NF1 相关周围神经鞘瘤：胚系–体细胞相互作用塑造分子分化Germline-Somatic Interplay Shapes Molecular Divergence in Peripheral Nerve Sheath Tumors.

Scope: open-access full text (CC BY 4.0) from Modern Pathology, including 5 figures and 3 tables; supplementary material not included. Superscript numbers are the original reference numbers; see the original article for the reference list.

Abstract

Peripheral nerve sheath tumors (PNSTs) arising in neurofibromatosis type 1 (NF1) show marked histological and clinical heterogeneity, ranging from benign to malignant PNSTs. Although germline NF1 variants underlie tumor predisposition, the mechanisms driving phenotypic and clinical heterogeneity remain incompletely defined. We performed integrated germline and somatic analyses of 208 samples (74 blood samples and 134 tumors) from a primary cohort of 74 patients with NF1. Using a custom NF-focused sequencing panel, we investigated NF1 variant types, co-occurring oncogenic alterations, and affected domains to assess genotype–phenotype associations. Germline analysis identified 55 pathogenic NF1 variants, including eight unreported variants, with recurrent NF1 microdeletions involving SUZ12 in high-grade and premalignant tumors. Among tumors from individuals with germline NF1 variants, biallelic NF1 inactivation occurred in 90/124 tumors (72.6%), most frequently through loss of heterozygosity. In addition, high-grade PNSTs frequently harbored co-alterations in CDKN2A, SUZ12, EED, or TP53. Analysis of the distribution of somatic NF1 variants across functional domains further revealed significant enrichment in the GRD in benign tumors, whereas variants involving the LRD, including the Sec14-PH subdomain, showed a trend toward enrichment in premalignant and malignant tumors. Notably, multi-lesion analysis of spatially and temporally distinct tumors from 30 individuals revealed pronounced intrapatient somatic heterogeneity: despite a shared germline NF1 background, individual tumors typically carried distinct NF1 second-hit events, consistent with lesion-specific molecular divergence rather than a single shared somatic driver across lesions. In contrast, microdissected histologically distinct components within individual tumors shared underlying NF1 alterations and showed sequential acquisition of additional alterations involving CDKN2A and EED, supporting stepwise molecular evolution during malignant progression. Together, these data reveal lesion-specific molecular divergence across distinct tumors and stepwise molecular evolution within individual tumors, underscoring the value of lesion-specific genomic evaluation in NF1-associated PNSTs.

Introduction

Neurofibromatosis (NF) syndromes are characterized by multiple neural tumors, particularly Schwann cell neoplasms, which manifest in different locations ¹. These syndromes include neurofibromatosis type 1 (NF1) and schwannomatosis. Among them, NF1 is an autosomal dominant genetic disorder with an estimated incidence of approximately 1 in 3,000 individuals worldwide ^2-4. It is caused by pathogenic variants in the NF1 gene, a tumor suppressor gene located on chromosome 17q11.2. ^5,6.

NF1 is a multisystem disorder characterized by significant phenotypic variability, ranging from benign peripheral nerve sheath tumors (PNSTs) to more severe malignancies, including malignant peripheral nerve sheath tumors (MPNSTs) ⁷. Due to the systemic involvement and tumor-suppressive function of the NF1 gene, individuals with NF1 are prone to developing diverse tumor types, such as breast cancer, gastrointestinal stromal tumors (GISTs), and neuroendocrine neoplasms ⁸.

NF1-associated PNSTs encompass a wide range of cutaneous neurofibromas (CNFs), plexiform neurofibromas (PNFs), atypical neurofibromatous neoplasms of uncertain biologic potential (ANNUBP), and MPNSTs ⁹. Nearly all NF1 patients develop CNFs, which are confined to cutaneous tissue and may number in the hundreds or thousands ¹⁰. Unlike CNFs, the surgical removal of PNFs is challenging because PNFs occur deeper and involve multiple nerve bundles. The FDA approved the MAPK/ERK pathway (MEK) inhibitor, selumetinib, for the treatment of children with symptomatic inoperable PNF, based on a clinical trial demonstrating significant tumor volume reduction in children with NF1 ^1,11. However, MEK inhibitors are not curative, and PNFs carry a risk of malignant transformation to MPNSTs ⁷. Approximately 8–13% of individuals with NF1 develop MPNSTs during their lifetime ¹². Because of their highly aggressive behavior, MPNSTs are treated by surgical resection with wide margins ¹³, radiation, and chemotherapy in a primary or adjuvant setting. Despite these approaches, the local recurrence rates remain high (∼32–65%), highlighting an urgent need for more effective treatments ¹⁴.

NF1 is caused by pathogenic variants in the NF1 gene, which encodes neurofibromin, a tumor suppressor protein that negatively regulates the RAS/MAPK signaling pathway. More than 3,000 disease-causing NF1 variants have been reported ¹⁵, contributing to the marked genetic and clinical heterogeneity of the disease. These variants include nonsense, frameshift, splice-site, and missense variants, as well as multi-exonic deletions, which can reduce neurofibromin function. Neurofibromin harbors multiple functional domains that mediate interactions with various cellular components, potentially contributing to the diverse clinical manifestations of NF1 ^16,17. Recently, clinically relevant genotype–phenotype correlations have been identified ^18,19, such as an association between NF1 microdeletions and MPNST development and three hotspot regions within neurofibromin, including the GAP-related domain (GRD, residues 1198–1530), cysteine/serine-rich domain (CSRD, residues 543–909), and Armadillo1 domain within the C-terminal domain (CTD, residues 2260–2818) ²⁰. These findings suggest that the type and location of NF1 variants may contribute to the phenotypic heterogeneity of NF1-associated tumors.

Despite substantial progress in germline genetic testing for NF1 diagnosis, the molecular mechanisms underlying the remarkable histological and clinical heterogeneity of PNSTs in NF1 remain incompletely understood. Although a recent study suggested that somatic second-hit events may occur independently in spatially distinct lesions in NF1 ²¹, whether multiple tumors arising within an individual or among members of the same family share common somatic drivers or develop independently has not been systematically investigated. Furthermore, correlations between the histopathology of PNSTs and genetic alterations remain underexplored, aside from a limited number of established genotype–phenotype associations. Thus, beyond advances in diagnostic genetics, comprehensive, systematic acquisition of human tumor specimens, coupled with integrated molecular analyses, is essential to delineate disease mechanisms and facilitate the development of effective therapies for NF1-associated PNSTs ⁹. In this study, we performed integrated germline and somatic analyses in patients with clinically diagnosed NF1, combining genomic profiling with detailed histopathological classification. This approach allowed us to characterize domain-specific patterns of NF1 alterations and to demonstrate that NF1-associated PNSTs develop through divergent molecular pathways, despite a shared germline predisposition. Our findings provide further insight into NF1-associated tumorigenesis and highlight the therapeutic implications of lesion-specific genomic heterogeneity.

Materials & Methods

Patient Sample Collection

Participants clinically diagnosed with NF1 who underwent surgery at the Samsung Medical Center (SMC) in Seoul, Korea, between December 2017 and June 2022 were enrolled. A total of 74 blood samples and 134 tissue samples were subjected to targeted sequencing. Immediately after surgery, a portion of each resected specimen was separated by the surgeon in the operating room and stored at –80 °C as fresh-frozen tissue for NGS analysis. The remaining tissue was submitted to the Department of Pathology and processed into formalin-fixed paraffin-embedded (FFPE) blocks for histopathological evaluation. The histopathological diagnoses were confirmed using H&E-stained sections prepared from the corresponding FFPE tissue obtained from the same surgical specimen. Additional microdissection-based whole-exome sequencing (WES) was performed on six histologically distinct regions from two patients: one patient from the primary cohort and one additional patient who was included exclusively in the exploratory microdissection-based WES analysis. All studies were approved by the SMC Institutional Review Board (IRB Nos. 2013-09-025 and 2026-08-095), and written informed consent was obtained from all patients in the primary cohort. The requirement for informed consent was waived by the IRB for one additional patient who was retrospectively included in the study.

Pathology Review

Clinical and radiological data were extracted from patients' medical records. The collected clinical information included age, sex, tumor location, date of excision, family history, associated symptoms, medical history, computed tomography and magnetic resonance imaging findings, follow-up duration, treatment history, and pathological diagnosis. Two board-certified pathologists reviewed all available slides and classified the tumors according to the proposed nomenclature from the consensus meeting on the pathology of NF1-associated atypical nerve sheath tumors held in October 2016 at the National Cancer Institute/National Institutes of Health ²² and the 2020 World Health Organization (WHO) classification of bone and soft tissue tumors ²³. We classified NF1-associated nerve sheath tumors into the following categories: CNF, PNF, diffuse-type neurofibroma, neurofibroma with degenerative atypia, ANNUBP, low- and high-grade MPNST. CNFs were defined as benign, localized dermal neurofibromas. PNFs involved multiple nerve fascicles and exhibited a tortuous, multinodular growth pattern. Neurofibromas showing nuclear atypia, without hypercellularity or increased mitosis, were termed neurofibroma with degenerative atypia. Tumors exhibiting at least two of the four features—such as cytologic atypia, loss of neurofibroma architecture, hypercellularity, and a mitotic index >1/10 and < 3/10 high-power fields (HPFs)—were classified as ANNUBP. Low-grade MPNST (or ANNUBP with increased proliferation) showed ANNUBP features, but with mitotic figures (mf) of 3–9/10 HPFs and no necrosis. If an MPNST had 3–9 mf/10 HPFs with necrosis or at least 10 mf/10 HPFs, it was classified as a high-grade MPNST. Diffuse-type neurofibromas were retained as a separate subtype because they can clinically mimic large malignant tumors but exhibit histological features distinct from classical neurofibromas ^24,25. Tumors exhibiting both the plexiform and diffuse growth patterns were classified as mixed-type neurofibromas. In addition, slides from other concurrent tumors occurring in patients with NF1 were reviewed to confirm pathological diagnoses.

NF-panel Design and NGS Sequencing

To investigate genetic alterations in NF1-associated tumors, we designed a custom-targeted sequencing panel comprising 62 genes (Supplementary Table S1). Genes were selected based on their known involvement in the tumorigenesis of NF-associated tumors, including core components of the RAS/MAPK pathway^7,26 (e.g., NF1, NRAS, KRAS, HRAS, MAP2K1/2), cell cycle regulation (CDKN2A, TP53, CCND2), chromatin remodeling (SUZ12, EED, CHD4, SMARCB1), and other tumor suppressors or oncogenes that are frequently altered in peripheral nerve sheath tumors. Celemics Inc. (Seoul, Korea) manufactured the panel as a custom target-capture panel. Capture probes were designed to cover all coding exons of selected genes and the 50-bp intronic regions adjacent to exon boundaries. DNA was extracted from fresh-frozen tissue samples using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) and from blood samples using the QIAamp DNA Blood Maxi Kit (Qiagen). Library preparation followed standard enzymatic steps (end repair, A-tailing, adapter ligation, and PCR amplification) using a Celemics Library Preparation Kit (Celemics, Seoul, Korea). DNA libraries were hybridized with capture probes in buffer using a Celemics Target Enrichment Kit (Celemics, Korea). After hybridization and washing, the captured libraries underwent post-capture PCR amplification. Pooled libraries were sequenced on an Illumina NextSeq 500 Sequencing System (Illumina, San Diego, CA, USA) using 2 × 151-bp paired-end reads.

Bioinformatics Analysis of Targeted Sequencing Data

Raw FASTQ files were processed using AdapterRemoval and aligned to the human reference genome (GRCh37) using BWA-MEM. Duplicate reads were removed using Picard, and subsequent sequence processing was performed using GATK. Copy number variation (CNV) values were calculated by dividing the normalized regional depth of the test sample by the average normalized regional depth of the control samples. Short-base variants were identified using VarDict and annotated using SnpEff based on their genomic locations and predicted coding effects. We analyzed genomic variants, including SNPs, INDELs, MNPs, CNVs, and translocations, for clinical significance using the ClinVar database. Germline variants with a population frequency of <0.1% were reviewed, and somatic variants with variant allele frequency (VAF) >5% were selected. Loss of heterozygosity (LOH) was determined by an increase in the VAF of the germline pathogenic variant in tumor samples compared to that in matched blood samples.

Whole-exome Sequencing and Bioinformatics Analysis of Microdissected FFPE Samples

Additional WES was performed on six histologically distinct regions from two patients, including diffuse-type neurofibroma, ANNUBP, low-grade MPNST, and high-grade MPNST. The corresponding regions were manually microdissected from unstained FFPE tissue sections. Exome libraries were prepared from 100 ng of genomic DNA using the Agilent SureSelect XT HS/Low Input protocol and SureSelect Human All Exon V8 kit and sequenced with 151-bp paired-end reads by Macrogen (Seoul, Korea). After adapter trimming with AGeNT Trimmer (version 3.0.6), reads were aligned to the GRCh38 human reference genome using BWA-MEM (version 0.7.17). Molecular barcode–based duplicate processing and base quality score recalibration were subsequently performed. Somatic SNVs and small insertions/deletions were called using GATK Mutect2 in tumor-normal paired mode and annotated using SnpEff (version 5.2).

Statistical Analysis

Oncoprint visualization was generated using the PyComplexHeatmap (version 1.7.4) in Python. Graphical analyses were conducted using Seaborn (version 0.11.0), and schematic figures were created using BioRender.com. The frequency of domain distribution was evaluated using Fisher's exact test. A two-sided P value less than 0.05 was considered statistically significant. Statistical analyses were performed using Rex 3.6.0, an Excel-based statistical analysis software package (RexSoft, Korea; http://rexsoft.org/) and R ver. 4.0.0 (R Foundation for Statistical Computing, Austria).

Results

Sample Collection and Histological Classification for Targeted Sequencing

A total of 83 individuals with a clinical diagnosis of NF1 who underwent surgical resection between December 2017 and June 2022 were initially enrolled for this study (Fig. 1). Eight patients were excluded because of ineligibility or the unavailability of blood samples, and one patient harboring a germline NF2 pathogenic variant was excluded from the NF1 cohort. Finally, we performed germline NF panel sequencing on peripheral blood samples from 74 patients. (Supplementary Table S2). To identify somatic mutations, tumor specimens from 68 patients were subjected to panel sequencing. Among them, 38 patients had a single tumor specimen, and 30 patients had a total of 96 tumor samples obtained either from multiple anatomical sites during a single surgical procedure or from serial resections over time. Six patients without tumor tissue available for analysis underwent germline analysis alone and were excluded from somatic profiling (Supplementary Table S3).

The clinical information and histopathological classification of each tumor sample are provided in Supplementary Table S4. A histopathological review of 127 PNST cases demonstrated marked phenotypic heterogeneity, encompassing the full histological spectrum from benign neurofibromas to MPNSTs (Fig. 2). The tumor specimens included 61 CNFs, including two neurofibromas with degenerative atypia, 26 PNFs, 15 diffuse-type neurofibromas, eight mixed-type neurofibromas, nine ANNUBP, one low-grade MPNST, and seven high-grade MPNSTs. Seven cases could not be histologically reviewed because the corresponding H&E slides were unavailable. Other concurrent tumors in patients with NF1 included three meningiomas, three GISTs, two neuroendocrine tumors, and two breast carcinomas.

CNFs were the most prevalent subtype, characterized by low cellularity and bland spindle cells with wavy nuclei. PNFs exhibited expanded tortuous nerve fascicles within a loose myxoid matrix, resulting in a lobulated growth pattern (Fig. 2A). Several lesions showed unusual histological findings not indicative of premalignant transformation, including scattered bizarre nuclei consistent with degenerative atypia in neurofibromas (Fig. 2B) and poorly circumscribed dermal infiltration in diffuse-type neurofibromas (Fig. 2C). Diffuse-type neurofibromas exhibited eosinophilic lamellar structures resembling Meissner corpuscles (Fig.2D) ²⁴.

ANNUBPs exhibited increased cellularity (Fig. 2E), mild nuclear atypia (Fig. 2F), and focal loss of p16 immunohistochemistry expression in atypical cells (Supplementary Fig. S1A, B), while retaining an overall neurofibromatous architecture. Low-grade MPNSTs showed partial preservation of the underlying architecture with atypical spindle and epithelioid cells (Fig. 2G) and a low mitotic index of 3−9 mf/10 HPFs without necrosis (Fig. 2H). In contrast, high-grade MPNSTs demonstrated dense cellular fascicles with marked cytological atypia, necrosis, and brisk mitotic activity (Supplementary Fig. S1C, D). These findings illustrate the histopathological continuum of NF1-associated PNSTs and provide a basis for integrating molecular and histopathological analyses described below.

Germline Variant Landscape of NF1 Patients

Germline variant analysis was performed on blood-derived DNA from 74 individuals clinically diagnosed with NF1, identifying pathogenic variants in NF1 in 64 patients (86.5%). Notably, 17 cases were clustered into eight distinct familial groups, with affected members of each family harboring the same NF1 variant, indicating shared inheritance. After accounting for familial clustering, we identified 55 distinct NF1 variants (Table 1). Among the 55 non-redundant variants detected, 21 (47.7%) were identified in patients with a positive family history, whereas 23 (52.3%) were classified as de novo; 11 cases remained unverifiable owing to a lack of parental data. This is consistent with a previous study, which found that approximately half of affected individuals had NF1 due to a de novo NF1 disease-causing variant ⁷. Of note, eight variants (14.6%) were novel and had not been reported in the ClinVar database (ClinVar release: 2024-08-06). Among all identified pathogenic NF1 variants, the most frequent molecular subtype was splice-altering mutations, which accounted for 16 of 55 variants (29.1%) (Fig. 3A). In addition to canonical splice site disruptions, we identified a deep intronic variant in patient NF26, which introduced a cryptic exon (r.288_289ins288+1018_288+1135), resulting in a frameshift, consistent with a previously reported pathogenic mechanism ²⁷. Moreover, two exonic point mutations, c.1885G>A (NF52) and c.3304T>G (NF54), were predicted to generate novel cryptic splice sites within coding exons^28,29, likely leading to aberrant splicing. The remaining variants comprised 14 nonsense mutations, 14 small indel-induced frameshift mutations, five missense mutations, five large deletions, and one in-frame deletion. All variants in this cohort were classified as pathogenic or likely pathogenic according to the ACMG/AMP guidelines. A distinct subset of patients harbored NF1 microdeletions, defined as large deletions involving the NF1 locus and its adjacent genomic regions. These deletions are typically mediated by nonallelic homologous recombination between low-copy repeats. Two recurrent types of microdeletions have been described: type-1 deletion (∼1.4 Mb), flanked by NF1-REPa and NF1-REPc, and type-2 deletion (∼1.2 Mb), with breakpoints located within SUZ12 and its pseudogene SUZ12P ^30,31. In our cohort, copy number variation (CNV) analysis identified five large deletions involving the NF1 gene. Patient NF60 exhibited a focal deletion within exon 8, leading to exon skipping, whereas the remaining four patients showed broader CNV losses affecting the entire NF1 gene. Importantly, deletion boundaries varied among these patients, with NF1 microdeletions extending to SUZ12 exons 3 (NF04), 4 (NF11), 8 (NF12), and the entire SUZ12 locus (NF55) (Supplementary Fig. S2). Based on these boundaries, three of the four microdeletions (75%) were consistent with type-2 deletions, a proportion higher than the previously reported frequency of 11% ³². Among the four patients carrying large germline deletions encompassing both NF1 and SUZ12, two (NF12 and NF55) developed high-grade MPNSTs, and two developed ANNUBPs, supporting the elevated lifetime risk of malignant or premalignant lesions associated with NF1 microdeletions, as previously reported ^31,33.

^# Detected by cDNA sequencing, found through medical chart review; *, stop gain; fs, frameshift, NA, not available; WGD, whole gene deletion; P, pathogenic; LP, likely pathogenic; VUS, variant of uncertain significance

Genetic Features Underpinning Tumor Development in NF1 Patients

To explore the relationship between germline genetic alterations and the broader tumor spectrum in NF1, we integrated germline sequencing results with the clinicopathologic information of PNSTs and other concurrent tumors (Table 2). In total, NF1-associated premalignant/malignant tumors or other concurrent tumors were identified in 23 patients. These tumors included 15 MPNSTs (13 MPNSTs arising in patients with germline NF1 pathogenic variants and two MPNSTs arising in patients without detectable germline NF1 variants), three GISTs, two breast cancers, three meningiomas, and two neuroendocrine tumors. These findings underscore the diverse tumor spectrum observed in individuals with NF1. Whereas GIST and breast cancer mainly occurred in patients aged ≥60 years, patient NF05 developed a GIST before 60 years of age and carried pathogenic variants in both NF1 and A2ML1, which is a known gene implicated in RASopathies³⁴. A pediatric patient (NF70, <20 years of age) presented with multiple cutaneous CNFs and meningiomas. Targeted panel testing did not detect a pathogenic NF1 variant; however, sequencing revealed a germline TP53 nonsense mutation (c.1175C>A; p.Ser392*). The Ser392 residue has been implicated in modulating TP53 stability and transcriptional activity through phosphorylation ^35,36. Although this variant has not previously been reported to be associated with an NF1 phenotype, it may have contributed to the development of multiple tumors in this case.

*This column shows the histological types of tumors collected in this study, as well as concurrent NF-associated tumors per patient; cutaneous neurofibromas are not listed. PNF, plexiform neurofibromas; DNF, diffuse-type neurofibromas; MNF, mixed-type neurofibromas; CNS, central nervous system; ANNUBP, atypical neurofibromatous neoplasms of uncertain biologic potential; MPNST, malignant peripheral nerve sheath tumor. ^# Neurofibromin domains affected by germline pathogenic NF1 variants. ND, not detected; CSRD, cysteine-serine rich domain; TBD, tubulin-binding domain; GRD, GAP-related domain; PH, pleckstrin homologous domain; LRD, leucine-rich domain; CTD, C-terminal domain.

Clonal Divergence of Somatic NF1 Mutations Across Tumor Sites and Time Points

We investigated the somatic mutational landscape of NF1-associated tumors by analyzing 134 tumor samples from 68 patients. All germline NF1 pathogenic variants identified in matched blood samples were confirmed in the corresponding tumor tissues, indicating concordant germline transmission. In addition, somatic mutations leading to biallelic NF1 inactivation were detected in 90 of the 124 tumors (72.6%) that carried a matched germline NF1 variant. Among these, LOH represented the most frequent mechanism of somatic inactivation (Fig. 3B). Of note, biallelic NF1 inactivation was observed in two MPNSTs without germline NF1 variants, reinforcing the central role of NF1 loss in MPNST pathogenesis, even in the absence of a detectable germline predisposition.

We identified additional recurrent somatic alterations in premalignant and malignant lesions. Homozygous copy number loss of CDKN2A was observed in three MPNSTs and one ANNUBP with loss of p16 immunohistochemical expression (Supplementary Fig. S1A, B), consistent with its role in malignant tumor progression from premalignancy ³⁷. Furthermore, LOH of EED and somatic mutations in SUZ12 and TP53 were identified exclusively in MPNSTs and premalignant lesions, but not in benign neurofibromas (Fig. 3C).

To investigate intertumoral heterogeneity of tumor development in patients with NF1, we analyzed multiple anatomically and temporally distinct tumors in 30 patients (Supplementary Table S5). Fig. 3D represents a representative patient from whom multiple tumor samples were obtained at different anatomical sites and time points. Patient NF69 highlights the spatial and temporal heterogeneity of somatic NF1 alterations in the presence of a shared germline NF1 pathogenic variant (c.7348C>T). A total of eleven tumors were obtained between February 2021 and May 2022 and encompassed various histological subtypes, including CNF, PNF, diffuse-type neurofibroma, cutaneous neurofibroma with degenerative atypia, and ANNUBP. Somatic alterations differed among the tumor samples, with distinct mutations identified in nearly every lesion, including LOH, a splice-site variant (e.g., c.2990+1G>A), small deletions (e.g., c.5789_5790delGT), and duplication events (e.g., c.1882dupT). In higher-grade lesions, such as ANNUBP, co-occurring alterations were observed, including LOH of NF1 and homozygous deletion of CDKN2A, implicating additional disruption of tumor suppressor genes in disease progression.

In most cases, somatic NF1 alterations differed among tumors arising from the same individual, indicating temporally and spatially independent clonal origins. Despite sharing an identical germline NF1 variant, each tumor harbored a distinct somatic variant, resulting in independent biallelic inactivation events. Only two patients (NF13 and NF23) harbored identical NF1 variants across all sampled tumors; both patients underwent debulking surgery for large, anatomically contiguous lesions, suggesting that the sampled tissues likely originated from a single clone. These findings highlight the critical role of somatic NF1 inactivation as a major molecular event in NF1-associated tumorigenesis, and further support independent clonal tumor development in NF1, in which separate tumors arise under a shared germline predisposition.

Microdissection-based WES Reveals Stepwise Molecular Evolution During Malignant Progression

To further investigate intratumoral molecular evolution of NF1-associated PNSTs, we performed microdissection-based WES from two representative patients (Supplementary Table S6). A total of six regions representing benign neurofibroma, ANNUBP, low-grade MPNST, and high-grade MPNST components were separately analyzed (Fig. 4). In patient NF79, NF1 inactivation and CDKN2A deletion were shared across the analyzed components, followed by acquisition of an EED mutation in the low-grade MPNST and subsequent EED loss of heterozygosity in the high-grade MPNST. In patient NF101, NF1 loss of heterozygosity was present in the diffuse-type neurofibroma, ANNUBP, and MPNST components, whereas a CDKN2A alteration emerged in the ANNUBP and was retained in the MPNST, with an additional EED mutation detected in the MPNST. These findings suggest stepwise disruption of CDKN2A and the polycomb repressive complex 2 (PRC2) pathway during malignant progression.

Domain-specific Distribution of NF1 Variants by Histological Subtype

The distribution of NF1 variants identified in our cohort is presented in Fig. 5 and was analyzed with respect to germline and somatic mutations based on their localization within the functional domains of neurofibromin (Supplementary Fig. S3). The two hotspots, CSRD and CTD, showed a germline-enriched pattern, although not statistically significant (P = 0.08 and P = 0.11, respectively), and germline variants were relatively evenly distributed across all domains. When the CSRD and CTD domains were combined, they showed significant germline enrichment relative to all other domains (OR = 3.90, P = 0.006), and a slightly higher frequency of germline CSRD and CTD variants was observed among patients presenting with multiple NF1-associated or other concurrent tumors (Table 2). In contrast, somatic mutations were most enriched in another hotspot, GRD (2.4-fold) (P=0.008), and underrepresented in the CTD (0.25-fold) (Supplementary Fig. S3C).

We also integrated tumor histological types with the domain locations affected by mutations, as summarized in Table 3. Importantly, the GRD domain had the highest mutation frequency in benign tumors (23.3% vs. 0%, P = 0.035) (Supplementary Table S7). In contrast, LRD-domain variants containing the Sec14-PH subdomain were more frequently observed in premalignant/malignant lesions than in benign lesions (25.0% vs. 8.2%), although this difference did not reach statistical significance (P = 0.076). Furthermore, LOH in the NF1 gene was observed at a notable rate across all histological types.

*Neurofibromin domains affected by somatic pathogenic NF1 variants. CSRD, cysteine-serine rich domain; TBD, tubulin-binding domain; GRD, GAP-related domain; Sec, Sec14 homologous domain; PH, pleckstrin homologous domain; LRD, leucine-rich domain; CTD, C-terminal domain; LOH, loss of heterozygosity; ANNUBP, atypical neurofibromatous neoplasms of uncertain biologic potential; MPNST, malignant peripheral nerve sheath tumor.

Discussion

To our knowledge, this is one of the largest studies to integrate germline and somatic alterations across spatially and temporally distinct NF1-associated PNSTs, with detailed histopathologic correlation from benign CNFs to high-grade MPNSTs.

Domain-specific mapping of NF1 variants revealed the enrichment of somatic mutations in the GRD of benign tumors, suggesting its potential role in early tumorigenesis, whereas LRD (Sec14-PH)-domain variants showed a non-significant trend toward enrichment in premalignant and malignant lesions. Given the limited number of premalignant and malignant tumors in our cohort, larger studies are needed to determine whether domain-specific NF1 alterations contribute to malignant progression. From a therapeutic standpoint, these findings suggest that RAS/MAPK pathway inhibition may be particularly relevant in benign lesions such as PNFs. In contrast, malignant progression may require additional therapeutic strategies beyond MEK inhibition. A slightly higher frequency of germline CSRD and CTD variants in patients with multiple or concurrent tumors suggests a possible association with broader tumor susceptibility. However, this observation requires validation in larger cohorts.

Analysis of multiple tumors from the same individual demonstrated substantial somatic heterogeneity, with distinct tumors arising independently on a shared germline background. This finding supports lesion-specific molecular divergence rather than a common somatic driver across tumors within an individual. The occurrence of diverse malignant and benign neoplasms—including MPNST, GIST, breast cancer, meningioma, and neuroendocrine tumors—highlights the multisystemic nature of NF1-associated tumor predisposition.

Our results underscore the pivotal role of NF1 inactivation in the pathogenesis of PNSTs ^20,38. Consistent with prior findings, NF1 microdeletions encompassing SUZ12 were associated with increased malignant potential ³³. The observation of MPNST and ANNUBP in patients with germline deletions encompassing both NF1 and SUZ12 is consistent with previous evidence that large deletions confer an elevated risk of malignant or premalignant transformation. The identification of a germline TP53 nonsense variant in an NF1 panel-negative pediatric case suggests that secondary genetic events may contribute to tumor development in selected patients. Moreover, the co-occurrence of somatic alterations in genes regulating chromatin remodeling (SUZ12, EED) and cell cycle control (CDKN2A, TP53) in high-grade lesions underscores the multistep nature of NF1-associated tumorigenesis. Microdissection-based WES in this study further illustrates the molecular evolution across histologically distinct components from NF1-associated neurofibroma to MPNSTs. Notably, an additional alteration involving EED, a core component of PRC2, was identified in a more advanced tumor component, supporting its acquisition during progression. Overall, these data support the notion that both NF1 inactivation and additional somatic or germline alterations cooperate to shape the tumor spectrum and clinical outcomes in individuals with NF1.

Unlike earlier studies that primarily focused on either germline variants or individual tumor types, our study integrated germline and somatic data across multiple histological subtypes of spatially and temporally distinct tumors, revealing significant intra- and inter-patient heterogeneity. According to a recent genotype-based study ³⁹, NF1 pathogenic variants may be more prevalent than previously estimated. They may occur in individuals lacking classical clinical features, often presenting as somatic mosaicism or reduced penetrance. Consistent with these findings, updated genetic diagnostic criteria have been proposed, with strong consensus recommending testing in patients with segmental manifestations and in individuals with multiple bilateral spinal nerve neurofibromas without other NF1 features ¹⁸. Our identification of eight previously unreported germline variants expands the known mutational spectrum of NF1 and may contribute to improved molecular diagnosis.

Some limitations of this study should be acknowledged. First, the single-institution design may limit the generalizability of the findings. Although our cohort was relatively large for NF1-associated PNST research, the number of premalignant lesions, MPNSTs, and multifocal samples was limited. Second, we acknowledge that the recently updated international consensus recommendations for NF1-associated tumors ⁴⁰ no longer support the term "low-grade MPNST". One case in our cohort had originally been diagnosed as low-grade MPNST according to the 2020 WHO classification and the pathologic criteria available at the time. We retained the original diagnosis for consistency with the historical classification while recognizing that such lesions may be classified differently under current consensus recommendations. Third, although our targeted 62-gene panel is comprehensive for NF1-related tumorigenesis pathways, it may have missed additional genomic alterations outside the panel. We did not functionally validate the identified variants, and non-coding or structural variants beyond the panel may contribute to tumor phenotypes in panel-negative cases. Finally, the cohort included too few MPNSTs without germline NF1 pathogenic variants for a meaningful comparison with NF1-associated MPNSTs; therefore, we could not directly compare the molecular profiles of NF1-associated and sporadic MPNSTs. Future studies including larger numbers of both groups are warranted to define their shared and distinct genomic characteristics.

In conclusion, our findings have several important clinical implications. First, domain-specific enrichment of somatic NF1 variants may inform genotype–phenotype correlations and aid in risk stratification, if validated in larger cohorts. Second, the lesion-specific molecular divergence highlights the necessity for lesion-specific genomic profiling when establishing personalized therapeutic strategies, particularly in patients with multiple or high-grade tumors. Third, the stepwise acquisition of alterations in chromatin remodeling and cell cycle regulatory genes during malignant progression highlights potential therapeutic targets beyond MEK inhibition in high-risk NF1-associated tumors. Future multi-institutional studies incorporating spatially resolved genomic approaches and broader sequencing strategies, including whole-genome sequencing, are needed to validate genotype–phenotype correlations and identify additional drivers of tumorigenesis, including non-coding and structural variants. These insights may guide the development of novel therapeutic approaches that combine RAS/MAPK pathway inhibitors with other agents targeting high-risk NF1-associated tumors.