1226-45. [Crossref] [PubMed]
23. Teng HW, Wang TY, Lin CC, et al. Interferon Gamma Induces Higher Neutrophil Extracellular Traps Leading to Tumor-Killing Activity in Microsatellite Stable Colorectal Cancer. Mol Cancer Ther 2024;23:1043-56. [Crossref] [PubMed]
24. Wu Q, Huang Q, Jiang Y, et al. Remodeling Chondroitin-6-Sulfate-Mediated Immune Exclusion Enhances Anti-PD-1 Response in Colorectal Cancer with Microsatellite Stability. Cancer Immunol Res 2022;10:182-99. [Crossref] [PubMed]
25. Nicolini A, Ferrari P. Involvement of tumor immune microenvironment metabolic reprogramming in colorectal cancer progression, immune escape, and response to immunotherapy. Front Immunol 2024;15:1353787. [Crossref] [PubMed]
26. Mazzeschi M, Sgarzi M, Romaniello D, et al. The autocrine loop of ALK receptor and ALKAL2 ligand is an actionable target in consensus molecular subtype 1 colon cancer. J Exp Clin Cancer Res 2022;41:113. [Crossref] [PubMed]
27. Hu C, Huang X, Chen J, et al. Dissecting the cellular reprogramming and tumor microenvironment in left- and right-sided Colorectal Cancer by single cell RNA sequencing. Transl Res 2025;276:22-37. [Crossref] [PubMed]
28. Lei JX, Wang R, Hu C, et al. Deciphering tertiary lymphoid structure heterogeneity reveals prognostic signature and therapeutic potentials for colorectal cancer: a multicenter retrospective cohort study. Int J Surg 2024;110:5627-40. [Crossref] [PubMed]
29. Jia Q, Chu H, Jin Z, et al. High-throughput single-cell sequencing in cancer research. Signal Transduct Target Ther 2022;7:145. [Crossref] [PubMed]
30. Hung CS, Huang CY, Hsu YW, et al. HSPB1 rs2070804 polymorphism is associated with the depth of primary tumor. J Cell Biochem 2020;121:63-9. [Crossref] [PubMed]
31. Choi SH, Lee HJ, Jin YB, et al. MMP9 processing of HSPB1 regulates tumor progression. PLoS One 2014;9:e85509. [Crossref] [PubMed]
32. Lee YJ, Lee HJ, Choi SH, et al. Soluble HSPB1 regulates VEGF-mediated angiogenesis through their direct interaction. Angiogenesis 2012;15:229-42. [Crossref] [PubMed]
33. Moussa AT, Cosenza MR, Wohlfromm T, et al. STIL overexpression shortens lifespan and reduces tumor formation in mice. PLoS Genet 2024;20:e1011460. [Crossref] [PubMed]
34. Wang M, Pan B, Hu Y, et al. STIL facilitates the development and malignant progression of triple-negative breast cancer through activation of Fanconi anemia pathway via interacting with KLF16. Transl Oncol 2024;46:102010. [Crossref] [PubMed]
35. Yu Q, Wang X, Wang L, et al. Knockdown of asparagine synthetase (ASNS) suppresses cell proliferation and inhibits tumor growth in gastric cancer cells. Scand J Gastroenterol 2016;51:1220-6. [Crossref] [PubMed]
36. Wang Z, Zhang W, Yin X, et al. Multi-omics analyses were combined to construct ubiquitination-related features in colon adenocarcinoma and identify ASNS as a novel biomarker. Front Immunol 2024;15:1466286. [Crossref] [PubMed]
37. Lin JF, Liu ZX, Chen DL, et al. Nucleus-translocated GCLM promotes chemoresistance in colorectal cancer through a moonlighting function. Nat Commun 2025;16:263. [Crossref] [PubMed]
38. Chen X, Chen J, Feng Y, et al. Prognostic Value of SLC4A4 and its Correlation with Immune Infiltration in Colon Adenocarcinoma. Med Sci Monit 2020;26:e925016. [Crossref] [PubMed]
39. Yu C, Li H, Zhang C, et al. Solute carrier family 4 member 4 (SLC4A4) is associated with cell proliferation, migration and immune cell infiltration in colon cancer. Discov Oncol 2024;15:597. [Crossref] [PubMed]
40. Parks SK, Pouyssegur J. The Na(+)/HCO3(-) Co-Transporter SLC4A4 Plays a Role in Growth and Migration of Colon and Breast Cancer Cells. J Cell Physiol 2015;230:1954-63. [Crossref] [PubMed]
41. Lim B, Tsolaki M, Soosaipillai A, et al. Liquid biopsy of cerebrospinal fluid identifies neuronal pentraxin receptor (NPTXR) as a biomarker of progression of Alzheimer’s disease. Clin Chem Lab Med 2019;57:1875-81. [Crossref] [PubMed]
42. Shinozuka T, Kanda M, Kodera Y. ASO Author Reflections: Evaluating NPTXR as a Prognostic Biomarker and Therapeutic Target in Esophageal Squamous Cell Carcinoma. Ann Surg Oncol 2024;31:5765. [Crossref] [PubMed]
43. Liu J, Zhang Q, He L, et al. Pan-Cancer Analysis Identifies YKT6 as a Prognostic and Immunotherapy Biomarker, with an Emphasis on Cervical Cancer. Onco Targets Ther 2025;18:107-27. [Crossref] [PubMed]
44. Zhang L, Wang S, Wang L. Comprehensive analysis identifies YKT6 as a potential prognostic and diagnostic biomarker in lung adenocarcinoma. BMC Cancer 2024;24:1235. [Crossref] [PubMed]
45. Sautès-Fridman C, Petitprez F, Calderaro J, et al. Tertiary lymphoid structures in the era of cancer immunotherapy. Nat Rev Cancer 2019;19:307-25. [Crossref] [PubMed]
46. Wang J, Zhang Y, Dou Z, et al. Knockdown of STIL suppresses the progression of gastric cancer by down-regulating the IGF-1/PI3K/AKT pathway. J Cell Mol Med 2019;23:5566-75. [Crossref] [PubMed]
47. Fridman WH, Meylan M, Petitprez F, et al. B cells and tertiary lymphoid structures as determinants of tumour immune contexture and clinical outcome. Nat Rev Clin Oncol 2022;19:441-57. [Crossref] [PubMed]
48. Helmink BA, Reddy SM, Gao J, et al. B cells and tertiary lymphoid structures promote immunotherapy response. Nature 2020;577:549-55. [Crossref] [PubMed]
49. Arquint C, Sonnen KF, Stierhof YD, et al. Cell-cycle-regulated expression of STIL controls centriole number in human cells. J Cell Sci 2012;125:1342-52. [Crossref] [PubMed]
50. Goel S, DeCristo MJ, Watt AC, et al. CDK4/6 inhibition triggers anti-tumour immunity. Nature 2017;548:471-5. [Crossref] [PubMed]
51. MacIver NJ, Michalek RD, Rathmell JC. Metabolic regulation of T lymphocytes. Annu Rev Immunol 2013;31:259-83. [Crossref] [PubMed]
52. Wu J, Li G, Li L, et al. Asparagine enhances LCK signalling to potentiate CD8(+) T-cell activation and anti-tumour responses. Nat Cell Biol 2021;23:75-86. [Crossref] [PubMed]
53. Cappellesso F, Orban MP, Shirgaonkar N, et al. Targeting the bicarbonate transporter SLC4A4 overcomes immunosuppression and immunotherapy resistance in pancreatic cancer. Nat Cancer 2022;3:1464-83. [Crossref] [PubMed]
54. He M, Ye H, Liu L, et al. Comparative analysis of tertiary lymphoid structures for predicting survival of colorectal cancer: a whole-slide images-based study. Precis Clin Med 2024;7:pbae030. [Crossref] [PubMed]
55. Le Rochais M, Morvan M, Bouzeloc S, et al. A Tertiary lymphoid structures-based pathological score predicts survival and recurrence in colorectal Cancer patients. Immunobiology 2025;230:152911. [Crossref] [PubMed]
56. Le Rochais M, Brahim I, Zeghlache R, et al. Automated classification of tertiary lymphoid structures in colorectal cancer using TLS-PAT artificial intelligence tool. Sci Rep 2025;15:9845. [Crossref] [PubMed]
57. Hope HC, Brownlie RJ, Fife CM, et al. Coordination of asparagine uptake and asparagine synthetase expression modulates CD8+ T cell activation. JCI Insight 2021;6:e137761. [Crossref] [PubMed]
58. Franklin CC, Backos DS, Mohar I, et al. Structure, function, and post-translational regulation of the catalytic and modifier subunits of glutamate cysteine ligase. Mol Aspects Med 2009;30:86-98. [Crossref] [PubMed]
59. D’Angelo JA, Dehlink E, Platzer B, et al. The cystine/glutamate antiporter regulates dendritic cell differentiation and antigen presentation. J Immunol 2010;185:3217-26.
60. Posch F, Silina K, Leibl S, et al. Maturation of tertiary lymphoid structures and recurrence of stage II and III colorectal cancer. Oncoimmunology 2018;7:e1378844. [Crossref] [PubMed]
61. Tokunaga R, Nakagawa S, Sakamoto Y, et al. 12-Chemokine signature, a predictor of tumor recurrence in colorectal cancer. Int J Cancer 2020;147:532-41. [Crossref] [PubMed]
62. Xu Z, Wang Q, Zhang Y, et al. Exploiting tertiary lymphoid structures gene signature to evaluate tumor microenvironment infiltration and immunotherapy response in colorectal cancer. Front Oncol 2024;14:1383096. [Crossref] [PubMed]
63. Guo Z, Feng H, Jia W, et al. Tertiary lymphoid structures signature predicts prognosis and clinical benefits from neoadjuvant chemotherapy and PD-1 blockade in colorectal adenocarcinoma. Int J Surg 2025;111:5088-104. [Crossref] [PubMed]
64. Amini P, Ettlin J, Opitz L, et al. An optimised protocol for isolation of RNA from small sections of laser-capture microdissected FFPE tissue amenable for next-generation sequencing. BMC Mol Biol 2017;18:22. [Crossref] [PubMed]