YTHDF3as a prognostic predictive biomarker of thyroid cancer and its correlation with immune infiltration

The latest research by Li Y.et al. [8] systematically analyzed the molecular alterations and clinical relevance of m⁶A regulators. In their study, 20 m⁶A RNA methylation regulators were screened out with more genetic possibilities, more cancer pathways, and better clinical relevance, which means they deserve further study. Therefore, these 20 m⁶A RNA methylation regulators, including IGF2BP1, RBM15, FTO, ZC3H13, KIAA1429, YTDHF3, YTHDC2, METTL14, YTHDC1, IGF2BP3, YTHDF1, ALKBH5, IGF2BP2, RBM15B, YTHDF2, HNRN2PB1, METTL3, HNRNPC, RBMX and WTAP, were included in our study for further analysis [6, 8].

RNA-seq data with clinical information of TC patients was downloaded from the TCGA database (https://​tcga-data.​nci.​nih.​gov/​tcga/​). Specifically, the mRNA expression profile of TCGA-THCA (the Cancer Genome Atlas Thyroid Cancer) was excavated for further study. Then, 496 TC tissues and 58 normal tissues were included in our research for further analysis, after excluding 14 TC tissues with unclear clinical characteristic information. None of the TC patients had been treated. The different expression levels of m⁶A methylation regulators between TC and normal tissue were analyzed by “limma” R package. (log 2-fold change (FC) absolute value > 1 and the adjusted p value < 0.05). The “Vioplot” R package was used to compare the differences of m⁶A methylation regulators expression. Search Tool for the Retrieval of Interacting Gene (STRING) database (https://​string-db.​org/​cgi/​input.​pl) was applied to construct the PPI network of 20 m⁶A RNA methylation regulators, with the score of interactive relationships greater than 0.4. Then, Cytoscape software was used to visualize the PPI network. The correlation coefficient is also calculated by the “corrplot” R package.

Cluster analysis was performed on m⁶A RNA methylation regulators using “Consensus ClusterPlus” R package. Furthermore, TC patients were divided into two subgroups (named Cluster1 and 2, respectively). And then principal component analysis (PCA) and chi-square test were performed to calculate the significance between Cluster1 and Cluster2. Lasso Cox model was established by “survival” R package and “glmnet” R package. The formula of the individual risk score is as follows: Risk Score = ∑coefficient (GENEi) × expression (GENEi). Here, GENEi presented the candidate gene. As a result, patients were classified into low-risk group and high-risk group, according to the median risk scores [9].

Univariate and multivariate Cox analyses were used to analyse the prognostic value of m⁶A RNA methylation regulators. The independent risk factors include risk scores and various clinical characteristics. Additionally, Multi-ROC curve was used to evaluate the specificity and sensitivity of multiple clinical indicators. Then, Kaplan-Meier curve with Log-rank test was applied to present the overall survival among high-risk and low-risk groups. The ‘survival’ R package was subjected to survival analysis.

To identify potential enriched pathways associated with the meaningful m⁶A RNA methylation regulators, gene set enrichment analysis (GSEA) was performed. And GSEA 3.0 was used for GSEA analysis. H gene sets (h.all.v6.0.symbol.gmt) was selected as hallmark gene set. p < 0.05 and FDR < 0.25 was significant.

TIMER is an interactive Web tool that provides comprehensive and flexible analysis of tumor-infiltrating immune cells, using deconvolution to infer the infiltration abundance of those in different cancer [10]. We analyzed the correlation between YTHDF3 and the abundance of immune infiltrates. Specifically, the immune infiltration data of B cells, CD4⁺ T cells, CD8⁺ T cells, neutrophils, macrophages, and dendritic cells was used in the analyses. These genetic markers have been cited in previous studies [11‐13]. Pearson correlation was used to calculate the relationship between risk scores and immune infiltration.

The human PTC cell line K1 (catalogue number: 92,030,501), human FTC cell line FTC-238 (catalogue number: 94,060,902), human ATC cell line 8305 C (catalogue number: 94,090,184), human MTC cells TT (catalogue number: 92,050,721) and normal human thyroid follicular epithelial cell line Nthy-ori3-1 (catalogue number: 90,011,609) were purchased from the European Collection of Animal Cell Cultures (ECACC). Dr. Robert Gagel (MD, Anderson Cancer Center, University of Texas) presented the PTC cell line W3.

K1 cells were cultured in DMEM-Ham’s F12-MCDB 105 (2:1:1) (Invitrogen) with 10% fetal bovine serum (FBS) (Gibco), 100 µg/mL streptomycin, and 100 U/mL penicillin. W3 cells were cultured in DMEM with 10% FBS, 100 µg/mL streptomycin and 100 U/mL penicillin. FTC-238 cells were cultured in DMEM-Ham’s F12(1:1) containing 10% FBS, 2mM glutamine, 100 U/mL penicillin and 100 mg/mL streptomycin. 8305 C cells were cultured in EMEM(HBSS) (Thermo Fisher Scientific) with 10% FBS, 1% nonessential amino acids (NEAA), 2mM glutamine, 100 µg/mL streptomycin and 100 U/mL penicillin. TT cells were cultured in Ham’s F12 with 10% FBS, 2mM glutamine, 1%NEAA,1% sodium pyruvate (NaP), 100 U/mL penicillin and 100 mg/mL streptomycin. Nthy-ori3-1 cells were cultured in RPMI-1640 (Invitrogen) containing 10% FBS, 100 U/mL penicillin, and 100 mg/mL streptomycin. The cells were cultured in incubators containing 5% CO2 at 37 °C.

Total RNA was isolated from cells using the TRIzol (Life Technologies), and reverse transcribed. Followed by qPCR with Power SYBR Green PCR Master Mix (Eppendorf), each gene’s relative expression levels were calculated and normalized to β-actin as an endogenous control using the 2-^△△CT method. Takara kit used for cDNA synthesis. Each sample was repeated at least three times. The forward primers sequences of YTHDF3 is 5’-CTGGAACAATCACTCGGCCA-3’, the reserve sequences is 5’-CCTTGCCCTTTAGGTCTCTGA-3’. The forward sequence of β-actin is 5’-CACCTTCTACAATGAGCTGCGTGTG-3’, the reserve sequences is 5’-ATAGCACAGCCTGGATAGCAACGTAC-3’. The primers (random hexamers) were synthesized by Shanghai Sangon (Shanghai, China).

Thirty pairs of TC (12 PTC, 6 FTC, 6 ATC, 6 MTC) and adjacent normal thyroid paraffin-embedded tissue were collected from the Shanghai General Hospital during the period of September 2017 to May 2020. The experiments were approved by the Ethical Committee of Shanghai General Hospital. All patients have signed written informed consent. All samples were incubated using rabbit monolyclonal anti-YTHDF3 antibody (1:100 dilution; Cat. ab220161; Abcam; USA) overnight at 4 °C. Rabbit Anti-Mouse IgG H&L (HRP; ab6728; Abcam; USA) was used as secondary antibody. The standard procedures of IHC were described in detail previously [14]. We scanned the slices by Pannoramic (3DHISTECH) and ran caseviewer (2.4.0.119028) to take pictures. In the quantization process, four classes (from 1 to 4) were assigned based on the estimated positive area, which corresponding positive area is 0%, < 10%, 10–50%, > 50%, respectively.

We analyzed different expression of 20 m⁶A RNA methylation regulators between TC tissues and normal tissue. As a result, ALKBH5 and YTHDF2 were down-expressed in TC tissue, while WTAP, YTHDF3 and ZC3H13 were up-expressed in TC tissue (P < 0.05) (Fig. 1A).

In order to infer the relationship between these 20 m⁶A RNA methylation regulators, the PPI network was constructed based on the STRING database (Fig. 1B) and their correlation was calculated by “corrplot” R package (Fig. 1C). The line connecting each m⁶A RNA methylation regulators means the association between two methylation factors. The more lines there are, the stronger the relationship between the two factors. Figure 1B showed that WTAP, ZC3H13 present the function of writer, and YTHDF3 presents the function of readers. In addition, METTL14, YTHDC1, FTO, ZC3H13, KIAA1429 and YTHDF3 were significantly correlated with each other in TC. Among them, the expression of YTHDF3 was significantly correlated with KIAA1429 (r = 0.83), YTHDC2 (r = 0.62) and YTHDF2 (r = 0.60) (Fig. 1C). What’s more, the expression of HNRNPA2B1 was significantly associated with YTHDC1 (r = 0.7) (Fig. 1C). These may suggest that YTHDF3 may be a methylation factor worth further studying.

To understand the risk characteristics of 20 m⁶A RNA methylation regulators in TC, LASSO regression analysis was used to find potential risk-related genes and assess their independent prognostic value. LASSO coefficient profiles of 20 m⁶A RNA methylation regulators were listed in Fig. 2A. And a coefficient profile plot generated against the log (lambda) sequence was listed in Fig. 2B. The coefficient of three risk -related genes were listed in Fig. 2C. As a result, IGF2BP2, YTHDF1, YTHDF3 were screened out as three candidate genes. According to the regression coefficient of these three genes, an individual risk rating model was set up, in order to assess and predict the risk of each TC patient. The formula is as follows: Risk score = (-0.111 × IGF2BP2) + (0.144 × YTHDF1) + (0.717 × YTHDF3). Then, in order to test the prognostic value of these three genes, TC tissues were divided into low-risk subgroup and high-risk subgroup according to the median risk score. Furthermore, Multi-ROC curve was used to evaluate the specificity and sensitivity of multiple clinical indicators, shown in Fig. 2D. The AUC of risk score was 0.863, which means our model had the better predictive power than other indicators, such as: gender (AUC = 0.613), pathological stage (AUC = 0.775), tumor size (T) (AUC = 0.767). In order to compare the clinical characteristics with TC, the patients with TC were divided into two groups: high risk group and low risk group. Then they mapped the relationship between risk scores (high- and low risk groups) and clinicopathological characteristics (including tumor site, metastasis, age, and sex). Understandably, most patients with TC from high-risk group had an increased expression of YTHDF3 and YTHDF1, and significant association was established with the regional lymph node(N), metastasis(M) and pathological stage (Table 1; Fig. 2E).

Then, the results of survival curve showed that the overall survival (OS) of the high-risk group decreased significantly, compared with the low-risk group (p < 0.001) (Fig. 3A). Next, univariate and multivariate COX regression analysis was conducted to compare clinicopathological characteristics (including age, gender, stage, tumor size (T), node (N), and metastasis (M)) in the predictive model, in order to find out whether these risk characteristics could be independent factors for TC. As a result, the age at diagnosis (p < 0.001), pathological stage (p = 0.003), stage T (p = 0.028) and risk score (p = 0.002) were correlated with OS in univariate Cox regression analysis (Fig. 3B), while only age (p = 0.001) and risk score (p = 0.035) remained significantly correlated with OS in multivariate Cox regression (Fig. 3C). Then, based on the results of univariate analysis, a subgroup analysis was conducted on the clinicopathological characteristics with significant significance in age, gender and pathological stage, so as to evaluate the predictive prognostic value of risk characteristics of TC patients. As shown in Fig. 3D-F, compared with the low-risk group, the OS of high-risk patients in each subgroup was significantly lower (p < 0.05), suggesting that the high-risk group related to the older, later stage, and higher T classification level has poor prognosis in TC. These results indicate that the risk model based on the expression of m⁶A RNA methylation regulators can evaluate and predict the prognosis of TC.

In order to find the hub regulator among the risk models, survival analysis among these three regulators was conducted. The results of survival analysis showed that the high expression group of YTHDF3 (p = 0.041) and YTHDF1 (p = 0.014) has lower survival rate (Fig. 4B, C). However, there was no significant difference between high and low expression group of IGF2BP2 (Fig. 4A). Combine the results of Fig. 1A, the analysis of mRNA expression showed that YTHDF1 was not significantly expressed among TC (p > 0.05). So, YTHDF3 was selected as hub gene and then included in Gene Set Enrichment Analysis (GSEA) for pathway analysis. Then, GSEA showed that the signaling pathways with high YTHDF3 expression were P53 pathway (NES = 1.94, NOM p = 0.006, FDR q = 0.068), Glycolysis (NES = 1.74, NOM p = 0.009, FDR q = 0.211) (Table 2; Fig. 4D, E). In a word, the prognostic value and potential mechanism of YTHDF3 were indicated.

In recent years, immune infiltration has investigated potential molecular characterization of tumor-immune interactions, and plays a more and more important role in cancer therapy. So, it’s necessary to analyse the immune infiltration among TC. The expression of YTHDF3 has significant positive correlations with infiltrating levels of B cells (r = 0.572, p = 3.81e-43), CD8⁺ T cells (r = 0.503, p = 1.21e-32), CD4⁺ T cells (r = 0.655, p = 3.49e-61), macrophages (r = 0.622, p = 1.39e-53), neutrophils (r = 0.326, p = 1.45e-13) and DCs (r = 0.261, p = 5.36e-09) in Fig. 5A. Moreover, we found that CD14 of monocytes (r = -0.128, p = 4.72e-03), NOS2 of M1 phenotype (r = 0.265, p = 2.81e-09), CD163 of M2 phenotype (r = 0.19, p = 2.30e-05) are significantly correlated with YTHDF3 expression in TC (p < 0.001) in Fig. 5B.

To further verify our analysis results, immunostaining of YTHDF3 in TC (including papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), medullary thyroid carcinoma (MTC), and anaplastic thyroid carcinoma (ATC)) and normal thyroid tissues were displayed in Fig. 6A, which proved that compared with normal tissues, YTHDF3 was up-regulated in TC. And immunohistochemistry can also be quantified in Fig. 6B. And then, the human PTC cell line K1, human FTC cell line FTC-238, human ATC cell line 8305 C, human MTC cells TT and normal human thyroid follicular epithelial cell line Nthy-ori3-1 were used in q-PCR. As a result, compared with normal thyroid tissue, the mRNA expression of YTHDF3 in PTC, FTC, ATC, and MTC was up-regulated by q-PCR (Fig. 6C), which is basically consistent with TCGA dataset analysis. Through these two validations, YTHDF3 was validated as a predictive biomarker.