Identification and validation of an m1A-score model to predict outcomes and immunomodulation in lung squamous cell carcinoma by integrated analysis of single-cell and bulk RNA sequencing

Abstract

Lung squamous cell carcinoma (LUSC) is a common and aggressive subtype of lung cancer, characterized by difficult treatment and poor prognosis. N1-methyladenosine (m1A) is an RNA modification that plays a crucial role in various biological processes. However, the mechanisms and clinical significance of m1A in LUSC have not been fully elucidated. In this study, we performed a comprehensive analysis of m1A regulators in LUSC. The results revealed significant dysregulation of m1A regulators in LUSC, despite a low frequency of genetic alterations. Through unsupervised clustering, we identified two distinct molecular subtypes associated with m1A modulators, exhibiting significant differences in biological entities and tumor microenvironments (TME). To better guide patient management, we developed an m1A scoring system that demonstrated modest predictive power and confirmability in an external cohort. Notably, the m1A score was positively correlated with cancer-associated fibroblasts (CAF), suggesting that CAF-induced immunosuppression may contribute to the poor prognosis, which was further validated through immunohistochemical analysis. Collectively, our findings offer insights into the potential role of m1A regulators in LUSC and may inform future research on their clinical implications. The m1A score model demonstrated predictive potential for prognosis and immunotherapy response in LUSC patients, though further validation in larger cohorts is needed.

Introduction

Lung cancer poses a major worldwide health challenge owing to its high occurrence and death rates [1]. Despite progress in surgical methods, the survival rate remains low, with a 5-year survival rate of merely 20% [2]. Lung squamous cell carcinoma (LUSC) is one of the most common subtypes of lung cancer, comprising approximately 20% to 30% of all lung cancers. Patients with LUSC tend to have a worse prognosis than patients with lung adenocarcinoma [3, 4]. The treatment of LUSC is complicated by the high mutation burden, which presents a significant challenge to drug therapy. The limited availability of drugs developed for specific receptor families, coupled with excessive toxicity, hinders therapeutic efficacy. To enhance efficacy, selecting patients with specific biomarkers is often necessary, whether in lung cancer or other malignancies [4, 5]. Therefore, the development and investigation of biomarkers are critical for individuals with LUSC. Additionally, epigenetic therapy is a primary focus in the treatment of LUSC, with the histone lysine methyltransferase NSD3 promoting histone H3 methylation and driving LUSC development [6]. Loss-of-function mutations in histone lysine N-methyltransferase 2D (KMT2D) are prevalent in LUSC, affecting up to 22% of cases, but can be reversed through KDM5 inhibition [7]. Based on these findings, further studies of LUSC epigenetic targets and robust biomarker predictions are warranted.

Epigenetic dysregulation is a hallmark of LUSC, and recent advances have revealed that RNA modifications represent a new layer of post-transcriptional control involved in tumorigenesis and therapeutic resistance [8, 9]. Among more than 170 known RNA modifications, N¹-methyladenosine (m1A) has emerged as a crucial regulator of RNA stability, translation, and structural conformation [10–12]. m1A modification occurs at the first nitrogen position of adenosine and is catalyzed by a dynamic network of methyltransferases (“writers”), demethylases (“erasers”), and binding proteins (“readers”). While individual m1A regulators such as ALKBH1, ALKBH3, and TRMT6/TRMT61A have been implicated in the pathogenesis of several cancers [13–15]. However, their collective contribution to LUSC biology remains largely unexplored.

Previous investigations have typically focused on single m1A genes or limited expression analyses, providing only fragmentary insights into the complex regulatory interactions and tumor microenvironment (TME) influences governed by the m1A machinery. However, LUSC is characterized by high genomic heterogeneity, extensive stromal remodeling, and profound immune infiltration [16–18], suggesting that isolated gene-level analyses may overlook critical cross-talk between epitranscriptomic regulation and the tumor–immune ecosystem. Therefore, a systematic, multi-omics exploration integrating bulk and single-cell RNA-sequencing data is urgently needed to comprehensively delineate the landscape, functional significance, and clinical implications of m1A regulation in LUSC.

The impact of m1A on the progression of LUSC is still not clear. To address this gap in our knowledge, a systematic analysis of m1A regulators in LUSC patients was performed for the current study, and the data were retrieved and downloaded from The Cancer Genome Atlas (TCGA) project. Additionally, a prognostic m1A model for patients with LUSC was constructed, and the relationship between m1A in LUSC and immunotherapy response and TME was elucidated.

Materials and methods

Data collection and preprocessing

We conducted the GDCquery procedure of the “TCGAbiolinks” package (v2.24.2) to retrieve the transcriptome matrix of 549 TCGA-LUSC samples. Clinical data, including age, sex, disease stage, overall survival (OS), and OS status, were downloaded using the "GDCprepare_clinic" function. LUSC patients without OS status or with OS of less than 1 month were excluded from subsequent analysis. We also retrieved three external datasets (GSE3141, GSE74777, and GSE157011) from the NCBI GEO database (https://www.ncbi.nlm.nih.gov/geo/) and two scRNA datasets from the TISCH2 database (http://tisch.comp-genomics.org/) [19]. The 10 m1A regulators were collected from a previous paper [20, 21].

Identification of m1A-related subtypes

To explore the underlying m1A pattern in LUSC patients, we divided LUSC samples into two m1A-related subtypes using the ConsensusClusterPlus package (v1.60) with the clusterAlg parameter set to "km", the distance parameter set to "euclidean", and innerLinkage set to "average". We then utilized t-distributed stochastic neighbor embedding (t-SNE) to visualize the distribution of the two m1A-related subtypes and implemented Kaplan–Meier (KM) product limit analysis to compare outcomes between the two m1A subtypes.

Assessment of activities of biological behaviors

We used the Gene Set Variation Analysis (GSVA) package (v1.44.5) to calculate enrichment scores of gene sets for LUSC samples, including the v2022.1 versions of Hallmark, KEGG, and GO-Biological Process gene sets from the Molecular Signatures Database (MSigDB, downloaded on January 27, 2023) [22]. Consistent with previous studies [16, 23], the limma software package was used to identify differences in biological processes between the distinct subtypes [24].

Tumor microenvironment (TME) and immunotherapy response analysis

We used multiple bioinformatics tools to comprehensively estimate immune and stromal cell infiltration in the TME of LUSC individuals, including “ESTIMATE” [25], “CIBERSORT” [26], “MCPcounter” [27], and “EPIC” [28]. Tumor immune dysfunction and exclusion (TIDE) is a powerful model for predicting the efficacy of immunotherapy based on multiple signatures [29]. We normalized each gene's expression based on the average values across tumor samples. By submitting the normalized data of LUSC patients to the TIDE website, we calculated TIDE scores and analyzed potential immunotherapy responders among the LUSC patients.

Function annotation and PPI network

The DEGs between the distinct molecular subtypes were obtained using the “limma” package with a threshold of |log₂-fold change (FC)|> 0.5 and adjusted P value < 0.05. To explore the underlying biological mechanisms, we performed the GO-BP enrichment method to explore the functions in which DEGs might be involved. The STRING tool (https://string-db.org/) was used to generate the protein‒protein interaction (PPI) network of m1A-related DEGs. We further executed the MCODE procedure to identify the hub subnetworks and chose the top 2 to visualize in Cytoscape software (v3.9.1).

Generation and evaluation of the 3-gene signature in LUSC

We implemented the univariate Cox method to evaluate the association of candidate genes with the prognosis of LUSC patients and selected 12 prognostic genes with a P value < 0.05. We used the LASSO Cox model to build the m1A-related 3-gene signature. Based on the correlation coefficients and expression levels of the 3 genes, the risk model was established as follows:

where Coef represents the risk coefficient of each gene and Exp represents the expression levels. The LUSC patients were divided into high-risk and low-risk groups based on the median risk score. The KM product limit analysis and log-rank test were executed to analyze the clinical outcomes of the two groups. The receiver operating characteristic (ROC) method was used to assess the model discrimination efficacy of the 3-gene model in predicting OS of 1-, 3-, and 5-years in LUSC.

External validation of the 3-gene signature

To assess the robustness of the 3-gene model, three GEO cohorts (GSE3141, GSE74777, and GSE157011) were analyzed. Risk scores were calculated based on the risk coefficients and gene expression values. The optimal cutoff method [30, 31] was applied to stratify patients into high- and low-risk groups for survival comparison.

Single-cell RNA (scRNA) analysis

We conducted scRNA analysis using the “Seurat” package (v4.3.0) and determined 11 types of cells based on the expression of canonical markers [32]. The AddModuleScore function was utilized to calculate the m1A scores across cell types.

Drug sensitivity analyses

To better guide the treatment of antitumor drugs for LUSC patients, we used the “oncoPredict” package (v0.2) to predict the half maximal inhibitory concentration (IC50) values of about 200 drugs for LUSC individuals [33].

Quantitative reverse transcription PCR

Total RNA was isolated from formalin-fixed and paraffin-embedded (FFPE) LUSC tissues using the RNA Isolation Kit (Qiagen). Complementary DNA synthesis was carried out using a Reverse Transcription Kit (Thermo Fisher). The primers for quantitative reverse transcription PCR (qRT-PCR) were synthesized by Sangon Biotechnology as follows:

ALKBH1 forward primer, AGAAGCGACTAAACGGAGACC.

ALKBH1 reverse primer, GGGAAAGGTGTGTAATGATCTGC.

TRMT61B forward primer, CAGGAGCAACCGAAGACAT.

TRMT61B reverse primer, ATATACAGCACATACACCACCAT.

β-actin forward primer, TCCATCATGAAGTGTGACGT.

β-actin reverse primer, GCTCAGGAGGAGCAATGAT.

qRT-PCR assays were performed with SYBR Green 2 × Real-time PCR Mix Kit (Takara) on an ABI7500 Thermal Cycler (Applied Biosystems), and the data were analyzed using the 2^−ΔΔCt method, with the β-actin gene as an internal control.

Specimen and immunohistochemistry

A total of 38 FFPE LUSC tissues and 16 matched normal tissues were obtained from patients who underwent surgical resection at Xiangya Hospital from 2019 to 2022. Tissues were diagnosed histopathologically and staged according to the TNM classification system. Sequential sections were cut from FFPE blocks for each case and mounted on glass slides. Immunohistochemistry (IHC) was performed as previously described with antibodies against ALKBH1, TRMT61B, and FAP [34]. Two pathologists independently evaluated the staining results using the immunoreactivity score (IRS), which was determined by multiplying the intensity of the staining by the staining percentage (Staining intensity was graded as negative, weak, moderate, or strong and received scores of 0, 1, 2, or 3, respectively; staining percentage: 1:1–25%, 2:26–50%, 3:51–75%, 4:76–100%). The protein expression was classified into low expression (< 7) and high expression (≥ 7) according to immunoreactivity score [35].

Statistical analysis

Statistical analyses and visualization were performed using GraphPad Prism software (v8.0.1) and R software (v4.1.3). We analyzed the underlying differences in survival probabilities of LUSC patients using the KM method and calculated the P value using the log-rank test. Unless otherwise noted, Student’s two-tailed t test was used to compare the differences between distinct LUSC subgroups, and a P value < 0.05 was considered statistically significant. The chi-square test was performed to assess the association between candidate gene expression and clinicopathologic characteristics.

Results

Characterization of m1A regulators in LUSC

In this study, we characterized the expression of 10 m1A regulators (TRMT10C, TRMT6/61A, TRMT61B, YTHDF1, YTHDF2, YTHDF3, YTHDC1, ALKBH1 and ALKBH3) in cancer and normal tissues from the TCGA-LUSC cohort and found that all regulators except YTHDC1 were expressed at higher levels in tumors than in normal tissues (Fig. 1A). To further reveal the relationship between m1A regulators, we constructed a correlation network through Spearman analysis, and the results showed that all regulators were positively correlated to exert the modification function of m1A (Fig. 1B). Next, we analyzed the mutation information of m1A regulators and found that there were low-frequency mutations of m1A regulators in LUSC samples. Among the 485 LUSC samples, 32 mutations occurred (approximately 6.6%). Seven genes (TRMT6, YTHDF1 YTHDF3, TRMT10C, YTHDF2, ALKBH1 and ALKBH3) had 1% mutations, and the remaining three genes (TRMT61B YTHDC1, TRMT6/61 a) had no mutations (Fig. 1C). In addition, we analyzed the copy number variation (CNV) information of the m1A regulators. All m1A regulators except YTHDF2 were dominated by copy gain, and only YTHDF2 was dominated by copy loss (Fig. 1D). In general, our analysis outlines the expression landscape of m1A regulators in LUSC.

Fig. 1.

Comprehensive analysis of m1A regulators in LUSC. A Differential expression analysis of 10 m1A regulators between normal and LUSC samples. B Correlation network of the m1A regulators based on Spearman analysis. Red indicates positive correlation; blue indicates negative correlation. C Mutation frequencies of 10 m1A regulators in 485 patients with LUSC from the TCGA cohort. D Lollipop plot showing CNV frequencies of the m1A regulators. ^*P < 0.05, ^****P < 0.0001

Two molecular subtypes related to m1A regulators

To better evaluate the clinical significance of m1A for LUSC patients, we implemented a consensus clustering algorithm to classify LUSC samples based on transcriptome profiles of 10 m1A regulators. The heatmap showed that there was a clear boundary between the two clusters of LUSC patients when k = 2, which was further supported by t-SNE analysis (Fig. 2A, B). The KM curves illustrated that cluster 2 had a significantly higher survival advantage than cluster 1 (Fig. 2C). We then performed GSVA on the 50 hallmark gene sets in the distinct m1A-related clusters. Our analysis revealed that cluster 1 exhibited significant enrichment in TNF-α signaling via NF-κB, TGF-beta signaling, hypoxia, apoptosis, NOTCH signaling, P53 pathway, etc., while cluster 2 was mainly enriched in the unfolded protein response, oxidative phosphorylation, MYC target, etc. (Fig. 2D), which might be the underlying mechanism contributing to the difference in prognosis of the two m1A subtypes. In addition, we employed the ESTIMATE algorithm to obtain TME scores of LUSC samples, which included stromal, immune, and ESTIMATE scores. The results showed that the cluster 2 scores were significantly lower than those of cluster 1 (Fig. 2E). We further analyzed the relative proportions of 22 immune cell types in each LUSC patient and found that five immune cell types (memory B cells, CD8 + T cells, resting memory CD4 + T cells, Tregs, and M0 macrophages) were significantly different between the two m1A subtypes (Fig. 2F). For example, cluster 1 had higher M0 macrophages but lower CD8 + T cells, indicating that the immune activity of cluster 1 might be suppressed. Regarding immune checkpoints, we found that the expression levels of PDCD1, CTLA4, LAG3, and HAVCR2 were significantly lower in cluster 2 (Fig. 2G). Based on these findings, it is plausible to suggest that m1A modification might have a crucial role in regulating immunity and the TME in LUSC patients.

Fig. 2.

Identification of two molecular subtypes related to m1A regulators. A Clustering heatmap of the consensus matrix when k = 2. B t-SNE analysis of two m1A clusters. C Kaplan–Meier curve for survival probabilities of two m1A clusters. D Enrichment scores of 50 hallmark pathways in two m1A clusters. Yellow represents the pathways that were enriched in cluster 1. E The difference analysis of TME scores in two m1A clusters. F Infiltration analysis of 22 types of immune cells in two m1A clusters. G Boxplots showing the differential expression of immune checkpoints (e.g., PDCD1, CTLA4, LAG3, and HAVCR2) in the two m1A clusters. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.0001, ns, no significance

Identification of gene clusters in LUSC

To further characterize the biological differences among the two m1A-modified clusters, we obtained 744 DEGs between cluster 1 and cluster 2 and constructed a PPI network using the STRING database. By constructing the MCODE module, we identified the top 2 ranked subnetworks (Fig. 3A, B). GO enrichment analysis revealed that these genes were significantly enriched in immune-related functions (Fig. 3C). We then divided LUSC patients into 2 gene clusters (clusters A and B) using the unsupervised clustering method, and the heatmap illustrated the Z scores of DEGs in the two DEG-related clusters and two m1A-related clusters (Fig. 3D). Similarly, cluster A had a better survival advantage than cluster B (Fig. 3E). Then, we calculated the enrichment scores of KEGG pathways for the two gene clusters by GSVA and implemented the “limma” package to identify the different pathways with a P value < 0.001. The heatmap analysis presented in Fig. 3F indicated that cluster B was enriched for several immune-related pathways, including antigen processing and presentation, the T-cell receptor signaling pathway, and the chemokine signaling pathway. Additionally, the TME infiltration analysis depicted in Fig. 3G demonstrated that cluster B had higher scores than cluster A. These results provide further evidence that m1A regulators may be associated with the TME in LUSC patients.

Fig. 3.

Gene clustering based on DEGs between clusters 1 and 2. A, B The top 2 hub networks of PPI using the MCODE procedure. C GO enrichment results based on genes from the hub networks. D Heatmap of DEGs in two gene clusters and two m1A clusters. The gene expression levels were transformed to Z scores by scale function. E Kaplan–Meier curves for survival probabilities of two DEG-related clusters. P values were calculated by the log-rank test. F Different KEGG pathways between two DEG-related clusters. Red represents higher enrichment scores. G Three kinds of TME scores in two DEG-related clusters. ^****P < 0.0001

Construction of the m1A-related 3-gene signature

To better predict the prognosis of LUSC patients, we performed the LASSO method to build an m1A-related risk model based on prognostic DEGs identified by univariate Cox regression analysis. The results showed that after tenfold validation and 1000 iterations, the optimal number of model genes was 3 (Fig. 4A). The risk scores of LUSC patients were calculated by the following formula: Risk score = (0.0590 * Exp SERPINE1) + (0.0903 * Exp ITGA3) + (0.0617 * Exp VSIG4). LUSC patients were then dichotomized into high-risk and low-risk groups according to the median risk score. K-M analysis revealed that the high-risk group had significantly shorter OS than the low-risk group (log-rank test, P value = 0.0043) (Fig. 4B). The ROC results suggested that the 3-gene signature showed moderate predictive ability, with area under the curve (AUC) values of 0.6, 0.63, and 0.61 for 1-, 3-, and 5-year survival, respectively (Fig. 4C). In three external cohorts (GSE3141, GSE74777, and GSE157011), the patients with high risk tended to have a shorter OS than those with low risk (Fig. 4D–F).

Fig. 4.

Generation and performance of the 3-gene m1A signature. A The line chart shows the partial likelihood deviance by the LASSO Cox model. B, C Kaplan–Meier curves for survival probabilities and ROC curves for 1-, 3-, and 5-year survival of the 3-gene signature. D–F Kaplan–Meier curves for survival probabilities of the 3-gene model in three validation sets (GSE3141, GSE74777, and GSE157011)

m1A score as a prognostic risk factor in LUSC

Our findings supported the association of m1A regulators with outcomes in LUSC. To better evaluate this influence, we carried out the ssGSEA algorithm to quantify the m1A score of each LUSC patient based on the 3-gene signature. In a similar manner to the risk model classification, we divided LUSC patients into high- and low-m1A subgroups based on the median value. The survival analysis results indicated that the low-m1A group had better outcomes than the high-m1A group, as depicted in Fig. 5A. Furthermore, the ROC curve analysis revealed that the AUC values for 2-, 4-, and 6-year survival were 0.61, 0.61, and 0.59, respectively, as shown in Fig. 5B. We executed the univariate Cox method to compare the HRs and P values of m1A with 50 hallmark gene sets, and the results suggested that m1A was the most significant prognostic risk factor (Fig. 5C). Combined with univariate and multivariate Cox regression analysis, it was demonstrated that the m1A score was an independent biomarker of outcomes after adjusting for other clinical variables (univariate: HR = 3.7388, P = 0.0005; multivariate: HR = 5.8483, P = 0.0001) (Fig. 5D, E). Furthermore, we validated the m1A-score model with two independent datasets (GSE3141 and GSE157011), and the KM curves showed that patients with low m1A scores had a favorable prognosis (Fig. 5F, G). These findings suggest that the m1A score may serve as a potential tool for predicting OS in LUSC.

Fig. 5.

Development and assessment of the m1A-score system for LUSC individuals. A Kaplan–Meier curves for survival probabilities of distinct m1A subgroups in TCGA-LUSC. B ROC curves of m1A subgroups in the TCGA-LUSC set. C The hazard ratio (HR) and P value of m1A scores and 50 hallmark gene sets. D, E Univariate Cox and multivariate Cox analyses of the m1A-score model and other clinical characteristics. F, G Survival curves for outcomes of distinct m1A subgroups in two independent cohorts (GSE3141 and GSE157011)

m1A score were correlated with cancer-associated fibroblasts (CAF)

The previous results suggested that m1A regulators had a significant correlation with the TME. To systematically assess the role of the m1A score in the TME of LUSC patients, we further analyzed the TME using single-cell RNA datasets. In GSE153935, we identified 11 types of cells in TME, including alveolar cells, CD8 + T, CD8 + Tex, endothelial and epithelial, fibroblasts, mast, mono/macro, myofibroblasts, plasma, and T-proliferation (Fig. 6A). The dotplot shows the marker genes across different cell types (Fig. 6B). For example, the markers of plasma cells were SLAMF7, IGKC, and JCHAIN. We carried out the AddModuleScore procedure to assess the m1A scores across cell types and found that fibroblasts had the highest scores, followed by myofibroblasts and endothelial cells (Fig. 6C, D). In another single-cell dataset, EMTAB-6149, we classified cells into immune cells, stromal cells, and tumor cells (Fig. 6E). Consistent with our previous findings, the results revealed that stromal cells had the highest m1A scores compared to the other cell types (Fig. 6F).

Fig. 6.

m1A was significantly correlated with CAF. A A total of 11 cell types were identified in GSE153935. B The dot plots show the canonical marker genes across cell types. C, D The distributions of m1A scores across distinct cell types. E Three major cell types were confirmed in EMTAB-6149. F The m1A scores were abundant in stromal cells

To validate the scRNA analysis results, we conducted immune infiltration analysis on the TCGA-LUSC dataset. Our Spearman correlation analysis revealed a positive correlation between m1A scores and stromal scores (Fig. 7A). Additionally, we employed two algorithms (EPIC and MCP-counter) to estimate the fractions of CAF and found that the high-m1A group exhibited higher levels of infiltration (Fig. 7B, C). Furthermore, we utilized the CIBERSORT method to evaluate the 22 types of immune cells in the TCGA-LUSC cohort. Interestingly, we observed a relatively quiescent immune microenvironment in the high m1A group, characterized by higher levels of resting CD4 + T memory cells and M2 macrophages, as well as lower levels of T follicular helper cells (Fig. 7D). Considering the important impact of the immune microenvironment on immunotherapy [36, 37], we used the TIDE algorithm to analyze the relationship between the m1A scores and immunotherapy responses in LUSC patients. We found that the patients with low m1A scores had lower TIDE scores and were more responsive to immunotherapy (Fig. 7E, F), which was well validated in the bladder cancer dataset (Fig. 7G, H). These findings indicate that CAF may contribute to an immunosuppressive microenvironment in LUSC patients with high m1A levels, which could be associated with poorer prognosis.

Fig. 7.

The two m1A subgroups had distinct TME patterns. A The m1A scores showed a highly positive correlation with the stromal scores. B Infiltration analysis of fibroblasts in the two m1A subgroups using the EPIC algorithm. C The high m1A group had higher CAF fractions than the low m1A group using the MCPcounter method. D The comparison of TME infiltration profiles between the two m1A subgroups. E, F Prediction of immunotherapy response based on the TIDE algorithm. G, H The efficacy of m1A scores for immunotherapy response in the IMvigor210 cohort. ^*P < 0.05, ^***P < 0.001, ^****P < 0.0001, ns, no significance

Correlation of m1A scores with drug sensitivity

To enhance the efficacy of treatment plans for LUSC patients, we employed the "oncoPredict" package to predict the susceptibility of ten antitumor drugs that are either clinically used to treat lung cancer or undergoing clinical trials. Our analysis revealed that patients with low m1A scores were more responsive to several antitumor drugs, including vinorelbine, docetaxel, erlotinib, axitinib, gefitinib, paclitaxel, rapamycin, and lapatinib (Fig. 8A–H). Conversely, patients with high m1A scores showed greater sensitivity to selumetinib and dasatinib (Fig. 8I, J). These findings highlight the potential of m1A regulators as therapeutic targets in LUSC and suggest that they may facilitate the development of personalized medicine.

Fig. 8.

Prediction of drug susceptibility. A–H The IC50 values of antitumor drugs were higher in patients with high m1A score. I, J The IC50 values of antitumor drugs were higher in the low m1A score patients

Validation of representative m1A regulators expression in LUSC tissues and its relationship with CAF

To validate that m1A regulator can serve as a prognostic marker for LUSC, we selected ALKBH1 and TRMT61B with the highest correlation among all m1A regulators for validation by using immunohistochemistry (IHC) or quantitative reverse transcription PCR (qRT-PCR). Immunohistochemical analysis was performed to evaluate the expression patterns of ALKBH1 and TRMT61B in LUSC tissues (n = 38) and adjacent normal tissues (n = 16). Representative staining is shown in Fig. 9A and B, and the results indicated that the expression of ALKBH1 and TRMT61B was significantly higher in LUSC tissues than in normal lung tissues, according to the immunoreactive score (IRS) (Fig. 9C, D). In addition, a subset of samples underwent qRT-PCR analysis to ascertain transcript levels, revealing a substantial upregulation of ALKBH1 and TRMT61B transcripts (Fig. 9E, F). Upon sorting by IRS, we found that the expression of both ALKBH1 and TRMT61B was significantly higher in a greater proportion of LUSC tissues than in adjacent normal tissues (Fig. 9G, H). Specifically, 81.5% (31/38) of tumor tissues exhibited high expression of ALKBH1, while 73.6% (28/38) showed high expression of TRMT61B. Further stratification analysis indicated that high expression of ALKBH1 and TRMT61B was related to higher pathologic grade (Fig. 9I, J). These results suggest that specific m1A regulators may have potential as prognostic and diagnostic indicators for LUSC, pending further validation.

Fig. 9.

Validation by immunohistochemistry and qRT-PCR. A, B Representative IHC images of ALKBH1 and TRMT61B in LUSC tissues and adjacent normal tissues. Scale bar: 50 μm. C, D The staining score of ALKBH1 and TRMT61B expression in LUSC tissues compared with adjacent normal tissues. E, F ALKBH1 and TRMT61B mRNA levels in LUSC tissues and normal lung tissues measured by qRT-PCR. G, H Proportion of ALKBH1 and TRMT61B expression in LUSC tissues and normal lung tissues. I, J Association of ALKBH1 and TRMT61B expression with pathological grade in the LUSC cohort. K Representative IHC images of FAP, ALKBH1 and TRMT61B in LUSC tissues. The staining levels were evaluated by IRS. Scale bar: 50 μm. L Correlation analysis of FAP and ALKBH1, as well as FAP and TRMT61B expression levels in LUSC tissues. The P value was determined by Pearson’s correlation test. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.0001

Beyond this, we evaluated the association between m1A regulators and CAF by immunohistochemical staining with fibroblast activating protein (FAP), a main biomarker of CAF. The expression pattern of FAP exhibited similarity with the expression levels of ALKBH1 and TRMT61B (Fig. 9K). As assessed by IHC staining scores and correlation analysis, the degree of concordance between FAP and ALKBH1 was high, as well as FAP and TRMT61B (Fig. 9L). The Pearson correlation analysis showed a significant positive correlation between them. Therefore, our data strongly support the relevance of m1A regulators and CAF, implying that CAF may lead to poor prognosis through immunosuppression.

Discussion

RNA modifications have recently emerged as pivotal regulators of gene expression and tumor progression. Among these, m1A represents an important but understudied layer of epitranscriptomic control. In this study, we identified in the TCGA-LUSC cohort that all m1A regulators except YTHDC1 were significantly highly expressed in cancer. The two most correlated m1A regulators, ALKBH1 and TRMT61B, were well validated by immunohistochemistry of LUSC tissue. YTHDC1, one of the readers of m1A, plays an important role in mRNA splicing. YTHDC1 is downregulated in clear cell renal cell carcinoma (ccRCC), inhibits the progression of kidney cancer cells by the ANXA1/MAPK pathway [38], and restricts the proliferation of glioma by inhibiting VPS25 [39]. YTHDC1 mediates miR-30d/RUNX1 to inhibit the occurrence of pancreatic tumors [40], which proves that YTHDC1 plays an inhibitory role in a variety of cancers, which is consistent with our data analysis. However, the effect of YDHTC1 on LUSC needs to be further tested.

In the 3-gene m1A scoring model, we used the angiogenic factors SERPINE1, integrin α3 (ITGA3), V-set and immunoglobulin domain-containing 4 (VSIG4). SERPINE1 is highly expressed in advanced lung cancer, is involved in lung cancer growth as a target of CD248, inhibits Wnt signal transduction pathways and is associated with worse outcomes in lung cancer patients [41]. In addition, SERPINE1 deficiency effectively inhibits the progression of gastric cancer (GC) [42]. Studies have shown that integrins are involved in the process of tumor growth and progression. In melanoma studies, it was found that miR-214 regulates ITAG3 and other surface proteins to coordinate the migration of melanoma [43]. Similarly, it was found in non-small cell lung cancer that ITAG3 was highly expressed and correlated with poor prognosis of patients [44]. VSIG4, a member of the B7 family, is specifically expressed in resting macrophages, negatively regulates the proliferation of human and mouse T cells [45], and inhibits the transcription of Nlrp3 and Il-1β through A20-mediated NF-κB pathway inactivation [46].

Importantly, our analysis expands upon and complements existing studies on m6A-based prognostic models. m6A methylation, which occurs at the sixth nitrogen atom of adenosine, primarily influences RNA splicing and degradation, whereas m1A modification alters RNA structure and translation initiation [47]. Several m6A-derived models have been established in lung cancer [48–51], highlighting the relevance of RNA methylation in tumor immunity. Yet, these models rely solely on bulk-level expression data and do not capture cellular specificity within the TME. By contrast, our m1A-score integrates bulk and single-cell information, thereby uncovering cell-type–resolved insights into stromal activation and immune exclusion. This integrated approach provides a complementary dimension to m6A-based studies, emphasizing the role of m1A-mediated fibroblast remodeling in shaping an immunosuppressive microenvironment.

The TME is an important factor that promotes tumor development and metastasis and affects drug therapy sensitivity [52]. The transformation of fibroblasts into CAF in the TME has received much attention due to its role as an immune cell recruitment regulator and tumor promoter [53]. However, recent studies have interpreted the role of fibroblasts in the TME from a different perspective. CAF can induce M2 polarization of macrophages through paracrine effects of M-CSF in pancreatic cancer [54]; similarly, CAF have been found to recruit monocytes and differentiate them into M2-like macrophages in breast cancer [55]. In addition, recent TCGA data analysis revealed that CAF score was positively correlated with CD4 + T-cell infiltration and resting NK cell infiltration [56, 57]. These imply that CAF can induce immunosuppression within the TME of various tumors through diverse mechanisms. While their impact on LUSC is similar, the precise underlying mechanisms necessitate additional investigation.

This study has several limitations. First, although the m1A-score model was validated using multiple public cohorts, the external sample sizes were relatively small, and the predictive performance was modest, which may restrict the generalizability of our findings. Second, the newly generated qRT-PCR and immunohistochemistry data were derived from a limited number of clinical samples at a single institution, and thus larger, multicenter cohorts will be needed to confirm these results. Third, most of the mechanistic insights into the relationship between m1A regulators, cancer-associated fibroblasts, and the tumor immune microenvironment were inferred from bioinformatics analyses; experimental validation in functional models is warranted. Finally, the predictive power of 3-gene m1A-score model remains modest. This limitation indicates that the model currently serves as a conceptual and exploratory framework for understanding m1A-related transcriptional heterogeneity in LUSC rather than as a ready-to-use clinical predictor. The integration of additional molecular variables (e.g., mutational, methylation, and proteomic profiles) and validation in larger multicenter cohorts will be essential to improve its robustness and translational potential.

Conclusions

In conclusion, this study provides evidence that m1A regulators are dysregulated in lung squamous cell carcinoma and that their expression patterns are associated with distinct molecular subtypes, tumor microenvironment features, and patient outcomes. By integrating bulk and single-cell transcriptomic data, we developed and preliminarily validated an m1A-score model that showed potential for predicting prognosis and immunotherapy response. Importantly, we also observed associations between high m1A scores, cancer-associated fibroblasts, and immunosuppressive microenvironmental patterns. While these findings highlight the possible clinical significance of m1A in LUSC, further large-scale and prospective studies are required to confirm their robustness and to explore underlying mechanisms. Taken together, our results offer a foundation for future research aimed at refining prognostic biomarkers and developing individualized therapeutic strategies for patients with LUSC.

Abbreviations

AAA

Abdominal aortic aneurysm

AUC

Area under the curve

CAF

Cancer-associated fibroblast

ccRCC

Clear cell renal cell carcinoma

CNV

Copy number variation

DEGs

Differentially expressed genes

FFPE

Formalin-fixed and paraffin-embedded

GC

Gastric cancer

LUSC

Lung squamous cell carcinoma

m1A

N1-methyladenosine

TME

Tumor microenvironment

scRNA-seq

Single-cell RNA-sequencing

PPI

Protein-protein interaction

GSVA

Gene set variation analysis

HCC

Hepatocellular carcinoma

IHC

Immunohistochemistry

IRS

Immunoreactivity score

ITGA3

Integrin α3

KMT2D

Histone lysine N-methyltransferase 2D

KM

Kaplan–Meier

LASSO

Least absolute shrinkage and selection operator

OS

Overall survival

PD1

Programmed cell death protein 1

ROC

Receiver operating characteristic

TIDE

Tumor immune dysfunction and exclusion

t-SNE

T-distributed stochastic neighbor embedding

VSIG4

V-set and immunoglobulin domain containing 4

Author contributions

H.G., H.Z. and L.Z. contributed to the idea for the article, performed bioinformatics analyses, and wrote the manuscript. Y.Z. performed the experimental verification of gene expression. D.X.,Y.Z.,and J.L. revised the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (No. 81920108004, 82270127, 82303573), the fellowship of China Postdoctoral Science Foundation (No. 2023M743967), the Natural Science Foundation of Hunan Province (No. 2024JJ6684, 2024JJ6646), the Natural Science Foundation of Changsha (No. kq2403025), and the Youth Science Foundation of Xiangya Hospital (2022Q15, 2023Q20).

Data availability

The data that support the findings of this study are available on public databases. Gene expression profiles and clinical data for TCGA-LUAD were obtained from the GDC database (https://xenabrowser.net/datapages/). The GEO dataset accessions analyzed in this study are GSE3141 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE3141), GSE3141 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE74777), and GSE31547 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc = GSE157011).

Declarations

Ethics approval and consent to participate

The study using clinical samples was approved by the Medical Ethics Committee of Xiangya Hospital (No. 2023111626) and conducted in accordance with the Declaration of Helsinki and applicable national regulations. Research involving human research participants have been performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients prior to surgery.

Consent for publication

Consent to publish has been obtained from all authors.

Competing interests

The authors declare no competing interests.