Integrated analysis of DNA methylation and mRNA expression profiles to identify key genes involved in the regrowth of clinically non-functioning pituitary adenoma

Abstract

Tumour regrowth is a key characteristic of clinically non-functioning pituitary adenoma (NFPA). No applicable prognosis evaluation method is available for post-operative patients. We aimed to identify DNA methylation biomarkers that can facilitate prognosis evaluation. Genome-wide DNA methylation and mRNA microarray analyses were performed for tumour samples from 71 NFPA patients. Differentially expressed genes and methylated genes were identified based on the regrowth vs non-regrowth grouping. There were 139 genes that showed alterations in methylation status and expression level, and only 13 genes showed a negative correlation. The progression-free analysis found that FAM90A1, ETS2, STAT6, MYT1L, ING2 and KCNK1 are related to tumour regrowth. A prognosis-prediction model was built based on all 13 genes from integrated analysis, and the 6-gene model achieved the best area under the receiver operating characteristic curves (AUC) of 0.820, compared with 0.785 and 0.568 for the 13-gene and 7-gene models, respectively. Our prognostic biomarkers were validated by pyrosequencing and RT-PCR. FAM90A1 and ING2 was found to be independent prognostic factors of tumour regrowth with univariate Cox regression. The DNA methylation and expression levels of FAM90A1 and ING2 are associated with tumour regrowth, and may serve as biomarkers for predicting the prognosis of patients with NFPA.

Keywords: DNA methylation; clinically non-functioning pituitary adenoma; regrowth prediction.

Figure 1

Flowchart of this study.

Figure 2

Differential analyses of gene methylation and expression status between regrowth and non-regrowth patients. (A) There are 3329 differentially methylated genes, of which 2788 hypomethylated genes (blue) and 541 hypermethylated genes (red) (B) The heatmap shows methylation profiles of 71 NFPA samples. The rows represent the different probes, and the columns represent each sample. The color in the heatmap represents the methylation level difference, which are hypermethylation (orange) and hypomethylation (green). The bar on the top shows the clinical and grouping information, and the sample ID is on the bottom. (C) The volcano plot shows 501 differentially expressed genes, and there are 438 upregulated genes(red) and 63 downregulated genes (blue). (D) The heatmap shows the expression profiles of the 71 NFPA samples. The rows represent the different genes, and the columns represent each sample. The color in the heatmap represents the expression level difference: upregulated (yellow) and downregulated (blue). The bar on the top shows the clinical and grouping information, and the sample ID is on the bottom.

Figure 3

Integrated analysis of DMGs and DEGs. (A) The Venn diagram shows 139 genes with both DNA methylation and expression level changes. (B) GO and KEGG pathway analyses of 139 genes. (C) Pearson analysis of 139 genes. There are 13 genes showing negative correlation (red), 78 genes showing positive correlation (blue) and 48 genes showing no correlation. The R value of 13 genes is shown. (D) GO and KEGG pathway analyses of 13 negative correlation genes.

Figure 4

Kaplan-Meier analyses of six significant genes in patients with NFPA. Patients with upregulation of FAM90A1, ETS2 and STAT6 are less likely to have tumour regrowth (A–C). Patients with downregulation of MYT1L, ING2 and KCNK1 are less likely to have tumour regrowth (D–F).

Figure 5

SVM regrowth prediction model. Three ROC curves using LOOCV show the comparisons of the AUC for the prediction of regrowth with 6 genes (A), 13 genes (B) and 7 genes (C). The red line shows the prediction model efficiency, and the blue line shows the permutation p-value of AUC was obtained from 1,000 randomization tests for testing the null hypothesis. The 6-gene model shows a better prediction accuracy.

Figure 6

Evaluation of DNA methylation and expression levels of selected genes. The DNA methylation status, expression levels and Pearson correlation of FAM90A1, MYT1L, ETS2, ING2, STAT6, KCNK1 are shown. Each dot represents the average DNA methylation and gene expression level for every sample. * p < 0.05, ** p < 0.01, *** p < 0.001.