Development and validation of a gastric cancer prognostic model utilizing lymphatic endothelial cell-related genes

Abstract

Background

Gastric cancer is one of the most common cancers worldwide, with its prognosis influenced by factors such as tumor clinical stage, histological type, and the patient’s overall health. Recent studies highlight the critical role of lymphatic endothelial cells (LECs) in the tumor microenvironment. Perturbations in LEC function in gastric cancer, marked by aberrant activation or damage, disrupt lymphatic fluid dynamics and impede immune cell infiltration, thereby modulating tumor progression and patient prognosis. Hence, we aimed to construct a prognostically discriminative model group in LECs-related factors.

Methods

Gene expression and clinical data of gastric cancer patients were obtained from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO), and Fudan University Shanghai Cancer Center (FUSCC). Using the Wilcoxon test, we assessed the relationship between LECs, angiogenesis, and the immunological milieu. Differentially expressed and prognostically significant LEC-associated genes were identified through “limma” R package-assisted analysis coupled with univariate Cox analysis. A prognostic model was developed, and LEC-associated gene signatures were refined through least absolute shrinkage and selection operator (LASSO)-Cox regression. Subsequently, the prognostic potential of this model was evaluated using ROC (receiver operating characteristic) curve analysis, Kaplan-Meier survival curve analysis and decision curve analysis (DCA).

Results

LECs exhibited association with angiogenesis, immune cell infiltration, immune escape, and epithelial-mesenchymal transition (EMT). Utilizing an 18-gene signature, gastric cancer patients from TCGA and GEO cohorts were stratified into high- risk and low-risk groups, with the former showing significantly poorer overall survival. Leveraging this gene signature, we designed a LECs-related gastric cancer prognostic model, demonstrating superior performance indicated by the area under the ROC curve (AUC) compared to existing models. Moreover, the nomogram and DCA underscored the clinical utility of our model in predicting the prognosis of GC patients.

Conclusions

Our prognostic signature, based on 18 LECs-related genes, holds promise for refining overall survival prediction in gastric cancer patients, offering a valuable tool for clinical decision-making.

Clinical trial number

Not applicable.

Background

Gastric cancer (GC) is a formidable malignancy characterized by diverse prognostic outcomes. According to global statistics from GLOBOCAN, GC contributed to 5.6% of new cancer cases worldwide in 2020, ranking fifth among 36 types of cancer. Furthermore, the mortality rates attributed to gastric cancer accounted for 7.7% of all cancer- related deaths, trailing behind lung, colorectal, and liver cancer [1]. Lymph node metastasis has been acknowledged as a crucial prognostic factor in gastric cancer patients [2]. Typically, tumor cells orchestrate an immunosuppressive milieu and subsequently disseminate from the primary site to nearby lymph nodes and various distant sites [3]. Previous study has demonstrated the pivotal role of lymphatic endothelial cells (LECs), integral components of the lymph node stromal cell (LNSC), in mediating immune cell trafficking and facilitating the transport of peripheral antigens to lymph nodes [4]. Furthermore, LECs directly inhibit the maturation of dendritic cells and the expression of peripheral tissue antigens (PTAs), which in turn diminishes the responses of autoreactive T cells [5]. Moreover, LECs actively contribute to maintaining peripheral immune tolerance by capturing exogenous antigens and presenting them to corresponding CD8 (+) T cells [6]. Notably, interactions between LECs and tumor cells have been linked to tumor progression and upregulation of genes associated with inflammation, cell proliferation, and migration [7]. Moreover, macrophages involved in the inflammatory response can induce lymphatic vessel formation in the tumor microenvironment through crosstalk with LECs [8].

In addition to immune evasion, angiogenesis, and epithelial-mesenchymal transition (EMT) represent pivotal determinants influencing gastric cancer growth and progression. Angiogenesis, a multi-step process, entails the proliferation and migration of activated endothelial cells, culminating in the formation of new capillary structures and vascular lumen through basement membrane synthesis and blood vessel maturation [9]. Conversely, EMT signifies a reversible phenotypic transition in epithelial cells, wherein they lose apical-basal polarity and intercellular tight junctions under the influence of various stimuli, transitioning into mesenchymal-like cells with invasive and migratory capabilities [10]. EMT is a precursor to invasion and metastasis, signifying heightened metastatic potential [11]. Moreover, gap junctions, pivotal mediators of inter- cellular communication between tumor and stromal cells, including endothelial cells, have been implicated in tumor adhesion and metastasis [12].

To elucidate the potential mechanisms through which LECs influence the prognosis and progression of gastric cancer, we employed xCell algorithm [13] to compute LEC scores utilizing the transcriptome data retrieved from TCGA, GEO, and FUSCC. Furthermore, we explored the relationship between LECs and tumor immunity, angiogenesis, and EMT in gastric cancer. Subsequently, we developed a LECs- related nomogram integrating risk scores and selected clinical prognostic parameters to comprehensively assess the clinical predictive utility of LECs.

Materials and methods

Extraction of mRNA expression and clinical data from GC patients

RNA expression data and clinical characteristics of 407 GC cases were obtained from the TCGA-STAD project, accessible through UCSC Xena (http://xena.ucsc.edu). Additionally, the GSE62254 dataset, comprising 300 GC samples and associated clinical data, was obtained from Gene Expression Omnibus (GEO) using the GPL570 platform (Affymetrix Human Genome U133 Plus 2.0 Array). The GSE62254 dataset was designated for validation, whereas the TCGA dataset utilized as the training cohort. Furthermore, mRNA expression data from 87 GC patients who underwent gastrectomy at FUSCC were included for analysis [14]. Table 1 presents the clinical traits of participants in the TCGA, GSE62254, and FUSCC cohorts.

Table 1.

Patient characteristics of TCGA, GSE62254, and FUSCC cohorts

Characteristics	TCGA (%)	GSE62254 (%)	FUSCC (%)
Gender	-	-	-
Male	263 (64.62)	199 (66.33)	47 (54.02)
Female	144 (35.38)	101 (33.67)	14 (16.09)
Unknown	0	0	26 (29.89)
Age	-	-	-
≤ 65	180 (44.23)	172 (57.33)	39 (44.83)
> 65	227 (55.77)	128 (42.67)	22 (25.29)
Unknown	0	0	26 (29.89)
Stage	-	-	-
I	59 (14.50)	30 (10.00)	46 (52.87)
II	126 (30.96)	96 (32.00)	15 (17.24)
III	156 (38.33)	95 (31.67)	0
IV	42 (10.32)	77 (25.67)	26 (29.86)
Unknown	24 (5.90)	2 (0.67)	0
T classification stage	-	-	-
T1	22 (5.41)	0	55 (63.22)
T2	91 (22.36)	188 (62.67)	0
T3	181 (44.47)	91 (30.33)	2 (2.30)
T4	105 (25.80)	21 (7.00)	6 (6.90)
Tx	8 (1.97)	0	24 (27.59)
N classification stage	-	-	-
N0	123 (30.22)	38 (12.67)	32 (36.78)
N1	108 (26.54)	131 (43.67)	22 (25.29)
N2	83 (20.39)	80 (26.67)	3 (3.45)
N3	74 (18.18)	51 (17.00)	6 (6.90)
Nx	19 (4.67)	0	24 (27.59)
M classification stage	-	-	-
M0	358 (87.96)	273 (91.00)	61 (70.11)
M1	27 (6.63)	27 (9.00)	26 (29.89)
Mx	22 (5.41)	0	0

In the FUSCC cohort, patients with ‘Tx’ or ‘Nx’ all had distant metastases and were therefore classified as stage IV, which was a known fact

The inclusion criteria of TCGA and GSE62254 patients: availability of complete mRNA expression profiles (e.g., FPKM/TPM values); definitive clinical staging information (AJCC [X] edition); accessible survival data (overall survival/progression-free survival); samples with complete clinicopathological characteristics (age, sex, grade, etc.). The exclusion criteria of TCGA and GSE62254 patients: samples missing clinical staging or survival data; duplicate samples; normal tissue or adjacent non-tumor samples (if study focused solely on tumor tissues); technical outliers (low-quality samples excluded via PCA or expression distribution analysis).

The inclusion criteria of FUSCC patients encompassed individuals diagnosed with gastric cancer who possessed complete clinical information, including but not limited to age, gender, pathological staging, and treatment his- tory. Additionally, inclusion necessitated the availability of complete mRNA expression data, with samples collected prior to any treatment initiation. Conversely, exclusion criteria for FUSCC patients involved individuals with diagnoses other than gastric cancer, those with incomplete or missing clinical information, and samples lacking mRNA expression data or collected after the commencement of treatment. Furthermore, patients with severe comorbidities or concurrent malignant tumors were excluded from the analysis.

Estimation of infiltrating cells proportion in the tumor microenvironment

The cell count within the TME was estimated using the xCell score derived from mRNA expression data. Pre- calculated TCGA data from the xCell scores were obtained from the official website of xCell (https://xcell.ucsf.edu/). Among the TCGA cohort, 373 samples with valid xCell score data and mRNA expression data were included for subsequent analysis.

Gene set expression analyses

Gene set enrichment analysis (GSEA) was performed utilizing the GSEA software (version 4.1, Broad Institute, Cambridge, MA, USA). Gene sets were considered statistically significant when they exhibited a false discovery rate (FDR) of lower than 25% and a p-value below 0.05 [15].

Construction of the LECs-associated gene signature

The process of constructing the LECs-related gene signature is illustrated in Fig. 1. Patients in the TCGA cohort were categorized into two groups based on their levels of LECs, using the median as a dividing point. This categorization facilitated a clear distinction between individuals with elevated LEC levels and those with lower LEC levels. The “limma” R package (version 3.50.0) was used to conduct a differential gene expression analysis to identify genes that show differential expression between the high LECs group and the low LECs group. Genes were considered differentially expressed if their absolute log2FoldChange (log2FC) exceeded 1 and their Benjamini-Hochberg-adjusted p-value (adj. p) was less than 0.05. A heatmap illustrating the differential expression of genes was generated using the “pheatmap” R package (version 1.0.12), while a volcano plot depicting the differentially expressed genes was created using the “ggplot2” R package (version 3.4.0).

Fig. 1.

Flowchart depicting the data collection and analysis process in the current study. (LECs: lymphatic endothelial cells)

Univariate Cox analysis and LASSO-Cox regression were then applied to construct the LECs-related gene signature. These statistical analyses were performed using the “glmnet” R package (version 4.1.4). The prognostic risk score for gastric cancer (GC) patients was calculated as the sum of the products of the coefficient index (βi) and the expression level (Ei) of each gene, where N denotes the total number of signature genes across both cohorts. Additionally, a nomogram to predict overall survival (OS) was created using the “rms” R package (version 6.3.0), according to the findings from the multivariate Cox regression analysis. The prognostic performance of the LEC-related nomogram was assessed by plotting ROC and DCA curves with the “timeROC” (version 0.4) and “dcurves” R packages (version 0.3.0), respectively.

Statistical analysis

Data analyses were performed using the R program (version x64 4.0.5, University of Auckland, Auckland, New Zealand). The cut-off value for LEC scores was determined based on the median value for correlation analysis between groups and difference analysis. To assess the relationship between different groups and overall survival (OS), Kaplan-Meier analyses were performed, and a log-rank test was utilized to calculate the corresponding p-value. Patients were classified into high-LECs and low-LECs groups based on cut-offs determined by the “cutp” function from the R package “survMisc” (version 0.5.6). The relationship between endothelial cell scores and clinicopathological characteristics was investigated using the Kruskal-Wallis or Wilcoxon test, with statistical significance set at a p-value < 0.05 for all analyses.

Results

Prognostic implications of abundance of LECs in GC

Utilizing the xCell algorithm, we evaluated 760 GC samples from the TCGA, GSE62254, and FUSCC cohorts and collected their corresponding clinical data. Firstly, Kaplan-Meier analyses were conducted to plot the survival curve by comparing high- and low-LECs in GC cohorts. Notably, elevated LEC abundance was consistently associated with poorer OS across all GC cohorts examined (Fig. 2A–C). Given the established prognostic significance of the TNM stage in GC, we compared GC patients with LECs at TNM stage. Our analysis indicated a relationship between LECs’ abundance and gastric cancer metastasis in the GSE62254 and FUSCC cohorts (Fig. 2D, E). Disappointingly, in the TCGA cohort, there was no significant correlation between stage and LECs.

Fig. 2.

Relationship between LECs’ abundance and clinical characteristics. (A–C) KM curves illustrating survival status and survival duration for the TCGA, GSE62254, and FUSCC cohorts. (D, E) Comparison of LEC levels across different clinical subgroups categorized by TNM stage in the GSE62254 (D) and FUSCC (E) cohorts. (lyE: lymphatic endothelial cells)

Association of lymphatic endothelial cells with angiogenesis-related genes

Given the integral role of LECs in lymphatic vessels, we evaluated the relationship between LEC abundance and vasculogenesis. Notably, we investigated the expression of representative genes associated with angiogenesis from a previous study [16], including VEGFA, VEGFB, VEGFC, VWF, TIE2/TEK, CDH5, and CLDN5. Our results indicated a statistically significant correlation among the relative expressions of all the above genes (p < 0.05) in the TCGA and FUSCC cohorts, with all exhibiting positive correlations except for VEGFA (Fig. 3A, C). Similarly, in the GSE62254 cohort, with the exception of VEGFA (p = 0.24) and VEGFB (p = 0.23), all other related genes demonstrated positive correlations with LECs (p < 0.05) (Fig. 3B).

Fig. 3.

Association between the abundance of LECs and genes associated with angiogenesis. (A–C) Boxplots displaying the expression levels of angiogenesis-related genes in high- and low-LEC groups across the TCGA (A), GSE62254 (B), and FUSCC (C) cohorts. (D, E) The comparison of pericyte abundance in different LEC groups within the TCGA (D) and GSE62254 (E) cohorts. (lyE: lymphatic endothelial cells)

Moreover, existing literature suggests a close structural and functional relationship between endothelial cells and pericytes in angiogenesis [17]. Consistent with these findings, our investigation revealed a statistically significant association between elevated LEC levels and increased pericytes in the TCGA and GSE62254 cohorts (Fig. 3D, E).

Correlation between lymphatic endothelial cells and immunocytes infiltration.

Additionally, we explored the potential interactions between LECs and immune cells in the tumor microenvironment (TME) of gastric cancer, taking into account the significant impact of immune cells on the TME and the dual role of LECs in facilitating angiogenesis and enabling tumor cells to evade immune responses [18]. Our analysis revealed that the high-LEC group exhibited a greater fraction of dendritic cells (DCs) and endothelial cells in the TCGA (p < 0.01), GSE62254 (p < 0.01) and FUSCC cohorts (p< 0.01) (Fig. 4A–C), whereas there was a reduced fraction of regulatory T cells (Tregs) in the TCGA cohort but an in- creased fraction in the GSE62254 cohort (p < 0.01). Kaplan-Meier analyses of relevant immune cells indicated that high endothelial cells correlated with worse overall survival (OS) across all cohorts (Fig. 4D–F), while high DCs correlated with better OS in the FUSCC cohort (Fig. 4F).

Fig. 4.

Association between the abundance of LECs and immunocytes infiltration. (A–C) The association among the abundance of LECs and the penetration of crucial immune cell subtypes in the neoplastic microenvironment was examined across different LEC groups in the TCGA (A), GSE62254 (B), and FUSCC (C) cohorts. (D–F) Kaplan-Meier survival curves were generated for immune cells in relation to the proportion of LECs in the TCGA (D), GSE62254 (E), and FUSCC (F) cohorts. (G–I) The expression levels of key immune factors were analyzed and compared between the high and low LECs groups in the TCGA (G), GSE62254 (H), and FUSCC (I) cohorts. (lyE: lymphatic endothelial cells; DC: dendritic cell)

Further investigation into the relationship between LECs and immunologic variables, such as lymphangiogenesis, selected based on previous literature [19]. The TCGA cohort’s high LECs group increased expression levels of MMP2, TGF-β, and IL-10 in the high-LEC group of the TCGA cohort (all p < 0.01) (Fig. 4G). Similarly, in the GSE62254 cohort, IL-10 and MMP2 positively correlated with LECs (both p < 0.01) (Fig. 4H). Moreover, in the FUSCC cohort, the high-LEC group exhibited increased levels of IL-10 (p = 0.021), MMP2 (p = 0.003), and TGFB1 (p < 0.01) alongside reduced IL18 (p = 0.00024) (Fig. 4I). These findings suggest potential mechanisms by which LECs influence GC tumor immunity, primarily through the modulation of inflammatory cytokines.

Correlation of lymphatic endothelial cells with immune checkpoints

Given the pivotal role of immune checkpoints in tumor immune evasion, our findings revealed elevated levels of Tim-3 in the high-LECs group of both TCGA and GSE62254 cohorts (Fig. 5A, B), as well as increased ex- pression of PDCD1 in the TCGA and FUSCC cohorts (Fig. 5A, C). In addition to these well-established immune checkpoints commonly targeted in clinical therapeutic approaches [20], we explored additional potential immune checkpoints relevant to our study hypothesis.

Fig. 5.

Association between the abundance of LECs and immune checkpoints. (A–C) Box plots demonstrating the expression levels of immune checkpoint markers within the elevated and reduced LECs groups in the TCGA (A), GSE62254 (B), and FUSCC (C) cohorts. (lyE: lymphatic endothelial cells)

Consistently, the high-LEC group exhibited higher NRP1 expression levels across all cohorts (Fig. 5A–C). Moreover, CD86 was enriched in the high-LEC groups of TCGA and GSE62254 cohorts (Fig. 5A, B), while CD276 was enriched in the high-LEC groups of TCGA and FUSCC cohorts (Fig. 5C). Notably, PDCD1 was only enriched in the high-LEC group of the FUSCC cohort (Fig. 5C). In summary, all the identified immune checkpoints promote immune escape, providing insights into potential mechanisms underlying the observed worse prognosis associated with LECs. These findings prompt further exploration into the intricate interplay between LECs and immune checkpoints in gastric cancer.

Association of lymphatic endothelial cells with EMT pathway

To investigate the potential link between LECS and key signaling, focusing on EMT, we conducted a GSEA analysis based on the median abundance of LECs. As shown in Fig. 6A–C, GSEA revealed enrichment of the EMT pathway in the high-LEC group compared to the low-LEC group in TCGA and GSE62254 cohorts. Conversely, the EMT pathway was downregulated in the low- LEC group compared to the high-LEC group in the FUSCC cohort, as illustrated in Fig. 6D–F (with “1” representing the low-LEC group in Fig. 6F).

Fig. 6.

GSEA conducted on gastric cancer (GC) samples from the TCGA, GSE62254, and FUSCC cohorts. (A–C) Signaling pathways that are notably enriched in the elevated-LECs group are shown for TCGA (A), GSE62254 (B), and FUSCC (C) cohorts. (D–F) Enrichment plots illustrate the enrichment of the epithelial-mesenchymal transition (EMT) signaling pathway in the high LECs group for the TCGA (D), GSE62254 (E), and FUSCC (F) cohorts

Development and validation of the gene signature related to LECs

Patients were classified into high-risk and low-risk groups according to their own risk scores, with the median value serving as the cutoff (Fig. 7D–F). Validation in the GSE62254 dataset, which contained 11 of the 18 LECs-related genes, confirmed that the risk score was associated with worse OS outcomes in the training (Fig. 7G) and validation sets (Fig. 7H). Furthermore, the relationship between the clinical features of GC patients and their risk scores in both cohorts was illustrated (Fig. 7I, J). Notably, the risk score was statistically correlated with LECs and TNM stages in both cohorts (Fig. 7K– N).

Fig. 7.

Development of the gene signature associated with LECs. (A) Volcano plot illustrating the differential gene expression between groups was characterized by high and low LEC levels. (B) LASSO coefficient profiles for 458 prognostic genes in gastric cancer (GC). (C) LASSO regression, employing 1000-fold cross-validation, identifying a total of 18 prognostic genes, was determined by selecting the minimum λ. (D–F) The distributions of risk scores, as well as overall survival status, were visualized for the training, validation, and combined groups. (G, H) Kaplan-Meier survival estimates for were constructed based on the LECs-related gene signature for the TCGA (G) and GSE62254 (H) cohorts. (I, J) Heatmaps displaying the expression of genes incorporated into the signature in the TCGA (I) and GSE62254 (J) cohorts. (K, M) Boxplots illustrating the distribution of risk scores among the high- and low-LECs groups in th overall survival (OS) TCGA (K) and GSE62254 (M) cohorts. (L, N) Risk scores for individuals diagnosed with gastric cancer at various stages were evaluated within the TCGA (L) and GSE62254 (N) cohorts

Development and verification of the signature-based nomogram

We employed a multivariate Cox regression model to a cohort comprising 330 samples from the TCGA dataset to enhance the prognostic prediction for GC patients. This model incorporated the LECs-related gene signature with factors such as age and TNM stage, resulting in the establishment of a LECs-related clinical prognostic indicator (LRCPI) calculated as (0.031 age) + 0.663 (stage = 3) + 1.312 (stage = 4) + 1.622 (risk score). Fig. 8A illustrates the components of this model.

Fig. 8.

A nomogram was constructed and validated based on the LECs-related risk score to predict overall survival (OS) in gastric cancer (GC). (A) Results from the univariate and multivariate Cox regression analyses for overall survival (OS) using the 18-LECs-related gene signature are presented. (B) The nomogram includes age, TNM stage, and the risk score. (C) Calibration curves of the nomogram for forecasting 1-, 3-, and 5-year OS in the GSE62254 cohort. (D, E) Time-dependent ROC curves of the nomogram, which includes age, TNM stage, risk score, and an age + TNM stage model for 1-, 3-, and 5-year OS in the TCGA (D) and GSE62254 (E) cohorts. (F, G) Kaplan-Meier estimates of OS based on the LECs-related nomogram in the TCGA (F) and GSE62254 (G) cohorts. (H, I) Kaplan-Meier estimates of OS derived from the combined risk score and TNM stage in the TCGA (H) and GSE62254 (I) cohorts. (J, K) Decision curve analysis (DCA) curves of the nomogram, which includes age, TNM stage, risk score, and an age + TNM stage model for predicting 1-, 3-, and 5-year OS in the TCGA (J) and GSE62254 (K) cohorts

A nomogram tailored for predicting overall survival (OS) in the GC training cohort was developed incorporating key factors, including risk score, patient age, and TNM stage, as part of the OS prediction model (Fig. 8B). The performance and clinical utility of the model were evaluated using measures such as the C-index, calibration curve, ROC curve, and DCA. These analyses pro- vided insights into the accuracy, calibration, and clinical relevance of the model.

In the TCGA cohort, the area under the receiver operating characteristic curve (AUC) for predicting overall survival (OS) at 1 year, 3 years, and 5 years was calculated to be 0.72, 0.73, and 0.73, respectively. Additionally, our nomogram exhibited a C-index of 0.708 for OS prediction. Calibration curves for the likelihood of OS at 1, 3, and 5 years aligned with actual observations and the nomogram predictions (Fig. 8C). Furthermore, for overall survival (OS) at 1 year, 3 years, and 5 years in both cohorts, the AUC values of the nomogram model exceeded those of age, indicating its superior predictive performance. The combined stage model, integrating TNM stage, risk score, and age, is visually represented in Fig. 8D, E.

Survival curves for the high-LRCPI and low-LRCPI groups exhibited significant differences, with the low- LRCPI group demonstrating better OS (Fig. 8F, G). Com- paring the OS of the same stage group (I/II or III/IV), the prognosis in the high-LRCPI and low-LRCPI groups was significantly different (Fig. 8H, I). Moreover, the combined nomogram provided a modest additional net benefit for overall survival (OS) at 1 year, 3 years, or 5 years probability compared to the model that included only clin- ical characteristics, highlighting its potential clinical utility (Fig. 8J, K). These results underscore the performance of this combined indicator in predicting poor prognosis in GC patients, indicating its clinical significance.

Discussion

The therapeutic landscape for GC remains challenging, presenting a significant global health concern. Our previous investigations highlighted the short survival time associated with distant metastasis in GC patients, under- scoring metastasis as a primary contributor to treatment failure [21]. In this study, we assessed the abundance of LECs in GC patients using transcriptomic data and the xCell algorithm, revealing close associations with clinical features and genomic expression signatures across the TCGA, GSE62254, and FUSCC cohorts. Kaplan-Meier survival analysis demonstrated that high-LEC patients in the three cohorts had lower OS, supporting our hypothesis that LECs may accelerate the progression and deterioration of gastric cancer.

Considering angiogenesis is a recognized mechanism in cancer progression [22], we initially compared the high- and low-LEC groups in both cohorts based on the expression of genes associated with angiogenesis. Notably, LECs exhibited a significant positive correlation with VEGFC ex- pression, suggesting that elevated VEGFC levels may promote LEC activation under the inflammatory conditions induced by cancer cells. Additionally, genes indicative of vessel stabilization, such as TIE1, TIE2/TEK, CDH5, and CLDN5, showed significant positive correlations with LEC abundance. Moreover, high LECs corresponded to increased pericyte abundance, which regulates vessel maturation, stabilization, and remodeling by interacting with endothelial cells [23]. These findings suggest that LECs, through their interaction with pericytes, may promote angiogenesis in tumors and enhance the stability of neovessels.

The interaction between cancer cells and stromal cells in the microenvironment is pivotal for tumor growth, and the role of the immunological environment in tumor gene- sis and progression has long been recognized [24]. In one of our published articles, we also revealed that mutations in DNA damage response genes and mismatch repair genes could be associated with immunotherapy outcomes in patients with gastrointestinal tumors [25]. Consequently, we investigated the correlation between LECs and 63 other immune cells, integrating analysis results from three datasets. The high fraction of tumor-infiltrating immune cells observed in the high-LEC group indicated that LECs may re- cruit more tumor-infiltrating immune cells, including immunosuppressive cells such as Tregs [26], by promoting lymphatic vessel generation.

Immune checkpoint ligands (ICLs) must be expressed within the tumor microenvironment to activate inhibitory signals through immune checkpoint receptors (ICRs) and promote immune evasion by tumor cells [20]. Indeed, immune checkpoint blockade therapy has shown clinical efficacy across various tumor types [27]. The observed positive association between LECs and immune checkpoint expression indicates that endothelial cells may inhibit the immune clearance of tumor cells by inducing the expression of immune checkpoints on immune cell surfaces. Notably, NRP1, expressed on endothelial cells, serves as a coreceptor for VEGFR-2, playing a crucial role in VEGF-driven angiogenesis and vasculogenesis [28, 29]. In prostate carcinoma cells, NRP1 enhances tumor angiogenesis and progression [30], suggesting that tumor-associated LECs may facilitate gastric cancer metastasis and progression by regulating NRP1 expression. The results indicate that immunotherapy aimed at chemokines and immune checkpoints may enhance the prognosis of GC patients exhibiting high LEC levels. Additionally, GSEA results revealed several enriched pathways, with the upregulation of the EMT signaling pathway as the most significant. This sheds light on the poor prognosis observed in the high-LEC group and suggests that EMT might contribute to the heightened metastatic potential of patients with elevated LEC levels.

Our previous studies have established prognostic signatures for predicting the prognosis of GC patients, prompting us to further investigate lymphatic endothelial cells (LECs) to develop another predictive model for GC patients [14, 31]. Moreover, alterations in the expression of specific genes influence the prognosis of GC patients by affecting the tumor immune microenvironment (TIM) [32]. In this study we developed a novel gene signature associated with LECs, which incorporates VWFP1, ST6GALNAC3, CCDC178, BAALC, MMP16, SNCG, ENSG00000259363, SERPINE1, P3H2, SERTM2, GDF6, SPESP1, BPI, APOD, RIMS1, GAD1, MAGEA11, and MAGEA3 to predict the prognosis of GC. The risk score model demonstrated favorable predictive performance in the training set TCGA and the validation set GSE62254. The effectiveness of this gene signature was evaluated through ROC curves and survival analysis, revealing its effectiveness. Notably, patients identified as high-risk based on this signature exhibited significantly poorer overall survival (OS).

The primary objective of this study was to establish and validate a robust prognostic model integrating LECs-related gene signatures with clinical parameters. Our LASSO-Cox regression identified 18 LECs-related genes (e.g., VWFP1, MMP16, SERPINE1), which collectively demonstrated strong prognostic power across TCGA and GSE62254 cohorts. The risk score derived from this signature effectively stratified patients into high- and low-risk groups, with significant survival differences. Notably, the model’s performance was further enhanced by incorporating clinical variables (age, TNM stage), achieving a C-index of 0.708 in the TCGA cohort and 0.682 in GSE62254, outperforming traditional TNM staging alone (C-index: 0.63–0.65). This study provided a practical tool for risk stratification in GC. For example, high-risk patients (LCRPI score ≥ 2.5) might benefit from intensified surveillance or adjuvant immunotherapy, while low-risk patients could avoid overtreatment. Nevertheless, this study has its limitations. The most critical limitation is the lack of experimental validation in this study. Additionally, the clinical information obtained from the TCGA database was incomplete and contained missing data.Then, numerous known prognostic factors, such as pathological classification and degree of differentiation, were not included in the nomogram. Finally, the retrospective study design of the nomogram necessitates validation through multi-center and large-scale prospective clinical trials to further validate its reliability and applicability.

Conclusions

Based on correlation analysis, our study has developed a prognostic prediction model incorporating LECs-related genes. This model offers a relatively straightforward approach for prognostic assessment compared to traditional clinicopathological features. Our study offers important insights into predicting the prognosis of gastric cancer patients and holds potential relevance for clinical applications.

Author contributions

SS, JZ, and WG contributed to the study’s conception. Data acquisition, analysis, and interpretation were carried out by SS, JZ, and WG. WG supervised the project. SS was responsible for writing, while SS and WG edited the manuscript. All authors participated in the critical revision for significant intellectual content and approved the final version for publication. Each author has made substantial contributions to the work, taking public responsibility for their respective sections and agreeing to be accountable for all aspects of the study, ensuring that any questions concerning its accuracy or integrity are appropriately addressed.

Funding

Not applicable.

Data availability

All data comes from http://xena.ucsc.edu, https://www.ncbi.nlm.nih.gov/geo/ and Fudan University Shanghai Cancer Center (FUSCC).

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Commit- tee of Fudan University Shanghai Cancer Center (Approval No.: 050432-4-1805 C). All experiments were performed in accordance with relevant guidelines and regulations such as the Declaration of Helsinki and the patients signed the informed consent form and agreed to be published.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.