Introduction

The pituitary gland, often referred to as the master regulator of the endocrine system, lies within the sella turcica at the base of the brain. It is composed of the adenohypophysis (anterior pituitary), arising from the oral ectoderm, and the neurohypophysis (posterior pituitary), originating from the neuroectoderm. The neurohypophysis consists primarily of axonal projections from the hypothalamus. Together with the hypothalamus, the pituitary gland plays a critical role in regulating essential physiological processes such as growth, puberty, metabolism, stress responses, reproduction, and lactation through hormone production and secretion.

Pituitary tumors, or Pituitary NeuroEndocrine Tumors (PitNETs), are the third most common type of intracranial tumors, accounting for approximately 15% of cases [1]. Although many PitNETs remain undiagnosed (with an autopsy prevalence of around 16%), the incidence of clinically relevant cases in the United States was estimated at 4.07 per 100,000 per year between 2012 and 2016 [2]. The vast majority of pituitary tumors (95%) are localized, with invasion of the surrounding tissue and a high risk of local recurrence, but metastatic lesions occur only in 0.1% to 0.2% of cases [3]. Clinically, PitNETs can present with severe symptoms due to excessive hormone secretion, inhibition of specific hormone secretion, or signs and symptoms related to an expanding sellar mass, such as visual deficits and headaches.

The most common treatment for PitNETs is transsphenoidal surgery, followed by radiotherapy and drug control for certain subtypes [4]. Additionally, immunotherapy with PD-L1 has been proposed as an alternative treatment [5]. A recent study aimed to investigate the association between PD-L1 expression and the radiological and pathological behavior of PitNETs to determine its suitability as a target in relapse cases [6].

Classification of these tumors follows the latest WHO guidelines from 2022 [7], which distinguish tumors of the anterior and posterior lobes, as well as other hypothalamic tumors. Key features used to determine the type and subtype of a tumor include transcription factors, hormones, and other biomarkers, such as low-molecular-weight cytokeratin to determine the cell of origin. For instance, somatotroph tumors are divided into densely and sparsely granulated subtypes, based on the distribution of secretory granules within the tumor. Densely granulated tumors generally secrete more hormones than their sparsely granulated counterparts. These tumors are further characterized by staining for the transcription factor PIT1 (POU1F1, a lineage determinant) and GH (GH1), along with the glycoprotein hormone γ-subunit (CGA). Additionally, cytokeratin staining shows perinuclear patterns in densely granulated tumors and more than 70% fibrous bodies in sparsely granulated subtypes.

Despite significant advancements in the accurate classification of PitNETs, their etiology remains poorly understood. Genetic studies have highlighted the role of inherited germline mutations in genes such as AIP in Familial Isolated Pituitary Adenoma [8], PRKAR1A in Carney Complex, MEN1 in multiple endocrine neoplasia syndrome type 1 (MEN 1) [9], CDKN1B in MEN 4 [10], and USP8 as an oncogene in sporadic corticotroph adenomas [11], as well as GNAS mutations in somatotroph tumors [12].

In this report, we present a potential novel driver for pituitary neuroendocrine tumors: the histone-lysine N-methyltransferase 2D (KMT2D) gene. Germline pathogenic mutations in KMT2D are known to cause Kabuki syndrome, characterized by distinct facial features, growth delay, and mental retardation [13]. Somatic mutations in this gene have been implicated in various cancers, including breast cancer [14], lung squamous cell carcinoma (LUSC) [15], and B-cell-derived lymphoma [16], making it one of the most frequently altered genes in cancer. However, KMT2D has not previously been reported as a driver for pituitary neuroendocrine tumors.

Patient information

Here, we present the case report of a male patient in his 50 s with a secreting mixed somatolactrotroph macrotumor causing acromegaly and elevated IGF-1 levels. Surgical intervention successfully alleviated acromegaly symptoms and normalized IGF-1 levels. Post-operative brain MRI confirmed complete tumor resection. The pathology report on the excised tissue mentions the loss of the usual acinar architecture on reticulin staining as well as cells having a densely granulated eosinophillic cytoplasm. No apparent sign of necrosis or mitosis. Cells express PIT1 (POU1F1), SF1, GH and PRL (Fig. 1). Proliferation index MIB1/Ki67 is around 1%. Very weak intensity staining with PD-L1 (CD274) for 40% of tumoral cells (Suppl. Fig. 1).Cytokeratin staining (not shown) was diffusely cytoplasmic without formation of fibrous bodies. The patient is referred as patient 16.

Fig. 1: Immunostaining markers for somatolactrotroph PitNETs.
figure 1

a Hematoxylin and eosin staining. b POU1F1 staining. c PRL staining. d GH staining.

Full size image

Results

In the context of a large collaborative research project, Pituitary Neuroendocrine Tumors (PitNETS) are fresh collected after surgery at the university hospital (CHUV) in Lausanne, Switzerland and processed for single-cell RNA sequencing (scRNA-seq) and Whole Genome Sequencing (WGS) in our laboratory.

Variant information

Variant analysis of matched germline (blood) and tumor WGS from the patient revealed 455 somatic single nucleotide mutations [cf. Methods], with the only plausible candidate to be a driver in the gene KMT2D. The mutation is NM 003482:exon32:c.8047-2 A > G, a splicing variant disrupting the acceptor site of exon 33. This variant is predicted as pathogenic by SpliceAI, ADA and RF [17, 18]. This variant is not reported in ClinVar and absent from gnomAD exome and genome databases. It is reported as rs2120513666 in dbSNP. Absent in germline, in the tumor tissue the variant presents with an allelic ratio of 50% indicating a tumor purity close to 100% in absence of structural variants in this region (Fig. 2a, b).

Fig. 2: Evidence of a somatic variant in KMT2D.
figure 2

a IGV output of Patient 16’s tumor WGS coverage showing the somatic mutation. b IGV output of Patient 16’s germline WGS coverage. c log2 Matched germline and tumor coverage ratio.

Full size image

Structural alterations

Indeed differential coverage analysis [cf. Methods] between germline and tumor DNA showed extensive alteration in chromosomes 1, 4, 15, 16, 18 and 22 (full chromosomal deletions) as well as chromosomes 7, 9 and 13 (full chromosomal amplifications) but not in chromosome 12 (Fig. 2c).

Single-cell RNA seq

Tumor gene count matrix was generated by CellRanger (version 6.1.0) and processed with Seurat for quality control, normalization, dimensionality reduction, clustering and annotation [cf. Methods]. 2928 cells were retained after stringent quality control with a mean number of detected genes of 2031 and mean unique molecular identifiers (UMIs) of 7348 per cell. Initial clustering displayed 13 clusters (Fig. 3a) which after automatic and manual annotation were separated into tumor cells (3 clusters), immune cells (6 clusters) and structural cells (4 clusters).

Fig. 3: Single cell clustering analysis.
figure 3

a Initial clustering with 13 clusters. b Annotated UMAP based on canonical markers and automated annotation pipeline. c Copykat heatmap representing chromosomal alterations. Blue for deletion and orange for amplification. Y axis bar represent predicted aneuploid and diploid cells. d Quantification of KMT2D-expressing cells across PitNET samples with paired scRNA-seq and WGS data, compared to healthy control tissue.

Full size image

In depth analysis enabled us to separate the tumor clusters into GH1 and PRL expressing cells, the immune cells into CD4 T cells, CD8 T cells and monocytes (with subcategories, cf. Immune cell analysis) and structural cells into stem cells, pericytes, fibroblasts and endothelial cells (Fig. 3b).

Alterations at single-cell level

We validated the chromosomal alterations using Copykat in single cells (version 1.1.0). This software predicts the aneuploid or diploid status of each individual cell allowing for a clear separation of tumor cells versus normal cells. The resulting heatmap showed essentially the same pattern of altered chromosomes we observed in WGS with little differences mostly due to the increased coverage in WGS with respect to single-cell transcriptomics (Fig. 3c).

KMT2D expression in PitNETs

The fraction of cells expressing KMT2D was computed in all PitNET scRNA-seq data sets with WGS (to exclude the presence of KMT2D germline or somatic mutations) and also in an integrated reference (somatotroph and lactotroph cells from the single-cell data set of 3 healthy anterior pituitary gland [19]) (Fig. 3d). A consistently lower fraction was detected in patient 16 compared to all the other tumors and reference indicating a lower transcript abundance due to the non-sense mediated decay of the aberrant spliced transcript.

Differential gene expression analysis (DGEA)

We performed a differential gene expression analysis between the integrated reference and the tumor cells using a pseudobulk approach [cf. methods]. Hallmark gene sets (H) from the Broad Institute were selected to detect up- and down-regulated pathways. Results showed up-regulation of multiple terms (Fig. 4a) which can be grouped into different categories; energy metabolism (Oxidative phosphorylation (NES: 4.18, adjusted P value: 1.26e-23), adipogenesis (NES: 2.39, adjusted P value: 1.1e-4), proliferation and cell growth (MYC targets (NES: 3.91, adj. p = 6.67e-20), MTORC1 signaling (NES: 2.73, adj. p = 2.16e-20), E2F targets (NES: 2.32, adj. p = 5.59e-4), DNA damage and repair mechanisms (DNA repair (NES: 2.9, adj. p = 9.61e-7), P53 pathway (NES: 2.4, adj. p = 3.4e-4), Apoptosis (NES: 2.25, adj. p = 1.51e-3)), stress response and survival (Unfolded protein response (NES: 2.41, adj. p = 4.33e-4), ROS pathway (NES: 2.33, adj. p = 7.21e-4)), immune response and inflammation (Interferon gamma (NES:

Fig. 4: In depth analyses of patient 16's tumor compared to other PitNET samples or healthy references.
figure 4

a Differential gene expression analysis results showing HALLMARKS up- and downregulated between patient 16 and reference. b Tumor and immune cells proportion in PitNET samples where scRNA-seq is available. c Immune cells composition in the PitNETs predicted by Azimuth.

Full size image

2.55,adj. p = 1.02e-4) and alpha response (NES: 2.4, adj. p = 6.19e-4), TNFA signaling (NES: 2.38, adj. p = 4.33e-4)) and cell signaling pathways (UV Response up (NES: 2.69, adj. p = 3.23e-5). Down-regulated pathways are KRAS signaling down (NES:-2.17, adj. p = 2.09e-3)) and UV Response down (NES: -2.11, adjusted P value: 2.89e-3)). Interestingly, the GO term ”KRAS signaling down” indicates a reduced expression of known KRAS downregulated genes: DCC, RYR2, KCNQ2, PRKN, CDKAL1, SIDT1, FGGY, DTNB, CPEB3. Indeed, KRAS appears up-regulated in the tumor (log2FC: 1.47, adjusted P value: 3.34e-35).

Immune cells analysis

While exploring the general composition of patient 16’s scRNA-seq, we noticed a peculiar distribution of cell-types compared to the other PitNETs (i.e. increased number of immune cells proportional to the tumor cells). In this specific single cell sample, there are more than 60% immune cells while generally in PitNETs, the proportion is around 5% (Fig. 4b). To determine the composition of immune cell subcategories, we used Azimuth (https://app.azimuth.hubmapconsortium.org) for label transfer (i.e. automatic annotation) was used. It allows for the prediction of the most likely cell type compared to a comprehensive list of peripheral blood mononuclear cells (PBMCs) [20]. Predicted cell-types were effector memory CD8 T cells (37.6%), central memory CD4 T cells (26.9%), CD14 monocytes (15.9%), effector memory CD4 T cells (6.8%), central memory CD8 T cells (4.8%) and different, low proportion immune cell-types that we grouped in an ”other” category (7.9%). Then, we performed the same analysis for the other PitNETs as comparison and results showed a lower fraction of monocytes, a higher fraction of central memory CD4 T cells as well as the presence of effector memory CD4 T cells (absent in all others). In addition, there is a general increase in cytotoxic activity (both effector and central memory CD8 T cells). Lastly, there is a complete absence of hematopoietic stem and progenitor cells (HSPCs - present in all others) (Fig. 4c). When we compare with ISH on CD4, CD8, CD3 and PD-L1 (Suppl. Fig. 1), we observe a relatively high number of CD8+ (33 cells × mm2 - 5.3 × hpf) and CD4+ (18 cells × mm2 - 2.9 × hpf) cells, being in the high range of the expected distribution as previously reported in [21]. This partially supports our scRNA-seq observations. However, those numbers are not matching with the percentages of immune single cells we detected in the single-cell assay. The reason might be ascribed to the heterogeneity of the tumor (different areas might have different infiltrated cellular densities) or the differences in the assays (protein vs RNAseq).

Methylation profile in PitNETs

To investigate the epigenetic landscape of pituitary tumors, we conducted a genome-wide DNA methylation analysis across our cohort of PitNETs. We focused our analysis on the 1000 most variable methylation probes to identify significant patterns. The boxplots illustrate the methylation levels of these probes across all tumor samples in our cohort (Fig. 5). We identified 796 differentially methylated regions, of which 483 were hypermethylated and 313 were hypomethylated. Notably, we observed hypomethylation in the promoter region of SPON2, an immune-related gene. This hypomethylated state has previously been reported to lead to SPON2 overexpression in prostate cancer [22] and colorectal carcinoma [23]. Our analysis confirmed the previously reported hypomethylation in POU1F1-PIT1 lineage [24], which includes somatotroph, lactotroph, and thyrotroph tumors. These tumors, including Patient 16’s somatotroph tumor, consistently exhibited lower overall methylation levels across the most variable probes compared to other tumor types. No significant DNA methylation differences between this tumor and other somatotroph tumors were detected.

Fig. 5: Boxplot representing beta values of the top 1000 most variable probes in 16 PitNETs where methylation data and WGS were available.
figure 5

POU1F1 lineage includes somato- and lactotroph tumors. TBX19 lineage produces corticotroph tumors and NR5A1 gonadotroph tumors (Non-functioning).

Full size image

Discussion

In this case report, we present a patient in his 50 s with a secreting mixed somato-lactotroph tumor of the pituitary gland exhibiting extensive chromosomal alterations. Variant calling from whole-genome sequencing (WGS) data revealed a splicing variant in KMT2D. This histone methyltransferase enzyme is associated with active gene transcription and chromatin regulation. In normal conditions, it acts as a tumor suppressor gene, preventing uncontrolled cell growth and division [25]. The loss of function of KMT2D disrupts histone methylation, leading to dysregulation of DNA damage repair and gene expression [26]. This disruption increases chromosomal instability, which can potentially lead to severe forms of cancer [27,28,29]. However, it has not been previously reported as a driver for pituitary neuroendocrine tumors (PitNETs). Subsequent single-cell RNA sequencing (scRNA-seq) analyses of the tumor revealed unusual results compared to other PitNETs. Firstly, a reduced number of KMT2D transcripts were detected compared to a healthy reference for the action of the nonsense mediated decay (NMD) pathway on the aberrantly spliced transcript. Differential gene expression (DGE) analysis showed up-regulation of several pathways associated with tumor maintenance and progression, such as oxidative phosphorylation, DNA repair mechanisms, inflammation and immune response, and KRAS signaling. This suggests similarities between this tumor and more aggressive carcinomas [30]. The extensive chromosomal alterations observed in this patient are the consequence of a significant chromosomal instability, which is often a hallmark of cancer progression. Chromosomal instability can lead to an increased mutation rate and the activation of oncogenic pathways. In this case, the up-regulation of KRAS related pathways is noteworthy. KRAS is a well-known oncogene, and its activation can drive tumorigenesis through various mechanisms, including promoting cell proliferation and survival, metabolic reprogramming, and evasion of immune surveillance. However, unlike in lung cancer where up-regulated KRAS disrupts the circadian gene PER2 leading to increased activation of glycolytic pathways [31], this disruption is not observed in our patient’s tumor. This difference may be tissue-specific, indicating that the role of KRAS activation varies across different tumor types. Immune cell analysis in single cell RNA between patient 16 and other PitNETs highlighted an inverse ratio (more immune cells than tumor cells), indicating an increased involvement of the tumor microenvironment. Moreover, immune infiltration was different, with increased immune surveillance and cytotoxic activity (higher numbers of CD4 and CD8 T cells). Although it has be shown that a higher proportion of T cells can be found in somatotroph tumors compared to other tumor types [21], it can’t explain these high levels (especially compared to other somatotroph tumors present in our cohort). We speculate that this might counteract a possible transition to malignancy and that the tumor microenvironment, including immune cell infiltration, plays a significant role in preventing the tumor from progressing to carcinoma. Immunohistochemical analysis revealed light PD-L1 expression in approximately 40% of tumor cells in this pituitary neuroendocrine tumor. While PD-L1 expression is often associated with immune evasion in various cancer types, its significance in PitNETs appears to be more nuanced. Our review of other pathology reports indicates that PD-L1 expression levels in PitNETs can be highly variable, with even some indolent tumors showing weak expression in up to 80% of cells. Thus, PD-L1 expression alone may not be a reliable indicator of tumor aggressiveness or invasiveness in this context. However, when considered alongside other findings—such as the KMT2D mutation, extensive chromosomal alterations detected by WGS, and the unusually diverse immune cell infiltrate observed in single-cell analysis—it contributes to a complex picture of tumor biology. While these collective findings don’t necessarily indicate malignancy, they highlight the tumor’s unique characteristics and emphasize the need for comprehensive, multi-modal analysis in tumor characterization. Our methylation analysis further supports this complex picture of tumor biology. We analyzed the 1000 most variable methylation probes across our cohort of pituitary tumors. Interestingly, we observed that somatotroph tumors, including Patient 16’s tumor, consistently showed lower overall methylation profiles compared to the other tumor types. This hypomethylation pattern in somatotroph tumors suggests a distinct epigenetic landscape that may contribute to their unique cellular characteristics and behavior. Notably, despite the presence of the KMT2D mutation and extensive chromosomal alterations in Patient 16’s tumor, we did not find significant methylation differences between this tumor and other somatotroph tumors. This suggests that the KMT2D mutation in this context may not dramatically alter the overall methylation profile beyond what is typically observed in somatotroph tumors. Intriguingly, despite harboring epigenetic alterations, potent drivers like KMT2D, and significant genetic and transcriptomic changes, pituitary neuroendocrine tumors (PitNETs) seem to resist malignant transformation. This resilience suggests a complex interplay between genetic mutations, epigenetic regulation, and the tumor microenvironment, potentially influenced by the unique anatomical location near but outside the brain. Unraveling the mechanisms behind this distinctive tumor behavior could pave the way for novel targeted therapies and diagnostic approaches in the future.

Methods

Tissue collection and dissociation

PitNETs are resected at the university hospital in Lausanne, Switzerland. After the operation, the department of pathology takes half of the fresh tumor to perform routine analyses and the clinical report. The other part is dedicated to this project. To dissociate the cells we are taking advantage of Worthington papain dissociation kit. In short, tumor is first minced to break the strongest adherence between tissue and cells. Then pieces are mixed with the digestion mix (EBSS, papain, DNAse and collagenase) and heat up at 37°C for 30 minutes in a shaker set to 60 RPM. Every 10 minutes, the liquid is passed through a 10 mL pipette to further separate the cells from the tissues. At the end of the incubation, remaining visible parts are mechanically dissociated with a 200 µl pipette tip set on a 1000 µl tip and pressed to the side of the tube to ensure maximum dissociation thank to laminar flow. Next, sample is filtered on a 40 µm cell strainer and centrifuged for 5 minutes at 300 g. The pellet is resuspended in 500 µl diluted 1/10 ovomucoid (trypsin inhibitor) to stop the digestion process. Next steps involve gradient centrifugation in concentrated ovomucoid for 6 minutes at 70 g to get rid of aggregates and dead cells. Pellet is then resuspended in 600 µl of cold HBSS and filtered a second time on a 40 µm cell strainer. Cells are counted before the second gradient centrifugation in order to resuspend them in the desire volume of HBSS to hit a range of 800-1200 cells per µl.

Single-cell capture, cDNA library preparation, and sequencing

For this project, we used the 10X Chromium workflow. A Chromium Next GEM Chip G (10X Genomics) is loaded with approximately 6’000 cells and sequencing libraries are prepared strictly following the manufacturer’s recommendations (manual CG000204 revD). Briefly, an emulsion encapsulating single cells, reverse transcription reagents, and cell barcoding oligonucleotides are generated. After the reverse transcription step, the emulsion is broken, and double-stranded cDNA are generated and amplified for 12 cycles in a bulk reaction. The cDNA is fragmented, a P7 sequencing adaptor is ligated, and a 3’ gene expression library generated by PCR amplification for 12-14 cycles depending on the initial cDNA amount. Libraries were quantified using a fluorimetric method (Q-Bit), and their quality was assessed on a Fragment Analyzer (Agilent Technologies). Sequencing was performed with DNBSEQTM technology in BGI Europe (Denmark). Primary data processing and assembly was performed with the Cell Ranger Gene Expression pipeline (version 6.1.0, 10X Genomics) with GRCh38 (refdata-gex-GRCh38-2020-A) as transcriptome reference.

DNA extraction and WGS

Extraction of genomic DNA is done with AllPrep DNA/RNA Mini Kit (Qiagen) following standard protocol. Tumor and matched blood samples have been whole genome sequenced by BGI Poland with PE150 30X/90 Gb on DBNSEQ machines. Alignment and mapping has been performed by a custom pipeline based on Sentieon v202308.03 [32] with GRCh38 human genome reference. For somatic variant discovery TNScope algorithm from Sentieon [33] was used in a tumor-normal matching sample mode. From resulting variant set any variant not passing any TNScope filter was removed as potentially of non-tumor origin. Annotation of resulting vcf files with tumor variants has been performed with Annovar annotation tool using dbSNP v150, SpliceAI, gnomad exonic v 2.11 and gnomad genomic v 3.0 databases [34].

Match germline tumor coverage ratio

Analysis of structural variations has been performed through tumor vs matched germline (blood) coverage ratio. Germline single nucleotide variants with allelic ratio = 0.5 and quality score > 400 have been matched with overlapping tumor variants and the log2 ratio of the normalized coverages per variant reported.

Sample processing

Data processing was done with R (version 4.2.1). For each samples, the same processing is used for quality control and downstream analyses. Quality control is performed using SingleCellExperiment (version 1.26.0), DropletUtils (version 1.24.0), scuttle (version 1.14.0), AnnotationDbi (version 1.66.0) and scDblFinder (version 1.18.0). Shortly, data object is created and genes are mapped to chromosomes using EnsDb.Hsapiens.v86 (version 2.99.0). Next, cells expressing less than 200 genes are removed from the analysis. Then to account for outliers in mitochondrial gene expression, isOutlier is used with default parameters (Median Absolute Deviation = 3). Afterward, to detect and remove putative doublets, scDblFinder with dbr.sd = 1 is used. Finally, the main processing was done with Seurat [35] (version 5.0.3) using high quality cells as input. Data normalization is performed with SCTransform() followed by linear dimension reduction (RunPCA(), default parameter), non-linear dimension reduction (RunUMAP(), with dims = 1:20, neighbors detection with Share Nearest Neighbor graph (SNN graph, FindNeighbors() with dims same as above) and finally cluster detection (FindClusters() with default parameters).

First round of cell-type assignment was done by i. extracting markers per cluster (FindMarkers(), default parameters) and comparing them with current knowledge, ii. literature reviewing.

Integration and differential gene expression analysis

Prior to perform differential gene expression analysis, integration of 3 data sets of healthy adult pituitary gland to create a unified data set is performed with Seurat. Briefly, individual data set previously processed with the above described workflow are merged into a single object. Then, object layers are integrated with IntegrateLayers() using RPCAIntegration as integration method. Rest of the processing is the same.

Differential gene expressions analysis is then performed in a pseudobulk because of an eccess of false positive in the default Seurat workflow [36]. Thus we used DESeq2 (version 1.44.0) package for bulk DGEA. Briefly, each data set (i.e. the integrated reference and patient 16) were split into four pseudo-replicates with an equivalent number of cells and all clusters were pulled together and considered as one single cluster. Then, DESeq2 with default parameter was applied. Genes not present in a least 20% of the cells were removed and differentialy expressed genes with adjusted P-value of < 0.01 and log2 fold change > 1 & < -1 are kept.

Finally, to perform gene ontology, fgsea [37] (version 1.30) was used with H: hallmark gene sets and C5: ontology gene sets from GSEA MSigDB collections from the Broad Institute (https://www.gsea-msigdb.org/gsea/msigdb/collections.jsp). Number of permutation (nPermSimple) was set to 10000 and pathways with an adjusted P-value < 0.05 are kept.

Azimuth label transfer

For the immune cell analysis, we annotated them using Azimuth webtool and the Human PBMCs reference (https://app.azimuth.hubmapconsortium.org/app/human-pbmc). Azimuth is a set of single-cell RNA and single-cell ATAC seq references that can be used to annotate cells of a query data set. For the PBMCs reference, it contains 3 layers of increased cell-type specificity (layer 1 is more general compared to layer 2 and so on). For our analysis, we used layer 2 and plot the data using ggplot2 (version 3.5.1).

Methylation analyis

The methylation assay was performed by the IGE3 genomics platform of Geneva University in Switzerland using Illumina Infinium Methylation Assay. In silico processing was done with the Sesame R package (1.23.8) using default parameters.