Development of m6A-related prognostic models for survival in lung squamous cell carcinoma with different PD-L1 expression levels

Abstract

Background

Programmed death-ligand 1 (PD-L1) is widely used in the clinical context of immune checkpoint inhibitor therapy, but its relationship with N6-methyladenosine (m6A) RNA methylation in lung squamous cell carcinoma (LUSC) has not been well defined. This study aimed to investigate the association between PD-L1 messenger RNA (mRNA) expression and m6A regulator expression patterns and to develop exploratory m6A-based prognostic models in LUSC.

Methods

Transcriptome data from 502 patients with LUSC were obtained from The Cancer Genome Atlas (TCGA). Patients were divided into PD-L1 high-expression (PHE) and PD-L1 low-expression (PLE) groups according to the median PD-L1 mRNA level. Differential expression and correlation analyses were performed for 30 m6A regulators. Transcriptome sequencing data from surgical specimens from 28 Asian patients with LUSC were used for expression-pattern comparison. Principal component analysis (PCA), univariate Cox regression, and least absolute shrinkage and selection operator (LASSO)-Cox regression were used to construct prognostic models in the TCGA cohort.

Results

In the TCGA cohort, the main differentially expressed m6A regulators between the two PD-L1 groups were YTHDF2 (P<0.001), IGF2BP3 (P<0.001), and YTHDC2 (P<0.001). In the Asian cohort, ALKBH5 (P=0.008) and ZC3H13 (P=0.03) showed significant differences. LASSO-Cox models were constructed for the overall LUSC cohort and for the PHE and PLE subgroups. The overall model included METTL3, HNRNPC, and CBLL1, with a 5-year time-dependent area under the receiver operating characteristic curve (AUC) of 0.579. The 5-year AUCs were 0.742 in the PHE subgroup and 0.652 in the PLE subgroup. The risk score remained independently associated with prognosis in multivariate Cox analysis.

Conclusions

In LUSC, PD-L1 mRNA status was associated with distinct m6A regulator expression profiles. In the TCGA cohort, the PHE subgroup showed higher expression of CBLL1, G3BP1, IGF2BP3, FMR1, and YTHDC2, but lower expression of VIRMA, YTHDF2, and PRRC2A compared with the PLE subgroup. In the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences (CICAMS) cohort, ALKBH5 and ZC3H13 were more highly expressed in the PHE subgroup. Moreover, m6A-based risk models were associated with survival outcomes, with significant prognostic separation in the overall TCGA cohort and the PHE subgroup, whereas the PLE subgroup showed a weaker survival separation.

Highlight box.

Key findings

• Distinct N6-methyladenosine (m6A) regulator expression patterns were observed between programmed death-ligand 1 (PD-L1) high- and low-expression lung squamous cell carcinoma (LUSC).

• m6A-based risk signatures were associated with prognosis in the overall LUSC cohort and in PD-L1-defined subgroups, with the strongest discrimination seen in the overall cohort and the PD-L1 high-expression subgroup.

What is known and what is new?

• PD-L1 is widely used in the clinical context of immune checkpoint inhibitor therapy, but its predictive value as a standalone marker is limited. Increasing evidence suggests that m6A RNA methylation is involved in tumor progression and immune regulation.

• This study examined the relationship between PD-L1 messenger RNA (mRNA) expression and m6A regulators in LUSC from both The Cancer Genome Atlas (TCGA) cohort and an Asian institutional cohort. It further developed m6A-based prognostic models for the overall cohort and for PD-L1 high- and low-expression subgroups, showing that the composition and prognostic performance of these signatures differed according to PD-L1 status.

What is the implication, and what should change now?

• These findings suggest that m6A-related features may provide additional prognostic information in LUSC patients stratified by PD-L1 mRNA expression. Future studies should validate these signatures in external cohorts, especially in patients receiving immune checkpoint inhibitors.

• At present, the results support prognostic risk stratification rather than direct prediction of anti-programmed cell death protein 1 (PD-1)/PD-L1 treatment response.

Introduction

Lung cancer remains one of the most common malignant tumors and the leading cause of cancer-related death worldwide (1,2). It is broadly divided into non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), and NSCLC accounts for about 85% of all cases (3). NSCLC mainly includes lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), and large-cell lung cancer (4,5). With the development of immunotherapy, treatment strategies for lung cancer have changed markedly, especially in advanced disease (6). At present, programmed death-ligand 1 (PD-L1) expression and tumor mutational burden (TMB) are among the biomarkers most often used to estimate the likelihood of benefit from immune checkpoint inhibitors (7).

PD-L1 is an imperfect marker, but it still has practical value in several tumors, including NSCLC and triple-negative breast cancer (8,9). Most previous studies have focused on PD-L1 protein expression, whereas the transcriptomic features associated with PD-L1 messenger RNA (mRNA) expression in LUSC remain incompletely characterized. In this study, PD-L1 was evaluated at the mRNA level to explore whether N6-methyladenosine (m6A)-related gene expression patterns differ according to PD-L1 mRNA status.

Although immune checkpoint inhibitors have changed the treatment landscape of lung cancer, the present cohorts were not designed as anti-programmed cell death protein 1 (PD-1)/PD-L1 treatment-response cohorts. Therefore, the present analysis does not directly evaluate immunotherapy response. Instead, we aimed to characterize m6A-related transcriptomic patterns and develop prognostic models in LUSC stratified by PD-L1 mRNA expression.

N6-methyladenosine (m6A) is the most common internal RNA modification in eukaryotic mRNA and plays an important role in post-transcriptional regulation (10,11). Growing evidence shows that m6A is involved in tumor development, progression, and treatment response. For example, ALKBH5 was reported to support PD-L1 expression by reducing m6A-mediated degradation of PD-L1 mRNA in intrahepatic cholangiocarcinoma (12), and FTO knockdown decreased PD-L1 mRNA and protein expression in colon cancer cells (13). These findings suggest that certain m6A regulators may positively modulate PD-L1 expression in specific tumor contexts. On this basis, we analyzed m6A-related features in LUSC stratified by PD-L1 mRNA expression and established m6A-based prognostic models for different PD-L1 expression groups. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0829/rc).

Methods

Patients

Clinical data were collected from patients with LUSC who underwent surgery at the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences (CICAMS) between 2018 and 2019. Twenty-eight patients were selected, and the corresponding pathological wax blocks were retrieved for transcriptome sequencing. Progression-free survival (PFS) was used as the study endpoint, and follow-up was completed in June 2021. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences (approval No. 17-151/1407). The clinical specimens and institutional cohort data were collected by the corresponding author, Fenglong Bie, at this institution while he was studying at Peking Union Medical College and conducting research there. He subsequently joined Shandong Provincial Hospital Affiliated to Shandong First Medical University, which is his current affiliation. Written informed consent was obtained from all patients. The institutional cohort size was determined by data availability, and all eligible cases with available formalin-fixed paraffin-embedded (FFPE) tissue and follow-up information during the study period were included.

TCGA data

Transcriptome data [fragments per kilobase of transcript per million mapped reads (FPKM)] for patients with LUSC were downloaded from TCGA (https://portal.gdc.cancer.gov/). Only tumor-tissue transcriptome data were retained (n=502). Overall survival (OS) was used as the study endpoint. Patients were divided into PD-L1 high-expression (PHE) and PD-L1 low-expression (PLE) groups according to the median PD-L1 mRNA expression level. Differentially expressed genes (DEGs) between the two groups were analyzed in R, and the corresponding heatmaps and distribution plots were generated. The TCGA cohort included all eligible tumor samples with available transcriptome and survival data. Standardized anti-PD-1/PD-L1 treatment information and immunotherapy response data were not available for the TCGA-LUSC cohort, and the CICAMS cohort was not analyzed as an anti-PD-1/PD-L1-treated cohort; therefore, this study evaluated prognostic endpoints rather than immunotherapy-response endpoints.

RNA sequencing

FFPE tissue samples from 28 patients with LUSC were used for whole-transcriptome RNA sequencing. RNA extraction and purification were performed from FFPE sections according to the manufacturers’ protocols. RNA quantity was measured using the Life Technologies Invitrogen Qubit 3.0/4.0 system (Life Technologies/Invitrogen, Carlsbad, CA, USA), and RNA quality was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). For library preparation, 50 ng of total RNA was processed with the SMARTer Stranded Total RNA-Seq Kit v2 (Takara Bio USA, Inc., Mountain View, CA, USA) using the low-throughput protocol. This workflow uses SMART complementary DNA (cDNA) synthesis technology and generates Illumina-compatible libraries through polymerase chain reaction amplification while preserving strand orientation. After enrichment and purification, libraries were quantified again and sequenced on the Illumina HiSeq X Ten platform (Illumina, Inc., San Diego, CA, USA).

Preprocessing and mapping of sequenced data

To ensure data quality, raw reads were preprocessed by removing low-quality reads, junction contamination, ribosomal RNA contamination, and other technical artifacts to generate clean reads. All downstream analyses were based on these clean reads. Reference genome sequences and annotation files were downloaded from Ensembl (http://www.ensembl.org/). Clean reads were aligned to the reference genome using HISAT2 (http://ccb.jhu.edu/software/hisat2/index.shtml). Gene-level read counts were estimated with HTSeq (http://www.huber.embl.de/users/anders/HTSeq/doc/overview.html), and gene expression levels were quantified using the FPKM method.

Analysis of m6A regulators

m6A methylation is dynamically regulated by methyltransferases, binding proteins, and demethylases. In this study, 30 m6A-related genes were included. The 30 m6A regulators included in the present analysis were independently classified as writers, erasers, and readers based on established functional categories (14), as summarized in Table 1. Tumor mRNA expression data from the TCGA LUSC cohort and the CICAMS cohort were divided according to PD-L1 expression. Heatmaps, violin plots, correlation heatmaps, and correlation circle plots were generated to describe the expression pattern of m6A-related genes. In the TCGA cohort, PCA and consensus clustering were further performed. Univariate Cox regression was used as an initial screen, and m6A regulators with P<0.5 were retained as exploratory candidates for least absolute shrinkage and selection operator (LASSO)-Cox regression. LASSO-Cox models were constructed with the glmnet package, and lambda.min from cross-validation was used to select the final model. For each patient, a linear predictor was calculated as the sum of the expression value of each selected gene multiplied by its coefficient, and the relative risk score was calculated as exp(linear predictor). The final formulas are provided in Table 2. Patients were classified into high-risk and low-risk groups according to the median risk score within the corresponding analysis set. Kaplan-Meier curves, clinicogenomic heatmaps, univariate and multivariate Cox regression analyses, and 5-year time-dependent receiver operating characteristic (ROC) analysis were then performed.

Table 1. Functional classification of the m6A regulators analyzed in this study.

Functional category	Regulators included in the present study	Number
Writers	METTL3, METTL14, METTL16, METTL5, WTAP, VIRMA, RBM15, RBM15B, ZC3H13, CBLL1, ZCCHC4	11
Erasers	FTO, ALKBH5	2
Readers	YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, HNRNPA2B1, HNRNPC, FMR1, EIF3A, IGF2BP1, IGF2BP2, IGF2BP3, ELAVL1, G3BP1, G3BP2, PRRC2A, RBMX	17

The regulator panel was independently organized according to functional categories for the present analysis. Gene symbols follow the official gene nomenclature. VIRMA is also known as KIAA1429. m6A, N6-methyladenosine.

Table 2. Final LASSO-Cox risk-score formulas and model performance in the TCGA cohort.

Analysis set	Final LP	Risk score and cutoff	5-year AUC	Kaplan-Meier P value
Overall TCGA LUSC	LP =−0.023381 × METTL3 − 0.002594 × HNRNPC − 0.001057 × CBLL1	Risk score = exp(LP); high risk if > median 0.7422	0.579	0.002
PD-L1 high-expression subgroup	LP =−0.010883 × METTL5 + 0.158548 × METTL16 − 0.005462 × HNRNPC − 0.084720 × CBLL1 − 0.004499 × HNRNPA2B1 − 0.013645 × IGF2BP3 + 0.006512 × PRRC2A − 0.036791 × YTHDC1 − 0.015740 × FMR1 − 0.026768 × IGF2BP1 + 0.010772 × ZC3H13 + 0.012650 × YTHDF2 + 0.006417 × RBM15B − 0.023819 × RBM15	Risk score = exp(LP); high risk if > median 0.4285	0.742	2.654×10−5
PD-L1 low-expression subgroup	LP =−0.114051 × METTL3 − 0.001443 × HNRNPC − 0.023418 × G3BP1 + 0.056580 × FMR1 − 0.016670 × RBM15B + 0.001843 × FTO + 0.075018 × ZCCHC4	Risk score = exp(LP); high risk if > median 0.4505	0.652	0.07

AUC, area under the curve; LASSO, least absolute shrinkage and selection operator; LP, linear predictor; LUSC, lung squamous cell carcinoma; PD-L1, programmed death-ligand 1; TCGA, The Cancer Genome Atlas.

Statistical analysis

All statistical analyses were performed in R software (R x64 3.5.2) with the corresponding packages. Differential expression analysis was performed using the limma package, correlation analysis using corrplot, survival analyses using the survival package, LASSO-Cox regression using glmnet, and time-dependent ROC analysis using survivalROC. The institutional CICAMS cohort was used for expression-pattern comparison with the TCGA cohort, whereas prognostic model development and performance assessment were performed in the TCGA cohort because complete survival data and sufficient sample size were available. Missing or unavailable clinical variables were not imputed. A two-sided P value <0.05 was considered statistically significant.

Results

Gene expression and correlation analysis

In this study, gene expression data of 30 genes closely related to m6A methylation in the transcriptome data of TCGA LUSC patients was screened out, grouped according to PD-L1 expression, and heatmap (Figure 1A) and violin plot (Figure 1B) related to m6A methylation gene expression were drawn. The results of m6A methylation gene expression in TCGA LUSC patients showed that VIRMA (also known as KIAA1429, P=0.01), YTHDF2 (P<0.001), and PRRC2A (P=0.002) had lower expression levels in the PHE group, while CBLL1 (P=0.006), G3BP1 (P=0.004), IGF2BP3 (P<0.001), FMR1 (P=0.008) and YTHDC2 (P<0.001) had higher expression. Plot the correlation heatmap (Figure 1C) and circle plot (Figure 1D) of m6A methylation gene expression. The gene pairs with the most apparent positive correlation are YTHDC1-HNRNPA2B1 (r=0.54), RBMX-HNRNPC (r=0.53), PRRC2A-RBM15B (r=0.52). The gene pair with the most obvious negative correlation is RBM15B-WTAP (r=−0.23).

Figure 1.

Expression and correlation analysis of m6A methylation-related genes in TCGA LUSC patients with different PD-L1 expressions. (A) Heatmap of m6A methylation-related genes in different PD-L1 expression groups. The abscissa represents patient ID, and the ordinate represents the m6A methylation-related gene names. The red bar in the first row represents the PLE group, and the blue bar represents the PHE group. The green represents low-expressed genes, and the red represents high-expressed genes. (B) Violin plot of m6A methylation-related genes in different PD-L1 expression groups. The abscissa represents m6A methylation-related gene names, and the ordinate represents gene expression. The green column represents the PLE group, and the red column represents the PHE group. (C) Correlation heatmaps of m6A methylation-related genes in TCGA LUSC patients. The abscissa and oblique coordinates represent gene names, and the ordinate represents the correlation coefficients. The red represents a strong correlation, and the blue represents a weak correlation. (D) Correlation circle plots of m6A methylation-related genes. The peripheral coordinate represents gene names. The red represents a strong correlation, and the green represents a weak correlation. *, P<0.05; **, P<0.01; ***, P<0.001. LUSC, lung squamous cell carcinoma; m6A, N6-methyladenosine; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression; TCGA, The Cancer Genome Atlas.

In this study, the transcriptome data of CICAMS LUSC patients were analyzed similarly to the above, grouped according to the expression of PD-L1, and the heatmap (Figure 2A) and violin plot (Figure 2B) were drawn related to the expression of m6A methylation genes. The m6A methylation gene expression results in CICAMS LUSC patients showed that ALKBH5 (P=0.008) and ZC3H13 (P=0.03) had higher expression levels in the PHE group. Plot the correlation heatmap (Figure 2C) and circle plot (Figure 2D) of m6A methylation gene expression. The gene pairs with the most obvious positive correlation were YTHDF2-IGF2BP3 (r=0.89), ELAVL1-IGF2BP3 (r=0.87), and G3BP2-YTHDF2 (r=0.86). The gene pairs with the most obvious negative correlation were IGF2BP3-FTO (r=−0.64), METTL5-FTO (r=−0.58) and IGF2BP1-ZCCHC4 (r=−0.54).

Figure 2.

Expression and correlation analysis of m6A methylation-related genes in CICAMS LUSC patients with different PD-L1 expressions. (A) Heatmap of m6A methylation-related genes in different PD-L1 expression groups. The abscissa represents patient ID, and the ordinate represents the m6A methylation-related gene names. The red bar in the first row represents the PLE group, and the blue bar represents the PHE group. The green represents low-expressed genes, and the red represents high-expressed genes. (B) Violin plot of m6A methylation-related genes in different PD-L1 expression groups. The abscissa represents m6A methylation-related gene names, and the ordinate represents gene expression. The green column represents the PLE group, and the red column represents the PHE group. (C) Correlation heatmaps of m6A methylation-related genes in CICAMS LUSC patients. The abscissa and oblique coordinates represent gene names, and the ordinate represents the correlation coefficients. The red represents a strong correlation, and the blue represents a weak correlation. (D) Correlation circle plots of m6A methylation-related genes. The peripheral coordinate represents gene names. The red represents a strong correlation, and the green represents a weak correlation. *, P<0.05; **, P<0.01. CICAMS, National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences; LUSC, lung squamous cell carcinoma; m6A, N6-methyladenosine; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression.

Principal component analysis (PCA) and consensus clustering of m6A methylation genes

The transcriptome data of TCGA LUSC patients were carried out PCA based on the expression of m6A methylation-related genes, separately for the total LUSC (Figure 3A), PHE (Figure 3B), or PLE (Figure 3C) groups. According to consensus clustering results, patients were divided into two clusters (the most suitable classification was 2 clusters). According to clustering results, draw corresponding scatter plots separately for patients in the total LUSC, PHE, or PLE groups (Figure 3D-3F).

Figure 3.

PCA and univariate Cox proportional hazard regression analysis of m6A methylation-related genes in TCGA LUSC patients with different PD-L1 expressions. (A-C) m6A methylation-related heatmaps of consensus clustering in different PD-L1 expression groups. (A) Heatmap of consensus clustering in the total LUSC group. (B) Heatmap of consensus clustering in the PHE group. (C) Heatmap of consensus clustering in the PLE group. (D-F) Scatter plots of m6A methylation-related PCA of LUSC patients with different PD-L1 expressions. (D) Scatter plot of m6A methylation-related PCA in the total LUSC group. (E) Scatter plot of m6A methylation-related PCA in the PHE group. (F) Scatter plot of m6A methylation-related PCA in the PLE group. (G-I) Forest plots of univariate Cox proportional hazard regression analysis of m6A methylation-associated genes in LUSC patients with different PD-L1 expressions. (G) Forest plot of univariate Cox proportional hazard regression analysis of m6A methylation-related genes in the total LUSC group. (H) Forest plot of univariate Cox proportional hazard regression analysis of m6A methylation-associated genes in the PHE group. (I) Forest plot of univariate Cox proportional hazard regression analysis of m6A methylation-associated genes in the PLE group. LUSC, lung squamous cell carcinoma; m6A, N6-methyladenosine; PCA, principal component analysis; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression; TCGA, The Cancer Genome Atlas.

LASSO regression analysis

The three patient sets, including the overall LUSC cohort (Figure 3G), PHE subgroup (Figure 3H), and PLE subgroup (Figure 3I), were analyzed separately by univariate Cox proportional hazards regression. Candidate m6A regulators with P<0.5 were then subjected to LASSO-Cox regression (Figure 4A-4F). The final overall LUSC model included METTL3, HNRNPC, and CBLL1. The PHE model included METTL5, METTL16, HNRNPC, CBLL1, HNRNPA2B1, IGF2BP3, PRRC2A, YTHDC1, FMR1, IGF2BP1, ZC3H13, YTHDF2, RBM15B, and RBM15. The PLE model included METTL3, HNRNPC, G3BP1, FMR1, RBM15B, FTO, and ZCCHC4. The final risk-score formulas and median cutoffs are shown in Table 2. Kaplan-Meier analysis showed poorer survival in the high-risk group in the overall cohort (P=0.002) and PHE subgroup (P=2.654×10⁻⁵), while the survival separation in the PLE subgroup was weaker (P=0.07) (Figure 4G-4I). The 5-year time-dependent areas under the ROC curve were 0.579 for the overall cohort, 0.742 for the PHE subgroup, and 0.652 for the PLE subgroup.

Figure 4.

m6A methylation-related LASSO regression and prognostic analysis of TCGA LUSC patients with different PD-L1 expressions. (A-C) Bar plots of LASSO regression analysis of m6A methylation-related genes in LUSC patients with different PD-L1 expressions. (A) Bar plot of LASSO regression analysis of m6A methylation-associated genes in the total LUSC group. (B) Bar plot of LASSO regression analysis of m6A methylation-related genes in the PHE group. (C) Bar plot of LASSO regression analysis of m6A methylation-related genes in the PLE group. (D-F) LASSO regression curves of m6A methylation-associated genes in LUSC patients with different PD-L1 expressions. (D) LASSO regression curve of m6A methylation-related genes in the total LUSC group. (E) LASSO regression curve of m6A methylation-related genes in the PHE group. (F) LASSO regression curve of m6A methylation-associated genes in the PLE group. (G-I) Survival curves of the risk score in LUSC patients with different PD-L1 expressions. The abscissa represents survival time, and the ordinate represents survival rate. The red line represents the high-risk group, and the blue line represents the low-risk group. (G) Survival curve of the risk score in the total LUSC group. (H) Survival curve of the risk score in the PHE group. (I) Survival curve of the risk score in the PLE group. LASSO, least absolute shrinkage and selection operator; LUSC, lung squamous cell carcinoma; m6A, N6-methyladenosine; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression; TCGA, The Cancer Genome Atlas.

Clinical correlation heatmap

Group patients according to the risk score and draw the corresponding clinical correlation heatmaps (Figure 5). For total LUSC patients (Figure 5A), the m6A methylation genes are included in the construction of risk score mainly including CBLL1, METTL3, and HNRNPC. The risk score has a specific correlation with clinical traits such as T stage, age, and patients’ survival status (fustat). In the high-risk score group, the expression levels of CBLL1, METTL3, and HNRNPC were lower. For patients in the PHE group (Figure 5B), the m6A methylation genes included in the construction of risk scores mainly include METTL5, HNRNPC, METTL16, CBLL1, IGF2BP3, FMR1, ZC3H13, HNRNPA2B1, YTHDC1, IGF2BP1, PRRC2A, RBM15B, YTHDF2, and RBM15. The level of risk score has a specific correlation with the survival status of patients. In the high-risk score group, the expression levels of METTL5, HNRNPC, CBLL1, IGF2BP3, FMR1, HNRNPA2B1, YTHDC1, IGF2BP1, and RBM15 were lower. The expression levels of METTL16, ZC3H13, PRRC2A, RBM15B, and YTHDF2 were higher. For patients in the PLE group (Figure 5C), the m6A methylation genes included in the construction of risk scores mainly include FTO, ZCCHC4, METTL3, HNRNPC, RBM15B, G3BP1, and FMR1. The level of risk score has a specific correlation with the survival status of patients. In the high-risk score group, the expression levels of METTL3, HNRNPC, RBM15B, and G3BP1 were lower. The expression levels of FTO, ZCCHC4, and FMR1 were higher.

Figure 5.

Clinical heatmaps of m6A methylation-related genes in LUSC patients with different PD-L1 expressions. The abscissa represents patient ID, and the ordinate represents the classifications of clinical factors and the names of m6A methylation-associated genes. The green means low expression, and the red represents high expression. (A) Clinical heatmap in the total LUSC group. (B) Clinical heatmap in the PHE group. (C) Clinical heatmap in the PLE group. *, P<0.05; **, P<0.01. LUSC, lung squamous cell carcinoma; M, metastasis; m6A, N6-methyladenosine; N, lymph node; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression; T, tumor.

Univariate and multivariate Cox proportional hazards regression analysis

Perform univariate (Figure 6A-6C) and multivariate (Figure 6D-6F) Cox proportional hazards regression analysis of risk scores and clinical traits. For total LUSC patients, the results of univariate Cox proportional hazards regression analysis (Figure 6A) showed that clinical stage (P=0.007), tumor (T) stage (P=0.001), metastasis (M) stage (P=0.009), and risk score (P<0.001) were factors affecting prognosis. The results of multivariate Cox proportional hazards regression analysis (Figure 6D) showed that T stage (P=0.02), M stage (P=0.01), and risk score (P=0.002) were independent factors affecting prognosis. For PHE patients, the univariate Cox proportional hazards regression analysis results (Figure 6B) showed that risk score (P<0.001) was a factor affecting prognosis. The results of multivariate Cox proportional hazards regression analysis (Figure 6E) showed that risk score (P<0.001) was an independent factor affecting prognosis. For PLE patients, univariate Cox proportional hazards regression analysis results (Figure 6C) showed that gender (P=0.043), T stage (P<0.001), M stage (P=0.008) and risk score (P<0.001) were factors affecting prognosis. The results of multivariate Cox proportional hazards regression analysis (Figure 6F) showed that gender (P=0.02), T stage (P<0.001), M stage (P=0.01), and risk score (P<0.001) were independent factors affecting prognosis.

Figure 6.

m6A methylation-related Cox proportional hazards regression analysis of LUSC patients with different PD-L1 expressions. (A-C) Forest plots of univariate Cox proportional hazard model analysis of risk score and clinical characteristics in LUSC patients with different PD-L1 expressions. (A) Forest plot of univariate Cox proportional hazard model analysis of risk score and clinical characteristics in the total LUSC group. (B) Forest plot of univariate Cox proportional hazard model analysis of risk score and clinical characteristics in the PHE group. (C) Forest plot of univariate Cox proportional hazard model analysis of risk score and clinical characteristics in the PLE group. (D-F) Forest plots of multivariate Cox proportional hazard model analysis of risk score and clinical characteristics in LUSC patients with different PD-L1 expressions. (D) Forest plot of multivariate Cox proportional hazard model analysis of risk score and clinical characteristics in the total LUSC group. (E) Forest plot of multivariate Cox proportional hazard model analysis of risk score and clinical characteristics in the PHE group. (F) Forest plot of multivariate Cox proportional hazard model analysis of risk score and clinical characteristics in the PLE group. LUSC, lung squamous cell carcinoma; M, metastasis; m6A, N6-methyladenosine; N, lymph node; PD‑L1, programmed death‑ligand 1; PHE, PD‑L1 high‑expression; PLE, PD‑L1 low‑expression; T, tumor.

Discussion

In this study, we characterized m6A-related expression patterns in LUSC patients stratified by PD-L1 mRNA expression and constructed m6A-based prognostic models. The main findings were that several m6A regulators differed between PD-L1 high- and low-expression groups, and that m6A-based risk scores were associated with survival outcomes in the overall TCGA cohort and in PD-L1-defined subgroups. Importantly, this study should be interpreted as a transcriptomic association and prognostic modeling study, not as a direct evaluation of response to anti-PD-1/PD-L1 therapy.

PD-L1 expression is widely used in the clinical context of immune checkpoint inhibitor therapy, but it has important limitations as a standalone biomarker (6,9). Its expression is heterogeneous across tumor types, sampling sites, and time points (15,16), and different immunohistochemistry assays and evaluation standards may produce inconsistent results (17). In the present study, we focused on PD-L1 mRNA expression rather than PD-L1 immunohistochemistry and explored whether m6A-related transcriptomic features differed according to PD-L1 mRNA status. This approach was intended to describe molecular heterogeneity and prognostic risk, rather than to directly select patients for immunotherapy. This interpretation is consistent with recent NSCLC immunotherapy biomarker reviews, which emphasize that PD-L1 alone is insufficient and that composite biomarker assessment may improve patient stratification (18,19).

RNA methylation is an important post-transcriptional modification, and m6A is the most common internal modification in eukaryotic mRNA (11). Zaccara et al. systematically reviewed the mechanisms by which m6A regulators write, read, and erase mRNA methylation (11). Aberrant expression of m6A regulatory genes can alter RNA methylation patterns and contribute to tumor progression (20). Zhang et al. reported that YTHDC2 expression was associated with prognosis and immune infiltration across multiple tumor types (21). Su et al. showed that FTO plays oncogenic roles in multiple cancers, especially in cancer stem cell self-renewal and immune evasion (22). These studies indicate that m6A modification is closely related to tumor progression and tumor immunity. A recent review focusing on NSCLC further supports the involvement of m6A regulatory networks in tumor progression, therapeutic resistance, and biomarker development (23).

Yi et al. explored the relationship between PD-L1 expression and m6A regulators at the transcriptome level, showing that PD-L1 expression was associated with m6A regulators in head and neck squamous cell carcinomas (24). Xu et al. analyzed the relationship between PD-L1 and m6A RNA methylation regulators and showed that upregulated m6A regulators were associated with high PD-L1 expression levels (25). This is consistent with our results. TCGA LUSC data showed that CBLL1, G3BP1, IGF2BP3, FMR1, and YTHDC2 were up-regulated in the PHE group, and only VIRMA, YTHDF2, and PRRC2A were down-regulated. CICAMS LUSC data showed that ALKBH5 and ZC3H13 were up-regulated in the PHE group, and there were no down-regulated genes. Overall, m6A regulators were mainly upregulated in the PHE group.

Gu et al. analyzed the prognostic effect and mutation characteristics related to m6A-related genes in LUSC and showed some genes (WTAP, YTHDC1, and YTHDF1) were independent prognostic factors in LUSC (26). The study by Zheng et al. showed higher CBLL1 expression was associated with a better prognosis in breast cancer (27). This is consistent with our research results, which showed that lower CBLL1 expression was associated with a higher risk of poor prognosis. Wang et al. showed elevated METTL3 expression could promote gastric cancer progression, indicating that METTL3 may be a potential therapeutic target for gastric cancer (28). In our model, lower METTL3 expression contributed to a higher-risk score, suggesting that reduced METTL3 expression may be associated with poorer prognosis in LUSC patients. This finding differs from the report in gastric cancer. This opposite conclusion may be due to different types of cancer, and a more precise conclusion needs further research. Xu et al. showed that in high-risk LUSC patients, the expression of m6A RNA methylation regulators such as HNRNPC was low. These patients were more sensitive to immunotherapy and chemotherapy (29). This is also consistent with our research results, suggesting that the high expression level of HNRNPC may be a protective factor.

The main strengths of this study are as follows. First, m6A-related expression profiles were described in both the TCGA LUSC cohort and an Asian institutional LUSC cohort, and m6A-related DEGs were compared between PD-L1-defined groups. Second, the relationship between PD-L1 mRNA expression and m6A-related regulators was assessed at the transcriptomic level, providing clues for further investigation of PD-L1-associated molecular heterogeneity. However, several limitations should be acknowledged. First, standardized anti-PD-1/PD-L1 treatment information and immunotherapy response data were not available in the two cohorts, so the present results cannot demonstrate predictive value for immune checkpoint inhibitor response. Second, the assessment was mainly based on mRNA expression data and lacked validation at the DNA, protein, or m6A modification levels. Third, the institutional cohort was small and was used for expression-pattern comparison rather than independent model validation. Internal validation, calibration assessment, and external validation were not performed in this exploratory analysis.

Taken together, this study mapped m6A-related expression patterns in LUSC patients stratified by PD-L1 mRNA expression and constructed exploratory prognostic models based on m6A regulators. These models may complement routine clinical indicators such as age, sex, and clinical stage for prognostic assessment. Their potential value for predicting benefit from PD-1/PD-L1 inhibitors remains unproven and should be tested in independent LUSC cohorts with documented immune checkpoint inhibitor treatment and response data.

Conclusions

In conclusion, this study demonstrated that m6A-related regulators display distinct expression patterns in LUSC patients stratified by PD-L1 mRNA expression. m6A-based risk models provided additional prognostic information in the overall TCGA cohort and PD-L1-defined subgroups, especially in the PHE subgroup. However, because standardized anti-PD-1/PD-L1 treatment-response data were unavailable, these findings should not be interpreted as direct evidence for predicting immunotherapy response. Further validation in independent and immunotherapy-treated LUSC cohorts is warranted before clinical application.

Supplementary

The article’s supplementary files as

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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences (approval No. 17-151/1407). Written informed consent was obtained from all patients.

Data Sharing Statement

Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0829/dss

tcr-15-08-598-dss.pdf ^{(95.1KB, pdf)}

DOI: 10.21037/tcr-2026-0829