Genome-wide DNA Methylation Differences in Nonfunctioning Pituitary Adenomas With and Without Postsurgical Progression.

CONTEXT: Tumor progression in surgically treated patients with nonfunctioning pituitary adenomas (NFPAs) is associated with excess mortality. Reliable biomarkers allowing early identification of tumor progression are missing.
OBJECTIVE: To explore DNA methylation patterns associated with tumor progression in NFPA patients.
METHODS: This case-controlled exploratory trial at a university hospital studied patients who underwent surgery for NFPA that had immunohistochemical characteristics of a gonadotropinoma. Cases included patients requiring reintervention due to tumor progression (reintervention group, n = 26) and controls who had a postoperative residual tumor without tumor progression for at least 5 years (radiologically stable group, n = 17). Genome-wide methylation data from each tumor sample were analyzed using the Infinium MethylationEPIC BeadChip platform.
RESULTS: The analysis showed that 605 CpG positions were significantly differently methylated (differently methylated positions, DMPs) between the patient groups (false discovery rate adjusted P value < 0.05, beta value > 0.2), mapping to 389 genes. The largest number of DMPs were detected in the genes NUP93 and LGALS1. The 3 hypomethylated DMPs and the 3 hypermethylated DMPs with the lowest P values were all significantly (P < 0.05) and individually associated with reintervention-free survival. One of the hypermethylated DMPs with the lowest P value was located in the gene GABRA1.
CONCLUSION: In this exploratory study, DNA methylation patterns in NFPA patients were associated with postoperative tumor progression requiring reintervention. The DMPs included genes that have been previously associated with tumor development. Our study is a step toward finding epigenetic signatures to predict tumor progression in patients with NFPA.

Nonfunctioning pituitary adenomas (NFPAs) are benign tumors with no clinically relevant hormonal secretion, representing 30% to 50% of all pituitary adenomas (1). They include different subgroups identified by immunohistochemistry (IHC) analysis, for example, gonadotropinomas and null cell adenomas (2). Although most NFPAs are benign, approximately one-third can still show invasive growth into nearby structures such as the cavernous sinus (3). Residual tumor is common, being reported in almost half of patients after primary surgical treatment (4). Progression of residual tumors has been described in 30% to 50% of patients (4, 5), which often requires additional tumor treatment such as reoperation or radiotherapy. Tumor progression after surgery is strongly associated with excess mortality (6). At present, there is no method that can accurately predict tumor progression after primary surgery.

Since most pituitary adenomas are sporadic and without any known genetic mutation driving tumorigenesis, more interest has been focused on epigenetic modifications (7). Epigenetic changes, including DNA methylation, histone modification, and noncoding RNAs, are heritable changes that can regulate transcription and translation without altering the DNA sequence; previous studies have suggested they are important in determining adenoma subtype differentiation as well as adenoma invasiveness and size (8-10). Epigenetics could also potentially play an important role in tumor behavior such as NFPA progression (7). In a targeted analysis of DNA methylation of the TERT promoter region, methylation of the promotor was associated with progression of pituitary adenomas (11); however, no genome-wide methylation studies investigating tumor progression in NFPAs have been published.

The aim of the present exploratory study was therefore to evaluate whether DNA methylation pattern differs in NFPAs between patients with residual adenoma with postoperative progression requiring reintervention and in patients with residual adenomas without any progression for at least 5 years follow-up after surgery. The study design with 2 distinctly different groups with the same IHC adenoma subtype (gonadotroph subtype) was chosen to increase the possibility of finding clinically relevant differences in methylation pattern.

Methods

In this case-control trial, patients who had been operated for a NFPA were investigated. All patients surgically treated within the western region of Sweden (Västra Götaland Region) between 1987 and 2014 were eligible and were reviewed for inclusion into the study. To ensure that all patients were screened for inclusion in the study, both The Swedish National Patient Registry and the computerized operating lists from the Neurosurgical Department of Sahlgrenska University Hospital, the sole provider of pituitary surgery in western Sweden, were used. A total of 340 patients who had undergone surgery for NFPA were identified. Their medical files were systematically reviewed to collect information about pituitary adenoma progression, hormone replacement therapy, and clinical course. The patients were followed until 2019, giving all patients a potential follow-up period of at least 5 years.

To study DNA methylation patterns predicting tumor progression, 2 distinctly different patient groups were created with separate inclusion and exclusion criteria. The reintervention group (cases) included patients with tumor progression requiring reintervention. Reintervention due to tumor progression was defined as surgery or radiation therapy due to growth of residual tumor. Exclusion criteria for the reintervention group were reintervention of a stable tumor remnant; postoperative radiotherapy as part of the primary treatment; and documented postoperative tumor progression not requiring reintervention. The radiologically stable group (controls) included patients with a radiologically stable tumor remnant for at least 5 years of follow-up. Patients with no residual tumor after primary surgery were therefore excluded. Exclusion criteria also included postoperative radiotherapy as part of the primary treatment.

Of the 340 patients, 66 fulfilled the criteria for the reintervention group and 55 for the radiologically stable group (Fig. 1). In order to perform the study on a homogenous adenoma subtype, only patients with a gonadotropinoma were included in the analysis. After also excluding those with insufficient tumor material for methylation analysis, the genome-wide DNA methylation analysis was performed on 28 tumors from the reintervention group and 21 tumors from the radiologically stable group (Fig. 1).

Figure 1.

Flowchart for the patients included in the study. Abbreviations: IHC, immunohistochemistry; NFPA, nonfunctioning pituitary adenoma; RT radiotherapy.

Ethics

The study was approved by the Regional Ethical Review Board in Gothenburg (Dnr: 100-15), Sweden, and by the National Board of Health and Welfare, Sweden.

Radiology

Three imaging scans were retrieved and evaluated for each patient: the preoperative scan, the first postoperative scan, and the scan before reoperation or radiation therapy in the reintervention group or the latest available scan in the radiologically stable group. All examinations were evaluated by a neuroradiologist (A.E.) regarding size, invasiveness (Knosp grade ≥ 3), and compression of the optic chiasm. Length, width, and height were measured, and tumor volume calculated according to the formula (length × width × height)/2. The imaging evaluations were performed by magnetic resonance imaging in 97% of the cases.

Immunohistochemistry

After surgery, tumor tissue samples were immediately fixed in 4% buffered formalin, and formalin-fixed paraffin-embedded (FFPE) tissue blocks were prepared and stored at the Department of Pathology, Sahlgrenska University Hospital. For each case, new 4-µm thick sections of FFPE tissue were used for IHC. One slide was stained with hematoxylin and eosin and used for adenoma tissue verification. Additional slides underwent IHC staining using standard methods with either an Agilent EnVision or Biocare detection system. The following primary antibodies were used at the dilutions indicated: adrenocorticotrophic hormone (ACTH) [dilution 1:100; product no. M3501/A2A3, Dako/Agilent RRID:AB_2166039, AB_2166039 https://antibodyregistry.org/search.php?q=AB_2166039], growth hormone (GH) [1:200; A0570, Dako/Agilent, RRID:AB_2617170, AB_2617170 https://antibodyregistry.org/search?q=AB_2617170], prolactin (PRL) [1:300; ab64377, Abcam, RRID:AB_1142327, AB_1142327 https://antibodyregistry.org/search.php?q=AB_1142327], follicle-stimulating hormone (FSH) [1:500, M3504/C10, Dako/Agilent RRID:AB_2079146, AB_2079146 https://antibodyregistry.org/search?q=AB_2079146], luteinizing hormone (LH) [1:50; M3502/C93, Dako/Agilent, RRID:AB_2135325, AB_2135325 https://antibodyregistry.org/search?q=AB_2135325], thyroid-stimulating hormone (TSH) [1:50; M3503/0042, Dako/Agilent, RRID:AB_2287785, AB_2287785 https://antibodyregistry.org/search?q=AB_2287785], pituitary-specific positive transcription factor 1 (Pit-1) [1:100; Anti-SLC20A1/HPA035834, Atlas Antibodies, RRID:AB_2674809, AB_2674809 https://antibodyregistry.org/search.php?q=AB_2674809], and T-box family member TBX19 (T-Pit) [1:200; Anti-TBX19/HPA072686, Atlas Antibodies, RRID:AB_2732209, AB_2732209 https://antibodyregistry.org/search?q=AB_2732209].

The tumors were subclassified regarding positivity for hormonal markers and/or transcription factors: thyrotroph (TSH, Pit-1), somatotroph (GH, Pit-1), corticotroph (ACTH, T-Pit), gonadotroph (LH/FSH), lactotroph (PRL, Pit-1), plurihormonal (multiple combinations), and null cell (none). Due to insufficient archival tumor material, IHC for steroidogenic factor 1 (SF-1) was not performed. Adenomas with inconclusive staining pattern, not allowing definite classification, were categorized as “adenoma not otherwise specified”, for example, adenomas that could be either null cell or gonadotroph were placed into this category and therefore not included in the study.

Genome-Wide DNA Methylation Analysis

To perform genome-wide methylation analysis, 4 × 5 µm-thick slices of FFPE tissue were placed in Eppendorf tubes and deparaffinization as previously described (12). DNA methylation analysis was performed for 49 samples using the Infinium MethylationEPIC 850K BeadChip platform (Illumina, San Diego, CA) following bisulfite modification using the EZ DNA methylation kit (D5001, Zymo Research, Orange, CA) and restoration using the Infinium HD FFPE DNA Restore Kit (WG-321-1002, Illumina) as previously described (12). Following quality control and data filtering, where probes at polymorphic positions and those from sex chromosomes were excluded, 754 940 probes remained for further analysis. Six of the samples were excluded due to the high number of failed CpG probes, leaving 43 samples in the final analysis [reintervention group (n = 26) and radiological stable group (n = 17); Fig. 1].

Statistical Analysis

Patient characteristics were compared between the reintervention and radiologically stable group using the Mann-Whitney U test or Student’s t test for continuous variables. Fisher’s exact test was used for comparison of categorical variables. The R-package ChAMP (13) was used for data preprocessing, normalization, and comparison between patient groups and used to calculate CpG sites that were differentially methylated between patient groups (differently methylated positions; DMPs). Inter-sample variations were assessed using Singular Value Decomposition (implemented in ChAMP) to evaluate the relative amounts of variation correlating with biological and technical factors. Technical batch effects were removed with Combat using ChAMP, leaving biological factors (ie, “group affiliation”) as top contributor to the variance, thus confirming the initial grouping. Hence, all the analyses were based on the initial grouping.

Comparison of the difference in methylation beta values between the patient groups were performed using limma (moderated t test) and Benjamini-Hochberg for adjustment of P values. Overall mean DNA methylation levels were compared between the patient groups with a 2-tailed Student’s t test. Kaplan-Meier curves, together with the log-rank test, were used to illustrate the influence of high vs low methylation (the cutoffs were set to the median value) of the 3 hypo- and hypermethylated DMPs with the lowest P value on tumor progression needing reintervention. Gene plots were generated using the Gviz R-package (14). Gene ontology (GO) analyses using the PANTHER bioinformatics resource were performed (15, 16).

Results

Bioinformatic analysis included 26 and 17 patients in the reintervention and radiologically stable groups, respectively (Fig. 1, Table 1). Mean age was lower in the reintervention group than in the radiologically stable group (56 vs 64 years; P = 0.018). There were no differences between the patient groups with respect to preoperative tumor volume, Knosp grade, and preoperative hormone deficiencies.

Table 1.

Baseline characteristics of patients with NFPAs by prognostic outcome

	Reintervention group	Radiologically stable group	P valuea
No.	26	17
Age, y	56 (13)	64 (9)	0.018
Gender, n (%)			1.0
Men	18 (69)	12 (71)
Women	8 (31)	5 (29)
Follow-up duration, mo	104 (78, 142)	129 (88, 164)	0.18
Time to reintervention, mo	61 (32, 92)	–
Preoperative hormone deficienciesb, n (%)
TSH	9 (35)	5 (29)	1.0
ACTH	5 (20)	6 (35)	0.30
Sex steroids	2 (8)	2 (12)	1.0
Diabetes insipidus	0 (0)	0 (0)	–
Tumor volumec, cm3	10.6 (7.9, 15.3)	9.9 (6.5, 18.4)	0.85
Reaches the chiasm, n (%)	23 (89)	14 (82)	1.0
Invasive (Knosp grade ≥ 3), n (%)	8 (31)	5 (29)	0.48

Data are shown as mean ± SD or median (25th percentile, 75th percentile).

Abbreviations: ACTH, adrenocorticotrophic hormone; mo, months; n, number of patients; TSH, thyroid-stimulating hormone; y, years.

aFor between-group comparison using Fisher’s exact test and Student’s t test or Mann-Whitney U test for categorical and continuous variables, respectively.

bNone of the patients were tested preoperatively for growth hormone deficiency.

cPreoperative tumor volume.

Unsupervised hierarchical clustering of the topmost variable CpG sites revealed 3 methylation clusters containing (1) mainly patients from the reintervention group (2), mainly patients from the radiologically stable group, and (3) a mix of patients from the 2 patient groups (Fig. 2A).

Figure 2.

(A) Unsupervised hierarchical clustering of the top 5000 most variable CpG sites revealed 3 methylation clusters containing (1) mostly patients in the reintervention group, (2) mostly patients in the radiologically stable group, and (3) a mix of patients from the 2 patient groups. (B) Volcano plot displaying 47 680 differently methylated positions with a P value cutoff below 0.05. Of these, 605 showed a difference in beta value > 0.2 (20%).

Differentially Methylated Positions

Mean overall DNA methylation levels did not differ in the reintervention and radiologically stable groups, respectively (0.573 vs 0.564; P = 0.16). Using ChAMP and a Benjamini-Hochberg false discovery rate (FDR)-adjusted P value cutoff below 0.05, we identified 47 680 differentially methylated positions (DMPs), of which 605 showed a difference in beta value (delta beta) methylation of more than 0.2 (20%). If instead an FDR-adjusted P value cutoff below 0.01 was used, 143 sites were identified with a delta beta of more than 0.2. The 605 DMPs were used for further analysis and 389 (64%) out of these mapped to a gene region. Of the 605 DMPs, 240 were hypomethylated and 365 hypermethylated in the reintervention group compared with the radiologically stable group (Fig. 2B). In comparison with the distribution of all CpG sites on the EPIC bead chip, DMPs were more commonly located in CpG shore (P < 0.001) and less frequently located in open sea (P < 0.001) (Fig. 3A). When DMPs were mapped in relation to gene position, DMPs were overrepresented in the TSS1500 region (P = 0.0011) and intergenic regions (P < 0.001), and less frequently positioned within a gene (P < 0.001) and in the TSS200 region (P = 0.0049) (Fig. 3B).

Figure 3.

Distribution of 605 differentially methylated positions (DMPs) across CpG features (A) and gene regions (B) compared to the distribution of non-DMPs of the EPIC array. Abbreviations: ExonBnd, within 20 bases of an exon boundary (ie, the start or end of an exon); IGR, intergenic regions; TSS, transcriptional start site; TSS200, 0–200 bases upstream of the TSS; TSS1500, 200–1500 bases upstream of the TSS.

There was an uneven distribution of DNA methylation in different regions when the reintervention group was compared with the radiologically stable group. Gene bodies (58% hypermethylated vs 42% hypomethylated) and the TSS1500 region (64% hypermethylated vs 36% hypomethylated), which are involved in regulation of gene expression, were more frequently hypermethylated in the reintervention group.

The 3 hypermethylated DMPs with the lowest P values in the reintervention group compared with the radiologically stable group were cg18778401 (adjusted P = 5.15 × 10–4, delta beta 0.24) located on chromosome 12 (5′ UTR of ZNF664-FAM101A), cg01742263 (adjusted P = 1.12 × 10–3, delta beta 0.23) located on chromosome 5 (gene body of SLC23A1), and cg18939363 (adjusted P = 1.27 × 10–3, delta beta 0.24) located on chromosome 5 (5′ UTR of GABRA1) (Fig. 4). Similarly, the 3 hypomethylated DMPs with the lowest P values in the reintervention group compared with the radiologically stable group were cg06624032 (adjusted P = 5.15 × 10–4, delta beta 0.23) located on chromosome 7 (gene body of CPED1), cg02457623 (adjusted P = 5.15 × 10–4, delta beta 0.38) located on chromosome 1 (the 5’UTR of ATP2B4), and cg14424181 (adjusted P = 7.20 × 10–4, delta beta 0.25) located on chromosome 20 (intergenic) (Fig. 4). These 6 positions all had an individual effect (log-rank P < 0.05 for all) on the reintervention-free survival in the whole cohort when analyzing both patient groups together (Fig. 4).

Figure 4.

Kaplan-Meier curves and box plots showing reintervention-free survival depending on methylation status of the 3 hypomethylated and the 3 hypermethylated DMPs with the smallest P values and the difference in beta values between the groups for each probe (A = radiologically stable group, B = reintervention group).

The genes containing the highest number of DMPs were the hypermethylated gene nucleoporin 93 (NUP93; chr16q13) and the hypomethylated gene galectin 1 (LGALS1; chr22q13.1). The gene plots for NUP93 and LGALS1 are shown in Fig. 5.

Figure 5.

Gene plots displaying mean methylation levels of the radiologically stable and reintervention groups for all probes on the Illumina EPIC platform for specific regions. (A) Hypomethylation in the reintervention group of the LGALS1 gene in the region 5′ to the transcription start site (TSS1500/CpG island shore region). (B) Hypermethylation in the reintervention group in the region surrounding the transcription start sites for one of the transcripts of the NUP93 gene.

Differentially Methylated Regions

To identify chromosomal stretches of aberrant methylation, ChAMP was used to analyze the dataset for differentially methylated regions (DMRs). This process identified 204 DMRs (P value cutoff < 0.05), spanning 55 to 1888 base pairs distributed over 1707 probes. The DMRs represented 193 annotated genes. The top 10 DMRs with the lowest P values are listed in Table 2. The highest number of DMRs were located on chromosome 1 (n = 20), and the fewest on chromosomes 9 and 21 (n = 2). When considering chromosome size, there was an uneven distribution of DMRs across the genome, with very few DMRs/Mbp on chromosome 9 and 14, and many DMRs/Mbp on chromosome 17 and 19.

Table 2.

Top 10 DMRs between NFPA patients by prognostic outcome

DMR location	DMR start	DMR end	DMR width	DMR P value	No. of probes in DMR	Gene
chr22	38071001	38071534	533	6.59E-06	7	LGALS1
chr1	203598761	203599154	393	3.30E-05	4	ATP2B4
chr6	28058724	28059208	484	3.96E-05	9	ZSCAN12L1
chr1	91301204	91302117	913	4.62E-05	9	Intergenic
chr16	85936409	85936666	257	5.27E-05	3	IRF8
chr1	2461278	2461929	651	7.91E-05	10	HES5
chr1	156829008	156829377	369	7.91E-05	3	NTRK1
chr16	2286287	2287294	1007	1.25E-04	15	DNASE1L2
chr16	56815558	56816138	580	1.85E-04	7	NUP93
chr2	45179312	45180462	1150	2.64E-04	12	Intergenic

Abbreviation: DMR, differentially methylated region.

Gene Ontology Analyses

GO analysis was performed to further explore the functions of the genes with DMPs (all DMPs = 47 680). Analyses were subdivided into groups of hypo- and hypermethylated genes (reintervention group compared to radiologically stable group) and chromosomal locations. Two genomic regions were of primary interest: gene body and a combined promoter region group (TSS200 and TSS1500). Searching for GO biological processes associated with genes with hypermethylation in promoter regions identified 262 GO terms that were significantly enriched with an FDR-adjusted P value < 0.05. GO terms are graphically displayed in Fig. 6A and the complete list can be found in the supplement (17). Among the most enriched GO terms were those related to regulation of mitotic cell cycle and mitotic cell cycle transition (GO:1901991, GO:2000045, GO:0045930, and GO:1901988) and to Wnt-signaling (GO:0198738, GO:0016055, and GO:0030111).

Figure 6.

The pie charts display the distribution of GO biological processes groups associated with genes with differently methylated positions in 2 different genomic regions. Panel A shows the distribution of the GO terms associated with genes with hypermethylation in promoter regions (TSS region 200 and 1500). Panel B shows the distribution of the GO terms associated with genes with hypomethylation in gene bodies.

Analysis of genes being hypomethylated in gene bodies revealed 479 GO terms that were significantly enriched (FDR-adjusted P value < 0.05). GO terms are graphically displayed in Fig. 6B and the complete list can be found in the supplement (17). Among the most enriched GO terms were those related to axonogenesis and axonogenesis regulation (GO:0050772 and GO:0007409) and to regulation of extent of cell growth (GO:0061387).

Discussion

This is the first study to explore genome-wide DNA methylation patterns as a marker of tumor progression in patients with NFPA. The study design with 2 distinctly different patient groups of the same adenoma subtype was chosen to reduce variability and increase the possibility of finding an epigenetic pattern associated with clinically relevant tumor progression. The analysis found methylation patterns associated with clinically significant tumor growth requiring reintervention (reoperation or radiation therapy). Interestingly, the differentially methylated genes LGALS1 and gamma-aminobutyric acid receptor alpha-1 (GABRA1), which have been previously reported to be involved in cancer development, were associated with tumor progression requiring reintervention.

Genome-wide methylation profiling studies in pituitary adenomas has previously focused on identifying different methylation patterns in different subclasses of adenomas such as functioning and nonfunctioning adenomas (8, 18, 19). It has also been suggested that the methylation profile differs with invasiveness as well as the size of the adenoma (9, 10). An overall higher DNA methylation level has been shown in NFPAs compared with normal pituitary tissue, indicating that aberrant DNA methylation plays a role in the pathogenesis of NFPA (20). However, NFPAs are not a homogenous group, and their different subtypes may have different and distinct genome-wide methylation patterns (21). We therefore created a homogenous cohort for the current study including only gonadotropinomas in order to minimize possible confounders.

DNA methylation information has been previously used to develop biomarkers to guide management of other intracranial tumor types. In glioblastoma, the most frequent malignant primary central nervous system tumor, epigenetic silencing of the MGMT gene by promoter methylation, has been shown to be a biomarker of sensitivity to alkylating chemotherapy (22). In meningiomas, methylation-based tumor classification has been suggested to be superior in predicting tumor recurrence and prognosis compared with the current WHO classification based on IHC analysis (23).

Our study found 605 differently methylated CpGs (DMPs) between the reintervention and the radiologically stable groups. The 3 hypermethylated positions with the lowest P value were linked to the genes ZNF664-FAM101A, SLC23A1, and GABRA1. Interestingly, GABRA1 has been suggested to be involved in pituitary tumor pathogenesis (24). Also, hypermethylation of GABRA1 has been reported to play an important role in colorectal cancer (25).

Among the 3 hypomethylated positions with the lowest P value, 1 of them was located in the gene for cadherin-like and PC-esterase domain-containing 1 (CPED1). The molecular role of CPED1 is still under investigation but its altered expression has been suggested to be associated with the progression of lung adenocarcinoma (26).

The 2 genes containing the largest number of DMPs were the hypermethylated gene NUP93 and the hypomethylated gene LGALS1, both showing multiple sites with differential methylation. Interestingly, increased expression of galectin 1 has been associated with tumor size, clinical stage, poor differentiation, and decreased overall survival in various cancers (27, 28).

In our study, the TSS1500 promotor region was more frequently hypermethylated (indicating downregulation) and relevant gene ontology terms were enriched, implying that the differently methylated genes in the 2 patient groups could potentially translate into different gene expression levels. Also, the 3 hypermethylated DMPs and 3 hypomethylated DMPs with the lowest P values all had an individual effect on tumor progression. Therefore, our data collectively would support a functional role for a difference in DNA methylation pattern between NFPAs that progress and those that do not progress during long-term follow-up.

This is the first study that describes a different genome-wide DNA methylation pattern between NFPAs with and without clinically significant tumor progression. Our study design with 2 well-defined, distinctly different patient groups with adenomas of only one subtype is one of its major strengths. First, the cases only consisted of patients with a clinically significant tumor progression requiring reintervention, which resulted in a clinically relevant endpoint. Secondly, the comparison was made to another adenoma group with residual tumors that remained indolent for at least 5 years, that is, truly nongrowing tumors. The limitations are the study size and the particular study group, a selection from a larger population, which could introduce bias. Also, some adenomas of the gonadotropinoma subtype may have been excluded since IHC for SF-1 was not performed. Moreover, a recent study has suggested that the current definition of the gonadotroph lineage is not sufficient, which might influence the epigenetic homogeneity of our cohort (21). The study is exploratory and, as such, confirmative studies are needed. As the focus was to identify biomarkers, the posttranscriptional effects of the DNA methylation pattern were not addressed in this study.

In conclusion, in this novel explorative study, we found DNA methylation patterns associated with postoperative tumor progression requiring reintervention in patients with NFPA. Among the differentially methylated genes identified, some have previously been described to be involved in pituitary adenoma differentiation or cancer development, suggesting the functional relevance of our findings. This study is a step toward deciphering epigenetic mechanisms involved in tumor progression and finding epigenetic signatures that might be used to predict postoperative tumor progression in NFPAs.