H3K27me3 expression and methylation status in histological variants of malignant peripheral nerve sheath tumours

Abstract Diagnosing MPNST can be challenging, but genetic alterations recently identified in polycomb repressive complex 2 (PRC2) core component genes, EED and SUZ12, resulting in global loss of the histone 3 lysine 27 trimethylation (H3K27me3) epigenetic mark, represent drivers of malignancy and a valuable diagnostic tool. However, the reported loss of H3K27me3 expression ranges from 35% to 84%. We show that advances in molecular pathology now allow many MPNST mimics to be classified confidently. We confirm that MPNSTs harbouring mutations in PRC2 core components are associated with loss of H3K27me3 expression; whole‐genome doubling was detected in 68%, and SSTR2 was amplified in 32% of MPNSTs. We demonstrate that loss of H3K27me3 expression occurs overall in 38% of MPNSTs, but is lost in 76% of histologically classical cases, whereas loss was detected in only 23% cases with heterologous elements and 14% where the diagnosis could not be provided on morphology alone. H3K27me3 loss is rarely seen in other high‐grade sarcomas and was not found to be associated with an inferior outcome in MPNST. We show that DNA methylation profiling distinguishes MPNST from its histological mimics, was unrelated to anatomical site, and formed two main clusters, MeGroups 4 and 5. MeGroup 4 represents classical MPNSTs lacking H3K27me3 expression in the majority of cases, whereas MeGroup 5 comprises MPNSTs exhibiting non‐classical histology and expressing H3K27me3 and cluster with undifferentiated sarcomas. The two MeGroups are distinguished by differentially methylated PRC2‐associated genes, the majority of which are hypermethylated in the promoter regions in MeGroup 4, indicating that the PRC2 target genes are not expressed in these tumours. The methylation profiles of MPNSTs with retention of H3K27me3 in MeGroups 4 and 5 are independent of mutations in PRC2 core components and the driver(s) in these groups remain to be identified. Our results open new avenues of investigation. © 2020 The Authors. The Journal of Pathology published by John Wiley & Sons, Ltd. on behalf of The Pathological Society of Great Britain and Ireland.


Introduction
Malignant peripheral nerve sheath tumours (MPNSTs) are characterised by nerve sheath differentiation and account for roughly 5% of all sarcoma subtypes [1]. Approximately 45% present in the context of neurofibromatosis type-1 (NF-1), a germline tumour of a monomorphic spindle cell sarcoma with alternating hypercellular and hypocellular areas and geographic tumour necrosis. Heterologous sarcomatous differentiation may also be seen, the most common being rhabdomyoblastic differentiation (Triton tumour) [1]. However, the histology is not always typical and MPNSTs may exhibit significant pleomorphism [3]. The differential diagnoses include, amongst others, spindle cell sarcoma not otherwise specified (NOS), synovial sarcoma, dermatofibrosarcoma protuberans, spindle cell rhabdomyosarcoma (SC-RMS), and melanoma [1].
Detection of somatic NF1 genetic alterations has been reported in sporadic and radiation-associated MPNST, but with variable detection rates ranging from 44% to 100% [4][5][6]. NF1 is a regulator of Ras protein activity and loss of function (LOF) of NF1 results in hyperactivation of the Ras pathway and excessive proliferation. MPNST is characterised by LOF of additional tumour suppressor loci, including p16 INK4A (CDKN2A) and TP53; however, these alterations are also seen in atypical neurofibroma, the presumed precursor of MPNST. In contrast, promotion of malignancy has been reported to be associated with recurrent somatic LOF alterations in genes encoding the polycomb repressive complex 2 (PRC2) core components, Embryonic Ectoderm Development (EED) and Suppressor of Zeste 12 (SUZ12). These have been identified in 23-70% of NF-1-associated, 36-92% of sporadic, and 90% of radiation-associated MPNSTs [4,5,7]. Subsequently, loss of H3K27me3 expression has been shown to correlate strongly with high-grade MPNST harbouring mutations in EED and SUZ12, consistent with the knowledge that the PRC2 complex trimethylates H3K27 [3]. However, in follow-up studies, the reported loss of H3K27me3 immunoreactivity has been variable, ranging from 34% [8] to 84% of cases [3,4,[9][10][11].
Röhrich et al reported that high-grade nerve sheath tumours clustered on the basis of their DNA methylation status, data that complemented the known role of the PRC2 complex in methylating H3K27. They identified two distinct clusters, the larger associated with loss of H3K27me3 immunoreactivity and the smaller group, sited in a paraspinal location, which retained expression of this protein [12]. Moreover, they and others demonstrated that loss of H3K27me3 immunoreactivity was valuable for distinguishing neurofibroma and low-grade MPNST from high-grade disease [3,4,8,9,13,14]. In this study, we investigate further the genomic and epigenomic landscape of MPNST and H3K27me3 expression in the context of morphologically distinct groups of MPNST.

Materials and methods
Ethical approval was obtained from the Cambridgeshire 2 Research Ethics Service (09/H0308/165).

Sample selection and identification of clinical information
All cases were retrieved from the Royal National Orthopaedic Hospital (RNOH) pathology archives between 1 January 2003 and 21 December 2015 (supplementary material, Table S1). A collection of undifferentiated pleomorphic sarcomas (UPSs), subjected to wholegenome sequencing (WGS) in a previous study, was also studied [15]. A spinal/paraspinal tumour was defined on imaging as arising from or involving a large spinal nerve route (supplementary material, Supplementary materials and methods).
Cases (totalling 176) where a diagnosis of MPNST had been considered in the differential diagnosis with a variable degree of confidence were selected for the study (by RT, DL, FA, and AMF). Additional IHC and molecular tests were performed where required to provide a more definitive diagnosis. Examples of the original reports of such cases included 'a spindle cell sarcoma NOS with features raising the possibility of a MPNST' or 'the diagnosis includes an MPNST with rhabdomyoblastic differentiation and an SC-RMS'. Following review, the diagnosis of the latter would have been confirmed by detection of the MYO-D1 single-nucleotide variant (SNV) p.L122R. Following this review, we classified the cases into ten histopathological groups (Table 1): 124 were considered to represent nerve sheath tumours in Histopathology Groups (HP Groups) 0, 1A, 1B, and 2 (

DNA sequencing
Whole-exome sequencing (WES) or WGS (n = 2) was performed on 51 of the 176 tumours for which frozen tumour and matched normal tissue were available (> 60% tumour-rich and < 20% necrotic). Table 2  EPIC array protocol and data analysis DNA methylation array assays were undertaken as published elsewhere [12]. DNA (500 ng) from frozen tumour samples was bisulphite-converted using a Zymo EZ DNA methylation Gold kit (Zymo Research Corp, Irvine, CA, USA) and hybridised to Infinium HumanMethylationEPIC BeadChip arrays (Illumina, San Diego, CA, USA). Details are provided in supplementary material, Supplementary materials and methods. RNOH and the Heidelberg group methylation array data [12] are available in the ArrayExpress database at EMBL-EBI under accession number E-MTAB-8864.

Statistical analysis
No statistical methods were used to predetermine sample size. All statistical analyses were performed using R [17]. Kaplan-Meier analysis was conducted using the R packages 'survminer' [18] and 'survival' [19], with survival time defined as the time from diagnosis to death or last follow-up: survival data were obtained in August 2019. P values less than 0.05 were considered statistically significant.

Diagnoses and genomic landscape of MPNST
The diagnosis of MPNST was excluded in 12/51 (27%) of the sequenced tumours following review of the pathology, new IHC, molecular tests, and clinical information. One sample (case 16, reclassified as UPS) failed the sequencing quality control ( Figure 1A, Table 2, and supplementary material, Table S1 show details of the refined diagnoses).

NF1 mutations in MPNST
Of 39 of the 51 sequenced tumours considered to represent MPNST, two were epithelioid MPNSTs and 16 were from patients exhibiting the clinical phenotype of NF-1, a finding substantiated by detection of germline NF1 mutations in 12/16 patients (Table 2, supplementary  material, Table S1, and Figure 1B). Nevertheless, the tumours in the four remaining patients (cases 4, 12, 21, and 23) were considered to represent bona fide MPNSTs on the following evidence: all had an NF-1 clinical phenotype; the tumour from case 12 arose from a large nerve and exhibited a low-grade component; and cases 21 and 23 contained a low-grade component, while case 4 revealed heterozygous alterations of core genes of PRC2. Of the remaining 21 tumours, three harboured biallelic somatic NF1 alterations; two had homozygous NF1 copy number loss (detected by visual inspection of read counts using the Integrative Genomics Viewer); three had non-synonymous NF1 mutations; and two had a heterozygous loss of NF1. In total, 27/37 (73%) MPNSTs had NF1 genetic alterations including loss of heterozygosity (LOH) ( Figure 1B and Table 2). Of the remaining ten MPNSTs in which an NF1 alteration was not detected, six exhibited typical morphological features (cases 1, 15, 18, 21, 28, and 36) and were categorised in HP Group 1A or 1B (supplementary material, Figure S1). Four tumours had a higher degree of nuclear pleomorphism than seen in the MPNST classical variant (HP Group 2), three of which (cases 19, 34, and 37) harboured LOH in EZH2. Neither NF1 nor PRC2 alterations were detected in case 26 (H3K27me3-retained), but this case arose from a paraspinal nerve, as did case 37.

Histopathological group (n)
Description HP Grp 0 (15) Benign nerve sheath tumours and MPNST grade 1 HP Grp 1A (46) Classical MPNST; fascicles with monomorphic spindle cells with alternating hypercellular and hypocellular areas and geographic tumour necrosis without heterologous elements HP Grp 1B (31) Classical MPNST; spindle cell sarcoma with heterologous element, e.g. rhabdomyoblastic, osteosarcoma; also, MPNST with a variable degree of cellular pleomorphism but in which a classical low-grade MPNST component was identified HP Grp 2 (41) MPNST in the absence of classical histological morphology, e.g. tumours exhibiting scattered atypia/pleomorphism, but where the diagnosis was supported either with additional phenotypic, genomic data pointing towards MPNST or arising from a large nerve HP Grp 3 (21) Spindle cell sarcoma NOS (DNA sequencing not available) which does not exhibit typical MPNST morphology and without other evidence, as listed in Grp 2, to support a diagnosis of MPNST HP Grp 4 (7) Sarcoma with specific diagnosis based on genetic alterations including SYT-SSX, COL1A1-PDGFB, and others HP Grp 5 (4) Epithelioid MPNST requiring loss of INI-1 HP Grp 6 (1) Undifferentiated pleomorphic sarcoma (UPS); tumour set had undergone WGS [15] HP Grp 7 (5) Spindle cell/sclerosing-RMS (SC-RMS) with MYOD1 mutation HP Grp 8 (5) Melanoma supported by publicly available genomic data [29] H3K27me3, DNA methylation, and histological variants in MPNST 153   Table S5).

At least 50% of MPNSTs have alterations in PRC2related genes
DNA sequencing of MPNSTs revealed nonsynonymous somatic mutations in EED (4/37; 11%) and SUZ12 (2/37; 5%). None of the 37 tumours displayed somatic mutations in EZH2. LOH was observed in EED in 12/37 (32%), in SUZ12 in 16/37 (43%), and in EZH2 in 10/37 (27%) cases ( Figure 1B), and TP53 was lost in 17/37 (46%) cases. Two tumours, cases 20 and 40, had a somatic mutation in addition to LOH (biallelic alterations) in EED and SUZ12, respectively. In total, 25/37 (68%) cases had at least one genomic aberration in EED, SUZ12 or EZH2 (including LOH) ( Table 3 and Figure 1B). Alterations in PRC2 core components were detected in 13/16 (81%) tumours in the setting of NF-1 and in 12/21 (57%) sporadic tumours. For completeness, visual inspection of other PRC2 complex genes, EPC1, AEBP2, and RBBP7, was undertaken but revealed no non-synonymous mutations. One case had a missense mutation in the H3K27me3 demethylase KDM6B (case 32, supplementary material, Table S2). A driver mutation analysis (supplementary material, Supplementary materials and methods) confirmed the importance of the PRC2 genes in MPNST by classifying EED and NF1 as significant driver genes. We investigated how H3K27me3 immunoreactivity correlated with the genomic data. Loss of the protein was observed in all cases (n = 6) in which an SNV in any of the PRC2 genes was detected (SUZ12, EED; p = 0.004, chi-square independence test). However, eight cases showed loss of H3K27me3 without an SNV being detected at these genes, although five of these had LOH in either SUZ12 or EED. In contrast, immunoreactivity was not associated with LOH of EZH2. In summary, loss of H3K27me3 expression was observed in 50% (n = 16) of NF-1-associated MPNSTs, 25% (n = 20) of sporadic MPNSTs, and in the single radiation-associated MPNST (case 35, Table 3), with an overall loss of H3K27me3 in 38% (14/37) of sequenced MPNSTs, including two grade 1 cases.

MPNST histology correlates with H3K27me3 immunoreactivity
To investigate the variable loss of H3K27me3 immunoreactivity reported in previous studies [3,4,8,9,11,13], we extended our tumour set to 176 cases, including the above-mentioned 51 sequenced cases. Following a review of the pathology, combined with additional molecular diagnostic testing and clinical information, we refined our original diagnoses in 30% (n = 52/176) of cases (Figure 2A and supplementary material, Table S1). Of the remaining 124 MPNSTs, 38% (n = 47/124) of MPNSTs revealed loss of H3K27me3 immunoreactivity. However, when classified according to the HP groups, we found that loss of H3K27me3   Table S1 and Figure S3) In view of the demethylation role of KDM6A/B, these loci were interrogated but revealed no evidence of biallelic inactivation; no association was found between the loss of copy number of these genes and the expression of H3K27me3 (supplementary material, Table S1).

Nerve sheath tumours are characterised by distinct methylation patterns
Loss of H3K27me3 expression caused by genetic alterations of the PRC2 genes is associated with a global change in methylation levels, a finding reflected in the Heidelberg publication [12]. This prompted us to analyse our DNA methylation profiles from the 102 cases from which DNA was available. We included the DNA methylation profiles of 53 previously reported UPS cases [15], thereby allowing comparison with

156
I Lyskjaer et al  H3K27me3, DNA methylation, and histological variants in MPNST 157 29 high-grade sarcomas NOS included in our 176 cases (supplementary material, Figure S4) and the Heidelberg methylation dataset: in total, a dataset from 309 cases. Unsupervised hierarchical clustering of the combined RNOH and Heidelberg cases revealed six methylation groups (MeGroups 1-6, Figure 3A and supplementary material, Table S7 Table S7).
Finally, the methylation data available from two of our five reclassified malignant melanoma cases were studied in combination with the TCGA melanoma methylation data. A t-SNE analysis supported a diagnosis of melanoma (supplementary material, Figure S6).

Polycomb target genes are differentially methylated between the two MPNST methylation groups
We analysed the differentially methylated regions (DMRs) and probes (DMPs) between MPNSTs in MeGroup 4 (n = 50) and MeGroup 5 (n = 33), of which 12 and 15 had been exome-sequenced, respectively. Methylation analysis revealed a total of 606 DMRs and 123,950 DMPs between MPNSTs in these two groups, of which 85,601 showed increased methylation in MeGroup 4 (supplementary material, Tables S8 and S9). Gene Set Enrichment Analysis (GSEA) showed strong enrichment for pathways associated with PRC2 biology (supplementary material, Table S10).
Analysis revealed that 1,200 of 1,350 (89%) possible PRC2 target genes (supplementary material, Supplementary materials and methods) were differentially methylated between the two MeGroups, 65% of which (784/1,200 target genes) were hypermethylated in the promoter regions (200 or 1,500 bp upstream of the transcription start site, TSS200 and TSS1500, respectively), indicating that the PRC2 target genes are not expressed in MeGroup 4. Analysis of the exome profiles failed to identify alterations that distinguished cases from the two MeGroups.

I Lyskjaer et al
Finally, when comparing only cases with retained H3K27me3 from MeGroup 4 (n = 11) and MeGroup 5 (n = 32), we found that 269 DMRs and 10,862 DMPs significantly separated these cases (supplementary material, Tables S11 and S12). 9042/10,862 DMPs were more methylated in the H3K27me3-positive cases in MeGroup 4 than in MeGroup 5, and 9,615 of these 10 862 DMPs were also differentially methylated in samples in the two groups. Of the 1,350 PRC2 target genes, the majority (89%) were differentially methylated between all samples in the two MeGroups, whereas only 409 were differentially methylated, when analysing only those samples which retained H3K27me3, of which 191 had hypermethylation of promoter regions. GSEA showed strong enrichment for terms associated MPNST with heterologous rhabdomyoblastic differentiation (Triton tumour). Note the plump cells with brightly eosinophilic cytoplasm and striations, indicative of a rhabdomyoblastic phenotype. These cells were strongly immunoreactive for desmin and myogenin (H&E, ×10). HP Grp 2: case 51. A spindle cell sarcoma arising from a nerve in a patient with neurofibromatosis type 1. The morphological features are not entirely classical but the associated clinicopathological features support the diagnosis (H&E, ×10). HP Grp 3: case 6. Spindle cell sarcoma, NOS. The morphological features are not typical for MPNST and there was no other supporting evidence to justify the diagnosis of MPNST (H&E, ×10). HP Grp 4: case 83. A tumour with definitive molecular evidence to exclude the diagnosis of MPNST. This case is a dermatofibrosarcoma protuberans with fibrosarcomatous transformation, confirmed by the presence of a COL1A1-PDGFB fusion on FISH (H&E, ×10). HP Grp 5: case 116. A tumour composed of sheets of cells with an epithelioid morphology. This tumour was strongly positive for S100 and has loss of INI-1 (H&E, ×10). HP Grp 6: a tumour composed of pleomorphic spindle cells and an undifferentiated immunophenotype, confirming the diagnosis of undifferentiated pleomorphic sarcoma (H&E, ×10). HP Grp 7: case 7. A tumour composed of monomorphic spindle cells with eosinophilic cytoplasm, embedded in a densely collagenous stroma. This tumour was strongly positive for desmin and myogenin, and had a MYOD1 p.L122R missense mutation, confirming the diagnosis of spindle cell/sclerosing rhabdomyosarcoma (H&E, ×10). HP Grp 8: case 17. A pleomorphic spindle cell tumour, focally positive for S100 and negative for melan-A and HMB45 with a mutational signature attributable to UV light, supporting the diagnosis of malignant melanoma (H&E, ×10).

H3K27me3 and genome doubling (GD) status as prognostic markers of survival
No correlation was found between survival and patients in MeGroups 4 (n = 28) and 5 (n = 25) (p = 0.71, supplementary material, Figure S7A). There was also no association between H3K27me3 immunoreactivity status and survival in MeGroups 4 and 5 (p = 0.88, supplementary material, Figure S7B), a finding validated on extending the analysis to 100 RNOH patients with high-grade MPNST for which survival data were available (p = 0.94, supplementary material, Figure S8 and Table S14). Given a recently published paper highlighting that GD is associated with poor prognosis across cancer types [23], we asked if such genomic events in MPNST correlated with clinical outcome. This analysis was performed on the 37 sequenced MPNSTs: GD was observed in neither of the two cases in HP Group 0, in 91% (10/11) of HP Group 1A, in 55% (6/11) of HP Group 1B, and in 69% (9/13) of HP Group 2. Despite a strong tendency, GD status and survival did not reach statistical significance (p = 0.06, supplementary material, Figure S9 and Table S14).

A subset of MPNSTs express SSTR2
In view of the lack of effective treatments for MPNST, we sought potential therapeutic targets. Our genomic data revealed that SSTR2 was amplified in 32% of cases and FISH confirmed amplification in 8/11 of these cases (supplementary material, Table S15 and Figure S10). The discrepancy in the three cases may be accounted

160
I Lyskjaer et al for by tumour heterogeneity; moreover, these cases were also aneuploid, making interpretation of the FISH challenging. Two of ten cases for which RNA-sequencing data were available showed increased expression of SSTR2 transcripts compared with the other eight cases (supplementary material, Figure S11). However, protein expression (SSTR2 IHC, supplementary material, Figures S12 and S13) showed little correlation with amplification on FISH or sequencing (supplementary material, Table S15), with only 12/55 (22%) cases being immunoreactive for SSTR2, only four of which were SSTR2amplified on WES and five by FISH. The expression of SSTR2 protein was neither associated with any of the HP Groups nor associated with the MeGroups (supplementary material, Table S15; SSTR2 IHC, Figure S14).

Discussion
In this study of well-characterised high-grade MPNSTs, we find that loss of H3K27me3 expression, occurring in approximately 40% of cases, is lower than reported by most groups, but similar to that reported by Bovée and co-workers [8]. We identified that H3K27me3 expression was largely dictated by the morphology of the tumours: 76% of the MPNSTs exhibiting the classical histological features (HP Group 1A) revealed loss of H3K27me3 expression, whereas MPNSTs with heterologous elements or with a low-grade component (HP Group 1B) revealed loss of H3K23me3 in just 23% of cases. Lastly, in those MPNSTs where a diagnosis could not be proffered confidently on morphology alone (HP Group 2), H3K27me3 loss was detected in only 14% of cases. We conclude that the variation of reported loss of H3K27me3 immunoreactivity is dictated by the different histological subgroups within MPNSTs, and propose that the reported differences in the frequency of H3K27me3 loss by others could be explained by the number of these subtypes represented in the different studies [3,4,[9][10][11]. Although our findings show that loss of expression of this marker is highly specific for MPNST, we consider that its sensitivity is variable and dependent on subtypes described herein. Consistent with previous publications, all of our radiation-associated MPNST cases showed loss of H3K27me3 expression [3]. However, in contrast to other reports [3,12], we did not find any association between H3K27me3 loss with germline or sporadic cases, and six of our paraspinal cases revealed H3K27me3 loss, three of which were associated with radiation therapy. Differences in H3K27me3 immunohistochemistry results across laboratories could reflect how the IHC was interpreted: 'partial loss' of the protein has been reported [13], whereas we reported cases as 'positive' or 'negative'. However, the number of cases reported with partial loss in the literature is unlikely to account fully for the wide variability in the published results.
Röhrich et al were the first to report the association between the methylation status of H3K27me3 and the immunoreactivity of this chromatin mark in MPNST. Our data confirm previous publications linking genetic alterations in one of the PRC2 complex components to tri-methylation of lysine 27 and loss of the protein. Methylation analysis of our combined data revealed, as reported previously [12], a distinct cluster containing atypical neurofibroma and low-grade MPNSTs (MeGroup 1), although six high-grade sarcomas were also in this cluster, highlighting that optimisation of the classifier is required prior to introducing methylation profiling into a diagnostic clinical service. The majority of high-grade MPNSTs formed two methylation groups, MeGroups 4 and 5, with Heidelberg and RNOH cases distributed across both MeGroups. MeGroup 4, composed largely of MPNSTs with classical features (HP Group 1A), was characterised by hypermethylation of PRC2 target genes and loss of H3K27me3 immunoreactivity in 78% of cases. In contrast, MeGroup 5 was composed of HP Group 2 cases, of which 94% retained H3K27me3 expression. Notably, most paraspinal MPNSTs clustered together within MeGroup 5, but did not form a distinct cluster as previously reported [12].
It is noteworthy that 30% of cases in MeGroup 4, which is composed largely of the classical variant of MPNSTs, show retention of H3K27me3 positivity. It is remarkable that these cases also showed a strong enrichment for GSEA terms associated with PRC2 biology similar to those cases in which H3K27me3 expression was lost. We speculate that this finding could be explained by activation of the non-canonical PRC2 pathway, a biological process that is well described in the regulation of gene transcription [24]. For example, there is considerable evidence that EZH2, the enzymatic subunit of PRC2, can act independently of the PRC2 complex, and in an independent enzymatic manner, to bring about transcriptional regulation through a variety of epigenetic mechanisms. This has in some cases been shown to involve non-core PRC2 proteins such as AE binding protein 2, JARID2, PHD finger protein 19, and long non-coding RNAs including HOTAIR and micro-RNAs, some of which have already been implicated in dysregulation of gene transcription in disease [25,26].
Virtually all MPNSTs in MeGroup 5, in contrast to MeGroup 4, expressed H3K27me3 and the morphology of the two groups is striking. MeGroup 5 tumours are often less clearly recognised as nerve sheath in origin. Furthermore, the clustering of these MPNSTs with UPS in MeGroup 5 underscores the morphological spectrum of tumours in this MeGroup and the difference to those in MeGroup 4. Perturbation of physiological and developmental processes that are normally tightly co-ordinated through the regulation of expression and repression of genes forming PRC2, a multisubunit epigenetic protein complex, could potentially account for these findings [27]. It is therefore interesting to speculate that the differences in MPNSTs in MeGroups 4 and 5 reflect their cell of origin and path of differentiation. However, the factors that determine the development of nerve sheath tumours rather than UPS remain unknown. Furthermore, the molecular H3K27me3, DNA methylation, and histological variants in MPNST 161 events that promote transformation of MPNST to highgrade disease in MeGroup 5 remain to be determined. The detection of SSTR2 amplification in 30% of our sequenced MPNSTs, which is supported by FISH and RNA sequencing but not immunoreactivity, is novel and of clinical interest, because somatostatin receptor antagonists are available in clinical practice [28]. However, the biological impact of SSTR2 is unclear; with evidence in the literature supporting both a pro-tumour survival signal and cancer growth inhibition [29]. In small cell lung cancer, there is an association between SSTR2 expression and poor outcome [29], whereas in this small cohort of patients with MPNSTs, we found no link between outcome and SSTR2 amplification status.
We have shown that reaching a diagnosis of MPNST is challenging, but accuracy is improved by excluding mimics with recognised molecular hallmarks, such as SYT-SSX fusion genes in synovial sarcoma. We consider that the most rigorous means of classifying MPNST is identification of biallelic NF1 alterations, which although not specific for MPNST, are considered diagnostic in the context of the relevant clinical/pathology context. However, the identification of all NF1 functional alterations is challenging because of the large gene size (60 exons), the absence of mutational hotspots, the presence of large structural alterations, and NF1 germline cryptic splice site variants [30], and is best achieved using WGS and RNA sequencing. Hence, despite our best efforts to classify MPNSTs correctly using WES, we accept that a rare case may not be an MPNST, and that occasional MPNSTs may have been classified as sarcoma NOS. The use of WES in this study is also likely to explain why NF1 germline alterations were identified in only 71% of patients with the NF-1 phenotype. Future large-scale comprehensive multi-omics analyses, and functional studies of MPNSTs and those suspected of being MPNST should provide a strong evidence base for classifying this disease and understanding its pathogenesis [31].