Identification and validation of m⁶A RNA methylation regulators with clinical prognostic value in Papillary thyroid cancer

Abstract

Background

Papillary thyroid cancer (PTC) is a type of malignant tumor with excellent prognosis, accounting for more than 80% of thyroid cancer. Recently, numerous studies illustrated the importance of N⁶-methyladenosine (m⁶A) RNA modification to tumorigenesis, but it has never been reported in PTC.

Methods

We downloaded data from The Cancer Genome Atlas (TCGA) and analyzed RNA expression, single nucleotide polymorphisms (SNPs) and copy number variations (CNVs) of 19 m⁶A RNA methylation regulators in PTC. Then we used nonnegative matrix factorization (NMF) to cluster patients into two m⁶A subtypes and compared them in overall survival (OS) and disease-free survival (DFS). The Weighted correlation network analysis (WGCNA) and univariate Cox proportional hazard model (CoxPH) were used to select genes for the construction of a m⁶A-related signature. The accuracy and prognostic value of this signature were validated by using receiver operating characteristic (ROC) curves, K-M (Kaplan–Meier) survival analysis, univariant and multivariant analyses.

Results

CNVs and differential expression of m⁶A regulators were observed in PTC patients. Especially IGF2BP2 (Insulin-like growth factor 2 mRNA binding protein 2), which was most significantly overexpressed in tumor tissue. We chose 4 genes in the m⁶A-related module from WGCNA: IGF2BP2, STT3A, MTHFD1 and GSTM4, and used them to construct a m⁶A-related signature. The prognostic value of this signature was validated, and risk scores provided by the signature was the independent prognostic factor for PTC. A nomogram was also provided for clinical usage.

Conclusions

We performed a comprehensive evaluation of the m⁶A RNA modification landscape of PTC and explored its underlying mechanisms. Our m⁶A-related signature was of great significance in predicting the DFS of patients with PTC. And IGF2BP2 was a gene worthy for further analysis as its strong correlation with DFS and clinical phenotypes of PTC.

Background

Thyroid cancer is one of malignant tumors whose incidence are rapidly increasing in the world for both men and women. It can be classified into several subtypes: PTC, follicular thyroid cancer (FTC) and medullary thyroid cancer (MTC) [1]. PTC is the most common type of thyroid cancer, accounting for more than 80% of all cases. Generally, prognosis of patients with PTC is excellent, with 5-year-survival rate over 97% [2]. The 10-year and 15-year survival rates of papillary microcarcinoma, PTC which is smaller than 1 cm, are even over 99% [3]. In previous studies, lymph node metastasis has been proved to increase the risk of local recurrence without influencing survival in PTC [4]. Wada et al. indicated that patients without lymph node metastasis has nearly no chance of recurrence, while the recurrence rate of patients with lymph node metastasis is over 16% [5]. As a result, it is of greater significance to explore prognostic factors for DFS than OS.

Generally, DNA and histone protein are considered to be essential participants of reversible epigenetic modification which can regulate gene expression in mammal cells [6]. In recent years, reversible RNA modification, especially methylation, has been demonstrated to be another important component of gene expression regulation. The m⁶A RNA methylation, which was discovered in the 1970s, was the first example of reversible RNA methylation and wildly distributed in long non-coding RNAs and polyadenylated mRNAs [7, 8]. m⁶A has been observed within introns, internal exons, 3ʹ untranslated regions (3ʹUTRs) and stop codons, suggesting its addition can be earlier or simultaneous with RNA splicing [9, 10]. There are 3 types of m⁶A RNA methylation regulators: methyltransferases (writers), RNA binding proteins (readers), and demethylases (erasers). Writers are composed of METTL3, METTL14, METTL16, RBM15, RBM15B, WTAP and KIAA1429. Readers are comprised of YTHDC1/2, YTHDF1/2/3, IGF2BP1/2/3, HNRNPA2B1 and HNRNPC. FTO and ALKBH5 serve as erasers which perform demethylation activity [11].

Numerous studies showed that m⁶A RNA methylation played a role in the occurrence and progression of multiple malignant tumors, including hepatocellular carcinoma, colorectal carcinoma, breast cancer, glioblastoma and clear cell renal cell carcinoma [12–15]. Yongsheng li et al. have concluded the characteristics of m⁶A RNA methylation across 33 types of cancer and predicted that the mechanism of m⁶A RNA modification might be related with the activation or depression of some oncogenic pathways such as PI3K-AKT-mTOR signaling, G2M checkpoint, KRAS and P53 pathways [16]. METTL3 and IGF2BP2 have been proved to be over-expressed in colorectal carcinoma and promote the progression of cancer [17]. The RNA transcripts of SOX2 were methylated by METTL3 and bonded with IGF2BP2, resulting in regulation of SOX2 degradation. Yunhao Chen et al. demonstrated that WTAP can lead to post-transcriptional suppression of its downstream effector, ETS proto-oncogene1 (ETS1), and further contribute to the proliferation of hepatocellular carcinoma [18]. However, there has been no research which specifically explored the landscape of m⁶A RNA methylation and its relationship with DFS in PTC. In addition to expression levels of m⁶A RNA methylation regulators, SNPs and CNVs may also have prognostic value for PTC.

In this study, we performed a comprehensive evaluation of the landscape of m⁶A RNA methylation in PTC and explored its underlying mechanisms. The expression level, CNVs, SNPs and correlated clinical phenotypes of m⁶A RNA methylation regulators were analyzed to confirm the significance of m⁶A modification in PTC. By applying NMF, we divided patients from the TCGA cohort into two clusters (cluster1 and cluster 2) according to the expression of 19 m⁶A RNA methylation regulators and validated their differences in OS and DFS. In order to screen out essential genes for constructing a m⁶A-related signature, we performed WGCNA, univariant and multivariant analyses. Finally, we validated the accuracy and explored the underlying mechanism of the m⁶A-related signature by a series of analyses to illustrate its prognostic value.

Methods

Data download

The transcriptome data, somatic mutation data and clinical information of PTC patients were obtained from the TCGA database via the GDC data portal (https://portal.gdc.cancer.gov/repository). We downloaded RNA-seq (level 3, HTSeq-FPKM data) of 493 PTC patients (493 primary tumor tissue and 58 solid normal tissue) with complete clinical information from the TCGA database. For SNP, we downloaded “Masked Somatic Mutation” subtype of somatic mutation data and used the VarScan software to process it. We used a R package called “maftools” [19] to analyze and visualize the Mutation Annotation Format (MAF) of somatic variants. For CNV, the loss and gain of copy-number have been identified using segmentation analysis and GISTIC2.0 algorithm. The microarray data of papillary thyroid cancer patients was downloaded from GSE58545 (normal = 18, PTC = 27) datasets in the Gene Expression Omnibus (GEO) database. Oncomine database (http://www.oncomine.org) was used to validated mRNA levels of m⁶A regulators in PTC. Human Protein Atlas (http://www.proteinatlas.org) was used to validate expression levels of m⁶A RNA methylation regulators by immunohistochemistry [20]. A list of antibodies which were used in IHC samples were provided in Additional file 1: Table S1. We used Genotype-tissue expression (GTEx) dataset to compare expression levels of m⁶A regulators among different tissues and genders.

Non-negative matrix factorization consensus clustering

To investigate the relationship between the expression of m⁶A regulators and clinical phenotypes in PTC, we clustered PTC samples from TCGA into 2 different clusters (cluster 1 and cluster 2) using NMF. The purpose of NMF was to identify potential characteristics in gene expression profiles by resolving the original matrix into two non-negative matrices [21]. Deposition was repeatedly performed, and its result was aggregated to acquire consensus clustering of PTC samples. The most suitable number of subtypes was decided according to cophenetic, dispersion and silhouette coefficients. NMF was performed by a R package called “NMF” [22]. The number of clusters k was chosen as 2, and the number of runs was set at 200. We also used a R package called “survival” to compare the OS and DFS between cluster 1 and cluster 2.

Construction of co-expression module networks

The WGCNA was performed to establish the gene co-expression network to find trait-related modules by the R package “WGCNA” [23]. All genes and samples were filtered by good genes or good samples test. Filtered genes were used to construct a scale-free network by calculating the connection strength between genes. Scale-free R² ranging from 0 to 1 was used to determine a scale-free topology model. To minimize effects of noise and spurious associations, the adjacency matrix was transformed into Topological Overlap Matrix (TOM). And TOM-based dissimilarity was used to form modules by dynamic tree cut. Here, we set minimal module size as 50 and cut height as 0.25. We used the KOBAS database to exert Kyoto Encyclopedia of Genes (KEGG) pathway enrichment and gene ontology (GO) analysis of the m⁶A-related module in WGCNA. When the P value was less than 0.05, the enriched pathway was considered to be statistically significant.

Construction of the m⁶A-related risk signature

The patients with PTC from TCGA were randomly divided into a training set (N = 241) and a testing set (N = 240) using a R package called “caret”. For the training set, the univariate CoxPH was used to identify genes whose expression levels were statistically correlated with DFS, among all genes in the m⁶A-related module (P < 0.05). The m⁶A-related risk signature gave patients in training and testing sets risk scores based on genes weighted value which was calculated by a linear combination of Cox coefficient and gene expression:

$$\text{Risk score}=\sum _{i=0}^N\left(\text{Expi * Coei}\right).$$

And patients were classified into low-risk and high-risk group according to the median of risk scores.

Validation of the m⁶A-related signature for colon cancer

The Chi square test was performed to confirm that there was no selection bias in classification of training and testing set. The univariate and multivariate analyses were conducted for both the m⁶A-related signature and clinical factors. The Kaplan–Meier (K–M) survival curves and log-rank test were generated to evaluate the difference in DFS between high‐risk group and low‐risk group in total TCGA PTC cohort, training set and testing set. We performed ROC curves to measure the prognostic capacity of our signature using a R package called “survivalROC”. A nomogram was used to predict cancer prognosis. In the TCGA datasets, all genes in the signature were included to generate the nomogram which can investigate the chancer of 1-, 3- and 5-year DFS of patients with PTC.

Statistical analysis

We exerted Mann–Whitney U tests to compare the expression levels of m6A RNA methylation regulators in different subgroups (normal tissues/primary tumor tissue, cluster1/cluster2). Chi square test was performed to confirm the difference in CNVs between normal and tumor tissues. Chi square test was also used to analyze the difference in clinical phenotypes between cluster 1 and cluster 2, as well as high-risk group and low-risk group. The relationships between m⁶A regulators and other genes in the m⁶A-related module were analyzed by calculating the Spearman correlation coefficients.

Results

m⁶A RNA methylation regulators had different expression level in PTC and normal tissues

Using transcriptome data from TCGA database, we analyzed the mRNA levels of 19 m⁶A regulators in PTC and para-tumor normal tissues (Fig. 1a, b, Additional file 2: Table S2). Except YTHDF2, 18 of 19 m⁶A regulators were differentially expressed in PTC and normal thyroid tissues. METTL3, YTHDC1, FTO, METTL14, RBM15, YTHDF3, WTAP, HNRNPA2B1, ALKBH5, METTL16, YTHDC2, KIAA1429, IGF2BP3, RBM15B and YTHDF1 had lower expression levels in tumor tissues, while IGF2BP2, IGF2BP1 and HNRNPC had higher expression level in tumor tissues. Among all of these genes, only IGF2BP2 had significantly higher expression pattern and distinguishable protein expression in tumor tissue compared to normal tissue (Fig. 1c). We also validated these results in the GEO database (Additional file 3: Table S3) and found 14 differentially expressed m⁶A RNA methylation regulators with 10 of them was the same as the analysis of TCGA. The expression of these genes was also further validated in the Oncomine database (Additional file 4: Figure S1) and Human Protein Atlas database (Fig. 1c and Additional file 5: Figure S2). Then we explored the relationships between 19 m⁶A RNA methylation regulators by Spearman correlation analysis (Fig. 1d). The relationships between each two of them were almost positively correlated, and YTHDF3 and KIAA1429 were most relevant (Cor = 0.82). However, there were also some genes which were negatively correlated, such as IGF2BP2 and ALKBH5 (Cor = − 0.43).

Fig. 1.

The overview of m⁶A RNA methylation regulators in PTC. a A Vioplot which visualized the differentially expressed m⁶A regulators in PTC (blue represents normal tissues and red represents tumor tissues). b The mRNA levels of 19 m⁶A regulators in PTC. Red means this gene is up-regulated while green means this gene is down-regulated (*P < 0.05, **P < 0.01, ***P < 0.001). c The validation of IGF2BP2 expression level by immunohistochemical staining. The result showed that the expression level of IGF2BP2 was higher in PTC tissue than normal thyroid tissue. d Spearman correlation analysis of the 19 m⁶A regulators in PTC. The “X” represents that the correlation between this pair of genes does not have statistical significance (P > 0.05). The number in every box is the Spearman correlation coefficient between two genes

CNVs and SNPs of m⁶A RNA methylation regulators can serve as prognostic factors for PTC

Among the 505 cases, CNVs of the 13 m6A RNA methylation regulators were frequently observed (Fig. 2a). In detail, two m6A “reader” genes HNRNPA2B1 and IGF2BP3 had the highest frequency of CNV events (2.04%, 2.04%) followed by YTHDC2 (1.63%) and METTl16 (1.22%). We performed Chi square test to analyze the difference in CNVs between normal tissues and PTC tissues, and found HNRNPA2B1 (P = 0.004235), IGF2BP3 (P = 0.004235) and METTL16 (P = 0.0040631) were of statistical significance (Additional file 6: Table S4). The chromosome position of HNRNPA2B1, IGF2BP3 and METTL16 were shown in Fig. 2b. Furthermore, we evaluated the correlation between the copy number and mRNA level of 19 m⁶A regulators, and found higher copy number of 4 genes were corresponded with higher expression level: ALKBH5 (P = 0.001), METTL16 (P = 2.287e−04), WTAP (P = 0.003) and YTHDF1 (P = 0.009) (Fig. 2d–g). SNPs of YTHDC1, RBM15, METTL14, HNRNPC, HNRNPA2B1 and FTO were found merely in 7 independent samples (Fig. 2c). CNVs and SNPs may influence the expression level and biological function of m6A RNA methylation regulators, and further effected the activities of RNA modification.

Fig. 2.

CNVs and SNPs analysis of m⁶A regulators. a The percentage of patients with CNVs in 13 m⁶A regulators. b The chromosome location of genes with frequently CNVs (HNRNPA2B1, IGF2BP3 and METTL16). c The SNPs of YTHDC, RBM15, METTL14, HNRNPC, HNRNPA2B1 and FTO were observed in 7 patients. d–g The copy numbers of ALKBH5, METTL16, WTAP and YTHDF1 were correlated with gene expression levels in PTC tissues

Two m⁶A subgroups were different in clinical phenotypes and DFS

The total TCGA cohort were clustered into 2 subgroups (cluster 1: n = 352 and cluster 2: n = 141) by applying NMF (Fig. 3a, b), according to expression levels of 19 m⁶A regulators in PTC samples. To better understand the clustering result and its relationships with survival outcomes and clinical phenotypes, we compared the OS and DFS between cluster 1 and cluster 2 and found cluster 2 had better DFS than cluster 1 (P = 0.034, Fig. 3c). But there was no statistical difference between the OS of cluster 1 and cluster 2 (P = 0.056, Fig. 3d). These results demonstrated that m⁶A RNA methylation may have strong correlation with DFS of PTC patients. As shown in the heatmap, IGF2BP1, WTAP, FTO, IGF2BP3 and ALKBH5 had higher expression level in cluster 2, while IGF2BP2, RBM15B, HNRNPC were significantly down-regulated in cluster 2 (Fig. 3e, Additional file 7: Table S5). We also found these 2 clusters were different in extrathyroidal extension (P < 0.01), T (P < 0.01) and N (P < 0.001) classifications, suggesting m⁶A RNA methylation may also related to clinical phenotypes and progression of PTC.

Fig. 3.

Identification of consensus clusters by m6A regulators. a The consensus map of NMF clustering results in the total TCGA cohort. Patients were divided into cluster 1 and cluster 2 according to the expression profiles of 19 m6A regulators. b The relationship between cophenetic, dispersion, evar, residuals, rss and silhouette coefficients with respect to number of clusters. c, d The survival curve of DFS (P = 0.034) and OS (P = 0.056) in cluster 1 and cluster 2. e The correlation analysis of m6A methylation regulators and clinical phenotypes in cluster 1 and cluster 2. The result showed that these 2 clusters has statistical differences in extrathyroidal extension, T and N classifications (*P < 0.05, **P < 0.01, ***P < 0.001)

Detection of DFS-related m⁶A regulator and its correlated module by WGANA

The univariant CoxPH was performed to identify m⁶A RNA methylation regulators which were prognostic factors for OS and DFS of patients with PTC (Table 1). We found that FTO (HR = 1.57, P = 0.044), RBM15 (HR = 3.84, P = 0.012), YTHDF3 (HR = 1.29, P = 0.009) and KIAA1429 (HR = 1.76, P = 0.042) were related with the overall survival rate, while only IGF2BP2 (P = 0.0006) was related with the DFS of PTC patients. As we mentioned before, the prognosis of PTC patient is excellent, and the recurrence of tumor is one of the biggest challenges at present. As a result, it is of greater value for us to study the mechanism of genes related to DFS of PTC patient. By WGCNA, we identified 22 co-expression modules and analyzed their association with 12 clinical phenotypes, including futime, fustat, TNM classification, stage, age, gender, new-event, new-event time, extrathyroidal extension and IGF2BP2 expression (High-expression and Low-expression) (Fig. 4a–c, Additional file 8: Figure S3). Except the grey module which contained non-clustering genes, the brown module was the most correlated module of IGF2BP2 expression (r = − 0.61, P = 7e−52). The brown module was also correlated with futime (r = − 0.15, P = 0.001), T (r = − 0.19, P = 3e−05), N (r = − 0.35, P = 1e−15), stage (r = − 0.14, P = 0.002), new-event (r = − 0.11, P = 0.01), new-event time (r = − 0.11, P = 0.02), age (r = 0.11, P = 0.02) and extrathyroidal extension (r = − 0.25, P = 3e−08). The result of KEGG and GO analyses showed that the brown module was related to membrane-bounded organelle, multiple metabolic process, thyroid hormone synthesis and HIF-1signaling pathway (Fig. 4d, e), especially metabolic pathways in the central carbon metabolism, such as pyruvate metabolism, citrate cycle, propanoate metabolism and glycolysis/gluconeogenesis.

Table 1.

Univariant CoxPH analysis of OS and DFS

uniCox analysis
Gene	Survival analysis				Disease free analysis
	HR	HR.95L	HR.95H	P-value	HR	HR.95L	HR.95H	P-value
METTL3	0.973958	0.742174	1.278129	0.849081	0.981439	0.836601	1.151353	0.818114
YTHDC1	1.036531	0.852548	1.26022	0.718934	0.977604	0.873975	1.093522	0.691972
FTO	1.573325	1.012161	2.44561	0.044044	0.78099	0.577311	1.056529	0.108867
METTL14	1.25211	0.688421	2.277356	0.461329	0.977691	0.685448	1.394532	0.900902
RBM15	3.843867	1.338806	11.03619	0.012343	0.858734	0.388698	1.897167	0.70649
IGF2BP2	0.919768	0.8431	1.003408	0.059652	1.097637	1.040737	1.157648	0.000603
IGF2BP1	4.527263	0.510412	40.156	0.175085	0.183686	0.000123	274.5852	0.649576
YTHDF3	1.290726	1.065857	1.563037	0.008975	0.999697	0.891851	1.120584	0.995846
WTAP	0.994085	0.811996	1.217007	0.954173	0.916739	0.821487	1.023036	0.120403
HNRNPA2B1	0.991969	0.959783	1.025234	0.631828	0.983031	0.963915	1.002526	0.0876
HNRNPC	0.975073	0.928021	1.024511	0.317143	1.020512	0.987934	1.054164	0.219977
ALKBH5	1.04919	0.991575	1.110154	0.095643	0.957849	0.913425	1.004433	0.075499
YTHDF2	1.088042	0.895735	1.321636	0.395142	0.939033	0.833287	1.058199	0.302091
METTL16	1.021759	0.712861	1.464507	0.906707	1.001206	0.802832	1.248598	0.991461
YTHDC2	1.812616	0.943702	3.481584	0.074104	1.020589	0.686415	1.517452	0.919788
KIAA1429	1.756042	1.020936	3.020446	0.041867	1.065358	0.779387	1.456256	0.691363
IGF2BP3	1.097294	0.926038	1.30022	0.283528	0.155158	0.002606	9.23819	0.371512
RBM15B	1.024573	0.807027	1.300762	0.841987	1.077647	0.933178	1.244482	0.30856
YTHDF1	1.066432	0.990667	1.147991	0.087156	1.013381	0.948002	1.083268	0.696068

Italic values are statistically significant

Fig. 4.

Detection and validation of m⁶A-related module by WGCNA. a Heatmap of the correlation between gene modules and the clinical phenotypes of PTC. The brown module was the most correlated module of IGF2BP2 expression level. b Hierarchical cluster analysis was performed to detect co-expression modules with corresponding colors. c The correlation analysis between membership (MM) in brown module and gene significance (GS) for IGF2BP2. d, e Bubble chart of GO and KEGG results of brown module

Construction and verification of the m⁶A-related risk signature

We randomly divided patients in the total TCGA cohort into training set and testing set. Then another univariant CoxPH was performed in the training set to filter genes which were related to DFS of PTC, among 796 genes in the brown module (Additional file 9: Table S6). We identified 4 genes: IGF2BP2 (HR = 1.19, P = 0.0002), STT3A (HR = 0.89, P = 0.033), MTHFD1 (HR = 1.32, P = 0.044) and GSTM4 (HR = 1.12, P = 0.049). The expression levels of STT3A, MTHFD1 and GSTM4 were strongly correlated with IGF2BP2 (Fig. 5a–c). These 4 genes were used to construct the m⁶A-related risk signature via multivariate CoxPH regression model (Additional file 10: Table S7). Risk scores of patients were calculated as follows:

$$\text{Risk score}=\left(0.390\times \text{MTHFD}1\right)+\left(0.167\times \text{IGF}2\text{BP}2\right)+\left(0.152\times \text{GSTM}4\right)+(-0.133\times \text{STT3A})$$

Fig. 5.

Construction and validation of m⁶A-related signature. a–c The Pearson correlation coefficients between IGF2BP2 and each gene in the signature. d–f The ROC curves of patients with PTC in the total TCGA cohort, training group and testing group. g–i Comparing DFS in high-risk and low-risk groups by performing K–M survival curves in the total TCGA cohort, training group and testing group

Patients were divided into high-risk and low-risk groups with the median risk score used as the cutoff value. The ROC curve analysis in total TCGA PTC cohort (AUC = 0.817, Fig. 5d), training set (AUC = 0.904, Fig. 5e) and testing set (AUC = 0.774, Fig. 5f), revealing promising prognosis value of the signature for PTC disease-free survival. The K–M survival curves were performed to illustrate the difference between the high-risk and low-risk groups in DFS: total TCGA PTC cohort (P = 8.166e−05, Fig. 5g), training set (P = 8.426e−04, Fig. 5h) and testing set (P = 1.473e−02, Fig. 5i). All of these analyses showed that patients in low-risk group had better prognosis than high-risk group and this m⁶A-related signature was of strong accuracy in predicting the DFS of patients with PTC. Univariant and multivariant CoxPH showed that T classification (HR = 1.691, P = 0.003), stage (HR = 1.49, P = 0.003) and risk score (HR = 1.001, P = 0.047) were prognostic factors for PTC (Fig. 6a), and only risk score was the independent prognostic factor (HR = 1.001, P = 0.04, Fig. 6b). Furthermore, a prognostic nomogram was constructed to predict DFS of individual patients with PTC (Fig. 6c). In Fig. 6d, we assessed whether there was statistical difference in clinical phenotypes between high-risk and low-risk groups by Chi square test. The heatmap indicated that high‐risk group was corresponded to advanced stage, higher level of T and N classifications, new tumor event and extrathyroidal extension in total TCGA PTC cohort. Finally, to better understand the expression of IGF2BP2 in human tissues, we used GTEx dataset to explore the landscape of IGF2BP2 in different genders and organs. The expression patterns of IGF2BP2 were similar in most organs of female and male, but were significantly different in blood vessel, brain, breast, skeletal muscle, skin and stomach (Fig. 7a–c). The tissue-specificity of IGF2BP2 is of great value to explore as it can provide clues for therapy and diagnosis. IGF2BP2 had high expression level in bone marrow and low expression level in brain, liver and skeletal muscle. These organs and malignant tumor originated from them can become potential objects of study (Fig. 7d).

Fig. 6.

Analysis of prognostic factors for PTC. a Univariant CoxPH of risk score and 6 clinical phenotypes (gender, T, N, stage, age, extrathyroidal extension). b Multivariant CoxPH of T, stage and risk score. c A nomogram performed based on prognostic factors found in CoxPH. d A heatmap which showed the risk scores, clinical phenotypes and gene expression (IGF2BP2, GSTM2, MTHFD1 and STT3A) of patient with PTC. A Chi square test was also performed to evaluate the relationships among risk score and 9 clinical phenotypes (*P < 0.05, **P < 0.01, ***P < 0.001)

Fig. 7.

The expression landscape of IGF2BP2 in different genders and organs. a, b The expression levels of IGF2BP2 in different organs of male and female. Red represented high level of IGF2BP2 while green represented low level. c Comparing the expression of IGF2BP2 in 31 tissues of male and female by Mann–Whitney test (*P < 0.05, **P < 0.01, ***P < 0.001). d The relative expression level of IGF2BP2 in 31 types of human tissues

Discussion

Over past decades, the occurrence of PTC has been proved to be correlated with external radiation exposure, dietary iodine content and resultant disturbance of thyroid stimulating hormone (TSH) level [24]. Nowadays, increasing investigations begin to focus on acquired genetic changes that can distinguish PTC from para-tumor normal tissue, which has greatly expanded our knowledge of the molecular pathogenesis of PTC. Several biomarkers have been used in clinic, such as RET/PTC rearrangement, PAX8-PPARγ rearrangement, BRAF and RAS mutations [25–27]. As a promising field of cancer biology, m⁶A RNA modification has been verified to participant in developing several types of malignant tumor. However, there has been no research which explore the role and mechanism of m⁶A RNA modification in the progression of PTC. Here, we discovered some m⁶A RNA methylation regulators whose expression level, CNVs and SNPs were strongly correlated with PTC, including IGF2BP2, HNRNPA2B1 and IGF2BP3. They are all “readers” which can selectively bind to and change the secondary structures of m⁶A-containing RNAs. This process results in regulated degradation of targeted RNAs and can be reversibly tuned via m⁶A methylation and demethylation [28]. Readers may also affect RNA splicing, storage, trafficking and translation [29]. Overexpression of IGF2BP2 has been indicated to be related to poor survival of patients with colorectal cancer, acute myelocytic leukemia and metaplastic breast cancer [30, 31]. Recently, K.Wang et al. demonstrated that the progression of thyroid carcinoma can be promoted by METTL3 and IGF2BP2 through m6A methylation on TCF1 mRNA and activation of Wnt signaling pathway in thyroid cancer [32]. The relationship between IGF2BP2 and PI3K/Akt signaling pathway has been discussed, suggesting up-regulated IGF2BP2 in pancreatic cancer plays a role in cell proliferation [33]. In addition, SNPs of IGF2BP2 and IGF2BP3 has been proved to promote the lymph node metastasis of esophagogastric junction adenocarcinoma [34]. In our GO and KEGG analysis, we noticed that IGF2BP2-related module was correlated with central carbon metabolism, thyroid hormone synthesis and HIF-1signaling pathway, which provided some potential regulatory metabolism of m⁶A RNA modification. In addition to thyroid gland, IGF2BP2 also showed high expression level in lung, small intestine and bone marrow, but the expression of IGF2BP2 did not have great difference between male and female. The next step is to analyze the expression pattern of IGF2BP2 in other organs and explore the potential relationship between IGF2BP2 and regulatory singling pathways, such as HIF-1 signaling pathway.

Furthermore, we explored and validated the prognostic value of m⁶A RNA methylation regulators in PTC. Although FTO, RBM15, YTHDF3, and KIAA1429 were correlated with OS of PTC, we chose IGF2BP2 for deeper analysis as it is the only gene which was correlated with DFS of PTC. We used WGCNA, univariant and multivariant CoxPH to select candidate genes (IGF2BP2, STT3A, MTHFD1 and GSTM4) for construction of a m⁶A-related signature. With the exception of IGF2BP2, other three genes were down-regulated in PTC, and their roles in tumorigenesis have been reported in previous researches. It is well-known that STT3A acts as an enzyme which catalyzes PD-L1 glycosylation and maintain PD-L1 stability, resulting in killing T cells [35]. As a result, low expression level of STT3A can support the immune activity in thyroid cancer tissue, which increases inflammatory mediators, cytokines, chemokines, reactive oxygen species in the tumor immune microenvironment and promotes tumor progression. MTHFD1 and GSTM4 are enzymes of folate metabolism and glutathione metabolism, and both of them have been reported to be related to immunodeficiency and tumor [36, 37]. After K-M plot, ROC curve, univariant and multivariant analyses, this signature showed its great value in predicting DFS of patients with PTC. The result was validated in different cohort (total TCGA cohort, training set and testing set) to ensure the accuracy. We can also notice that risk scores were correlated with T and N classifications, new tumor event and extrathyroidal extension of PTC. These clinical phenotypes were considered to indicators of recurrence and lymph node metastasis, both of which were regarded as determinants of DFS and particularly contribute to the exacerbation of PTC. We also provided a nomogram that reduce the m⁶A-related signature into a single numerical estimate of the probability of an event, such as death, 1-, 3-, 5-year DFS and recurrence, predicting the prognosis of every individual patient. For further study, we prepare to evaluate the clinical prognostic value of this signature by applying to patients who are not limited to internet databases. To deeply explore the mechanism of m⁶A modification, cell and animal experiments are urgently needed to search for downstream target of m⁶A RNA methylation regulators.

Conclusions

In this study, we performed a comprehensive evaluation of the landscape of m⁶A RNA methylation in PTC by analyzing the RNA expression level, CNVs, SNPs and correlated clinical phenotypes of 19 m⁶A RNA methylation regulators. In NMF clustering analysis, we found that cluster1 and cluster 2 were significantly different in DFS, stage and age, suggesting the important role of m⁶A modification in PTC. After WGCNA, univariant and multivariant CoxPH, IGF2BP2, STT3A, MTHFD1 and GSTM4 were used as candidates for construction of a m⁶A-related signature. This signature was capable to predict the DFS of Patients in different cohort and served as an independent prognostic factor for PTC. It was also correlated with T and N classifications, new tumor event and extrathyroidal extension of PTC. To sum up, IGF2BP2 is a possible biomarker for diagnosis and prognosis of PTC and our m⁶A-related signature is of great significance in predicting DFS of PTC patients.

Abbreviations

PTC

Papillary thyroid cancer

m6A

N⁶-methyladenosine

SNP

Single nucleotide polymorphism

CNV

Copy number variation

OS

Overall survival

DFS

Disease-free survival

WGCNA

Weighted correlation network analysis

CoxPH

Cox proportional hazard model

FTC

Follicular thyroid cancer

MTC

Medullary thyroid cancer

3ʹUTRs

3ʹ Untranslated regions

TSH

Thyroid stimulating hormone

TCGA

The Cancer Genome Atlas

GEO

Gene expression omnibus

MAF

Mutation annotation format

KEGG

Kyoto Encyclopedia of Genes and Gene Ontology

GO

Gene ontology

K-M

Kaplan–Meier

ROC

Receiver operating characteristic

IGF2BP2

Insulin-like growth factor 2 mRNA binding protein 2

Authors’ contributions

YXL, XYW and XRF designed the study. XYW and JJZ collected the mRNA transcriptome data and clinical information from TCGA, GEO and GTEx. XRF and XYW performed analyses on TCGA data. YXL and CFX performed statistical analyses. SYZ wrote the manuscript. YXL, XYW, XRF, JJZ, CFX and SYZ reviewed and revised the manuscript. All authors read and approved the final manuscript.

Funding

The work was supported by The National Natural Science Foundation of China (nos. 81660294, 81560397 and 81660403).

Availability of data and materials

Additional data not presented in the manuscript can be obtained by contacting the authors.

Ethics approval and consent to participant

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that there are no conflicts of interest.

Supplementary information

Supplementary information accompanies this paper at 10.1186/s12935-020-01283-y.