Cdkn2a Loss in a Model of Neurofibroma Demonstrates Stepwise Tumor Progression to Atypical Neurofibroma and MPNST

Neurofibromatosis type 1 (NF1) is an autosomal dominant tumor predisposition syndrome that affects about 1:3,000 individuals worldwide (1). The NF1 gene encodes neurofibromin, a GTPase-activating protein (GAP) that accelerates the intrinsic GAP activity of Ras proteins (2). Therefore, after cell activation, loss of NF1 correlates with increased signaling through the RAS pathway, a key driver of cancer (3). Nearly all (90%) of NF1 patients develop small benign cutaneous neurofibromas. Plexiform neurofibromas occur in 25%–50% of paients with NF1 and are histologically benign peripheral nerve sheath tumors (PNST) associated with nerve trunks that can cause substantial morbidity including pain and neurologic deficit (4, 5). Surgical resection is often impossible due to tissue invasion, large size, and/or association with critical peripheral nerves (5).

Plexiform neurofibromas can serve as precursor lesions for atypical neurofibromas (ANF), which can be symptomatic. Imaging studies show most atypical neurofibromas as distinct nodules that may be fluorodeoxyglucose–PET avid and/or painful (6). Clinically, atypical neurofibromas are frequently slow-growing tumors, and many atypical neurofibromas, unlike plexiform neurofibromas, show genomic loss at the CDKN2A locus (7). At least some atypical neurofibromas are atypical neurofibromatous neoplasms of uncertain biologic potential (ANNUBP), diagnosed by histology as hypercellular PNSTs composed of cells that also show nuclear atypia and/or mitoses (8). Atypical neurofibromas show few mutations apart from those in NF1 and CDKN2A/B (9).

Atypical neurofibromas themselves may be precursor lesions to malignant PNSTs (MPNST). In a recent study, 4 of 18 unresected atypical neurofibromas transformed to MPNST over an interval studied; others developed MPNST distant from a resected atypical neurofibromas (6). MPNST are highly aggressive soft-tissue sarcomas incurable with chemotherapy, for which the only effective treatment is complete surgical resection (6, 10). MPNST is the leading cause of early death in adults with NF1, and better therapies are desperately needed.

NF1 is a tumor suppressor gene on human chromosome 17q11. Loss of function in both NF1 alleles is detected in cutaneous and plexiform neurofibromas Schwann cells (SC; refs. 11–13). Although cutaneous and plexiform neurofibromas are composed of SCs, mast cells, fibroblasts, endothelial cells, and macrophages (14, 15), only SCs show biallelic NF1 alterations (12, 16). SC loss of NF1 function drives neurofibroma formation, as in genetically engineered mice (GEM) conditional loss of Nf1 in the SC lineage, driven by any of several different Cre driver alleles, results in neurofibroma formation (17–19). In the Nf1^fl/fl;DhhCre model, loss of Nf1 in SCs at embryonic day 12.5 is sufficient for benign plexiform neurofibromas formation, and in this GEM, transformation to more aggressive GEM-ANF or GEM-PNST is not observed (18).

Most atypical neurofibromas and 70% of MPNST show genetic heterozygous or homozygous loss of cyclin-dependent kinase inhibitor 2A gene (CDKN2A in human; Ink4a/Arf in mouse) (7, 9). The CDKN2A gene located at 9p21 encodes two proteins, p16^INK4a and p19^ARF (p14 in human; refs. 20, 21). The p16^INK4a protein inhibits the CDK4–6 kinases, maintaining the Rb protein in an unphosphorylated, growth-suppressive state, arresting the cell cycle (20, 22). p19^ARF binds the double minute 2 homolog, thereby stabilizing TP53, arresting cell proliferation, or leading to apoptosis (22–24). Loss of Ink4a/Arf can also compromise the senescence checkpoint, facilitating transformation to malignancy (25); senescent cells in cutaneous neurofibromas were postulated to contribute to their slow, benign, growth (26). Supporting a role in nerve tumor progression, Ink4a/Arf−/−;Nf1+/− mice (26%) developed GEM-PNST (27). Also, viral delivery of Cre recombinase into Ink4a/Arf^fl/fl;Nf1^fl/fl peripheral nerve led to aggressive nerve tumors (28). In these models, neither plexiform neurofibromas nor atypical neurofibromas were reported, suggesting that different precursor cells are affected by loss of Nf1 or Ink4a/Arf to drive the tumor types, and/or that simultaneous loss of both genes bypasses benign tumor formation. In mice expressing periostin-Cre;Nf1^fl/fl and Arf^fl alleles beginning at embryonic day 10 in early SC precursors (SCP), rare atypical neurofibromas formed and mice died subsequent to GEM-PNST, correlating with loss of the senescence checkpoint (29).

Here, we imposed loss of Ink4a/Arf in mice that develop paraspinal neurofibroma. Half of these tumors in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice were GEM-neurofibroma (plexiform neurofibromas), and half GEM-ANF (atypical neurofibromas). High-grade GEM-PNST formed at distant sites. Nerves in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice show reduced senescence, and tumors show RNA expression similar to their human counterparts. Transplantation of atypical neurofibromas into immunocompromised and immunocompetent mice results in GEM-PNST, providing new progression models useful for molecular, prevention, and treatment studies.

All experimental procedures were approved by the Institutional Animal Care and Use Committee (IACUC) at Cincinnati Children's Hospital Medical Center (Cincinnati, OH). Mice were housed in temperature and humidity-controlled facilities with free access to food and water, on a 12-hour light–dark cycle. Ink4a/Arf−/− mice were from the B6.129-Cdkn2a^tm1Rdp NCI; MMHCC strain #01XB1 (RRID:IMRS_NCIMR:01XB1), backcrossed to C57BL/6J females for five generations before generating compound mutant mice (30). We used Nf1^fl/fl;DhhCre (18), athymic nude (Envigo, 069), and C57BL/6J UBC GFP (Jackson 007076) mice. The DhhCre allele was maintained on the male. Except for athymic nudes, mice were maintained on the C57BL/6J background.

We performed Nf1 genotyping using oligonucleotides (5′ to 3′) CTTCAGACTGATTGTTGTACCTGA and ACCTCTCTAGCCTCAGGAATGA to detect the wild-type (WT) allele, and TGATTCCCACTTTGTGGTTCTAAG to detect the targeted allele. Ink4a/Arf mice (B6.129-Cdkn2a^tm1Rdp/NCI) were genotyped according to Mouse repository MMHCC strain #01XB1. DhhCre genotyping was performed using the forward primer ACCCTGTTACGTATAGCCGA and reverse CTCCGGTATTGAAACTCCAG.

We perfused mice intracardially with Karnovsky fixation solution [4% paraformaldehyde (PFA) and 3% glutaraldehyde in 0.1 mol/L phosphate buffer, pH 7.4–7.6]. Saphenous nerves were dissected, embedded, and ultrathin plastic sections were evaluated as described previously (31).

We homogenized tumor tissue in ice-cold buffer: 50 mmol/L Tris-Cl, 150 mmol/L NaCl, 0.1% SDS, 0.5% sodium deoxycholate, and 1% Triton x-100 (RIPA) containing protease/phosphatase inhibitor (Thermo Fisher Scientific, A32959). We measured protein concentration using a FlexStation3 molecular device, and denatured lysates in 5X sample buffer containing 10% 5 mol/L dithiothreitol (Sigma-Aldrich, 10197777001). We loaded samples into 4%–20% Criterion TGX precast gels (Bio-Rad, 5671093). We transferred gels to PVDF membrane (Millipore, IPFL00010), which was blocked in 5% blotting grade blocker (Bio-Rad, 1706404) in 0.1 mol/L TBS with 0.1% Tween, then incubated with anti-p16 antibody (Invitrogen, MA5–17142, 1:1,000), phospho-Rb (Ser807/811) antibody (Cell Signaling Technology, 8516S, 1:1,000), or horseradish peroxidase (HRP)-conjugated GAPDH antibody (Cell Signaling Technology, 3683S) overnight at 4°C. Secondary anti-mouse (Cell Signaling Technology, 7076S, 1:1,000) or anti-rabbit (Cell Signaling Technology, 7074S, 1:1,000) IgG was added for 1 hour at room temperature prior to Immobilon Western Chemiluminescent HRP Substrate (Millipore, WBKLS0500), and chemiluminescence imaging on an Azure Biosystems c500 chemi doc gel imager.

Genomic DNA was isolated from tumors with the DNeasy blood and tissue Kit (Qiagen, 69504) and amplified using TaqMan genotyping primers (4400291) for Ink4a/Arf gene expression. qPCR was performed using a Quant Studio 6 Flex Real-Time PCR system by Applied Biosystems. Each sample was run twice, with four technical replicates each.

MRI was performed on anesthetized Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice of both sexes on a 7T Bruker Biospec, and tumor volume quantified as described previously (32).

We sacrificed mice via isoflurane inhalation, removed tumors, and centrifuged them in L-15 medium (Corning Cellgro, MT-10–045-CV; 1,000 g at 5 minutes at room temperature). Tumors were placed into fresh L-15, chopped into pieces <1 mm³, and dissociated in L-15 containing 1% penicillin/streptomycin (Gibco, SV30010), Collegenase Type I (Worthington, LS004196), and Dispase Protease II (Gibco, 04942078001; 37°C at 170 rpm for 2–4 hours). DMEM (Gibco, 11965118) with 10% FBS (Gemini bioproducts, 100–500) was then added, and tumor cell suspensions collected by centrifugation (1,700 g for 5 minutes). Pellets were resuspended in DMEM with 10% FBS, filtered with a 40 μm BD Falcon Cell Strainer (catalog no. C4040), washed and resuspended in serum-free complete mouse sphere medium (27). Cells from each tumor were separately injected into 1–3 athymic nude mice (Envigo, 069) or into immunocompetent C57BL/6J GFP-expressing mice, at approximately 1 × 10⁵ cells/mouse in two-thirds complete mouse sphere media and one-third matrigel (BD Biosciences, 354234), in 150 μL. We measured tumors twice weekly with digital calipers (Thermo Fisher Scientific), measured volumes by (length × width²) × (π/6), and sacrificed mice when tumors approached 2,000 mm³ per IACUC standards.

We excised tumors from anesthetized mice, fixed them in 4% PFA overnight, dehydrated, and embedded in paraffin. Tumor paraffin sections (4 μm) were stained with hematoxylin and eosin (Richard-Allan ScientificSeries Hemotoxylin 1 7221 and Eosin-Y 7111, Thermo Fisher Scientific), Toluidine Blue (Thermo Fisher Scientific, T-161, 1:1,000), anti-S100 (Dako, Z0311, 1:10,000), Ki67 (Cell Signaling Technology, 12202, 1:250), or anti-neurofilament (Biolegend, SM312, 1:2,000). Images were captured with a SPOT Insight 4 Mp CCD camera (Spot Imaging).

We homogenized frozen tumors in Qiazol (Qiagen, 79306), isolated RNA using RNeasy Mini Kits (Qiagen, 74104), and sequenced bulk RNA using an Illumina NovaSeq 6000 (RNA polyA stranded, paired 100 bases, read depth 20M). We calculated raw read counts from BAM files with featureCounts and normalized them using Bioconductor-edgeR trimmed mean of M-values (TMM)-normalization (https://bioconductor.org/packages/release/bioc/html/edgeR.html). We performed principal component analysis (PCA) and Clara clustering analysis using R-cluster on whole transcriptomes and used R-fatctoextra to predict optimal cluster numbers. Hierarchical clustering (HC) analysis used 5,000 multiscale bootstrap resampling with R-pvclust (https://cran.r-project.org/web/packages/pvclust). We calculated Euclidean distances among samples, then built final hierarchical clusters by the average-linkage method for each cluster, and computed an “approximately unbiased (AU)” value by 5,000 multiscale bootstrap resampling. AU P >90% corresponds to P <0.1. We extracted pathways and gene sets associated with human atypical neurofibroma (9; Supplementary Fig. S6). We visualized differential expression patterns across sample groups using log₂-transformed TMM-normalized gene counts (after z-score conversion). Data are available in the Gene Expression Omnibus database (GSE 148249).

We excised cervical spinal cord with attached dorsal root ganglia and peripheral nerve from 2-month-old mice, and immediately froze tissue in OCT (Thermo Fisher Scientific, 4585). We cut, air dried, and stained 10 μm tissue section with a β-galactosidase senescence Cell Signaling Kit (Cell Signaling Technology, 9860; ref. 33). We immersed sections in ice-cold fixative for 5 minutes, rinsed them in PBS, and incubated in β-gal staining solution overnight at 37°C. Slides were washed in PBS and counterstained with 0.1% Nuclear Fast Red Aluminum Sulfate for 5–10 minutes (Poly Scientific, 49091). Images were captured with a SPOT Insight 4 Mp CCD camera (Spot Imaging). We counted blue cells per high-powered field (HPF; ×40 magnification).

All statistical analyses were conducted in GraphPad Prism. Significance was set at P ≤ 0.05. Statistical tests and post hoc tests are indicated in figure legends. Error bars represent the SEM unless otherwise noted.

Nf1^fl/fl;DhhCre mice form neurofibromas and require sacrifice at 9 to 15 months when benign paraspinal tumors compress the spinal cord, causing paralysis and inability to feed (18). More aggressive atypical neurofibromas or GEM-PNST do not form in this model system. Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice similarly died between 8 and 13 months (Fig. 1A), while Ink4a/Arf−/−;Nf1^fl/fl;DhhCre mice were dead by 6 months of age. Half the mice in the Ink4a/Arf+/−;Nf1^fl/fl;DhhCre and Ink4a/Arf−/−;Nf1^fl/fl;DhhCre cohorts required sacrifice due to NF1-related superficial GEM-PNST or paralysis secondary to paraspinal tumor impingement (Fig. 1B and see below). Recalculation of time to sacrifice after omitting mice found dead in cage did not alter overall survival times (Fig. 1C). Control double heterozygous Ink4a/Arf+/−;Nf1^fl/+;DhhCre mice did not form paraspinal nerve tumors. Most (80%) were found dead in cage and showed splenomegaly and/or hepatomegaly, consistent with the prevalence of sarcomas/leukemia/lymphomas in mice lacking one copy of Ink4a/Arf (21), and 25% developed GEM-PNST, similar to GEM-PNST incidence in global Nf1+/−;Ink4a/Arf−/− mice (27).

To determine whether paraspinal neurofibromas present in the Nf1^fl/fl;DhhCre model are present in Nf1^fl/fl;DhhCre mice heterozygous for Ink4a/Arf loss, we performed MRI. We detected paraspinal tumors in each mouse (Fig. 2A). Imaging a cohort of Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice at 4 months of age (n = 10), and surviving mice at 7 (n = 7) and 9 (n = 2) months of age showed that these paraspinal tumors have variable growth rate between scans, a characteristic of Nf1^fl/fl;DhhCre neurofibromas (Fig. 2B; ref. 32). At the 7-month scan, 1 of 7 mice showed a T2-intense tumor mass within a preexisting neurofibroma, postulated to be a more aggressive tumor (red arrow, Fig. 2A). On dissection, each mouse had 3–5 paraspinal tumors. All tumors showed similar gross appearance. Figure 2C summarizes the number and percent of paraspinal tumors in each model that showed plexiform neurofibroma histology and atypical neurofibroma histology, defined below.

Like their human counterparts, Nf1^fl/fl;DhhCre neurofibromas are characterized by progressive dissociation of SCs from small unmyelinated axons (17, 34). Progressive disruption visible by electron microscopy (EM) develops after 2 months of age. We performed EM on saphenous nerves of 4-month-old Ink4a/Arf−/−;Nf1^fl/fl;DhhCre and Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice (Supplementary Fig. S1A and S1B). Ink4a/Arf−/−;Nf1^fl/fl;DhhCre mice and Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice were intermediate between WT nerve, in which 60% of unmyelinated SC ensheathe >6 axons, and age-matched Nf1^fl/fl;DhhCre nerve, in which SC rarely ensheathe >2 axons. Loss of Ink4a/Arf may enhance survival and/or prevent death of Nf1 nerve SCs, retarding disruption of Remak bundles, or play role(s) in other pathway(s) that contribute to Remak bundle disruption.

We undertook a histologic analysis of paraspinal and superficial tumors, applying criteria used for human NSTs (8, 35) and criteria established for murine models of PNSTs (36). Of paraspinal tumors analyzed in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre and Ink4a/Arf−/−;Nf1^fl/fl;DhhCre mice, 63% were plexiform neurofibromas (GEM grade 1 neurofibroma; Fig. 2D,–H). These tumors had rare mitoses (<1 mitosis/50 HPF with or without hypercellularity). SC nuclei were thin and elongated, with minimal variability. Nuclear atypia was rare, and neurofibroma architecture, and S100+ was maintained (8). Toluidine blue metachromatic mast cells with characteristic granules were present. In the remaining 37% of Ink4a/Arf+/−;Nf1^fl/fl;DhhCre and Ink4a/Arf−/−;Nf1^fl/fl;DhhCre paraspinal tumors, histologic analysis showed increased cellularity, and nuclear atypia (larger round nuclei and mildly altered chromatin) or nuclear variability, but not both (Fig. 2I–M). These were designated GEM-ANF; tumors maintained fascicular nerve architecture, but demonstrated disrupted microarchitecture with disorganization, irregular cell crowding, and occasionally disrupted perineurium. These GEM-ANF showed <3 mitoses per 10 HPF. There was no necrosis or hemorrhage.

Ink4a/Arf+/−;Nf1^fl/fl;DhhCre and Ink4a/Arf−/−;Nf1^fl/fl;DhhCre mice often developed large subcutaneous tumors (Fig. 3A) requiring euthanasia due to tumor size and/or ulceration, per IACUC standards. These tumors developed in 40% of Nf1^fl/fl;DhhCre mice heterozygous or homozygous for Ink4a/Arf mutation (Fig. 3B), typically occurring between 5 and 10 months of age along the dorsum, shoulders, and rarely craniofacially (Fig. 3C). The histology of these superficial tumors was aggressive, and similar to human MPNSTs (GEM-PNST; Fig. 3D,–F). Tumor cells demonstrated marked nuclear atypia, spindle-cell architecture, and tumors contained areas of necrosis. Notably, like their human counterparts, GEM-PNST showed only patchy expression of S100. Nuclei were plump and pleomorphic. Collagen was only regionally retained, and there was extensive infiltration of cells into surrounding tissues including fat and muscle. Nuclear variability and atypia were present. Many (72%) of GEM-PNST contained nerves with S100+ SCs and neurofilament+ neuronal axons (Fig. 3G,–I). These nerves showed features of GEM grade 1 neurofibroma and/or GEM-ANF, suggesting the presence of precursor lesions, and 2 (12.5%) demonstrated distinct regions showing transition of more aggressive tumor from plexiform neurofibromas. Of note, 1 of 11 GEM-PNST in the Ink4a/Arf+/− cohort was consistent with an epithelioid MPNST, with small, round epithelioid cells with hyperplasia, discohesion and increased mitotic figures (19/10 HPF). Notably, this tumor had very large nuclei with extensive karyorrhexis diffusely with necrosis, and near complete loss of S100. We found that 3 of 5 tumors in the Ink4a/Arf−/− cohort were also epithelioid GEM-PNST.

An important goal is to define the percentage of atypical neurofibromas (tumors with atypical morphology and/or CDKN2A mutations), which are MPNST precursors. This is important, as current recommendations suggest that atypical neurofibromas be removed when feasible, as they might transform (6). As in previous studies, bulk cells from Nf1^fl/fl;DhhCre paraspinal tumors (GEM grade I neurofibroma) do not grow after subcutaneous injection into immunocompromised nude mice (n = 4 tumors; each into 1–2 mice; Fig. 4A). Cells from superficial GEM-PNST from Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice (n = 3 tumors; each injected into 1–2 mice) grow rapidly, with recipient mice requiring sacrifice at 40 days due to tumor burden (Fig. 4A). To determine whether paraspinal tumors from Ink4a/Arf+/−;Nf1^fl/fl;DhhCre grow in this setting, we removed tumors from 6- to 7-month-old mice (n = 20 tumors; each into 1–2 mice); dissociated cells from each tumor were injected separately into the flanks of nude mice. Importantly, of the paraspinal Ink4a/Arf+/−;Nf1^fl/fl;DhhCre tumors tested, 12 of 20 (60%), like Nf1^fl/fl;DhhCre neurofibromas, did not grow. Of tumors that grew, 4 reached large sizes (∼1,000 mm³; Fig. 4A); overall 8 of 20 (40%) exhibited growth, after a lag of 90 days; this 40% is similar to the percentage of paraspinal tumors identified as atypical neurofibromas by histological criteria (37%), suggesting the atypical neurofibromas can progress to higher grade GEM-PNST.

We repeated the allograft study in C57BL/6J syngeneic mice. Two of three transplanted GEM-PNST from both Ink4a/Arf+/−;Nf1^fl/fl;DhhCre and Ink4a/Arf−/−;Nf1^fl/fl;DhhCre grew rapidly, with recipient mice requiring sacrifice at approximately 60 days due to tumor burden (Fig. 4B). We determined whether Ink4a/Arf+/−;Nf1^fl/fl;DhhCre paraspinal tumors would grow in immune-sufficient hosts. Growth of transplanted paraspinal tumors occurred after a prolonged delay (175 days) and at low incidence of 2 of 9 (20% vs. 40% in immunocompromised mice; Fig. 4B). Low incidence could result from insufficient numbers grafted, or, given the delay, because an intact immune microenvironment delays transformation in immunocompetent mice.

Histologic analysis of allografted Ink4a/Arf+/−;Nf1^fl/fl;DhhCre paraspinal tumors mice was consistent with GEM-PNST (Fig. 4C), consistent with the idea that tumors showing atypical neurofibroma histology have increased potential for transformation to higher grade MPNST-like tumors. We observed regions of variable grade throughout each tumor. In small lower grade areas, nuclei were thin, with spindled cells and dense collagen matrix, while in the majority of the tumor, nuclei were plump, pleomorphic, with many mitoses (>3/10 HPF). These higher grade areas showed marked hypercellularity, loss of collagen matrix, invasion of muscle tissue, fat cells and notable, multiple, occurrences of entrapment of host peripheral nerves (Fig. 4D,–F). Necrosis was sometimes present and occasional hemorrhage was seen. In sum, similar to superficial GEM-PNST, all allografted paraspinal tumors that grew contained large regions of GEM-PNST histology (Fig. 4G,–I), and areas of lower grade, likely representing remnants of the lower grade tumor from which the higher grade tumors evolved.

GEM-PNST allografts exhibited plump, pleomorphic nuclei, and frequent mitoses. In contrast to paraspinal tumor allografts, the collagen matrix in GEM-PNST allografts was disturbed, and tumor invaded surrounding normal tissue. Features of transformation were absent; high-grade features were present throughout the tumor, multifocally, with necrosis. In both allografted paraspinal and GEM-PNST tumors, Ki67 staining (cell proliferation) was higher in higher grade areas of tumor (Fig. 4H), while S100 decreased. Large nerve bundles were prominent (Fig. 4I).

In human, most atypical neurofibromas show hemizygous loss of the region of chromosome 9p containing CDKN2A. Only 1 of 16 and 2 of 15 atypical neurofibromas showed homozygous loss in independent studies (7, 9). To determine whether this genetic feature is recapitulated in the mouse model, we isolated genomic DNA from 7 superficial GEM-PNST and 5 paraspinal tumors from Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice. We amplified the Ink4a/Arf locus using qPCR. Of the samples tested no atypical neurofibromas and only one GEM-PNST (8.3%) exhibited loss of heterozygosity (LOH; Fig. 5A). Of tumor allografts, 4 of 6 (30.8%) grafts from superficial GEM-PNST tumor but no paraspinal tumor grafts (0/7), exhibited LOH (Fig. 5B). To determine whether the WT allele might be silenced by other mechanisms we analyzed Western blots for expression of p16^INK4a (hereafter, p16), the product protein of the Ink4a locus. Consistent with the genetic data, paraspinal tumors maintained p16 expression, and some GEM-PNST tumors exhibited p16 loss. Interestingly, all tested allografted tumors showed low levels of p16, comparable with those in an Ink4a/Arf−/− mouse tumor (Fig. 5C). These data suggest that p16 expression is silenced posttranscriptionally in many aggressive nerve tumors, as they progress. p16^ink4a and p19^Arf regulate the cell cycle and apoptosis through effects on Rb and p53, respectively. Loss of Ink4a/Arf is expected to decrease phosphorylation of Rb, and apoptotic signaling via p53. While phosphorylated Rb is often observed in the absence of p16, we found no detectable expression in any of tested tumors (Fig. 5C).

RAS dysregulation and Nf1 loss, can increase cellular senescence (26). To test whether Nf1 loss in the Nf1^fl/fl;DhhCre model causes senescence, we used senescence-associated β-galactosidase (SA-β-gal) histochemical stain. In tissue sections of 2-month-old dorsal root ganglion and nerve, numbers of SA-β-gal⁺ cells were low in WT mice, elevated in Nf1^fl/fl;DhhCre nerve, and WT levels restored in double mutant nerve (Fig. 5D–F). Thus, before tumor formation Nf1^fl/fl;DhhCre nerve contains senescent cells and heterozygous loss of Ink4a/Arf together with Nf1 loss in SCs abrogates the senescence checkpoint caused by loss of Nf1.

We hypothesized that neurofibromas that develop in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice would show gene expression changes compared with Nf1^fl/fl;DhhCre plexiform neurofibromas, while GEM-ANF in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice would have further changes that more significantly resemble GEM-PNST and human MPNST. A recent study provided RNA sequencing data from human neurofibromas and atypical neurofibromas for comparison (9). Thus, we set out to define transcriptomic changes in GEM-ANF, prior to transformation to GEM-PNST. Unfortunately, neither neurofibromas nor atypical neurofibromas in Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice are large enough for both histologic and transcriptomic analysis. Therefore, we dissected unclassified paraspinal tumors from Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice for RNA sequencing. Results were compared with existing data from confirmed plexiform neurofibromas (Nf1^fl/fl;DhhCre, n = 12) and to superficial GEM-PNST (Ink4a/Arf+/−;Nf1^fl/fl; DhhCre, n = 4). PCA and clustering analysis suggested optimal cluster numbers of 3 or 4 from the total transcriptomes. Nf1^fl/fl;DhhCre neurofibromas, Ink4a/Arf+/−;Nf1^fl/fl;DhhCre paraspinal tumors, and GEM-PNST grouped into three major distinct clusters, with two subclusters in the paraspinal tumor samples (Fig. 6A). HC also predicted two subclusters (i.e., #1 and #2) among the paraspinal tumor samples (Fig. 6B). HC found that, overall, the 4 paraspinal tumors resembled neurofibromas (Nf1^fl/fl;DhhCre) more closely than GEM-PNST (Ink4a/Arf+/−;Nf1^fl/fl;DhhCre). This finding is consistent with the minimal genomic variability in atypical neurofibromas (13) and the wholesale gene expression and chromosomal changes in human MPNST.

We next analyzed the expression profiles of genes associated with human atypical neurofibromas to MPNST progression in the mouse subclusters. Pathway-based gene expression analysis verified the different behaviors of the two paraspinal tumor subclusters. Gene expression patterns of neurofibroma and paraspinal tumor subcluster #1 were similar in PRC and RAS/MAPK pathways (Fig. 6C), consistent with an absence of changes in the PRC complex in human atypical neurofibromas [9 (their Supplementary Fig. S6)]. However, paraspinal tumor subcluster #2 and GEM-PNST tumors showed similar expression of PRC, TP53/Trp53, RAS/MAPK, and SWI/SNF pathway genes. Expression of the PRC complex gene Suz12 was reduced; Suz12 is a PRC2 complex member shown to predispose to progression to MPNST (37). The SWI/SNF gene Smarca2 was also downregulated in pathway in GEM-PNST and subcluster #2. SMARCA2 is frequently mutated in atypical neurofibromas and GEM-PNST (9). Among RAS-MAPK pathway genes, the expression of Nf1 is reduced in paraspinal subcluster #2 tumors, likely attributed to increased proportions of Nf1^fl/fl tumor cells in atypical neurofibromas versus neurofibroma. Rb1 and Cdkn1a tumor suppressor gene expression was also reduced, with increases in expression of proliferation-associated genes Ccnd3, Cdk1, Ccne1, and E2f5, consistent with the slightly increased cell proliferation in some GEM-ANF. Overall, these results suggest that the Ink4a/Arf;Nf1^fl/fl;DhhCre model provides a surrogate for processes driving human neurofibroma progression to atypical neurofibromas.

Loss of CDKN2A in most human atypical neurofibromas and MPNST is believed to play a major role in NF1-associated nerve tumor progression. Intercrossing Ink4a/Arf mutant mice with Nf1^fl/fl;DhhCre mice generated a model that accurately recapitulates the progression from plexiform neurofibromas to atypical neurofibromas to MPNST, as assessed by similar histology and gene expression to human tumors. In addition, our in vivo transformation assays provide a robust preclinical platform to address the current lack of therapy for atypical neurofibromas or MPNST.

In previous Nf1-driven GEM-neurofibroma models, neurofibromas formed in all mice but GEM-PNST formed only at low (10%–15%) incidence, and in only some models (38, 39). In our study, half of paraspinal tumors were slowly growing atypical neurofibromas, which did not form GEM-PNST (which only formed at distant sites). In contrast, combining periostin-Cre–driven Nf1 loss with Arf heterozygosity or homozygosity in early SCPs (29), mice formed paraspinal GEM-neurofibromas, but only 10% were rapidly growing GEM-ANF, and GEM-MPNST were frequently paraspinal. The relatively reduced aggression of the new Ink4a/Arf;Nf1^fl/fl;DhhCre model may reflect reduced transformation potential of late SCPs, driven by DhhCre at E12.5, as compared with SCP in which Periostin-Cre induces Nf1 loss in cells, just after transition from neural crest cell to SCP, 2 days earlier. We posit that the timing of Nf1 loss contributes to transformation potential of neurofibroma to atypical neurofibromas. Also, although MPNST cells express neural crest cell makers (40, 41), we confirm that neural crest cells are not a required cell of origin for GEM-PNST; DhhCre is not active in neural crest cells (18, 42).

In the Ink4a/Arf;Nf1^fl/fl;DhhCre model, Ink4a/Arf loss occurs in the germline, before Nf1 Ioss. When Nf1 and Arf alleles are lost concurrently in the periostin-driven model (29) the complete spectrum of tumors also forms, albeit with different prevalence. Together these studies show that whether Ink4a/Arf loss precedes Nf1 loss, or if the genes are lost simultaneously, the full spectrum of tumors can form. In contrast, if Ink4a/Arf loss occurs in the germline, before Nf1 Ioss, in Ink4a/Arf+/−;Nf1+/− mice GEM-PNST (not neurofibroma) form (27). The late loss of Nf1 in the absence of Ink4a/Arf thus seems to enable the formation of more aggressive tumors. In human tumors, progression of neurofibroma to atypical neurofibroma is characterized by inactivation of NF1 function followed later by loss of CDKN2A, with increased loss of the second CDKN2A allele in MPNST (7). Together, these mouse studies are consistent with the idea that the timing of complete Ink4a/Arf loss regulates the atypical neurofibromas to MPNST transition. Simultaneous loss of both genes in adult Ink4a/Arf^fl/fl;Nf1^fl/fl mice injected with Cre-recombinase, however, generates GEM-PNST (not neurofibroma; ref. 28); it will be of interest to determine whether in this model, at earlier time points, less aggressive tumors are detectable.

In the Ink4a/Arf;Nf1^fl/fl;DhhCre model, Remak bundle disruption characteristic of neurofibromas occurs, but is reduced versus Nf1^fl/fl;DhhCre nerve. Remak SCs depend upon ligand-stimulated receptor tyrosine kinase signaling for homeostasis (43). Nf1 mutant SCs produce many factors (44, 45) that may influence surrounding healthy nerve cells, causing Remak bundle disruption; these factors may change, in variety and/or amount, in Ink4a/Arf;Nf1^fl/fl;DhhCre SC, or in Nf1; Tp53 (NPCis) mice, which form GEM-PNST, but no Remak bundle disruption (46). Neurons are necessary to maintain differentiated SCs (47) and for neurofibroma development (48). Atypical neurofibromas allografts transforming to GEM-PNST show abundant nerve recruitment, likely recruited by factor(s) from Nf1−/− SCs.

About half of human NF1-associated MPNST show homozygous loss of CDKN2A (49). Using LOH as a readout, this frequent genetic loss was not recapitulated in GEM-PNST; only 8.3% of the GEM-PNST exhibited Ink4a/Arf LOH. Increases in tumor stroma can decrease mutation detection, and/or the second Ink4a/Arf allele can be silenced epigenetically, by point mutations, or by small indels, rather than via LOH. We assessed p16 protein levels and found that silencing occurs in at least 50% of tumors, consistent with the absence of p16 staining in some human MPNST (50), and may be relevant to some human MPNST. Future studies are needed to investigate the mechanisms of p16 regulation, and the reasons for the absence of detectable p-Rb in the tumors in this model.

Heterozygous loss of Ink4a/Arf, together with loss of Nf1 in SCs, was sufficient to abrogate the senescence checkpoint caused by loss of Nf1, consistent with previous data that loss of NF1 induces SC senescence in cutaneous neurofibroma (26), in plexiform neurofibromas, and in Nf1−/− SCs in vitro (29). High levels of Ras-GTP drive senescence; Ink4a/Arf is often lost to circumvent Ras-mediated oncogene-induced senescence (25). Markers of senescence beyond low p16 and SA-β-gal are needed to define the senescence phenotype, which varies among cell types. Importantly, the absence of senescent cells in the Ink4a/Arf+/−; Nf1^fl/fl;DhhCre nerves was insufficient by itself to cause cell malignant transformation; paraspinal GEM-PNST did not develop. Instead, a subset of Ink4a/Arf+/−;Nf1^fl/fl;DhhCre paraspinal plexiform neurofibromas could undergo malignant transformation after a period of time, when transplanted as allografts into secondary mice. Paraspinal grafts into a syngeneic host grow at lower incidence and undergo malignant transformation later than in grafts into immunocompromised mice, so the intact immune system may restrain malignant transformation of atypical neurofibromas–like tumors. This will be an important aspect of the model to study, enabling assessment of the microenvironment in malignant transformation.

Ink4a/Arf−/−;Nf1^fl/fl;DhhCre and Ink4a/Arf+/−;Nf1^fl/fl;DhhCre mice demonstrated half plexiform neurofibromas and half atypical neurofibromas. The paraspinal tumors from the Ink4a/Arf;Nf1^fl/fl;DhhCre model demonstrated perturbation of pathways that are disrupted in their human counterparts, as assessed by RNA sequencing. However, atypical neurofibromas in this model may more closely resemble an as-yet unidentified stage in progression toward MPNST, revealed in the GEMM uniform genetic background. Constitutional heterozygosity for Ink4a/Arf may also alter tumor gene expression in paraspinal tumors (in tumor cells or stroma), affecting transformation but not tumor histology. Few human or mouse atypical neurofibromas have been analyzed to date; increasing sample numbers will allow more complete comparisons.

Given that half of paraspinal tumors in the Ink4a/Arf+/−;Nf1^fl/fl;DhhCre model show atypical neurofibromas histology and half generate GEM-PNST on allografting, we suspect that most or all GEM-ANF transform in this model. This new model provides a tractable platform upon which to study mechanisms of transformation and to design studies aiming to treat atypical neurofibromas and/or prevent transformation to GEM-PNST.

K.E. Chaney reports grants from NIH (R01 ns086219 to N. Ratner) during the conduct of the study. L.J. Kershner reports grants from NIH (R01 NS086219 to N. Ratner) and NIH (Ruth L. Kirschstein NRSA; 2T32CA117846-11A1 to self) during the conduct of the study; grants from NIH (Ruth L. Kirschstein NRSA; 2T32CA117846-11A1 to self) outside the submitted work. J. Wu reports grants from NIH during the conduct of the study. T.A. Rizvi reports grants from NIH (RO1 NS086219 to N. Ratner) during the conduct of the study. S. Szabo reports grants from NIH and DOD during the conduct of the study; personal fees and nonfinancial support from Parkview Mirro Cancer Center, IN (symposium speaker) and Michigan State University, MI (lecturer) outside the submitted work. D.A. Largaespada reports personal fees from B-MoGen Biotechnologies (chair of the board; sold to Bio-Teche summer 2019); grants from Genentech; other compensation from NeoClone Biotechnology (ownership interest), ImmuSoft (previously Discovery Genomics, Inc.; ownership interest), B-MoGen Biotechnologies (ownership interest, sold to Bio-Teche summer 2019), and Recombinetics, Inc. (ownership interest) outside the submitted work; and acted as Chief Science Officer for Surrogen, Inc. (uncompensated position). N. Ratner reports sponsored research contracts with Revolution Medicine and Boehringer Ingelheim outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

K.E. Chaney: Data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. M.R. Perrino: Data curation, formal analysis, validation, investigation, visualization, writing-original draft, writing-review and editing. L.J. Kershner: Formal analysis, validation, investigation, writing-original draft, writing-review and editing. A.V. Patel: Conceptualization, supervision, methodology, writing-review and editing. J. Wu: Supervision, investigation, methodology. K. Choi: Data curation, software, formal analysis, investigation, visualization, writing-original draft, writing-review and editing. T.A. Rizvi: Data curation, investigation. E. Dombi: Investigation, visualization, methodology. S. Szabo: Data curation, investigation, visualization, methodology, writing-review and editing. D.A. Largaespada: Conceptualization, resources, funding acquisition, writing-review and editing. N. Ratner: Conceptualization, resources, supervision, funding acquisition, visualization, project administration, writing-review and editing.

We thank Annmarie Ramkissoon for assistance with genomic PCR, and Robert Coover and Craig Thomson for assistance with Western blotting.

This work was supported by R01 NS086219 (to D.A. Largaespada and N. Ratner) and American Cancer Society, ACS Research Professor Award #123939 (to D.A. Largaespada). L.J. Kershner is supported by NIH Ruth L. Kirschstein NRSA (2T32CA117846-11A1).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.