Sellar region neurocytomas exhibit a CIMP and neuroendocrine-like epigenetic signature distinct from other intra-axial neurocytomas

Abstract

Sellar region neurocytoma (SELN) is a rare neoplasm whose relationship to other neurocytomas within the central nervous system (CNS) has remained unclear. Prior reports have variably classified SELN as a variant of extraventricular neurocytoma (EVN), while immunohistochemical and ultrastructural studies have suggested a hypothalamic origin. Here, we performed unsupervised clustering of DNA methylation data across a large pan-cancer reference set and identified SELN (n = 20) as distinct from other neurocytomas and regional mimics, as well as clustering with neuroendocrine tumors from other organ sites. SELN exhibited a CIMP-like phenotype, TTF1 negativity (0/8), and AVP (vasopressin) promoter hypomethylation, implicating a magnocellular hypothalamic cell of origin. In evaluable cases, a neuronal/neuroendocrine immunophenotype was observed (synaptophysin 8/8, chromogranin A 5/5) with absent pituitary transcription factor expression (PIT1 and TPIT negative 0/4). DNA sequencing (n = 5) and RNA-based fusion profiling (n = 4) did not detect recurrent mutations or gene fusions, respectively. Patients often presented with visual disturbances or headaches and spanned pediatric and older age groups (median 42 years, range 12.5–75), with no sex predilection. Despite locally aggressive imaging features in some cases (cavernous sinus invasion, carotid encasement, hydrocephalus), disease-free survival (n = 13) was comparable to central neurocytoma, with no disease-related deaths during the limited follow-up. Together, these findings support SELN as a molecularly distinct hypermethylated neuroendocrine-like epitype and clinicopathologic entity.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-026-03070-x.

Introduction

Among central nervous system (CNS) neoplasms, ‘neurocytoma’ refers to a group of low-grade neuroepithelial tumors with round, regular nuclei and finely dispersed chromatin arranged within a fibrillary neuropil-rich matrix [26, 27, 80]. Requisite neuronal differentiation is demonstrated through ultrastructural examination (synaptic junctions, neurosecretory vesicles) [26] and/or immunohistochemical studies (synaptophysin, NeuN, neurofilament protein) [27, 68]. Central neurocytomas (CN) are a distinct tumor type that typically arise within the lateral ventricles, often in proximity to the septum pellucidum or interventricular foramina (of Monro), and predominantly affect young adults [27, 37]. Morphologically similar tumors arising outside the ventricles are classified as extraventricular neurocytoma (EVN) and exhibit a wider histologic spectrum than CN [8], a greater tendency for ganglionic differentiation [8], and frequently harbor recurrent FGFR1::TACC1 fusions [66]. A recent series has proposed reclassifying EVN as "glioneuronal tumor with neurocytic differentiation" to reflect its glioneuronal lineage and occurrence at both extra- and intraventricular sites [47]. Cerebellar liponeurocytomas (LIPN) are mature neuronal neoplasms largely restricted to the posterior fossa and share similar immunophenotypic features with their supratentorial counterparts, with a marked propensity for lipomatous change [22, 25, 69]. More recently, unsupervised clustering of DNA methylation data has shown clear separation of CN, EVN, and LIPN as biologically distinct, methylation-defined classes whose composition reflects epigenetic profile rather than anatomic location or histology [10, 47, 66]. CN, EVN, and LIPN are recognized tumor types in the 5th edition of the WHO Classification of Central Nervous System Tumors and correspond to grade 2 [80].

Rare examples of neurocytoma-like neoplasms arising in the sella and hypothalamic region have been reported. Maguire et al. first described a suprasellar mass exhibiting histologic features typical of neurocytoma, ultrastructural evidence of neurosecretory granules [26, 44], and vasopressin immunoreactivity [44]. Approximately 40 cases have since been published, largely as case reports or small series [3, 23, 35, 38, 43, 45, 54, 56, 72, 73, 77, 79, 84–87]. Visual impairment is the most frequently reported clinical finding [3, 43, 56, 77, 87], and imaging features often lead to diagnostic confusion with pituitary neuroendocrine tumors (PitNET)/pituitary adenomas [43, 54, 72, 86]. Histologically, neurocytomas arising in the sellar region exhibit features typical of intra-axial neurocytomas, including expression of synaptophysin with neuropil formation, and are consistently negative for pituitary hormones and transcription factors [3, 43, 72, 86]. A neuronal cell-of-origin has been posited based on the aforementioned ultrastructural features, as well as reported focal expression of TTF1 [3, 43, 72, 86] and vasopressin [3, 44], suggesting hypothalamic lineage. Despite extensive morphologic and immunophenotypic characterization, the nosologic relationship of sellar neurocytomas with other intra-axial neurocytomas (CN, EVN, LIPN) remains unclear.

DNA methylation profiling has become an increasingly important tool for resolving CNS tumors into clinically meaningful types, including those that are morphologically ambiguous or histologically reminiscent of established entities [10, 59]. In a recent report, DNA methylation profiling of a sellar tumor with neurocytoma-like features demonstrated an epigenetic signature that was distinct from established (“Heidelberg”) CNS tumor classes [23]. Notably, t-SNE embedding of the index case with a large reference set revealed a group comprised predominantly of sellar, fronto-basal, and sinonasal region tumors that were epigenetically distinct from both PitNET and olfactory neuroblastoma (ONB), suggesting a previously unrecognized methylation class.

Here, we report the clinicopathologic and molecular features of sellar region neurocytomas (SELN) identified through unsupervised clustering of genome-wide DNA methylation array data. We show that SELN exhibits a signature distinct from other recognized CNS tumor types, clusters with pan-system neuroendocrine tumors, and shows concerted CpG-island hypermethylation (CIMP-like), supporting their classification as a distinct pathologic entity.

Materials and methods

Samples and datasets

This study included cases from publicly available resources and those profiled through the clinical consult service at the National Institutes of Health (NIH), the National Cancer Institute (NCI), Laboratory of Pathology. The research received ethical clearance from the NIH/NCI Central Institutional Review Board, allowing the use of biospecimens with associated anonymized clinical data.

Masked DNA methylation signal intensities from the Childhood Cancer Data Initiative (CCDI) [32] were downloaded from the Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/). Whole-slide images (WSI) for candidate cases were identified through the CCDI Hub Explore Dashboard and DICOM files accessed via the Childhood Cancer Clinical Data Commons (C3DC) [31]. Raw DNA methylation array data for previously published cases were downloaded from the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). Cohort cases from the following accessions were identified: GSE283928 [5], GSE211634 [82], and GSE189778 [33].

Bulk tumor samples compiled for the pan-system dataset were downloaded and annotated from publicly available sources: DNA methylation array data were obtained from the GDC Data Portal (https://portal.gdc.cancer.gov/) across the multiple programs and projects, with files downloaded via the GDC Data Transfer Tool (https://gdc.cancer.gov/access-data/gdc-data-transfer-tool) or GDC API. Additional pediatric CNS tumor data were obtained from the Children's Brain Tumor Network (CBTN) and the Pediatric Brain Tumor Atlas (PBTA) through the Pediatric Cancer Data Commons and the OpenPedCan project repository (https://github.com/PediatricOpenTargets/OpenPedCan-analysis), accessible via PedcBioPortal (https://pedcbioportal.kidsfirstdrc.org/).

Histology and immunohistochemistry

Histologic sections and immunohistochemical studies were performed at the respective referring institutions as part of the initial diagnostic work-up. Immunohistochemical slides were either centrally reviewed at the NIH/NCI Laboratory of Pathology as part of the consultation service or, when slides were not available for central review, the outside interpretations were incorporated into the analysis.

Nucleic acid extraction and sequencing

For samples processed at the NCI, genomic DNA and RNA were extracted from 5 μm sections of formalin fixed paraffin embedded tissue (FFPE) sections mounted on slides using the QIAamp DNA FFPE Tissue Kit (Qiagen, Hilden, Germany), RNeasy FFPE Tissue Kit (Qiagen), or AllPrep DNA/RNA FFPE Kit (Qiagen). Next-generation sequencing (NGS) was performed using commercial TruSeq RNA Exome panel (Illumina). Exome RNA sequencing libraries were prepared with 100 ng tumor RNA using the Illumina RNA Prep with Enrichment (L) Tagmentation kit (Illumina) with Exome Probe Panel (Illumina). Final enriched libraries were sequenced on NextSeq 550DX or NovaSeq 6000 (Illumina). After sequencing, the FASTQ files were aligned to human reference genome hg19 (GRCh37) using STAR aligner [13] to generate BAM files. The resulting BAM files were used by Arriba tool [75] to predict fusion calls. The filtered fusions (VCF file) were uploaded to the QIAGEN Clinical Insight (QCI; Qiagen) for annotation, classification, and interpretation. All variants were manually reviewed by visualizing the raw sequencing read alignments using the Integrative Genomics Viewer [63].

DNA methylation profiling

Samples with tumor purity over 60% estimated from histopathology slides, when possible, were selected for DNA methylation profiling. Genomic DNA (250 ng as the standard) was bisulfite-converted (EZ DNA Methylation Kit, Zymo Research D5001). Sample was also profiled when the total amount of DNA available was less than 250 ng. Bisulfite-converted FFPE DNA was processed with the Infinium FFPE DNA Restore kit (Illumina, USA) and assayed on the Infinium MethylationEPIC or MethylationEPIC v2.0 BeadChip (Illumina, USA), according to the Infinium HD FFPE Methylation Assay automated protocol (Illumina, USA).

Bioinformatic analyses

Detailed methods, including parameters, masking criteria, model architectures, training procedures, and method-specific thresholds, are provided in the Supplementary Methods. In short, methylation array data processing and downstream analyses were performed in R (v4.5.0). Raw IDAT files from HM450, EPIC, and EPICv2 arrays were processed with the SeSAMe package (v1.26.0) [89] using GRCh38-aligned manifests, with ELBAR [41] detection p-value masking, noob background correction, and non-linear dye bias correction; EPICv2 replicate probes were collapsed to a single representative locus using a sequential hierarchy based on bead count, detection, and published probe-quality annotations. Per-sample quality control metrics—including detection rate, signal intensity, and bisulfite conversion—were computed on both platform-specific and cross-platform probe sets after applying a general probe mask for mapping, sequence, and SNP-related exclusions. Sample sex was assigned by consensus across four independent inference methods. Masked probes were imputed using an in-house denoising autoencoder trained on the HM450/EPIC/EPICv2 common probe set.

For cross-platform analyses, beta values were converted to M values and batch-corrected across array platforms using ComBat. Dimensionality reduction was performed using PaCMAP [78] (pan-cancer embedding) and t-SNE [76] (sellar region, sinonasal and neuroendocrine cohort). Prior to hierarchical clustering, samples were purity-filtered (> 0.6) and subsampled within terminal clusters (capped at 30 per cluster, with all SELN retained) for visualization; clustering used Ward.D2 on Euclidean distance of the 2D embedding, with topology groups assigned by dendrogram cutting. Orthogonal embedding of high-purity (> 0.6) neuroendocrine and sellar region neoplasms was performed using PHATE [49]. Genome-wide copy-number profiles were generated using ASCAT.sc [70] with sex-matched reference panels. Tumor purity was estimated using an in-house symmetric gradient boosting model trained on consensus whole-exome sequencing-derived purity labels from > 10,000 pan-cancer samples across five GDC cohorts. Cell-of-origin fractions were estimated using EpiSCORE [90, 91] with a custom reference panel constructed from public scRNA datasets. Differential methylation analysis was performed with limma on M values using a design of ~ 0 + tumor_group + purity + platform, with significance called at adj.P.Val < 0.01 and |Δβ|> 0.20 (Δβ from group-mean β differences). Functional enrichment of significant CpG sets used missMethyl::gometh [58]. A per-sample CIMP index was computed as the fraction of CGI probes with β > 0.30. Mean CGI and open-sea β values were computed. Group means were estimated with platform-adjusted linear models (metric ~ group + platform) and compared using estimated marginal means (emmeans), with pairwise contrasts against SELN and within each CIMP-positive/CIMP-negative reference pair.

Statistical analysis

For comparison of cell-type fraction estimates across tumor groups, the Kruskal–Wallis test was used, followed by pairwise one-sided Wilcoxon rank-sum tests with Benjamini–Hochberg correction. Pairwise comparisons of AVP promoter β values used one-sided Wilcoxon rank-sum tests. DFS and OS were estimated by the Kaplan–Meier method. Pairwise comparisons of DFS across SELN, ONB, and CN were performed using log-rank tests. Age at diagnosis was compared using six pairwise Wilcoxon rank-sum tests; only comparisons with p < 0.05 are annotated in Fig. 5a.

Fig. 5.

Clinical and imaging features. Age at diagnosis across SELN, CN, EVN, and LIPN (a). Sex distribution across the four neurocytoma groups (b). Representative pre-operative MR imaging of two SELN cases (c). Sagittal (left) and coronal (right) T1-weighted post-contrast (top row) and T2-weighted (bottom row) images of Case 4 demonstrate a heterogeneous, mixed solid–cystic sellar mass with suprasellar extension, abutting the optic chiasm and effacing the normal pituitary gland. Case 17 (far right column) shows a large expansile sellar/suprasellar mass with obstructive hydrocephalus from third ventricular compression. Kaplan–Meier survival curves (d). Left: disease-free survival did not differ significantly between SELN (dark blue), ONB (light blue), and CN (green). Right: overall survival trended favorably in SELN compared with ONB (p = 0.099), with no SELN-related deaths during the available follow-up interval. *p < 0.05; **p < 0.01; ***p < 0.001

Results

Sellar region neurocytomas are epigenetically distinct and cluster with pan-system neuroendocrine tumors

A distinct group of 20 sellar region neoplasms was identified through unsupervised dimensionality reduction (DR) of genome-wide DNA methylation data from a large, harmonized pan-cancer cohort (> 80,000 samples). Following strict probe- and sample-level quality control (see Supplemental Methods), samples were embedded using Pairwise Controlled Manifold Approximation Projection (PaCMAP) [78], which improves global preservation of data structure in low-dimensional space [30]. Using this method, we were able to observe coarse grouping of tumor types by expected cell-of-origin signatures including partitioning of neuroepithelial, mesenchymal, and epithelial tumors (Fig. 1a). This newly discovered group showed unambiguous clustering with well-characterized neuroendocrine tumors from multiple tissue types, including lung, small bowel, pancreas, spinal cord (cauda equina), and pituitary gland (Fig. 1a); olfactory neuroblastoma (ONB) and pheochromocytoma/paraganglioma (PCPG) also clustered within the pan-neuroendocrine group. In contrast, other intra-axial neurocytomas (CN, EVN, LIPN), as well as TTF1-expressing neoplasms, including posterior pituitary (neurohypophyseal) neoplasms (PPN) and chordoid gliomas (CHGL), localized to larger neuroepithelial family clusters (EVN, LIPN) or were isolated at this coarse resolution (CN) (Fig. 1a). Furthermore, non-neuroendocrine neoplasms arising in the sellar region (CPH, AT/RT, MNG, PA) were not observed within the pan-neuroendocrine clade (Fig. 1a, outer track), confirming that region-specific non-neoplastic tissue (i.e., tumor impurity) was not driving clustering. Orthogonal embedding of high-purity samples using PHATE, a diffusion-based method that preserves continuous lineage-related trajectory structure [49], similarly placed these samples within the pan-neuroendocrine manifold alongside pituitary blastoma (PitBL), Merkel cell carcinoma (MCC), ONB, PCPG, and the PitNET subtypes, with CN, EVN, and LIPN occupying distinct, non-overlapping branches (Fig. S1). Notably, TTF1-expressing neuroepithelial neoplasms (PPN, CHGL) clustered together and occupied a distinct region of the embedding.

Fig. 1.

Hierarchical clustering of dimensionality-reduction coordinates from a purity-filtered, subsampled cohort of bulk cancer samples across organ systems (see Table S5). Sellar region neurocytomas (SELN, dark blue) cluster with the neuroendocrine tumor family (including olfactory neuroblastoma, ONB), distinct from central (CN), extraventricular (EVN), and cerebellar liponeurocytomas (LIPN); non-neuroendocrine tumors of the sella (red bars, outer track) are shown clustering with their respective epigenetic group (a). t-SNE embedding of SELN with tumors sharing overlapping morphology (PitNET, ONB, CN, EVN, NB FOXR2), localization (PitNET, ONB, PPN, CPH, PA, MNG, AT/RT), and terminology (CN, EVN, LIPN), as well as select neuroendocrine(-like) and sinonasal neoplasms (b). Importantly, this embedding illustrates the distinctness of SELN from its mimics; inter-group distances do not reflect biological relatedness. SELN, sellar region neurocytoma; CN, central neurocytoma; EVN, extraventricular neurocytoma; LIPN, cerebellar liponeurocytoma; PitNET ACTH, corticotroph pituitary neuroendocrine tumor/adenoma; PitNET GON, gonadotroph pituitary neuroendocrine tumor/adenoma; PitNET STH, somatotroph pituitary neuroendocrine tumor/adenoma; PitNET TSH, thyrotroph pituitary neuroendocrine tumor/adenoma; PitNET PRL, lactotroph pituitary neuroendocrine tumor/adenoma; PitNET USP8, USP8-mutant pituitary neuroendocrine tumor/adenoma [53]; PitBL, pituitary blastoma; CPH PAP, papillary craniopharyngioma; CPH ADM, adamantinomatous craniopharyngioma; AMBL, ameloblastoma; GLPC, sinonasal glomangiopericytoma [33]; MNG, meningioma^‡‡; PA, pilocytic astrocytoma^‡‡; NECL (IDH), Neuroendocrine carcinoma-like, IDH2 mutant [33]; NECL (SWI/SNF), Neuroendocrine carcinoma-like, SMARCA4/ARID1A enriched [33]; AT/RT^‡‡, atypical teratoid/rhabdoid tumor*; PCPG, pheochromocytoma/paraganglioma; EWS, Ewing sarcoma; MCC, Merkel cell carcinoma; ONB, olfactory neuroblastoma/esthesioneuroblastoma; PPN, posterior pituitary neoplasm; CHGL, chordoid glioma; CeNET, cauda equina neuroendocrine tumor; NB FOXR2, FOXR2-activated neuroblastoma. ^‡‡MNG, PA, and AT/RT samples located in the sella were selected from the main tumor methylation class and do not correspond to a specific epitype. ^§Methylation class subtype nomenclature derived from Capper et al. [10]. ^§§Methylation class subtype nomenclature derived from Jurmeister et al. [33]. ^‡Methylation class subtypes observed but not currently defined

The clinical and histologic impression of samples in this new group included neurocytoma, olfactory neuroblastoma, and pituitary adenoma/neuroendocrine tumor (Table 1, Table S1). Notably, two cases initially designated as ONB originated from the Childhood Cancer Data Initiative (CCDI), a recently released GDC repository providing comprehensive molecular data from newly diagnosed childhood cancers [20]. Additionally, three samples were previously characterized as a distinct hypermethylated (non-ONB) cluster in a study reclassifying ONB [9]. Supervised classification with the Heidelberg CNS Classifier (v12.8) [67] showed the majority of samples had low to moderate calibrated probability scores for ONB (range 0.1–0.77; Table S1), indicating epigenetic similarity but distinction from this class. Similarly, class probabilities from a separate machine learning classifier of sinonasal tumors [33] revealed sub-threshold scores for ONB (range 0.56–0.85). Unsupervised embedding within a curated set of tumor samples with shared histology, nomenclature, and neuroanatomic location, demonstrated that these sellar region neurocytomas (SELN) retained distinct clustering in the low-dimensional space using t-SNE (t-distributed Stochastic Neighbor Embedding) (Fig. 1b), supporting their distinction from sinonasal and sellar region mimics, such as ONB and PitNET, respectively.

Table 1.

Clinical characteristics of sellar region neurocytomas with available data^a

Case	Age	Sexb	Presentation	Imaging	Clinicopathologic impression	Outcome (PFS, months)
1	57	M	n/a	n/a	Neurocytoma of the sella vs. PitNET, null cell	Progression (144)
2	27	M	Vision changes	n/a	PitNET, null cell	No progression (2)
3	50	F	Vision changes	n/a	Ganglioneurocytoma	Progression (22)
4	74	F	Diplopia	Heterogeneous, mixed solid-cystic sellar mass with suprasellar extension, abutting the optic chiasm and effacing the normal pituitary gland (Fig. 5c)	PitNET, null cell	Progression (99)
5	13	F	Bitemporal hemianopia, CN6 paresis, ptosis	Solid-appearing, partially calcified, heterogeneously enhancing intra- and parasellar mass, 6.1 × 4.3 × 4.2 cm	PitNET with neurocytic differentiation	No progression (23)
6	55	M	Progressive visual disturbance, bitemporal hemianopia	n/a	Sellar region neurocytoma	Progression (15.6)
7	39	F	Refractory headaches + n/v	Large sellar/suprasellar tumor, measuring approximately 5.0 × 2.6 × 2.2 cm; moderately restricted diffusion, intense heterogeneous contrast enhancement, invading cavernous sinuses and enveloping internal carotid arteries bilaterally	Neurocytoma of the sella	No progression (32.6)
8	42	M	Progressive headaches	Well-defined heterogeneous enhancing mass with restricted diffusion measuring 2.1 × 2 × 2 cm; suprasellar extension and mass effect on optic nerve/chiasm/tract; cavernous sinus extension	High-grade neuronal tumor r/o olfactory neuroblastoma	No progression (8)
9	69	F	Headaches, diplopia, falls	n/a	PitNET	Progression (147)
10	42	F	n/a	n/a	Olfactory neuroblastoma	n/a
11	17	F	n/a	n/a	Olfactory neuroblastoma	n/a
12	23	M	n/a	n/a	Olfactory neuroblastoma	n/a
13	40	M	n/a	Mixed cystic and solid with heterogeneous moderately hyperintense T2/FLAIR signal and avid heterogeneous enhancement measuring 3.8 × 1.7 × 3.5 cm; encasing left cavernous internal carotid artery; partially effacing the left Meckel's cave and indenting the left parahippocampal gyrus, while displacing the optic chiasm and prechiasmatic optic nerve	Neurocytoma of the sella	No progression (12.8)
14	n/a	M	n/a	n/a	n/a	n/a
15	47	F	Headache, dizziness and lightheadedness	n/a	PitNET, null cell	No progression (2)
16	72	F	n/a	n/a	n/a	n/a
17	43	M	Hydrocephalus	Expansile intrasellar mass with suprasellar extension, measuring 4.8 × 3.5 × 3.8 cm; extension posteriorly into the interpeduncular and prepontine cisterns adjacent to the dorsum sella; suspicion for right cavernous sinus invasion (Fig. 5c)	Olfactory neuroblastoma	n/a
18	12.5	M	n/a	n/a	Olfactory neuroblastoma	No progression (12.6)
19	16.5	F	n/a	n/a	Olfactory neuroblastoma	n/a
20	75	F	Headache and jaw pain	n/a	n/a	No progression (5.2)

PitNET, pituitary neuroendocrine tumor; PFS, progression-free survival; n/a, not available; n/v, nausea and vomiting; r/o, rule out

^aAdditional clinical information available in Table S1

^bMicroarray-based consensus chromosomal sex inference (see Supplemental Methods)

Sellar region neurocytomas share developmental programs with ONB and exhibit a hypermethylator phenotype (CIMP-like)

Unsupervised hierarchical clustering across a five-group cohort confirmed that SELN and ONB share a predominantly hypermethylated landscape that is distinct from CN, EVN, and LIPN (Fig. S2). To further investigate the epigenetic similarities between SELN and ONB, as well as its distinction from other intra-axial neurocytomas, we performed SELN-anchored pairwise differential methylation comparisons. We identified significant CpG sets (FDR < 0.01, |Δβ|> 0.20) in all comparisons that were strongly skewed towards hypermethylation in SELN, with hyper-to-hypomethylated CpG ratios ranging from 3.7 (vs EVN) to 15.3 (vs LIPN), consistent with global hypermethylation (Fig. S3a). The top differential CpGs across all four comparisons were enriched at hypothalamic and neuroendocrine loci (Fig. S3a), including neuropeptide cargo genes (AVP, OXT, SST), neuropeptide receptors (AVPR1B, SSTR5), pan-neuroendocrine and hypothalamic-lineage transcription factors (e.g., INSM1, ASCL1), and neuroendocrine biosynthetic and secretory machinery (CHGA, PCSK2). While a majority of these loci were hypermethylated, including hypothalamic-lineage transcription factors, a discrete subset was hypomethylated in SELN, most notably in AVP (vasopressin) and ASCL1. Neurocytoma type-specific analysis confirmed this hypothalamic/neuroendocrine program as an SELN-defining signature, distinct from the SVZ developmental program of CN (PRDM16), the cortical TF program of EVN (CUX1), and the adipogenic program of LIPN (ZFPM1, NFATC1) (Fig. S3b).

Among these comparisons, the SELN-ONB set overlapped the least (mean Jaccard index 0.34 versus 0.42–0.47 among neurocytomas), consistent with ONB being the nearest epigenetic neighbor. To define the SELN-ONB shared epigenetic programs, we identified a strongly hypermethylated signature of 13,549 CpGs (11,102 hyper-/2447 hypomethylated relative to other neurocytomas; Fig. 2a), that were enriched for neuron development (p = 1.9 × 10⁻⁸), neuron differentiation (p = 5.8 × 10⁻⁸), and cell morphogenesis (p = 1.1 × 10⁻⁷; Fig. 2b), with recurrent loci at neural developmental transcription factors. Despite these shared programs, the AVP (vasopressin) promoter remained selectively hypomethylated in SELN relative to all four other groups including ONB (all pairwise p < 0.001; Fig. 2c); consistent with this, reference-based DNA methylation deconvolution (EpiSCORE) [90, 91] using a custom panel of olfactory, hypothalamic, and adenohypophyseal cell types demonstrated an enriched hypothalamic neuron signature in SELN compared to ONB (p < 0.01) and PitNET (p < 0.001; Fig. S4). Together, these findings indicate that SELN and ONB share epigenetic programs at neuronal-differentiation loci that distinguish them from other intra-axial neurocytomas.

Fig. 2.

Sellar region neurocytomas share epigenetic programs with olfactory neuroblastoma and exhibit a CIMP-like hypermethylator phenotype. Heatmap of the 13,549 CpGs differentially methylated between SELN+ONB and other neurocytomas (CN, EVN, LIPN) (a). Gene Ontology Biological Process (GO BP) enrichment of differentially methylated CpG-associated genes (b). Top: terms shared between SELN and ONB. Bottom: terms differentially enriched between SELN and ONB. β value distribution at the AVP (vasopressin) promoter across SELN, ONB, CN, EVN, and LIPN (all pairwise p < 0.001) (c). Per-sample CIMP index, defined as the fraction of CpG-island promoter loci (TSS200, TSS1500, 5′UTR, 1st exon) with β > 0.30 (d). Left of divider: SELN compared with intra-axial neurocytomas (CN, EVN, LIPN), ONB, and non-neoplastic regional controls (cerebral cortex [CTX], cerebellum [CBM], adenohypophysis [ADHYP], hypothalamus [HYPTH]). Right of divider: cross-cancer reference panel of putative or published CIMP-positive/CIMP-negative tumor pairs spanning 11 cancer types. *p < 0.05; **p < 0.01; ***p < 0.001

To evaluate the global hypermethylation observed in SELN, we computed a per-sample CIMP index, defined as the fraction of CpG-island (CGI) promoter loci (TSS200, TSS1500, 5′UTR, 1st exon) with β > 0.30, and compared SELN against non-neoplastic regional controls (cerebral cortex, cerebellum, adenohypophysis, hypothalamus) and published CIMP-positive/CIMP-negative tumor pairs as a CIMP index control. SELN exhibited a CIMP index of 0.118 [95% CI 0.107–0.129], significantly elevated relative to ONB (p = 4.1×10⁻¹⁴), intra-axial neurocytomas (vs CN p = 1.1×10⁻²⁴; vs EVN p = 1.7×10⁻¹⁵; vs LIPN p = 2.4×10⁻¹²), and all four non-neoplastic regional controls (all p < 1.4×10⁻⁴; Fig. 2d). Notably, SELN displayed the highest mean open-sea beta-value (0.77) of all 29 groups, a pattern of CGI hypermethylation with preserved open-sea methylation characteristic of IDH/SDH-mutant CIMP [42, 55, 74], rather than the CGI hypermethylation set against global/repetitive-element hypomethylation seen in most cancer methylomes [17].

Somatic alterations

Copy-number profiles inferred from non-allele-specific methylation array signals showed predominantly diploid genomes with only occasional low-level gains and losses; no recurrent focal CNAs were identified in the SELN cohort (Fig. S5). At the cohort level, SELN profiles resembled other neurocytomas (CN, EVN), including a subset of cases with whole-chromosome 19 loss (Fig. S5). By contrast, ONB showed near-universal loss of chromosomes 1, 3, 4, 10, and 12, with gains of chromosomes 5, 7, and 14. SELN was also readily distinguished from PitNETs, particularly lactotroph and somatotroph adenomas/tumors (Fig. S5). DNA sequencing for single-nucleotide variants (SNVs) was performed in five cases (UCSF500v3, TSO500, 170-gene solid-tumor panel: n = 3; whole-exome sequencing [CCDI MCI], n = 2). No recurrent deleterious somatic alterations were identified within the detection limits of the assays. Whole-transcriptome profiling (RNA-seq) was performed in three cases, and one case profiled with a targeted fusion assay (Archer FusionPlex Pan-Solid Tumor v2); all were negative for recurrent or biologically meaningful fusion transcripts (Tables S2–4). Notably, no fusions were detected involving the long non-coding RNA (lncRNA) genes LMCD1-AS1 or GRM7-AS1, as previously reported [43].

Histology and immunophenotype

Whole-slide images (WSI) of H&E-stained slides were available for review in 10/20 cases. Sellar region neurocytomas recapitulated the cytomorphologic features typical of neurocytomas occurring elsewhere in the CNS and were congruent with the previous reports. Tumors were composed of monotonous sheets and nests of small, round-to-oval cells with finely stippled "salt-and-pepper" chromatin, indistinct nucleoli, and scant cytoplasm embedded within a fibrillary neuropil-rich matrix (Fig. 3a, b). Rare cases showed a focal ganglion-like appearance, with scattered larger cells exhibiting more abundant eosinophilic cytoplasm and prominent nucleoli (Fig. 3c). Architectural features included diffuse sheet-like growth and lobular arrangements separated by delicate fibrovascular septa (Fig. 3d). Mitotic activity was low but variable, ranging from rare to 6 per 2 mm² (Fig. 3e, arrowheads). Necrosis was observed in two cases (Fig. 3f) and sinonasal involvement was present in a subset (Fig. 3g).

Fig. 3.

Histologic features of sellar region neurocytomas. Monotonous sheets and nests of small, round-to-oval cells with finely stippled "salt-and-pepper" chromatin, indistinct nucleoli, and scant cytoplasm embedded within a fibrillary, neuropil-rich matrix (a, b). Focal ganglion-like differentiation (c). Patterns included lobular arrangements separated by delicate fibrovascular septa (d). Mitotic activity (e) and necrosis (f) were observed in a subset of cases. Involvement of sinonasal mucosa (g). Scale bars = 50 µm

Sellar region neurocytomas demonstrated a uniform neuroendocrine/neuronal immunophenotype (Fig. 4). Synaptophysin (8/8) and chromogranin A (5/5) were diffusely expressed (Fig. 4a, b), as was INSM1 in one case evaluated (Fig. 4c). Somatostatin receptor 2a (SSTR2A) showed strong membranous labeling in evaluable cases (3/3) (Fig. 4d). Neuronal markers were variably expressed: NeuN was positive (diffuse or focal) in 6/7 cases (Fig. 4g), and neurofilament protein highlighted a delicate fibrillary network within neuropil-rich areas in one case (Fig. 4e). GFAP was negative in 4/5 cases, with scattered staining in one case (Fig. 4h), and OLIG2 demonstrated weak patchy nuclear staining in 3/5 cases (Fig. 4i). The S100 stain showed patchy positivity in 1/3 evaluable cases (Fig. 4f). The Ki-67 proliferation index, available in 11 cases, ranged from 3 to 15%.

Fig. 4.

Immunophenotype of sellar region neurocytomas. Representative immunohistochemical stains supporting a neuroendocrine phenotype with variable neuronal and glial/epithelial/adenohypophyseal lineage marker expression. Diffuse expression of cytoplasmic synaptophysin (a) and chromogranin A (b). Strong nuclear INSM1 (c) in one case evaluated. Strong, membranous SSTR2A labeling (d). Neurofilament protein highlighting a delicate fibrillary network within neuropil-rich areas (e). Patchy S-100 staining (f). Scattered NeuN (g) and GFAP (h) were present in some cases. OLIG2 with weak patchy nuclear staining (i). TTF1 was negative in all cases evaluated (8/8) (j). Summary of immunohistochemical findings across evaluable cases (k). Images of immunohistochemical stains are contributor-submitted representative fields and original magnification was not uniformly available

Adenohypophyseal lineage markers were uniformly absent. None of the tested cases expressed pituitary transcription factors PIT1 (0/4) or TPIT (0/4); SF1 was negative in 3/4 cases, with weak nuclear positivity noted in one. Anterior pituitary hormones, including LH, FSH, TSH, GH, PRL, and ACTH, were negative in three cases evaluated. Notably, TTF1 was interpreted as negative in all 8 cases evaluated (Fig. 4j). Cytokeratins (CAM 5.2 0/4, AE1/AE3 0/2, pan-cytokeratin 0/1, EMA 0/1) and GATA3 (0/3) were also negative. A summary of select immunohistochemical findings is shown in Fig. 4k.

Clinical characteristics and outcomes

In this cohort, of the 12 patients with clinical data, half (n = 6) presented with visual disturbances, including diplopia, bitemporal hemianopia, and cranial nerve palsies, while five presented with refractory or progressive headaches (Table 1). The median age at diagnosis was 42 years (range 12.5–75; Fig. 5a), with a wide distribution including pediatric (n = 4, < 18 years) and older (n = 4, >65 years) age groups. SELN occurred at a significantly older age than CN and EVN, with no significant difference between SELN and LIPN (Fig. 5a). The SELN cohort showed no sex predilection (9 male, 11 female), in contrast to a female predominance in LIPN not previously reported (Fig. 5b).

On imaging, SELN was consistently characterized as a well-defined, heterogeneously enhancing mass centered in the sella but extending into multiple adjacent compartments, including suprasellar, parasellar, cavernous sinus, and, in some cases, interpeduncular and prepontine cisterns, supporting a broader designation of "sellar region" (Fig. 5c). Furthermore, a subset presented as large (> 4 cm), locally aggressive lesions with Knosp grade 3–4 cavernous sinus invasion, internal carotid artery encasement, and obstructive hydrocephalus from third ventricular compression (e.g., Case 5: a 6.1 cm partially calcified intra/parasellar mass; Case 8: a 5.0 cm mass with bilateral ICA encasement; Case 17: a 4.8 cm expansile mass with obstructive hydrocephalus). In cases with available data, surgical management consisted of transsphenoidal or endonasal endoscopic resection (n = 5). Postoperative panhypopituitarism and SIADH/hyponatremia were notable complications (n = 4). Adjuvant radiotherapy or stereotactic radiosurgery was administered in five patients for indications that included subtotal resection, residual cavernous sinus disease, or progression.

Disease-free survival did not differ significantly between SELN, ONB [1, 7, 12], and CN [37] by pairwise log-rank testing (SELN vs ONB p = 0.71; SELN vs CN p = 0.24; ONB vs CN p = 0.42; Fig. 5d, left). In patients with available follow-up (n = 13), 5 had recurrence over a wide latency range (15.6–147 months) including one patient who recurred 12 years following initial transsphenoidal resection and stereotactic radiosurgery (Case 1). Overall survival trended favorably in SELN compared to ONB (p = 0.099), with no SELN-related deaths observed during the available follow-up interval (Fig. 5d, right), although interpretation is limited by the shorter median follow-up duration in the SELN cohort (median 15.6 vs ~100 months in ONB).

Discussion

The 5th edition of the WHO Classification of Central Nervous System Tumors [80] recognizes three tumor types that include the "neurocytoma" label: central neurocytoma, extraventricular neurocytoma, and cerebellar liponeurocytoma. Despite multiple reports of neurocytoma-like tumors arising from the sellar or hypothalamic region, its relationship to these established types remains uncertain, having been variably regarded as a sellar variant of EVN [35, 38, 73, 77, 79, 86]. Notably, use of the "extraventricular" designation has recently been contested as lacking diagnostic specificity, with a proposal for revision based on shared genetics and immunophenotype [47]. Here, we show that sellar region neurocytomas (SELN) are epigenetically distinct from other intra-axial neurocytomas (CN, EVN, LIPN), as well as from morphologically similar tumors arising in this region (e.g., PitNET, ONB). Furthermore, we also demonstrate that SELN exhibits a CIMP-like phenotype and a neuroendocrine-like signature, as well as support the use of DNA methylation profiling as a diagnostic adjunct for these tumors in difficult cases.

DNA methylation changes in cancer have been reproducibly linked to cell-of-origin [29] and persist through clonal evolution [2, 21, 46]. This epigenetic fingerprint has been leveraged in resolving cancers of unknown primary (CUP) with high sensitivity and specificity [50], and a recent large study of neuroendocrine neoplasms (NEN) demonstrated site-of-origin-specific methylation signatures [24]. Here, we show that neuroendocrine tumors (NET) share a common epigenetic signature when clustered across a broad range of human cancer types. This NET methylation family includes entities with recently revised terminology that explicitly reflects their neuroendocrine origin, such as PitNET [14] and cauda equina neuroendocrine tumor (formerly paraganglioma) [61]. Pituitary blastomas, thought to be derived from oral ectoderm (Rathke’s pouch) and folliculo-stellate cells [52, 64], and extra-CNS paragangliomas also embed within this family [19, 81], consistent with their established neuroendocrine identity and supporting the fidelity of global epigenetic clustering. Olfactory neuroblastomas similarly localize to the NET family, congruent with recent evidence of their NET-like transcriptional programs and clear separation from adrenal neuroblastomas, prompting the authors’ proposal to include ONB in neuroendocrine basket trials [18].

Sellar neurocytoma is a recent diagnostic entry to the 5th edition of the WHO Classification of Endocrine and Neuroendocrine Tumors [81]. It is hypothesized that these tumors may arise from the hypothalamus due to reports of vasopressin and thyroid transcription factor-1 (TTF1) protein expression [3, 44]. In the adult brain, TTF1 marks neurons of the basal hypothalamus (arcuate, ventromedial, and mammillary nuclei) and the neurohypophysis, but expression is absent from the magnocellular vasopressin- and oxytocin-producing neurons of the supraoptic (SON) and paraventricular (PVN) nuclei [36, 39]. This dichotomy mirrors sellar tumor biology: TTF1-positive pituicytes (modified glial cells of the neurohypophysis) give rise to posterior pituitary tumors (PPN) [40], and TTF1-positive ependymal cells of the organum vasculosum of the lamina terminalis are thought to give rise to chordoid glioma [6]. In contrast, the TTF1-negative AVP/OXT magnocellular region had previously lacked a tumor analog. Given both the uniform negativity in our limited series and the biology underpinning their putative magnocellular cell of origin, the utility of TTF1 as a diagnostic marker in SELN may warrant further study.

This study has important limitations. The molecularly defined cohort described here was assembled through multi-institutional collaboration, and a standardized comprehensive work-up was not feasible in every case. Although we report uniform TTF1 negativity across evaluable cases, variability in antibody clones (e.g., SPT24 vs 8G7G3/1) was not evaluated, and clone-dependent false-negative staining cannot be entirely excluded. Whole-transcriptome profiling was performed in only a small subset of cases, and the retrospective nature of the cohort precluded an exhaustive immunohistochemical panel. Future studies might include immunophenotypic comparison of SELN with its principal histologic mimics (ONB, null cell PitNET), as well as possible magnocellular-lineage markers such as SIM1 [48, 83]. Finally, despite locally aggressive imaging features observed in our series (cavernous sinus invasion, internal carotid artery encasement, obstructive hydrocephalus), consistent with prior reports [38, 86, 87], no SELN-related deaths occurred and disease-free outcomes were comparable to CN; nevertheless, the favorable overall survival trend compared to ONB (p = 0.099) requires validation given the substantially shorter median follow-up in our cohort.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Thomas Pearce: review and editing. Patrick J. Cimino: data collection and review. Calixto-Hope G. Lucas: data collection and review. Daniel Marker: data collection and review. Scott Kulich: data collection and review. John Skaugen: data collection and review. Julia Kofler: review and editing. Benjamin B. Cho: data collection and review. Kyle Kurek: data collection, clinical data contribution, review. Kyle Conway: data collection and review. Joshua Klonoski: data collection and review. Miguel Guzman Munoz: data collection and review. Sarra M. Belakhoua: data collection and review. Sofia Asioli: data collection and review. Naoko Inoshita: data collection and review. Omkar Singh: data collection and review. Zied Abdullaev: data collection and review. Katrin B. M. Frauenknecht: data collection and review. Arie Perry: data collection and review. Felipe Andreiuolo: data collection and review. Kenneth Aldape: data collection and review. Valérie Rigau: data collection and review. Romain Appay: data collection and review. Emmanuelle Uro Coste: data collection and review. Melike Pekmezci: data collection and review. Matija Snuderl: data collection and review. Caterina Giannini: data collection and review. Leonille Schweizer: data collection, clinical data contribution, and review. David Capper: data collection and review. Martha Quezado: data collection and review. M. Beatriz Lopes: data collection, clinical data contribution, review, and editing. Drew Pratt: conceptualization, methodology, formal analysis, writing—original draft, writing—review and editing, and supervision.

Data availability

Masked intensities for sellar neurocytoma samples have been made available for download at the Gene Expression Omnibus (GEO) repository under the accession number GSE331022 (https://www.ncbi.nlm.nih.gov/geo/).

Declarations

Conflict of interest

K.A.: Servier (Advisory Board). M.S.: scientific advisor and shareholder of Heidelberg Epignostix and Halo Dx, a scientific advisor of Arima Genomics, and InnoSIGN, consulted for Servier Pharmaceuticals, and received research funding from Lilly USA. D.C.: inventor on the patent ‘DNA-methylation based method for classifying tumor species’ (EP3268492B1) filed by Deutsches Krebsforschungszentrum Stiftung des öffentlichen Rechts and Ruprecht-Karls-Universität Heidelberg; shareholder in and co-founder of Heidelberg Epignostix GmbH. Other authors declare no conflict of interest.

Ethics statement

This study was approved by the National Institutes of Health Institutional Review Board (IRB ID: 000983/MOD013986), which granted a waiver of informed consent given the use of existing data, documents, records, or specimens and the minimal risk to subjects.