A genetic mouse model with postnatal Nf1 and p53 loss recapitulates the histology and transcriptome of human malignant peripheral nerve sheath tumor.

BACKGROUND: Malignant peripheral nerve sheath tumors (MPNST) are aggressive sarcomas. Somatic inactivation of NF1 and cooperating tumor suppressors, including CDKN2A/B, PRC2, and p53, is found in most MPNST. Inactivation of LATS1/2 of the Hippo pathway was recently shown to cause tumors resembling MPNST histologically, although Hippo pathway mutations are rarely found in MPNST. Because existing genetically engineered mouse (GEM) models of MPNST do not recapitulate some of the key genetic features of human MPNST, we aimed to establish a GEM-MPNST model that recapitulated the human disease genetically, histologically, and molecularly.
METHODS: We combined 2 genetically modified alleles, an Nf1;Trp53 cis-conditional allele and an inducible Plp-CreER allele (NP-Plp), to model the somatic, possibly postnatal, mutational events in human MPNST. We also generated conditional Lats1;Lats2 knockout mice. We performed histopathologic analyses of mouse MPNST models and transcriptomic comparison of mouse models and human nerve sheath tumors.
RESULTS: Postnatal Nf1;Trp53 cis-deletion resulted in GEM-MPNST that were histologically more similar to human MPNST than the widely used germline Nf1;Trp53 cis-heterozygous (NPcis) model and showed partial loss of H3K27me3. At the transcriptome level, Nf1;p53-driven GEM-MPNST were distinct from Lats-driven GEM-MPNST and resembled human MPNST more closely than do Lats-driven tumors.
CONCLUSIONS: The NP-Plp model recapitulates human MPNST genetically, histologically, and molecularly.

Postnatal Nf1;p53 cis-deletion in NP-Plp mice results in tumors similar to MPNST.

The transcriptomes of Nf1;p53-driven and Lats-driven MPNST models are distinct.

NP-Plp model resembles human MPNST genetically, histologically, and molecularly.

MPNST are aggressive sarcomas with a poor prognosis and limited treatment options. Existing mouse models of MPNST do not recapitulate some of the key genetic features of human MPNST. To model the somatic, possibly postnatal, mutational events seen in MPNST patients, we generated a mouse MPNST model by combining 2 genetically modified alleles, an Nf1;Trp53 cis-conditional allele and a Plp-CreER allele. Our histologic and transcriptomic analyses showed that this NP-Plp model resembles human MPNST genetically, histologically, and molecularly—more so than the widely used NPcis model and the recently published Lats-driven model. The NP-Plp model is genetically simple, making it easy to maintain and an ideal platform for preclinical studies. Given its tamoxifen-inducible nature, this model can be used to study the time/stage dependency of the tumorigenic potential of Schwann cells.

Malignant peripheral nerve sheath tumors (MPNST) are aggressive sarcomas arising from cells of the Schwann cell lineage. MPNST have a poor prognosis with limited treatment options. About half of MPNST occur in patients with the cancer predisposition syndrome neurofibromatosis type I (NF1), which is caused by germline loss-of-function (LOF) mutations in one copy of NF1 (encoding neurofibromin). Other cases occur sporadically or after radiotherapy in the general population. Somatic mutations in NF1 with co-occurring LOF alterations in CDKN2A, polycomb repressive complex 2 (PRC2) genes (SUZ12 and EED), and/or TP53 (Trp53 in mice, encoding p53) are found in the vast majority of MPNST regardless of their etiology.[1]

Genetically engineered mouse (GEM) models of human diseases are useful for studying disease mechanisms and testing diagnostic/therapeutic approaches. Although several GEM-MPNST models have been published,[2,3] they do not recapitulate some of the key genetic features of human MPNST. For example, the first and one of the most frequently used GEM-MPNST models harbors germline cis-heterozygous LOF mutations in Nf1 and Trp53 (NPcis mice),[4,5] whereas MPNST patients do not carry germline TP53 mutations, except for rare occurrences in Li-Fraumeni syndrome.[6] Most, if not all, conditional GEM-MPNST models use Cre lines that initiate gene deletion in embryonic Schwann cell lineage cells, such as Dhh-Cre,[7]Periostin-Cre,[8] and GFAP-Cre,[3,9] although it is unclear whether somatic alterations in patients occur in Schwann cell precursors or immature Schwann cells during fetal development or in mature Schwann cells later in life. Injecting viruses expressing Cre or Cas9-gRNA into the sciatic nerve has the advantage of temporally controlled gene deletion[3,10]; however, surgery is required to expose the nerve and may cause nerve injury.[11] Although these existing GEM-MPNST models are useful for many studies, they are not ideal for addressing questions such as those regarding tumor cell-of-origin.

Most GEM-MPNST models were created based on the knowledge that nearly all MPNST harbor NF1 mutations. Wu et al.[12] reported that deleting Lats1 and Lats2 (Lats1;2) of the Hippo pathway in the Schwann cell lineage generated tumors histologically similar to MPNST. It was not shown, however, whether these Lats-driven GEM-MPNST resembled human MPNST or Nf1-driven GEM-MPNST at the transcriptome level.

Here, we used conditional alleles of Nf1[13] and Trp53[14] to generate a cis-conditional allele, [Nf1;Trp53]. Mice carrying [Nf1;Trp53] and the inducible Plp-CreER,[15] which targets multiple glial cell types in the peripheral nervous system,[16] developed GEM-MPNST when [Nf1;Trp53] was deleted postnatally. This new model recapitulated the histopathologic features of human MPNST better than the NPcis model. Concurrent with the study by Wu et al.,[12] we generated Lats-driven GEM-MPNST models and found that the transcriptomes of Nf1-driven and Lats-driven GEM-MPNST were distinct. Cross-species transcriptomic comparison suggests that Nf1;p53-driven GEM-MPNST models resemble human MPNST more closely than do Lats-driven models. Therefore, the NP-Plp model most closely recapitulates human MPNST genetically, histologically, and molecularly.

Materials and Methods

Animals

NPcis (Stock #008191), Nf1 (#017640), Trp53 (#008462), Plp-CreER (#005975), and Nestin-Cre (#003771) were obtained from The Jackson Laboratory. Lats1;Lats2 were obtained from Randy Johnson.[17] NPcis, Nf1, Trp53, and [Nf1;Trp53] were maintained on the B6;129 background obtained from Jackson, Plp-CreER and Nestin-Cre on the C57BL/6J background, and Lats-Nes and Lats-Plp on a mixed background.

Tamoxifen was dissolved in corn oil. For P7 mice, 15 µl/g body weight of a 5 mg/ml solution was injected intraperitoneally once daily for 2 consecutive days. For 4- to 8-week-old mice, 5 µl/g body weight of a 20 mg/ml solution was administered by oral gavage once daily for 3 days. All procedures were approved by the St. Jude Children’s Research Hospital Animal Care and Use Committee. Male and female mice were used in a ~1:1 ratio.

Histology

Tissues were fixed in 10% formalin or 4% paraformaldehyde overnight. Tumors with adjacent bones were decalcified in 10% formic acid. Tissues were embedded in paraffin and sectioned at 4-µm or 6-µm thickness. Hematoxylin and eosin staining was performed per routine methods. For immunostaining, slides were subjected to antigen retrieval at 95°C for 30 minutes in 10 mM sodium citrate (pH 6.0) and incubated with primary antibody (Supplementary Table 1) overnight at 4°C. Images were acquired using a Keyence BZ-X700 or an Olympus BX46 microscope with an Olympus SC180 camera and processed in Photoshop. For Ki67 and H3K27me3 quantifications, each data point represents an individual tumor and is the average of measurements from 3 182-μm × 182-μm regions per tumor.

Western Blot

Tumor and cell lysates were prepared in 20 mM HEPES (pH 7.4), 150 mM NaCl, 2% SDS, and 5% glycerol supplemented with AEBSF and Halt protease and phosphatase inhibitors. Lysates were subjected to SDS-PAGE, probed with primary antibodies (Supplementary Table 1) and HRP-conjugated secondary antibodies.

RNA Isolation and Sequencing

Total RNA was isolated using the Direct-zol RNA Miniprep Kit (Zymo). Libraries were prepared using the TruSeq Stranded Total RNA Library Prep Kit (Illumina), analyzed for insert size distribution using the 4200 TapeStation D1000 ScreenTape Assay (Agilent), and sequenced on Illumina NovaSeq 6000, yielding 100 million 100-bp paired-end reads. Sequences are deposited in GEO (GSE172221).

RNA-seq Data Analysis

Mouse and human sequences were mapped to the mm10 and hg38 genomes, respectively, with STAR aligner. Gene-level quantification was determined using RSEM and based on GENCODE M22 gene annotation. Noncoding and GENCODE level 3 genes were excluded. Differential expression was modeled using the voom method in the limma R package. Voom-normalized counts were analyzed using the lmFit and eBayes functions in limma. The false discovery rate (FDR) was estimated using the Benjamini–Hochberg method. Gene set enrichment analysis (GSEA) was performed using MSigDB gene sets as described.[18] Single-sample GSEA (ssGSEA) was performed as described.[19]

Cross-Species Transcriptome Comparison

RNA-seq data from the Lats;Dhh-Cre model (GSE99040), Nf1 models (GSE148249), MPNST samples from The Cancer Genome Atlas (TCGA), and the MPNST and PNF cohort (GSE145064) were summarized to fragments per kilobase per million mapped fragments based on Ensembl genes (v93). These datasets were integrated with our mouse tumor samples and our (SJ collection) human samples by filtering genes to mouse–human orthologues that were present in all datasets. ComBat[20] was used to correct for batch and species differences. ssGSEA scores[19] were derived using a list of orthologous gene sets created by retaining all gene sets from MSigDB that contain ≥80% mouse–human orthologues, ≥15 genes, and ≤400 genes.

To assess the resemblance of murine tumor models to human tumors in an unsupervised manner, we used uniform manifold approximation and projection (UMAP) based on Pearson’s distance of the top 500 most varied gene sets across all murine models. A supervised analysis was performed by training a random forest classifier to predict whether human tumors were MPNST or PNF/NF based on the ssGSEA scores. The human tumor samples in the SJ collection and in GSE145064, comprising 34 MPNST and 34 PNF/NF samples, were split in a 70–30 ratio to generate the training and testing datasets. The top 20 most informative gene sets were used as features for training, which achieved a test set sensitivity of 0.825, specificity of 0.86, and accuracy of 0.80. This model was then used to predict the MPNST class probability for murine models and the independent test set of TCGA MPNST samples.

Results and Discussion

Postnatal Somatic Nf1;Trp53 Loss Results in GEM-MPNST

Mouse Nf1 and Trp53 both locate on chromosome 11. To mimic somatic loss of NF1 and TP53 in MPNST patients, we generated a chromosome 11 harboring the conditional alleles of both Nf1 and Trp53, designated [Nf1;Trp53], by crossing Nf1;Trp53trans-heterozygous mice with wild-type (WT) mice and screening for progeny carrying both conditional alleles as a result of meiotic recombination that placed them on the same chromosome. Mice homozygous for this chromosome were healthy and fertile, suggesting that the chromosome integrity was not affected by the recombination event. To enable temporal control of [Nf1;Trp53] deletion, we chose the tamoxifen-inducible Cre line, Plp-CreER, which targets multiple glial cell types in the peripheral nervous system.[16] We then combined the [Nf1;Trp53] and Plp-CreER alleles with a germline WT or a null allele of Nf1 (Nf1−), generating [Nf1;Trp53];Plp-CreER and [Nf1;Trp53];Plp-CreER mice, respectively, to mimic the genetics of sporadic and NF1-associated MPNST (Figure 1A).

Figure 1.

Postnatal somatic Nf1;Trp53 loss results in malignant peripheral nerve sheath tumors. (A) Nomenclature of Nf1;Trp53 genotypes and corresponding genetic alterations on mouse chromosome 11. (B) Kaplan–Meier survival curves of tamoxifen (TAM)-treated mice. Log-rank Mantel–Cox test. P, postnatal day. (C–G) Examples of dermal tumors, showing a [Nf1;Trp53];Plp-CreER mouse with an ulcerated dermal tumor (C, red arrow) and histology of a grade II tumor stained with H&E (D, E) and S100 and SOX10 antibodies (F, G, brown signal). Boxed area in D is enlarged in E. Boxed areas in E, F, G are enlarged in E′, F′, G′. Arrows in E and F indicate a nerve-like structure. (H–L) Examples of subcutaneous tumors, showing an [Nf1;Trp53];Plp-CreER mouse with a tumor (H, red arrow) and a grade III GEM-MPNST stained with H&E (I, J) and S100 and SOX10 antibodies (K, L). Boxed area in I is enlarged in J, showing regional heterogeneity. Solid-boxed area in J is enlarged in J′, showing a poorly differentiated area. Yellow arrowheads indicate examples of nuclear atypia. Dashed-boxed area in J is enlarged in J′′, showing spindle cells forming interlacing fascicles and storiform formations. Solid-boxed areas in K and L are enlarged in K′ and L′, and dashed-boxed areas are enlarged in K′′ and L′′.

As an initial test, we administered tamoxifen on postnatal days (P) 7 and P8. Six of 10 [Nf1;Trp53];Plp-CreER and 6 of 7 [Nf1;Trp53];Plp-CreER mice developed 1 or 2 palpable masses at 4–9 months of age (Figure 1B). Six of the 16 masses were firm and grew relatively slowly, and the mice had to be euthanized within 4 weeks of tumor appearance because of tumor ulceration (Figure 1C). These tumors were all in the skin. Of the other 10 masses, 6 were subcutaneous, associated with skeletal muscles, and 4 were in the abdominal cavity. These masses were soft, grew rapidly, and mice had to be euthanized within 1 week of tumor appearance because of tumor size (Figure 1H).

We performed histopathologic analyses of these tumors according to criteria established for mouse models of PNST[7,21] and criteria used for human nerve sheath tumors.[22,23] All 5 dermal tumors analyzed exhibited features typical of GEM-MPNST, comprising spindle cells forming interlacing fascicles and storiform formations with high cellularity and occasional pleomorphism and nuclear atypia (Figure 1D and E). Two were diagnosed as grade II GEM-MPNST based on their low mitotic rate and low-to-moderate levels of Ki67 staining. The remaining 3 were diagnosed as grade III GEM-MPNST based on the presence of regions with a higher mitotic rate. The 8 fast-growing, internal tumors analyzed were all diagnosed as grade III GEM-MPNST. These tumors had growth patterns similar to those of the grade II GEM-MPNST but had more frequent pleomorphism and nuclear atypia, a high mitotic rate, and areas of necrosis. They more frequently contained poorly differentiated areas composed of smaller cells with a high nucleus-to-cytoplasm ratio (Figure 1I and J). Both grade II and III tumors showed multifocal nuclear and occasionally cytoplasmic S100 immunolabeling (Figure 1F and K). The diagnosis of GEM-MPNST was further supported by moderately positive immunolabeling for SOX10 in 2 of 4 tumors tested (Figure 1G and L) and widespread positive Nestin immunolabeling in all tumors (data not shown). None of these tumors were diagnosed as atypical neurofibroma or ANNUBP (atypical neurofibromatous neoplasms of uncertain biologic potential).

Thus, postnatal somatic Nf1;Trp53 loss resulted in GEM-MPNST that resembled human MPNST. Because we detected no difference in tumor onset, histology, or survival between [Nf1;Trp53];Plp-CreER and [Nf1;Trp53];Plp-CreER mice—which might be caused by the small number of mice analyzed—we designated these mice/tumors collectively as NP-Plp mice/tumors.

NP-Plp GEM-MPNST Is Histologically Distinct From NPcis GEM-MPNST

Next, we compared our NP-Plp model to the NPcis model, one of the most frequently used GEM-MPNST models. Twelve of the 13 NPcis mice analyzed developed palpable subcutaneous tumors at 4–6 months of age (Figure 2A). These tumors were soft and fast-growing, and mice had to be euthanized within 1 week of tumor appearance because of tumor size. Gross dissection revealed that 2 mice harbored an additional soft tumor internally. The remaining mouse developed 3 firm, slow-growing dermal tumors at 4 months of age and was euthanized 1 month after tumor appearance because of ulceration (Figure 2F).

Figure 2.

Histopathologic comparison of NP-Plp and NPcis GEM-MPNST. (A–E) Examples of subcutaneous NPcis grade III GEM-MPNST, showing an NPcis mouse with a tumor (A, red arrow), H&E staining (B, C), and S100 and SOX10 immunostaining (D, E). Boxed area in B is enlarged in C. Boxed areas in C, D, and E are enlarged in C′, D′, E′. Yellow arrowheads in C′ indicate nuclear atypia. (F–J) Examples of dermal NPcis grade II GEM-MPNST, showing an NPcis mouse with a tumor (F, red arrow), H&E staining (G, H), and S100 and SOX10 immunostaining (I, J). Boxed area in G is enlarged in H. Boxed areas in H, I, and J are enlarged in H′, I′, and J′. (K–N) Ki67 immunostaining and quantification (mean ± SEM) in grade III tumors. Unpaired, two-tailed t test.

Five of the 7 soft, fast-growing tumors analyzed exhibited nearly identical histopathologic features (Figure 2B and C). These subcutaneous tumors were homogeneous, comprising small cells with mild spindling. The cells were more pleomorphic and undifferentiated and had a higher nucleus-to-cytoplasm ratio than is typical in NP-Plp tumors. The nuclei were often hyperchromatic and enlarged with marked atypia. The tumors had a high mitotic rate. Although the spindle-cell morphology and fascicular and storiform patterns typically found in MPNST were present only occasionally in these tumors and SOX10 immunolabeling was weak, most tumors showed occasional S100 immunosignal along with Nestin immunolabeling (Figure 2D and E and data not shown) and were, therefore, diagnosed as grade III GEM-MPNST. The remaining 2 soft, fast-growing tumors were diagnosed as lymphoma and were excluded from further analyses. The 2 dermal tumors analyzed exhibited typical MPNST features (Figure 2G–J), including spindle cells forming interlacing fascicles and storiform patterns, had a low mitotic rate, and were diagnosed as grade II GEM-MPNST. Quantification of Ki67 signals in grade III NP-Plp and NPcis GEM-MPNST showed a significantly higher proliferation rate in NPcis tumors than in NP-Plp tumors (Figure 2K–N).

LATS1/2 Loss in the Schwann Cell Lineage Results in GEM-MPNST

Inactivation of LATS1/2 of the Hippo pathway was recently shown to cause GEM-MPNST.[12] We made the same finding independently. Using a Nestin-Cre line that targets progenitor cells in the central and peripheral nervous system,[24] we found that mice lacking 3 alleles of Lats1;2, Lats;Lats2;Nestin-Cre and Lats1;Lats2;Nestin-Cre (referred to as Lats-Nes mice), all developed tumors at 4–6 months of age. Tumors were most frequently found in the skin, with each mouse typically having multiple firm, slow-growing tumors, and the mice had to be euthanized 2–4 months after tumor appearance, usually because of tumor ulceration and occasionally because of limb paralysis (Figure 3A). Gross dissection revealed additional firm, internal tumors that were often associated with nerves and neural ganglia. Histopathologic analysis revealed typical features of GEM-MPNST. All internal tumors were diagnosed as grade III GEM-MPNST, whereas some dermal tumors as grade III and others as grade II GEM-MPNST (Figure 3B and C). To confirm that GEM-MPNST formation was due to Lats1;2 loss in the peripheral glial lineage, we used Plp-CreER to delete Lats1;2. Tamoxifen administration to 1- to 2-month-old Lats1;Lats2;Plp-CreER (Lats-Plp) mice resulted in dozens of firm, slow-growing dermal tumors in each mouse and occasional internal tumors with 100% penetrance (Figure 3D). All Lats-Plp tumors were grade II GEM-MPNST (Figure 3E).

Figure 3.

LATS1/2 loss causes GEM-MPNST formation and comparison to Nf1;p53-driven GEM-MPNST. (A) Survival curves of Lats-Nes mice and littermate controls. Log-rank Mantel–Cox test. (B) A large infiltrative grade III paraspinal tumor stained with H&E. Boxed areas in B and B′ are enlarged in B′ and B′′, respectively. Arrow in B indicates the spinal cord. Yellow arrowhead in B′′ indicates nuclear atypia. (C) A grade III dermal tumor stained with H&E. (D) Survival curves of Lats-Plp mice and littermate controls. Log-rank Mantel–Cox test. (E) A grade II Lats-Plp GEM-MPNST stained with H&E. (F) Western blot analysis of key oncogenic drivers. SN, normal sciatic nerves from Lats1;Lats2;Nestin-Cre mice; SW10, an immortalized Schwann cell line. *Indicates a possible truncated neurofibromin product. All samples were grade III tumors inferred based on tumor location. 20 μg of SN and tumor lysate or 5 μg of SW10 cell lysate was loaded per lane. (G) mRNA counts obtained from RNA-seq analysis of tumor samples and fold changes (FC) in NP (NP-Plp and NPcis) versus Lats (Lats-Nes and Lats-Plp) tumors. Cpm, counts per million. (H) mRNA counts obtained from RNA-seq analysis of E12.5 Lats1;Lats2;Nestin-Cre double knockout (dKO) and control (no Cre) telencephalons. (I) TAZ immunostaining. Boxed areas in the upper-row images are enlarged in the lower row.

Differential Expression of Driver Tumor Suppressors in Nf1;p53-Driven and Lats-Driven GEM-MPNST Models

We compared the expression of the key oncogenic drivers in Nf1;p53-driven (NP) and Lats-driven (Lats) GEM-MPNST models. As expected from the genotype, neurofibromin and p53 proteins were undetectable in NP tumors but were present in at least some Lats tumors (Figure 3F). Surprisingly, LATS1/2 protein levels were very low in NP tumors, similar to those in Lats tumors in which the Lats1;2 genes were deleted. LATS1/2 phosphorylate transcriptional coactivators YAP and TAZ (YAP/TAZ) and prevent them from entering the nucleus and activating transcription. LATS1/2 loss was expected to cause YAP/TAZ activation. Indeed, NP and Lats tumors all contained high levels of TAZ protein and abundant nuclear TAZ immunosignal (Figure 3F and I). Compared to the sciatic nerve, p16INK4a and p19ARF levels were elevated in NP tumors but not in Lats tumors.

We also examined the mRNA levels of these oncogenic drivers using our RNA sequencing (RNA-seq) data (see below). Consistent with the genotype, Nf1 and Trp53 mRNA levels were markedly lower in NP tumors than in Lats tumors (Figure 3G). However, Lats1 mRNA levels were only slightly lower in Lats tumors than in NP tumors. The reason for this is unclear, as our previous RNA-seq data[25] showed that Lats1 mRNA levels were markedly reduced in Lats1; Lats2;Nestin-Cre double knockout (dKO) brains compared to control brains (Figure 3H). Lats2 mRNA levels were higher in Lats tumors than in NP tumors, just as they were higher in Lats1;2 dKO brains than in control brains (Figure 3G and H). This was probably because of the feedback activation of the Lats2 gene by activated YAP/TAZ.[26]Yap and Taz mRNA levels were similar in NP and Lats tumors.

In summary, whereas neurofibromin and p53 proteins were absent in NP tumors but present in Lats tumors, NP and Lats tumors both lacked LATS1/2 proteins and contained high levels of TAZ. These results raise the possibility that signaling downstream of LATS1/2 is similarly perturbed in Lats and NP tumors but that signaling downstream of neurofibromin and p53 is perturbed to a greater extent in NP tumors than in Lats tumors.

Partial H3K27me3 Loss in NP-Plp and Lats GEM-MPNST, but not in NPcis GEM-MPNST

PRC2 catalyzes the di- and tri-methylation of histone H3 at lysine 27. Somatic LOF mutations in PRC2 components are common in MPNST, leading to H3K27me3 loss.[27-29] Immunostaining revealed that both grade II and III NP-Plp tumors showed varying degrees of H3K27me3 loss, which often exhibited intratumor heterogeneity (Figure 4A). In contrast, H3K27me3 signal was retained nearly uniformly in grade II and III NPcis tumors. The mechanism underlying H3K27me3 loss is unclear; we detected no significant reduction in Suz12, Eed, or Ezh1;2 mRNA levels in NP-Plp tumors as compared to NPcis tumors, nor consequential changes in their mRNA sequences. Lats-Nes and Lats-Plp tumors also showed varying degrees of H3K27me3 loss (Figure 4A and B).

Figure 4.

Partial H3K27me3 loss in NP-Plp and Lats GEM-MPNST, but not in NPcis GEM-MPNST. (A and B) H3K27me3 immunostaining and quantification (mean ± SEM). Unpaired, two-tailed t test. Boxed areas in the upper-row images are enlarged in the lower row. (C) Key histopathologic features of grade III GEM-MPNST in different models.

Figure 4C summarizes the key histopathologic features of grade III GEM-MPNST in these models. NP-Plp and Lats-Nes grade III tumors exhibited very similar features, with the only notable difference being less-frequent nuclear atypia in Lats-Nes tumors. However, NPcis tumors were quite distinct from NP-Plp and Lats-Nes grade III tumors, exhibiting higher cellularity, higher mitotic rates, and more frequent nuclear atypia while being less differentiated.

Nf1;p53-Driven and Lats-Driven GEM-MPNST Have Distinct Transcriptomes

We compared the transcriptomes of NP and Lats tumors. Because many tumors used for RNA-seq analysis lacked matching histologic specimens, we inferred tumor grade based on observations from histopathologic analyses (Supplementary Table 2). Both principal component analysis and unsupervised hierarchical clustering segregated NP and Lats tumors (Figure 5A and B). However, these analyses did not segregate tumors of the 2 NP models or of the 2 Lats models based on their genotype. Neither did these analyses segregate tumors based on tumor grade, with the caveat that the tumor grades were inferred and were undetermined for many samples. These results suggest that the transcriptomes of NP and Lats tumors are distinct but that the differences between the 2 NP models (Supplementary Table 3) and between the 2 Lats models are small.

Figure 5.

Nf1;p53-driven and Lats-driven GEM-MPNST have distinct transcriptomes. (A) Principal component analysis of NP and Lats tumors based on the top 3000 most varied genes. UD, tumor grade undetermined. (B) Dendrogram of unsupervised clustering of tumor samples based on the top 2000 most varied genes and the corresponding heatmap. (C) MA plot of differential expression comparing all NP tumors to all Lats tumors. Black dots denote genes with FDR ≤0.05. The numbers of significantly up- and downregulated genes with a fold change (FC) of ≥2.0 or ≤−2.0 are shown. (D) GSEA results showing differentially regulated Hallmark gene sets in NP vs Lats tumors. (E) MA plot of differential expression comparing grade III NP-Plp to Lats-Nes tumors. (F) GSEA results showing differentially regulated Hallmark gene sets in grade III NP-Plp vs Lats-Nes tumors. No Hallmark gene sets were upregulated in Lats-Nes tumors. (G) Unsupervised clustering of our NP and Lats tumors and the published Lats-Dhh tumors based on the top 2000 most varied genes. (H) Unsupervised clustering of tumor samples based on single-sample GSEA signatures.

Nearly 3000 genes were significantly (FDR≤0.05) up- or downregulated with a ≥2-fold change between NP and Lats tumors (Figure 5C). GSEA using the Hallmark gene sets revealed upregulation of gene sets related to growth, proliferation, and cancer, such as MYC_targets, Glycolysis, and PI3K_AKT_MTOR_signaling, in NP tumors (Figure 5D). Comparison of only grade III NP-Plp and Lats-Nes tumors found nearly 2000 genes significantly up- or downregulated (Figure 5E). Most Hallmark gene sets upregulated in NP versus Lats tumors were also upregulated in grade III NP-Plp versus Lats-Nes tumors (Figure 5F). These results further support that Nf1;p53-driven and Lats-driven GEM-MPNST have distinct transcriptomes and suggest that Nf1;p53-driven GEM-MPNST express higher levels of cancer-related genes than do Lats-driven GEM-MPNST, even when only grade III tumors are compared.

To confirm that the transcriptomes of NP and Lats GEM-MPNST were distinct, we analyzed the data for grade III Lats-Dhh (Lats1;2 deleted with Dhh-Cre) GEM-MPNST.[12] Unsupervised clustering clustered Lats-Dhh tumors with our Lats tumors, not with NP tumors (Figure 5G), suggesting that the transcriptome of Lats-Dhh tumors is similar to that of our Lats tumors and distinct from NP tumors. We further tested this by ssGSEA, which calculates a separate enrichment score for each pairing of a sample and a gene set, thereby transforming a sample’s gene expression profile to a gene set enrichment profile that represents the activity levels of biological processes and pathways in that sample.[19] For the gene sets, we used genes that were significantly regulated ≥4-fold in the NP versus Lats comparison or in the grade III tumor comparison with expression levels ≥3 log2 counts per million. Lats-Dhh tumors again clustered with our Lats tumors, not with NP tumors (Figure 5H). Together, both our own models and published Lats model indicate that Nf1;p53-driven and Lats-driven GEM-MPNST have distinct transcriptomes.

Nf1;p53-Driven GEM-MPNST Resemble Human MPNST More Closely Than Do Lats-Driven GEM-MPNST at the Transcriptome Level

Next, we asked whether Nf1;p53-driven or Lats-driven GEM-MPNST were more similar to human MPNST at the transcriptome level. We performed RNA-seq analysis of human MPNST, plexiform neurofibroma (PNF), and neurofibroma (NF) samples. We also included published RNA-seq data for human MPNST and PNF samples,[30] as well as for GEM-neurofibroma from Nf1;Dhh-Cre mice and paraspinal tumors and GEM-MPNST from Nf1;Ink4a/Arf;Dhh-Cre mice[7] (Supplementary Table 2). We first transformed the gene expression profile of each sample to ssGSEA signatures, which are less sensitive to technical effects and species differences than are gene-level comparisons. We then took the top 500 most varied gene sets across all mouse tumors as inputs for unsupervised dimensionality reduction of all human and mouse tumors with UMAP (Figure 6A and B). This analysis showed that most Nf1-driven GEM-MPNST clustered together, whereas most Lats-driven GEM-MPNST clustered with GEM-neurofibroma from Nf1;Dhh-Cre mice. Moreover, most Nf1-driven GEM-MPNST clustered with the majority of human MPNST samples, whereas Lats-driven GEM-MPNST clustered with most human PNF/NF samples.

Figure 6.

Cross-species transcriptomic comparison of mouse and human nerve sheath tumors. (A and B) UMAP plots of human and mouse tumors based on the top 500 most varied gene sets across all mouse tumors. Mouse and human tumors were analyzed together but are shown on separate plots for visual clarity. MPNST, malignant peripheral nerve sheath tumor; PNF, plexiform neurofibroma; NF, neurofibroma. (C) Unsupervised clustering of tumor samples based on single-sample GSEA signatures of the top 20 gene sets selected by the machine learning classifier that distinguishes human MPNST from PNF/NF samples. (D) Class probability scores obtained by the machine learning classifier. Tumors scoring >0.5 are likely to be MPNST; those scoring <0.5 are likely to be PNF/NF. All comparisons are against MPNST-TCGA. Unpaired, two-tailed t test.

To further quantify the similarity between the mouse models and human MPNST, we performed a supervised analysis based on machine learning to discriminate MPNST from PNF/NF samples in our SJ collection and the samples in GSE145064[30] based on ssGSEA signatures and used this analysis to derive MPNST class probability scores for the mouse models. The machine learning classifier was trained by using the top 20 gene sets that were most discriminative between MPNST and PNF/NF samples (Figure 6C). Using MPNST samples from TCGA as an independent baseline for comparison, our analysis found that the NP-Plp and NPcis models showed no statistical difference in MPNST class probabilities when compared to TCGA MPNST samples, whereas the other GEM-MPNST models all showed significant differences (Figure 6D). Together, the results of our unsupervised and supervised analyses suggest that Nf1;p53-driven GEM-MPNST resemble human MPNST at the transcriptome level more closely than do Lats-driven GEM-MPNST.

Conclusions

We established a new mouse model of MPNST by combining an Nf1;Trp53 cis-conditional allele and Plp-CreER. By allowing postnatal Nf1 and p53 loss, this NP-Plp model recapitulates the genetic events in MPNST patients better than existing mouse models, with the caveat that it does not recapitulate the frequent loss of CDKN2A/B in human MPNST.[1] It also recapitulates the histopathologic features of MPNST better than the NPcis model, which showed only occasional spindle-cell morphology and fascicular and storiform patterns typically found in MPNST, and closely resembles human MPNST at the transcriptome level (Supplementary Tables 4 and 5). This model is genetically simple, making it easy to maintain and an ideal platform for preclinical studies. Given its tamoxifen-inducible nature, this model can be used to study the time/stage dependency of the tumorigenic potential of Schwann cells.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.