Identification of pathology-specific regulators of m⁶A RNA modification to optimize lung cancer management in the context of predictive, preventive, and personalized medicine

Abstract

Relevance

Lung cancer is the most common malignant tumor with high morbidity (11.6% of the total diagnosed cancer cases) and mortality (18.4% of the total cancer deaths), and its 5-year survival rate is very low (20%). Clarification of any molecular events and the discovery of effective biomarkers will offer increasing promise for lung canner management. N⁶-methyladenosine (m⁶A) modification is one of the important RNA modifications that are closely associated with lung cancer, and are tightly regulated by m⁶A regulators. Elucidation of pathology-specific m⁶A regulators will directly contribute to lung cancer medical services in the context of predictive, preventive, and personalized medicine (PPPM).

Purpose

To investigate pathology-specific regulators of m⁶A RNA modifications in lung cancer and further inspect the m⁶A regulator gene signature as useful tools for PPPM in lung cancers.

Methods

The gene expression data of 19 m⁶A regulators (m⁶A-methyltransferases—ZC3H13, KIAA1429, RBM15/15B, WTAP, and METTL3/14; demethylases—FTO and ALKBH5; and m⁶A-binding proteins—HNRNPC, YTHDF1/2/3, YTHDC1/2, IGF2BP1/2/3, and HNRNPA2B1) and clinical data of 1013 lung cancer patients [511 lung adenocarcinoma (LUAD) and 502 lung squamous carcinoma (LUSC)] and 109 controls (Con) were obtained from the TCGA database. Quantitative real-time PCR (qRT-PCR) was used to verify m⁶A regulators in lung cancer cell lines. Protein–protein interaction (PPI), gene co-expression, survival analysis, and heatmap were used to analyze these m⁶A regulators in this set of lung cancer clinical data. Lasso regression was used to optimize the pathology-specific m⁶A regulator gene signature. Gene set enrichment analysis (GSEA) was used to reveal the functional characteristics of m⁶A regulators.

Results

Those 19 m⁶A regulator profiling was significantly differentially expressed in lung cancer tissues relative to control tissues, which was also verified in lung cancer cell lines. Those m⁶A regulators interacted mutually, and those regulator-based sample clusters were correlated with clinical traits, including survival status, gender, tobacco smoking history, primary disease, and pathologic stage. Further, lasso regression based on the 19 m⁶A regulators optimized and identified a three-m⁶A-regulator signature (KIAA1429, METTL3, and IGF2BP1) as independent prognostic factor, which classified 1013 lung cancer patients into high-risk and low-risk groups according to median value (0.84) of the lasso regression risk scores. This three-m⁶A-regulator signature profiling was significantly related to lung cancer overall survival, cancer status, and the above-described clinical traits. Further, GSEA revealed that KIAA1429, METTL3, and IGF2BP1 were significantly related to multiple biological behaviors, including proliferation, apoptosis, metastasis, energy metabolism, drug resistance, and recurrence, and that KIAA1429 and IGF2BP1 had potential target genes, including E2F3, WTAP, CCND1, CDK4, EGR2, YBX1, and TLX, which were associated with cancers.

Conclusion

This study provided the first view of the pathology-specific regulators of m⁶A RNA modification in lung cancers and identified the three-m⁶A-regulator signature (KIAA1429, METTL3, and IGF2BP1) as an independent prognostic model to classify lung cancers into high- and low-risk groups for patient stratification, prognostic assessment, and personalized treatment toward PPPM in lung cancers.

Electronic supplementary material

The online version of this article (10.1007/s13167-020-00220-3) contains supplementary material, which is available to authorized users.

Introduction

Unsatisfactory statistics for current lung cancer management

Lung cancer is a common malignant tumor with high morbidity and mortality, which has aroused a wide concern in the world. Currently, lung cancer is the most commonly diagnosed cancer (11.6% of the total cases) and the leading cause of cancer death (18.4% of the total cancer deaths) [1]. Despite progressions in surgery, radiotherapy, and chemotherapy that have been achieved in lung cancer, its survival improvement is still a big challenge, with a low 5-year survival rate (20%) [2]. The main reasons that result in its low 5-year survival rate are as follows: (i) The early symptoms of lung cancer are commonly concealed, which causes most patients diagnosed in middle or advanced stages; thus, only a partial percentage (30%) of patients has a chance to undergo radical surgery in an early stage [3]. (ii) Lymphatic metastasis and distant metastasis are very common after surgery, and about 50% of the patients have recurrence and metastasis within 5 years [4]. (iii) The number of targeted therapy and immunotherapy is very limited although targeted therapy and immunotherapy are very effective and have achieved some progresses, with the discovery of many molecular targets, especially for the treatment of advanced stage of lung cancers, for example, EGFR-TKI drugs (erlotinib, gefitinib, and afatinib) that inhibit tyrosine kinase at the epidermal growth factor receptor and immunotherapy drugs (pembrolizumab, nivolumab, and durvalumab) that target immune checkpoint (PD-1/PD-L1) for the late stage nonsmall cell lung cancer [5]. There is an urgent need to develop more novel molecular targets for lung cancer.

Role of m⁶A RNA modification and m⁶A regulators in pathology-specific molecular signature

The epigenetics about the modifications occurring at DNAs and RNAs have been widely studied in tumor pathogenesis [6, 7]. Of them, m⁶A modification was one of the emerging research frontiers, which modifies different types of RNAs, including transfer RNAs (tRNAs), messenger RNAs (mRNAs), ribosomal RNAs (rRNAs), small nuclear RNA (snRNA), and long noncoding RNAs (lncRNAs), and others, and which is the most abundant posttranscriptional modification in eukaryotes [8]. A study has found that m⁶A modification is present in more than 7600 mRNAs and in more than 300 noncoding RNAs (ncRNAs) [9]. m⁶A modification can change RNA processing such as splicing, export, intracellular distribution stability, and translation [10]. The cellular m⁶A status is tightly regulated by a series of regulators, including m⁶A methyltransferases (named as “writers”), demethylases (named as “erasers”), and m⁶A-binding proteins (named as “readers”) [11]. The methyltransferase complex contains ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, and METTL14, which catalyzes m⁶A methylation [12]. Demethylases include FTO and ALKBH5, which catalyze oxidative demethylation of m⁶A [13]. The m⁶A-binding proteins include HNRNPC, YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, IGF2BP1, IGF2BP2, IGF2BP3, and HNRNPA2B1 [14]. The current technical methods used to detect m⁶A modification included high-throughput sequencing [15], colorimetric method, and liquid chromatography–mass spectrometry (LC-MS) [16]. Common methods included MeRIP-seq, miCLIP-seq, LC-MS/MS, and colorimetric method [17]. The multiple functions of m⁶A modification are associated with the occurrence and worsening of numerous human diseases, including cancer [18]. For example, the m⁶A eraser FTO promotes the growth of lung cancer cells by regulating the m⁶A level of ubiquitin-specific protease 7 (USP7) mRNAs, and a study has found that the overexpressed USP7 mRNA level is positively correlated with FTO mRNA level in human lung cancer tissues, and that FTO decreases the m⁶A level and increases mRNA stability of USP7, which was dependent on the demethylase activity of FTO [19]. The m⁶A reader IGF2BP1 promotes SRF-dependent transcription in cancer by N6-m⁶A-dependent manner in cancer, which results in the increased SRF-dependent transcriptional activity and enhances tumor cell proliferation and metastasis [20]. The m⁶A writer METTL3 promotes tumor proliferation by accelerating pri-miR221/222 maturation in the m⁶A-dependent manner [21]. Further, a study found that m⁶A regulators take part in m⁶A modification of many ncRNAs, such as microRNAs, lncRNAs, and circular RNAs. A review reports that m⁶A is closely associated with numerous cancers, including lung cancer, prostate cancer, stomach cancer, breast cancer, pancreatic cancer, kidney cancer, colorectal carcinoma, mesothelioma, sarcoma, and leukemia [22]. A study has found that these m⁶A regulators are the potential biomarkers for prognostic prediction and new therapeutic targets in cancer [23]; for example, the m⁶A regulator METTL3 is found to associate with mitogen-activated protein kinase cascades, ubiquitin-dependent process, RNA splicing, and regulation of cellular process, which suggests that METTL3 is a potential target for enhancing therapeutic efficacy in pancreatic cancer [24]. Indeed, the findings that m⁶A regulators are dysregulated and play important roles in various types of cancer suggest that m⁶A RNA modification could be involved in therapy resistance development [25]. Inhibitors of m⁶A regulators hold the potential to be combined with chemotherapy or radiotherapy to erase cancer stem cells and achieve complete remission [26]. Such data suggest that m⁶A RNA modification signatures may serve as predictive markers for personalized cancer treatment and shed light on how to overcome therapy resistance in cancer. The m⁶A writer KIAA1429 is revealed as an oncogenic factor in breast cancer and hepatocellular carcinoma, but has not been reported in lung cancer [27, 28]. The m⁶A writer METTL3 shows low expression in lung adenocarcinomas (LUAD) and squamous cell carcinoma (LUSC), when compared to normal lung epithelia, in the TCGA and GTEx data (p < 0.0001) [29]. The m⁶A reader IGF2BP1 is significantly increased in LUAD tissues (n = 188) compared to adjacent noncancerous tissues (n = 188), and a high expression of IGF2BP1 is correlated with a poor prognosis for lung cancer patients [30]. While m⁶A plays important roles in different types of cancers, the clinical trait–related m⁶A regulator study in lung cancer is insufficient. It emphasizes the important scientific merits of m⁶A regulators in cancer pathogenesis and the scientific necessity of in-depth insights into the relationship of m⁶A regulators and lung cancers.

Working hypotheses

We hypothesize that the m⁶A regulator profiling is significantly associated with pathological characteristics of lung cancer, and the optimized m⁶A regulator signature has benefits for patient stratification, prognostic prediction, and personalized management of lung cancer.

Study design

The mRNA expressions of 19 m⁶A regulators described above and the complete clinical data were obtained from 1013 human lung cancer tissues (511 LUADs and 502 LUSCs; these tissues were obtained from thoracic surgery of white American or African-American patients with lung cancer) and 109 normal control lung tissues (these tissues were postmortem tissues from donors who were healthy white American or African-American before death) from the TCGA database, and 19 m⁶A regulators were verified with qRT-PCR in 2 LUAD cell lines and 2 LUSC cell lines relative to the normal human bronchial epithelial cell line (HBE) cells. Protein–protein interaction (PPI), gene co-expression, survival analysis, and headmap were used to analyze the relationship of these m⁶A regulators and lung cancer clinical data. Lasso regression was used to optimize pathology-specific m⁶A regulator gene signature. Gene set enrichment analysis (GSEA) was used to reveal the functional characteristics of m⁶A regulators.

Expected impacts in the context of predictive, preventive, and personalized medicine

We expect the lasso regression-optimized pathology-specific m⁶A regulator gene signature will be helpful for predictive diagnosis or disease prognosis, stratifying patients, and guiding clinical therapy for personalized treatment of lung cancers in the context of predictive, preventive, and personalized medicine (PPPM) [31].

Materials and methods

TCGA data of lung cancer patients and preprocessing

TCGA (http://cancergenome.nih.gov/) was a data sharing platform for large-scale cancer patient information, including RNA expression, protein expression, mutation, RNA splicing, and corresponding clinical data. The mRNA expression data of those 19 m⁶A regulators and clinical data were obtained from 1013 lung cancer patients (LUAD: n = 511; and LUSC: n = 502) and control patients (n = 109) from lung cancer level 3 RNA-seq V2 transcriptomics data in the TCGA database. Those 19 m⁶A regulators included 7 writers (ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, and METTL14), 2 erasers (FTO and ALKBH5), and 10 m⁶A-binding proteins (HNRNPC, YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, IGF2BP1, IGF2BP2, IGF2BP3, and HNRNPA2B1). Reversible m⁶A modification on mRNA was shown (Fig. 1). The clinical data included primary disease (LUAD and LUSC), age (from 33 to 90 years), gender (male and female), tobacco smoking history (current reformed smoker with ≤ 15 years’ smoking, current reformed smoker with > 15 years’ smoking, current reformed smoker, and lifelong nonsmoker), pathologic T (T represented tumor size, including T1, T2, T3, T4, and TX), pathologic M (M represented tumor metastasis, including M0, M1, and MX), pathologic N (N represented tumor lymph node metastasis, including N0, N1, N2, and NX), pathologic stage (stages I, II, III, and IV), cancer status (tumor and tumor-free), tumor residual (R0, R1, R2, and RX), and status (alive and dead). PPI was performed with the String network (http://string-db.org/cgi/input.pl). The co-expressions among 19 m⁶A regulators were analyzed by Spearman correlation coefficient with R package.

Fig. 1.

Reversible m⁶A modification on RNAs. Methyltransferase complex (writers) contained ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, and METTL14. Demethylases (erasers) included FTO and ALKBH5. The m⁶A-binding proteins (readers) included HNRNPC, YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, IGF2BP1, IGF2BP2, IGF2BP3, and HNRNPA2B1. ZC3H13, zinc finger CCCH-type containing 13; KIAA1429, Vir-like m⁶A methyltransferase associated; RBM15, RNA-binding motif protein 15; RBM15B, RNA-binding motif protein 15B; WTAP, WT1-associated protein; METTL3, mRNA mA methyltransferase; METTL14, methyltransferase-like 14; FTO, FTO alpha-ketoglutarate-dependent dioxygenase; ALKBH5, AlkB homolog 5, RNA demethylase; HNRNPC, heterogeneous nuclear ribonucleoprotein C; YTHDF1, YTH N6-methyladenosine RNA-binding protein 1; YTHDF2, YTH N6-methyladenosine RNA-binding protein 2; YTHDF3, YTH N6-methyladenosine RNA-binding protein 3; YTHDC1, YTH domain containing 1; YTHDC2, YTH domain containing 2; IGF2BP1, insulin-like growth factor 2 mRNA-binding protein 1; IGF2BP2, insulin-like growth factor 2 mRNA-binding protein 2; IGF2BP3, insulin-like growth factor 2 mRNA-binding protein 3; HNRNPA2B1, heterogeneous nuclear ribonucleoprotein A2/B1

Cell lines and cell culture

Two LUAD cell lines (A549 and LTEP-a-2), two LUSC cell lines (H520 and SK-MES-1), and one normal HBE were purchased from the Cell Bank of Chinese Academy of Science (Shanghai, China). A549 was cultured in F-12K medium with 10% fetal bovine serum. LTEP-a-2 and H520 were cultured in RPMI-1640 medium with 10% fetal bovine serum. SK-MES-1 was cultured in Eagle’s minimum essential medium with 10% fetal bovine serum. HBE was cultured in DMEM medium (Corning, NY, USA) supplemented with 10% fetal bovine serum (FBS, GIBCO, South America, NY, USA). All these cells were maintained in 5% CO₂ atmosphere at 37 °C.

RNA extraction and quantitative real-time PCR verification

Total RNAs of each cell line (A549, LTEP-a-2, H520, SK-MES-1, and HBE) were extracted with TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. The cells were seeded in 6-well plates at 30–50% density. After the cell density was up to 90%, RNAs were isolated as follows: (i) medium was discarded; (ii) cells were washed with PBS (2×); (iii) a volume (1 ml) of TRIzol reagent and 200 μl chloroform were added in each well for 10 min; (iv) TRIzol reagent was collected on a tube followed by standing on ice for 15 min; (v) samples were centrifuged (15 min, 12,000g, 4 °C); (vi) supernatant was mixed with equal volume of isopropanol followed by standing on ice for 15 min; (vii) samples were centrifuged (15 min, 12,000g, 4 °C); (viii) supernatant was discarded followed by adding a volume (1 ml) of 75% ethyl alcohol; (ix) samples were centrifuged (5 min, 8000g, 4 °C); and (x) supernatant was discarded followed by adding 20 μl RNA enzyme-free water. Total RNAs were reversely transcribed into cDNAs, and then cDNAs were used to perform quantitative real-time PCR (qRT-PCR) with SYBR Premix ExTaq (TaKaRa). GAPDH was used as an internal control for gene quantification. Primer sequences were synthesized by Sangon Biotech (Shanghai, China) for 19 m⁶A regulators’ mRNAs (RBM15, KIAA1429, ZC3H13, HNRNPC, YTHDC2, METTL14, YTHDC1, WTAP, METTL3, FTO, ALKBH5, YTHDF1, YTHDF2, RBM15B, YTHDF3, HNRNPA2B1, IGF2BP1, IGF2BP2, and IGF2BP3) (Table 1) and were used for qTR-PCR analysis of those 19 m⁶A regulators in these 5 cell lines.

Table 1.

Primer sequences for qRT-PCR analysis of 19 m⁶A regulators in lung cancer cell lines (note: F = forward, R = reverse). GAPDH was used as control

Primer name	Primer sequence (from 5′ to 3′)
ALKBH5-F	GCAAGGTGAAGAGCGGCATCC
ALKBH5-R	GTCCACCGTGTGCTCGTTGTAC
FTO-F	GTTCACAACCTCGGTTTAGTTC
FTO-R	CATCATCATTGTCCACATCGTC
HNRNPA2B1-F	GCTTAAGCTTTGAAACCACAGA
HNRNPA2B1-F	GCTTAAGCTTTGAAACCACAGA
IGF2BP1-R	GGGGTGGAATATTTCGGATTTG
IGF2BP1-F	GATGAAGGCCATCGAAACTTTC
IGF2BP2-F	GATGAACAAGCTTTACATCGGG
IGF2BP2-R	GATTTTCCCATGCAATTCCACT
IGF2BP3-F	GAGGCGCTTTCAGGTAAAATAG
IGF2BP3-R	AATGAGGCGGGATATTTCGTAT
KIAA1429-F	GCAACTTCAGGCATTAAGTTCA
KIAA1429-R	GTATTGCCTTGTCGAATCTGTC
METTL14-F	CAGGCTGGCTCACAGTTGGAC
METTL14-R	TTCCACCTCTTCCTCCACCTCTG
METTL3-F	CTTCAGCAGTTCCTGAATTAGC
METTL3-R	ATGTTAAGGCCAGATCAGAGAG
RBM15-F	GGCTGCCTGAGGAGAGTGGAG
RBM15-R	CGGCTACTGCTCAATTCTGGACTG
RBM15B-F	ATCTTTCAGAGTACGCTCAGAC
RBM15B-R	CTAGGATATGCATAGACGTGGG
WTAP-F	CTGACAAACGGACCAAGTAATG
WTAP-R	AAAGTCATCTTCGGTTGTGTTG
YTHDC1-F	AGTGACTCTGGTTCTGAATCTG
YTHDC1-R	CTGGTTTGATCTTTTCGGACAG
YTHDC2-F	GAGAATTGGGCTGTCGTTAAAG
YTHDC2-R	TGAAGCAGGATGAAATCGTACT
YTHDF2-F	ACTTCTCAGCATGGGGAAATAA
YTHDF2-R	TATTCATGCCAGGAGCCTTATT
YTHDF3-F	TCAACCACCACAACCACAGCAG
YTHDF3-R	TGAAGCACTGACAGGTACAACACC
ZC3H13-F	GATCAGTTAAAGCGTGGAGAAC
ZC3H13-R	CTCTCTGTCGTGTTCATATCGA
YTHDF1-F	ATGACAATGACTTTGAGCCCTA
YTHDF1-R	AGGGAGTAAGGAAATCCAATGG
HNRNPC-F	ACAGATCCTCGCTCCATGAACTCC
HNRNPC-R	TTCTGCCATCCTCTCCTGCTACAG
GAPDH-F	CTGCACCACCAACTGCTT
GAPDH-R	TTCTGGGTGGCAGTGATG

Consensus clustering of lung cancer tissue samples

Lung cancer tissues with mRNA expression information of those 19 m⁶A regulators and corresponding clinical data were clustered with hierarchical agglomerative consensus clustering based on the method of Ward’s linkage and Euclidean distance. Unsupervised clustering methods used the proportion of ambiguous clustering (PAC) as a simple and powerful method to infer optimal K (K-means) for identifying and classifying patients for further analysis [32]. Cluster analysis was performed using the ConsensuClusterPlus R package with cycle computation for 1000 times to ensure the stability and reliability of classification [33]. The Kaplan–Meier method was used for the overall survival analysis in different clusters. Clinic correlation was performed with pheatmap R package.

The m⁶A regulator gene signature in lung cancer identified with lasso regression

Lasso regression based on 19 m⁶A regulators was used to identify the m⁶A regulator gene signature that was associated with high lung cancer risk, which was performed with the glmnet R package that fits a GLM with lasso or elasticnet regularization and a generalized linear model via penalized maximum likelihood. Lasso regression was a regression analysis method that performed both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model it produced. The best subset selection and the connections between lasso coefficient estimates can be identified to construct the prognostic model [34]. The lasso regression-identified m⁶A regulator gene signature model generated a risk score for each lung cancer tissue sample to associate with pathology-related clinical characteristics. Thus, 1013 lung cancer patients were divided into high- vs. low-risk groups based on the median value = 0.84 of all risk scores. Further, clinical characteristics associated with overall survival were analyzed in lung cancer patients using Cox regression in univariate and multivariate models, and the Kaplan–Meier method was used to evaluate the availability of the prognostic model.

GSEA of KIAA1429, METTL3, and IGF2BP1 in lung cancer tissues

GSEA was widely used to analyze the genomic or proteomic data, which linked disease phenotypes with many different functional gene sets. According to median RNA expression value of each m⁶A regulator (KIAA1429, METTL3, and IGF2BP1), 1013 lung cancer tissue samples were divided into the corresponding high vs. low expression groups based on the expression median value of KIAA1429 (median value = 10.36), METTL3 (median value = 9.18), or IGF2BP1 (median value = 3.46), respectively, followed by GSEA based on these two high vs. low expression groups of each m⁶A regulator. GSEA was also conducted based on those two high- vs. low-risk groups that were derived from the lasso regression model (median value = 0.84 of risk scores), which was actually the combination effects of the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and IGF2BP1).

Statistical analysis

Spearman correlation coefficient was calculated for the molecule pair between m⁶A regulator genes. Student’s t test in SPSS 13.0 (SPSS Inc., Chicago, USA) was used to assess the expression differences of each m6A regulator in LUAD and LUSC tissues compared to normal tissues and in different lung cancer cells compared to the control HBE cell. Each experiment was repeated at least three times. The relationship between clusters and different risk score groups was studied with chi-square test. In all cases, p < 0.05 was considered as statistically significant. Benjamini–Hochberg for multiple testing and false discovery rate (FDR) were calculated to correct the p value for GSEA.

Results

The mRNA expression profile of m⁶A regulators in lung cancer tissues and cell lines

The mRNA expressions of 19 m⁶A regulators (Supplementary Table 1), namely 7 methyltransferases—writers (ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, and METTL14), 2 demethylases—erasers (FTO and ALKBH5), and 10 m⁶A-binding proteins—readers (HNRNPC, YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, IGF2BP1, IGF2BP2, IGF2BP3, and HNRNPA2B1), were used to construct the expression heatmap between 1013 lung cancer and 109 control tissues (Fig. 2a). KIAA1429, RBM15, HNRNPC, YTHDF1, HNRNPA2B1, IGF2BP1, IGF2BP2, and IGF2BP3 were significantly increased in lung cancer compared to control tissues (Fig. 2b). ZC3H13, METTL3, METTL14, and FTO were significantly decreased in lung cancer compared to control tissues (Fig. 2b). Moreover, qRT-PCR analysis also found that the mRNA expression profile of 19 m⁶A regulators was changed in human LUAD cells (A549 and LTEP-a-2) and LUSC cells (H520 and SK-MES-1) compared to the control HBE cell (Fig. 3). These results clearly demonstrated the dysregulated gene expression profile of m⁶A regulators between human lung cancer and control tissues and between human lung cancer and control cells.

Fig. 2.

The mRNA expression profile of 19 m⁶A regulators in 1013 lung cancer tissues (T) and 109 controls (N). a The heatmap of mRNA expressions of 19 m⁶A regulators (ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, METTL14, FTO, ALKBH5, HNRNPC, YTHDF1, YTHDF2, YTHDF3, YTHDC1, YTHDC2, IGF2BP1, IGF2BP2, IGF2BP3, and HNRNPA2B1) in lung cancer tissues based on the TCGA database. T = lung cancer (red). N = normal controls (green). The color from green to red bar on the right means the level of gene expression, gradually changed green bar means downregulation gene, and gradually changed red bar means upregulation of gene. b The mRNA expression levels of each m⁶A regulator among normal (n = 109), lung squamous cell carcinoma (n = 502), and lung adenocarcinoma (n = 511) tissues. *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 3.

qRT-PCR analysis of 19 m⁶A regulators in human lung adenocarcinoma cells (A549 and LTEP-a-2) and lung squamous cell carcinoma cells (H520 and SK-MES-1) compared to control HBE cells. NS = no significance. *p < 0.05; **p < 0.01; ***p < 0.001

Further, PPI and correlation analysis revealed the relationships among those 19 m⁶A regulators (Fig. 4). Some highly interacting m⁶A regulator protein pairs (combined score ≥ 0.99) were identified, including IGF2BP1 and IGF2BP3, WTAP and KIAA1429, HNRNPC and HNRNPA2B1, WTAP and ZC3H13, METTL14 and METTL3, KIAA1429 and ZC3H13, METTL14 and WTAP, WTAP and METTL3, METTL14 and KIAA1429, and METTL3 and KIAA1429 (Supplementary Table 2; Fig. 4a). Some highly correlated m⁶A regulator gene pairs (|correlation coefficient| ≥ 0.4, p < 0.05) were identified, including ZC3H13 and YTHDC1, METTL14 and ZC3H13, KIAA1429 and YTHDF3, HNRNPA2B1 and YTHDC1, IGF2BP1 and IGF2BP2, IGF2BP1 and IGF2BP3, and IGF2BP2 and IGF2BP3 (Fig. 4b).

Fig. 4.

The relationships of 19 m⁶A regulators in lung cancers. a Protein–protein interaction (PPI) network of 19 m⁶A regulator proteins. b Co-expressions among 19 m⁶A regulator genes. X = no statistical significance

Consensus clustering of lung cancer tissue samples based on the mRNA expressions of 19 m⁶A regulators

To determine the optimal cluster number, clustering stability was assessed with the ConsensusClusterPlus package (Supplementary Fig. 1; Fig. 5a), which supported the existence of two robust lung cancer sample clusters (Fig. 5a). When the consensus matrix k value was equal to 2, there was no crossover between lung cancer samples; thus, 1013 lung cancer tissue samples were grouped into two clusters (Supplementary Table 3; Fig. 5a). Correlations between clinical feature and m⁶A regulators were tested with chi-square test based on those two robust lung cancer sample clusters (Supplementary Table 3; Fig. 5a). The sample clusters were significantly related to those clinical characteristics, including primary disease (LUAD and LUSC) (p < 0.001), gender (male and female) (p < 0.001), tobacco smoking history (current reformed smoker with ≤ 15 years’ smoking, current reformed smoker with > 15 years’ smoking, current reformed smoker, and lifelong nonsmoker) (p < 0.001), pathologic stage (stages I, II, III, and IV) (p < 0.001), and status (alive and dead) (p < 0.05) (Fig. 5b). However, the overall survival analysis showed no statistical significance between clusters 1 and 2 (Fig. 5c). Therefore, it was necessary to select other optimization methods.

Fig. 5.

Consensus clustering of lung cancer tissue samples based on the mRNA expressions of 19 m⁶A regulators according to the optimized consensus matrix k = 2. a The optimal cluster number was two using the ConsensusClusterPlus package. b The heatmap of clusters and lung cancer-related clinical characteristics, including primary disease, age, gender, tobacco smoking history, pathologic T (T represented tumor size, including T1, T2, T3, T4, and TX), pathologic M (M represented tumor metastasis, including M0, M1, and MX), pathologic N (N represented tumor lymph node metastasis, including N0, N1, N2, and NX), pathologic stage, cancer status, tumor residual, and status. c Overall survival analysis between cluster 1 and cluster 2. *p < 0.05; **p < 0.01; ***p < 0.001

Lasso regression identified the prognostic model of the three-m⁶A-regulator gene signature

To determine the optimal prognostic model, lasso regression was performed with the glmnet R package. Lasso regression was a generalized linear model, and the adjustment degree of lasso regression complexity was controlled by lambda. The optimal prognostic model of the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and IGF2BP1) was identified when log(lambda) was between − 3 and − 4 (Fig. 6 a and b). The lasso regression coefficient of the three-m⁶A-regulator gene signature was 0.170 for KIAA1429, − 0.160 for METTL3, and 0.036 for IGF2BP1. The univariate Cox regression analysis also revealed that KIAA1429 (hazard ratio [HR] 1.33; 95% confidence interval [CI] 1.05–1.68; p = 0.018), METTL3 (HR 0.77; 95% CI 0.64–0.92; p = 0.005), and IGF2BP1 (HR 1.34; 95% CI 1.07–1.67; p = 0.011) were correlated significantly with a poor overall survival (Table 2). It clearly demonstrated that the results of lasso regression were consistent with the results of the univariate Cox regression. The univariate Cox regression analysis confirmed the three-m⁶A-regulator gene signature model. The relationship between the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and IGF2BP1) and lung cancer risk was calculated; thus, each lung cancer sample obtained a risk score, and finally, lung cancer samples were grouped into high-risk vs. low-risk groups according to the median value = 0.84 of all risk scores (Supplementary Table 4). The overall survival analysis showed statistical significance between high-risk and low-risk groups (p = 1.723E-04) (Fig. 6c). Further, the sample risk groups were significantly related to those clinical characteristics, including primary disease (LUAD and LUSC) (p < 0.001), gender (male and female) (p < 0.001), tobacco smoking history (current reformed smoker with ≤ 15 years’ smoking, current reformed smoker with > 15 years’ smoking, current reformed smoker, and lifelong nonsmoker) (p < 0.001), pathologic stage (stages I, II, III, and IV) (p < 0.001), cancer status (tumor and tumor free) (p < 0.01), and status (alive and dead) (p < 0.05) (Fig. 7).

Fig. 6.

Lasso regression analysis of 19 m⁶A regulators to identify the prognostic model of three-m⁶A-regulator gene signature (2 = KIAA1429, 9 = METTL3, and 17 = IGF2BP1) between log(lambda) value from − 4 to − 3. a, b Lasso regression complexity was controlled by lambda using the glmnet R package (1 = RBM15; 2 = KIAA1429; 3 = ZC3H13; 4 = HNRNPC; 5 = YTHDC2; 6 = METTL14; 7 = YTHDC1; 8 = WTAP; 9 = METTL3; 10 = FTO; 11 = ALKBH5; 12 = YTHDF1; 13 = YTHDF2; 14 = RBM15B; 15 = YTHDF3; 16 = HNRNPA2B1; 17 = IGF2BP1; 18 = IGF2BP2; 19 = IGF2BP3). c Overall survival analysis between the high-risk score and low-risk score group derived from lasso regression

Table 2.

Association between each m⁶A regulator and overall survival in TCGA lung cancer patients analyzed with Cox regression

m6A regulator	HR	95% confidence intervals (low)	95% confidence intervals (high)	p value
RBM15	1.10	0.87	1.39	0.434
KIAA1429	1.33	1.05	1.68	0.018*
ZC3H13	1.12	0.91	1.36	0.283
HNRNPC	1.08	0.84	1.41	0.542
YTHDC2	0.90	0.73	1.10	0.308
METTL14	1.08	0.84	1.38	0.565
YTHDC1	0.86	0.64	1.17	0.337
WTAP	1.09	0.87	1.35	0.456
METTL3	0.77	0.64	0.92	0.005**
FTO	1.07	0.87	1.31	0.523
ALKBH5	1.01	0.79	1.29	0.943
YTHDF1	0.92	0.71	1.19	0.519
YTHDF2	0.90	0.67	1.20	0.463
RBM15B	1.04	0.80	1.35	0.758
YTHDF3	1.13	0.91	1.39	0.265
HNRNPA2B1	1.04	0.79	1.37	0.769
IGF2BP1	1.34	1.07	1.67	0.011*
IGF2BP2	1.03	0.97	1.09	0.330
IGF2BP3	1.03	0.98	1.08	0.199

HR = hazard ratio

*p < 0.05; **p < 0.01

Fig. 7.

The heatmap analysis of relationships between clinical characteristics and the model of the three-m⁶A-regulator gene signature in 1013 lung cancer tissue samples. The risk score of each lung cancer sample was derived from lasso regression risk scores of the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and 17 = IGF2BP1). These clinical characteristics included primary disease, age, gender, tobacco smoking history, pathologic T (T represented tumor size, including T1, T2, T3, T4, and TX), pathologic M (M represented tumor metastasis, including M0, M1, and MX), pathologic N (N represented tumor lymph node metastasis, including N0, N1, N2, and NX), pathologic stage, cancer status, tumor residual, and status. *p < 0.05; **p < 0.01; ***p < 0.001

Moreover, the lasso regression risk score of each lung cancer sample based on the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and IGF2BP1) (Supplementary Table 4) was used as a risk factor in combination with clinical features to perform univariate and multivariate Cox regression analyses. The univariate Cox regression analysis revealed that pathologic N (HR 1.350; 95% CI 1.204–1.514; p = 3.00E-07), pathologic T (HR 1.432; 95% CI 1.249–1.643; p = 2.00E-07), pathologic stage (HR 1.469; 95% CI 1.309–1.649; p = 5.00E-11), cancer status (HR 4.096; 95% CI 3.152–5.321; p = 4.72E-26), residual tumor (HR 1.276; 95% CI 1.088–1.497; p = 2.70E-04), and risk score (HR 6.374; 95% CI 2.671–15.211; p = 3.00E-05) were correlated significantly with a poor OS (Fig. 8a). The multivariate Cox regression analysis revealed that pathologic T pathologic stage (HR 1.262; 95% CI 1.015–1.569; p = 3.60E-02), cancer status (HR 3.884; 95% CI 2.875–5.247; p = 9.53E-19), and risk score (HR 5.082; 95% CI 1.460–17.685; p = 1.10E-02) were significantly correlated with a poor overall survival (Fig. 8b). It clearly demonstrated that the risk score factor based on the optimized prognostic model of the three-m⁶A-regulator gene signature was significant with both univariate and multivariate Cox regression analyses.

Fig. 8.

The univariate and multivariate Cox regression analyses of risk factors in lung cancer. a The univariate analysis of risk factors in lung cancer. b The multivariate analysis of risk factors in lung cancer

GSEA revealed pathways and potential target genes

According to the GSEA enrichment results of KIAA1429, METTL3, and IGF2BP1 (Supplementary Table 5), gene sets were significantly enriched in drug resistance (antibiotics), proliferation, RNA metabolism, metastasis, energy metabolism (gluconeogenesis), cell cycling, apoptosis, recurrence, DNA replication, inflammatory mediators pathway (IL6 and prostaglandin E2), and epithelial–mesenchymal transition (EMT). It clearly showed that m⁶A regulator genes affected tumor biology behaviors by several biological mechanisms. Further, the potential target genes in GSEA results of two m⁶A regulators IGF2BP1 (Fig. 9a) and KIAA1429 (Fig. 9b) included E2F3, WTAP, CCND1, CDK4, EGR2, YBX1, IL6, and TLX, which indicated the potential mechanism of m⁶A regulators and that m⁶A widely regulated mRNAs and ncRNAs [20].

Fig. 9.

Gene set enrichment analysis (GSEA) results based on mRNA expressions (high expression vs. low expression) of each m⁶A regulator (IGF2BP1 and KIAA1429). a The GSEA results of IGF2BP1. b The GSEA results of KIAA1429

GSEA was also carried out in different sample risk score groups derived from the lasso regression prognostic model of the three-m⁶A-regulator gene signature (KIAA1429, METTL3, and IGF2BP1) according to the median value = 0.84 of all risk scores. The GSEA results showed that the high-risk score group was significantly related to the multiple important biological processes, including ubiquitin-mediated proteolysis, hypoxia via ELK3, adrenocortical tumor markers, liver cancer survival, metastasis, stemness, and TGFB1 targets (Supplementary Table 6 and Fig. 10), which were closely associated with tumorigenesis and development.

Fig. 10.

Gene set enrichment analysis (GSEA) results based on risk score (high-risk score vs. low-risk score) derived from the lasso regression model. a Ubiquitin-mediated proteolysis pathway enriched by GSEA. b Hypoxia via ELK3 pathway enriched by GSEA. c Liver cancer survival enriched by GSEA. d Metastasis pathway enriched by GSEA. e Stemness pathway enriched by GSEA. f TGFB1 targets pathway enriched by GSEA

Discussion

This study analyzed the mRNA expression profile of 19 m⁶A regulators in 1013 lung cancer tissues relative to 109 normal tissues and in 4 lung cancer cell lines relative to normal control cell line and further analyzed the associations of these m⁶A regulators and clinical characteristics of lung cancers. An optimized lasso regression prognostic model of the three-m⁶A-regulator gene signature (IGF2BP1, METTL3, and KIAA1429) was achieved, which was significantly associated with important clinical characteristics of lung cancer and with the poor overall survival of lung cancer. It was a potential biomarker pattern for prognostic prediction, patient stratification, and personalized treatment of lung cancer toward its PPPM practice.

Roles of m⁶A RNA modification and its regulators in multiple cancers

The m⁶A modification is the most meaningful and pervasive modification of human RNA molecules. Further studies on the biological functions of m⁶A RNA modifications have not been paid enough attention until around 2012, when major progress has been made in the transcriptome profiling of m⁶A RNA modifications through antibody-based immunoprecipitation and high-throughput sequencing [35]. The m⁶A antibody-based immunoprecipitation has not been reported in lung cancer. The m⁶A RNA modification is a reversible process through regulation of m⁶A regulators, including methyltransferases (writers), demethylases (erasers), and m⁶A-binding proteins (readers) (Fig. 1). The m⁶A is installed by the methyltransferase complex (ZC3H13, KIAA1429, RBM15, RBM15B, WTAP, METTL3, and METTL14). The identity of demethylases (FTO and ALKBH5) provides reliable evidence that m⁶A RNA modification is a reversible posttranscriptional modification [36]. m⁶A RNA modification influences the properties of RNA transcripts, including secondary structure, base-pairing, protein–RNA interactions, and charge, which, in turn, affects gene expression through altering RNA localization, translation, processing, decay, and eventually metabolism. The altered expression levels of m⁶A regulators substantially affect phenotypes in a cell and organism [37]. Emerging data have demonstrated the important roles of m⁶A regulators (writers, erasers, and readers) in human cancers [38], including acute myeloid leukemia (METTL3, METTL14, FTO, and YTHDF2), breast cancer (METTL13, METTL14, FTO, and ALKBH5), nonsmall cell lung cancer (METTL3, FTO, ALKBH5, YTHDF1, and YTHDF2), gastric cancer (METTL3, FTO, ALKBH5, and IGF2BP3), bladder cancer (METTL3, METTL14, and ALKBH5), prostate cancer (METTL3 and YTHDF2), osteosarcoma (METTL3), glioblastoma (METTL3, METTL14, FTO, and ALKBH5), renal cell carcinoma (METTL3 and METTL14), hepatocellular carcinoma (METTL3, METTL14, WTAP, VIRMA, YTHDF1, YTHDF2, and IGF2BP1/2/3), pancreatic cancer (METTL3, FTO, ALKBH5, YTHDF2, and IGF2BP2), colorectal cancer (METTL3, METTL14, YTHDF1, YTHDF3, and IGF2BP2), gynecological cancer (FTO, YTHDF2, IGFBP1/2/3, METTL3, METTL14, and ALKBH5), and melanoma (METTL3 and FTO) [25, 39]. It clearly demonstrated the important scientific merits of m⁶A regulators in different types of cancers. The identification of pathology-specific m⁶A regulator pattern might directly serve cancer management in the context of predicting prognosis, stratifying patients, and personalizing medical services.

Role of m⁶A RNA modification and its regulators in lung cancer

Recent studies have found that m⁶A RNA modification plays important roles in human cancer biology [40]. EMT was the initiation of metastasis in cancer progression. In the EMT process, epithelial cells tend to lose cell–cell adhesion and cell polarity, and gain capacities of migration and invasion to become mesenchymal stem cells. The previous study highlighted the critical roles of m⁶A RNA modification in regulation of EMT in cancer cells [41]. m⁶A sequencing in combination with functional analysis confirmed that Snail was modified by m⁶A in coding the sequence region, which triggered polysome-mediated translation of m⁶A-regulated EMT by Snail in cancer cells [41]. It is well-known that viruses have close relationships with some kinds of cancers, such as the Epstein–Barr virus (EBV), human papilloma virus (HPV), hepatitis virus B (HBV), and human T cell leukemia/lymphoma virus I (HTLV-1) [42]. Due to the expanded knowledge of m⁶A, m⁶A was identified in transcripts encoded by many kinds of viruses, which indicated widespread control over life cycles of viruses [42]. It innovated our comprehension and cognition of the regulatory mechanisms about viral replication [43]. In this study, GSEA showed that the alerted m⁶A regulators in lung cancers were related to the activations of various biological behaviors, including drug resistance, proliferation, metastasis, energy metabolism, DNA replication, inflammatory mediators pathway, RNA metabolism, cell cycling, apoptosis, recurrence, and EMT. These findings were consistent with many previous studies [20, 34]. With the advancement of tumor immunology and molecular biotechnology, immunotherapy became an effective clinical therapy in the combination therapy of tumor. The relevant roles of m⁶A RNA methylation in immunoregulation have received great attention. Anti-PD-1/PD-L immune checkpoint therapy was the most popular targets for new anticancer drugs. The m⁶A demethylase FTO has been proved to promote tumorigenesis and has function of anti-PD-1 resistance. This result indicated that FTO inhibition in combination with anti-PD-1 blockade might be the possible solution to immunotherapy resistance in cancer [44]. Energy metabolism alteration of tumor cells has been proved to be an emerging sign of tumorigenesis since it was closely related to some malignant behaviors of cancer cells. Alternative splicing of energy metabolism-related enzymes may modulate enzyme activity to impact energy reprogramming, for example, the ratio between M1 and M2 isoforms of pyruvate kinase in aerobic glycolysis and splicing programs at different levels of 2-oxoglutarate-dependent oxidases. The effect of m⁶A on the splicing profile of target RNAs has been identified; for example, the m⁶A reader YTHDC1 affects splicing isoforms by recruiting splicing factors of the SR family, and the m⁶A erasers FTO and ALKBH5 were directly involved in RNA splicing [45]. One study also found that m⁶A regulators (METTL3, FTO, ALKBH5, YTHDF1, and YTHDF2) significantly contributed to the pathogenesis of nonsmall cell lung cancer [25]. These findings clearly demonstrated that m⁶A regulators play important roles in lung cancer pathogenesis.

The three-m⁶A-regulator gene signature as biomarker in the context of PPPM in lung cancer

In this study, lasso regression identified the prognostic model of the three-m⁶A-regulator gene signature (IGF2BP1, METTL3, and KIAA1429) in 1013 human lung cancer tissues. First, many studies showed that IGF2BP1 enhanced tumor progression and metastasis in LUADs and that the increased IGF2BP1 expression was correlated with lower overall survival in lung carcinoma [46]. IGF2BP1 promoted the expression of the transcriptional regulator SRF in a conserved and m⁶A-dependent manner through competing with the miRNA-dependent manner of SRF mRNA degradation [20]. IGF2BP1 that regulates various target genes in the m⁶A-dependent manner could be a conserved oncogenic driver network in lung cancer. Second, the METTL3 is one of the key components in the m⁶A-methyltransferase complex responsible for m⁶A RNA modification. The endogenous METTL3 change and subsequent alteration of related gene expression profiles might directly influence the growth of human nonsmall cell lung carcinoma cell line (H1299) in vitro and in vivo [47]. Third, the m⁶A writer KIAA1429 was never reported in lung cancer, but studies in other kinds of cancers were consistent with our findings. KIAA1429 was highly expressed in breast cancer tissues compared to noncancerous breast tissues. The high expression of KIAA1429 was a predicting factor of poor prognosis and was associated with cancer proliferation and metastasis by regulating CDK1 in an m⁶A-independent manner [27]. The activities and molecular mechanisms of m⁶A-methyltransferases–writers were an important field for lung cancer research. Here we have to note that it does not make sense to take a single molecule as biomarker for PPPM practice in cancer because cancer is a very complex and multifaceted disease [48, 49]. The significant hub-molecule signature model was constructed to assess cancer survival risks, which benefited for patient stratification, prognostic prediction, and personalized treatment of cancer [50]. In our study, the high-risk vs. low-risk groups based on the lasso regression prognostic model of the three-m⁶A-regulator gene signature (IGF2BP1, METTL3, and KIAA1429) were significantly related to overall survival and clinicopathologic characteristics, which might be an effective biomarker pattern for predictive diagnostics or prognostic assessment, patient stratification, and personalized treatment of lung cancers. However, the sensitivity, specificity, and accuracy of this optimized prognostic model must be investigated further in the large-scale clinical cohort study toward PPPM practice in lung cancers.

The potentially targeted genes of m⁶A regulators in lung cancer

The m⁶A RNA modification might influence multiple cellular signaling pathways and lots of RNA (mRNAs and ncRNAs). In this study, GSEA identified a few tumor-related gene sets of proliferative signaling, metastasis, apoptosis by CDKN1A via TP53, and mitochondrial transcription activity in lung cancer. Meanwhile, many studies showed that m⁶A-related gene knockdown inhibited total RNA m⁶A methylation level, as well as capacity of cancer cell proliferation and migration [51]. BNIP3, a pro-apoptosis gene, was identified as a downstream target of eraser FTO-mediated m⁶A RNA modification by m⁶A RNA immunoprecipitation sequencing [52]. It clearly showed that m⁶A RNA modification had functional significance in the apoptosis process. Even though the relationship of m⁶A and mitochondrial transcription in mammal has not been revealed, comprehensive characterization of m⁶A modification patterns has been studied in the Arabidopsis mitochondrial transcriptomes, which showed that over 86% of the transcripts were methylated by m⁶A in the mitochondria [53]. The m⁶A modification patterns in plant mitochondrial transcription were involved in mitochondrial transcription activity in an m⁶A-dependent manner. Moreover, in this study, GSEA identified a few tumor-related gene targets, including E2F3, WTAP, CCND1, CDK4, EGR2, YBX1, IL6, and TLX, which have been reported to be associated with lung cancer. For example, CCND1 and CDK4 promoted tumorigenicity of lung cancer [54, 55]. The oncogene YBX1 also proved to affect migration, invasion, and vasculogenic mimicry of lung cancer cells [56]. However, it is not clear whether E2F3, WTAP, CCND1, CDK4, EGR2, YBX1, IL6, and TLX were in the m⁶A-dependent manner to influence the malignant behavior of lung cancers.

Strength and limitations

m⁶A RNA methylation that extensively occurs at different types of mRNAs and ncRNAs is a new frontier research hotspot, which is closely associated with cancer pathogenesis and has extensive therapeutic implications in cancer [12–15, 25, 38, 39]. m⁶A RNA methylation is tightly controlled by the m⁶A regulator system that includes 7 methyltransferases—writers, 2 demethylases—erasers, and 10 m⁶A-binding proteins—readers (Fig. 1). The investigation of the relationships between the m⁶A regulator system and clinically pathological characteristics will directly contribute to clinical research and practice toward PPPM in cancer. The present study was the first link of the m⁶A regulator system and clinically pathological characteristics in lung cancer and identified the three-m⁶A-regulator signature (KIAA1429, METTL3, and IGF2BP1) as an independent prognostic model to classify lung cancers into high- and low-risk groups for patient stratification, prognostic assessment, and personalized treatment toward PPPM in lung cancers. Also, the mRNA expressions of 19 m⁶A regulators were analyzed in a large set of human tissue samples with the corresponding clinical data, including 1013 human lung cancer tissues (511 LUADs and 502 LUSCs) and 109 normal control lung tissues from the TCGA database, and in two LUAD cell lines and two LUSC cell lines relative to the normal HBE cells by qTR-PCR, which assured the reliability of the results in this study. Here, we would like to mention that not every tissue sample had the complete clinical data (Supplementary Table 3), which might be a limitation in this aspect, but from a statistical point of view, it did not play a significant role in this study. Also, these tissue samples in this study were from white and African Americans; thus, whether the findings in this study can be exactly applied in other races, it might need to further study the difference among different races. The status “alive/dead” of lung cancer tissue samples was the follow-up data of patient after thoracic surgery, which might not affect the overall outcome of our findings in this study. Furthermore, one must realize that, in order to deeply clarify the molecular mechanisms and biological roles of the m⁶A regulator system in lung cancer, (i) the protein expression profile of 19 m⁶A regulators should be further investigated in lung cancer tissues and cell models, (ii) the targeting RNA molecules and modification sites of m⁶A regulators should be investigated with m⁶A RNA sequencing and m⁶A antibody-based RNA immunoprecipitation in lung cancer, and (iii) the m⁶A RNA modification data may integrate the large-scale clinical characteristics and other multiomics data to achieve more meaningful findings for lung cancer.

Conclusions and expert recommendations

m⁶A RNA modification extensively occurs at mRNAs and ncRNAs to regulate transcription and affects the functions of RNAs, which is closely associated with cancer pathogenesis [12–15, 38, 39] and has extensive therapeutic implications in cancer [25]. m⁶A RNA modification is tightly regulated by m⁶A regulators. The identification of a pathology-specific m⁶A regulator pattern might directly serve cancer management in the context of predicting diagnosis, stratifying patients, and personalizing medical services. This study, for the first time, investigated the dysregulated mRNA expression profile of these m⁶A regulators in lung cancers and analyzed the relationship of the dysregulated mRNA expression profile of m⁶A regulators and clinical characteristics in human lung cancers. GSEA showed that the alerted m⁶A regulators in lung cancers were related to the activations of various biological behaviors, including DNA replication, RNA metabolism, EMT, cell cycling, proliferation, apoptosis, energy metabolism, inflammatory mediators pathway, drug resistance, metastasis, and recurrence. Lasso regression optimized and identified the prognostic model of the three-m⁶A-regulator gene signature (IGF2BP1, METTL3, and KIAA1429), which was significantly associated with clinical characteristics (primary disease, gender, tobacco smoking history, pathologic stage, cancer status, and alive/dead status) and poor overall survival in lung cancer. The lasso regression risk score factor based on this three-m⁶A-regulator gene signature was a comprehensive independent prognostic factor for prognostic assessment/prediction to stratify lung cancer patients into high- vs. low-risk score groups for personalized management in the context of PPPM practice. We conclude that pathology-specific regulators of m⁶A RNA modification are effective biomarkers and have important scientific merits in lung cancers.

We strongly recommend to emphasize and popularize the scientific importance of, and strengthen its studies of, the m⁶A regulator system (writers, erasers, and readers) in combination with clinical characteristics in lung cancer, and transit m⁶A regulator gene signature into clinical practice in the context of PPPM in lung cancer. We propose, for the further PPPM development and practical application of the altered m⁶A regulator pattern in lung cancer, the following three aspects:

• (i)Potential utility of the m⁶A regulator gene signature for predictive diagnosis: Many studies have demonstrated that the m⁶A regulator system plays important roles in cancer pathogenesis [38, 39]. It is a promising approach to integrate the abnormal expression profile of the m⁶A regulator system (writers, erasers, and readers) and clinical characteristics, for the discovery of a pathology-specific m⁶A regulator gene signature to optimize lung cancer management in the context of PPPM practice. This study identified a three-m⁶A-regulator gene signature (IGF2BP1, METTL3, and KIAA1429), which was significantly associated with clinical characteristics and poor overall survival. It is a potential biomarker pattern for predicting prognosis, stratifying patient, and personalizing treatment of lung cancer [57].
• (ii)Potential utility of the m⁶A regulator system for targeted prevention: The m⁶A regulator system (writers, erasers, and readers) tightly controls the m⁶A RNA modification to affect the functions and structures of mRNAs and ncRNAs, and is extensively associated with cancer pathogenesis [25]. The abnormal m⁶A regulator system (writers, erasers, and readers) is the potential therapeutic targets [25] for targeted drug therapy or targeted drug prevention of lung cancer. A study found that abnormal m⁶A regulators might promote or inhibit cancer [58]. The changed m⁶A regulators might be the novel therapeutic targets for lung cancer [25]. Thus, it would be an effective therapeutic strategy to restore the abnormal m⁶A regulators for targeted prevention or personalized therapy of lung cancers.
• (iii)Potential utility of the m⁶A regulator system for personalized treatment algorithms: m⁶A methylation extensively occurred in mRNAs and ncRNAs to affect the activities of transcripts and closely associate with cancer pathogenesis. The m⁶A regulator system (writers, erasers, and readers) is the central molecular machine to control the occurrence of m⁶A methylation at different mRNAs and ncRNAs. The development of different algorithms of the m⁶A regulator pattern in combination with clinical pathology characteristics will directly contribute to patient stratification for personalized treatment in lung cancer. Furthermore, m⁶A RNA sequencing and m⁶A antibody-based RNA immunoprecipitation are the effective methods to discover m⁶A RNA modification in lung cancer, and the m⁶A RNA modification data may integrate the large-scale clinical characteristics and other multiomics data of lung cancer [48–50], from a systematic point of view toward PPPM practice.

Abbreviations

ALKBH5

AlkB homolog 5, RNA demethylase

CCND1

cyclin D1 [Homo sapiens]

CDK4

cyclin-dependent kinase 4

DNA

deoxyribonucleic acid

E2F3

E2F transcription factor 3

EBV

Epstein–Barr virus

EGFR-TKI

epidermal growth factor receptor tyrosine kinase inhibitors

EGR2

early growth response 2

EMT

epithelial–mesenchymal transition

FDR

false discovery rate

FTO

FTO alpha-ketoglutarate dependent dioxygenase

GSEA

gene set enrichment analysis

HBV

hepatitis virus B

HNRNPA2B1

heterogeneous nuclear ribonucleoprotein A2/B1

HNRNPC

heterogeneous nuclear ribonucleoprotein C

HPV

human papilloma virus

HR

hazard ratio

HTLV-1

human T cell leukemia/lymphoma virus I

IGF2BP1

insulin-like growth factor 2 mRNA-binding protein 1

IGF2BP2

insulin-like growth factor 2 mRNA-binding protein 2

IGF2BP3

insulin-like growth factor 2 mRNA-binding protein 3

IL6

interleukin 6

KIAA1429

Vir-like m⁶A methyltransferase associated

lncRNAs

long noncoding RNAs

LUAD

lung adenocarcinoma, LUSC lung squamous carcinoma

m⁶A

N⁶-methyladenosine

METTL14

methyltransferase-like 14

METTL3

mRNA mA methyltransferase

mRNAs

messenger RNAs

ncRNAs

noncoding RNAs

PAC

proportion of ambiguous clustering

PD-1

programmed cell death 1

PD-L1

CD274 molecule

PPI

protein–protein interaction

qRT-PCR

quantitative real-time polymerase chain reaction

RBM15

RNA-binding motif protein 15

RBM15B

RNA-binding motif protein 15B

RNA

ribonucleic acid

rRNAs

ribosomal RNAs

snRNA

small nuclear RNA

TCGA

The Cancer Genome Atlas

TLX

nuclear receptor subfamily 2 group E member 1

tRNAs

transfer RNAs

USP7

ubiquitin-specific protease 7

WTAP

WT1-associated protein

YBX1

Y-box binding protein 1

YTHDC1

YTH domain containing 1

YTHDC2

YTH domain containing 2

YTHDF1

YTH N6-methyladenosine RNA-binding protein 1

YTHDF2

YTH N6-methyladenosine RNA-binding protein 2

YTHDF3

YTH N6-methyladenosine RNA-binding protein 3

ZC3H13

zinc finger CCCH-type containing 13

Abbreviations for all particular genes and proteins can be found at the following link:

https://www.ncbi.nlm.nih.gov/gene/

Authors’ contributions

N.L. performed bioinformatic analysis, carried out the experiments, prepared the figures and tables, and designed and wrote the manuscript. X.Z. conceived the concept, instructed experiments, supervised results, coordinated, critically revised/wrote manuscript, and was responsible for its financial supports and the corresponding works. All authors approved the final manuscript.

Funding information

The authors acknowledge the financial supports from the Shandong First Medical University Talent Introduction Funds (to X.Z.), the Hunan Provincial Hundred Talent Plan (to X.Z.), and the Central South University Graduate Student Exploration Innovative Project 2019 (Grant No. 206021701).

Compliance with ethical standards

Competing interests

The authors declare that they have no competing interests.

Ethical approval

All the patients were informed about the purposes of the study and, consequently, have signed their “consent of the patient” form. All investigations conformed to the principles outlined in the Declaration of Helsinki and were performed with permission by the responsible Medical Ethics Committee of Xiangya Hospital, Central South University, China.