Spatiotemporal Loss of NF1 in Schwann Cell Lineage Leads to Different Types of Cutaneous Neurofibroma Susceptible to Modification by the Hippo Pathway

Neurofibromatosis type 1 (NF1) is one of the most common genetic disorders of the nervous system and occurs in about 1 in 3,000 births. More than 25 years ago, independent laboratories discovered that mutations in the NF1 gene were responsible for causing NF1 in patients (1, 2). Subsequent genetic and biochemical analyses reveal that the protein product neurofibromin acts as a tumor suppressor and negative regulator of the RAS signaling pathway. The clinical spectrum of NF1 is very broad and affects many organ systems. Interestingly, the origin of most of the tissues affected in patients with NF1 can be traced back to neural crest cells, a stem cell population originating from the separation of the neural tube and ectoderm at the end of the neurulation phase in embryologic development. Examples of cells with a neural crest origin include Schwann cells, melanocytes, and glial cells. Indeed, loss of NF1 function in the Schwann cell lineage leads to neurofibroma formation, NF1^−/− melanocytes lead to café-au-lait macules (CALM; ref. 3) and Lisch nodules (iris hamartoma), and NF1^−/− in glial cells leads to astrocytomas (4).

There are several subtypes of neurofibromas. Most clinicians subdivide them according to their distinctive location in relation to the skin dermis and propensity for malignancy. Dermal or cutaneous neurofibromas (cNF) are strictly confined to the dermis and do not progress to frank malignant sarcoma lesions. They can be as numerous as thousands of tumors that cover most of the body surface. Consequently, daily activities such as showering, appearing in public, and mobility can be a significant burden. The only effective treatment options are surgical approaches, which are not practical in patients with numerous cNFs (5). Neurofibromas involving the internal nerve plexus are called plexiform neurofibromas (pNF). They are located below the dermis and have an increased risk of malignancy. In addition to their specific locations, the onset is drastically different. pNF is congenital, but cNF rarely appears before puberty. When a cNF is observed on a relatively large area of skin but still confined in the dermis and it is perfectly covered by skin pigmentation, it is called a diffuse cNF (6, 7). In this case, the skin is much thicker, the predominant body location is the head and neck and upper arms, and the pigment is localized in the dermis. Diffuse cNF is different from diffuse pNF which shares most of the diffuse cNF features except that it also involves the nerve plexus under the skin. Other subtypes of pNF include paraspinal and nodular. These are well-circumscribed sphere-like tumors that grow in the dorsal root ganglion (DRG) and peripheral nerve, respectively. Importantly, it is currently not known what causes different types of neurofibromas to develop. Thus, multiple subtypes of both cNF and pNF are described in the literature and vary according to their gross morphology, pigmentation, body location, and propensity to malignancy. The development of a cNF mouse model that faithfully recapitulates the human clinical scenario would allow researchers to study the progression of cNF and test potential antineurofibroma drugs, which is not feasible at the moment.

The report of the NF1 gene sequence started the race to build the first NF1 mouse model. Functional inactivation of Nf1 in all cell types is embryonically lethal in mice, but Nf1^+/− littermates (mimicking the genetic context of NF1 in patients) live long enough for spontaneous tumor formation studies (8). Disappointingly, the cardinal features of NF1 such as Lisch nodules, skin pigmentation, and neurofibromas did not develop in Nf1^+/− mice (8). These results suggest that loss of function of the second Nf1 allele may be a rate-limiting step in the cell of origin of neurofibroma. Coupled with the advent of the Cre-lox technology, dramatically increasing the number of Nf1^−/− cells in a tissue-specific manner in Nf1^+/− mice yields a robust mouse model of neurofibroma (9). However, all neurofibromas were of the plexiform type in this model. This suggests that cNFs have a different developmental origin than pNF.

As a first step toward the identification of the cell of origin of cNF, Le and colleagues (10) ablated Nf1 at adult stage in a localized manner in the skin by taking advantage of the topical activation of a nonspecific inducible system (CMV-Cre-ERT2;Nf1^f/−). It resulted in cNF-resembling tumors found in patients with NF1 and demonstrated that the cell of origin of cNF persists at adult stage and is present in the skin vicinity. One well-established source of neural crest–derived stem cells residing in the adult dermis is skin-derived precursor (SKP; ref. 11). To further investigate the cell of origin of cNF, Le and colleagues (10) isolated SKPs from CMV-Cre-ERT2;Nf1^f/− mice and successfully generated neurofibromas in an orthotopic mouse model. At this point, the next logical step was to build a germline mouse model using a Cre driver specific to SKPs to definitively identify the subpopulation of stem cells responsible for cNF. However, SKPs comprise heterogeneous populations, and as such SKP-specific Cre do not exist. As SKPs originate from the neural crest, several neural crest–specific Cre drivers (Wnt1, Pax3, and Mpz) were bred to Nf1^f/− mice to investigate their ability to drive neurofibroma development. Unfortunately, deletion of Nf1 in those models resulted in premature death, precluding any neurofibroma formation analysis (12).

We reasoned that the spatial expression (identity and number of cells involved) was not adequate as it may have hit too broad and vital a population of cells (13), although the timing (before neural crest cells bifurcate into cNF and pNF cell of origin) of the Cre tested was accurate (14). Therefore, our strategy was to identify and take advantage of mouse neural crest Cre lines that are expressed in the subpopulation of SKPs that give rise to cNF while sparing most of the vital neural crest–derived cells, as a balance between the right time and the right location. Here, we report the use of Hoxb7-Cre mice to lineage trace the specific cell of origin identity for cNF. We discovered that ablating Nf1 in the HOXB7 lineage–derived cells faithfully recapitulates both human cutaneous and plexiform neurofibroma. In addition, we discovered that the modulation of the Hippo pathway acts as a “modifier” to the development of neurofibroma.

To identify a putative cNF cell of origin lineage marker, we extensively searched the literature for a mouse Cre line with the highest chance of being expressed in a restricted population of neural crest cells giving rise to cNF. Classic studies on the morphogenesis of IX-X ganglionic complexes suggest that the cells expressing the HOXB7 protein derive from migrating neural crest cells (15). To lineage trace the HOXB7-expressing cell progeny, we crossed Hoxb7 mice with R26-LacZ and submitted E9.5–E12.5 whole-mount embryos to X-Gal staining (Fig. 1A). At E9.5, the neuropore has recently closed, which coincides with the initial delamination of neural crest cells. We observed HOXB7 expression mainly in the trunk neural crest as well as some punctate staining in the cranial region (Fig. 1A). At E10.5–11.5, some migratory neural crest cells stop at the future site of what will be the DRG, whereas other cells further migrate along the peripheral nerve. Note the apparent lower signal intensity due to the thickening of the skin at this developmental stage. At E12.5, some HOXB7 lineage–derived cells start to “reappear,” especially in the cranial region (Fig. 1A). This is because the migratory HOXB7 lineage–derived cells have now reached the skin surface; hence, the signal is visible once again. To confirm the results obtained with X-Gal staining, we performed immunohistochemistry using anti-HOXB7 antibodies on histologic sections from the chest and cranial level at E12.5. HOXB7 is expressed specifically in a subset of cells in the DRG (Fig. 1A, middle and bottom plots).

Next, we determined the fate of HOXB7 cells postnatally. As expected, cells originating from Hoxb7-derived lineage populated the DRG, the peripheral nerve as well as the nerve ending in the skin (Fig. 1B, left plot). In agreement with our observation in Fig. 1A, we obtained very little signal from the ventral motor nerve as compared with the dorsal sensory nerve when we studied the whole-mounted Hoxb7-Cre;LacZ X-Gal–stained mice (Fig. 1B, right plot). Histologic sections of DRG, intercostal nerve (IN), and skin confirm these findings. The LacZ reporter overlapped with the Schwann cell (S100β, GAP43) and the neuronal (βIII-tubulin) markers in the peripheral nervous system from DRG to dermis twigs (Fig. 1C). Finally, we confirmed the expression of HOXB7 in Schwann cells from human cNF (Fig. 1D). As the signal from the Schwann cell marker S100β is cytoplasmic and HOXB7 is nuclear, higher-magnification images confirm the colocalization of Schwann cell marker S100β and HOXB7 (Fig. 1D, insert). Altogether, these data suggest that cells from the HOXB7 lineage may be the cell of origin of cNF.

To demonstrate that the HOXB7 lineage in the skin can give rise to neurofibromas, we crossed the Hoxb7-Cre;LacZ mice to Nf1^f/f to generate Hoxb7-Cre;Nf1^f/f;LacZ. Next, we isolated, cultured, and implanted SKPs from Hoxb7-Cre;Nf1^f/f;LacZ newborn and control littermates into the sciatic nerves of nude mice (Fig. 2A; refs. 10, 16). After 3 to 4 months, mice implanted with SKPs from Hoxb7-Cre;Nf1^f/f;LacZ mice consistently generated neurofibromas that were LacZ positive (Fig. 2A), whereas none developed using cells derived from control littermates. The histologic hallmarks of neurofibroma such as disorganized nerve structure, hypercellular Schwann cells, and abundant collagen matrix were faithfully recapitulated (Fig. 2A). To further prove that the cells of origin of cNF are derived from the HOXB7 lineage, we decided to purify the HOXB7 lineage–positive SKPs by FACS. To do so, we crossed Hoxb7-Cre;Nf1^f/f to the fluorescent reporter mice R26-YFP, and isolated, cultured, and submitted these SKPs to FACS analysis. However, the fraction of HOXB7 lineage–positive SKPs recovered was about 10%, precluding the direct testing of this subpopulation of cells in our in vivo neurofibromagenesis assay as it required a larger amount of cells. This was consistent with the PCR analysis of DNA from bulk Hoxb7-Cre;Nf1^f/f SKPs showing a faint band for Nf1 ablation (Fig. 2A). As an alternative approach, we decided to use the bulk population of SKPs isolated from the Hoxb7-Cre;Nf1^f/f;YFP mice and quantified the number of YFP-positive cells before and after implantation (in the final neurofibroma tumor). We predicted that if the cNF cells of origin are truly derived from the HOXB7 lineage, then there should be an accumulation (enrichment) of YFP⁺ cells in neurofibroma. As expected, up to 80% of neurofibroma cells were YFP⁺ (Fig. 2B), providing additional support to our hypothesis that HOXB7 lineage–derived cells can give rise to neurofibroma when Nf1 is deleted and that the cell of origin of cNF is within the HOXB7 lineage population of SKPs.

To definitively demonstrate that cells derived from the HOXB7 lineage are the cell of origin of cNF, we analyzed the skin phenotype of a cohort of Hoxb7-Cre;Nf1^f/f (n = 16) and Hoxb7-Cre;Nf1^f/− (n = 9) mice hereafter called H7;Nf1mut mice. Strikingly, a large fraction (16 of 25, 64%) of H7;Nf1mut mice developed skin lesions by 1 year of age (Supplementary Table S1). The skin lesions had a relatively large surface of very thick skin, were fully pigmented, and developed mainly in the head and neck and upper limb areas (Fig. 3A), all characteristics reminiscent of the diffuse type of neurofibroma. To determine whether or not these lesions were restricted to the dermis (diffuse cNF) or connected to a nerve plexus below the skin (diffuse pNF), we carefully dissected the dorsal skin and lifted it up. All tumors examined (16 of 16, 100%) were exclusively confined to the dermis, as expected for diffuse cNF (Fig. 3B). Thus, the H7;Nf1mut mouse model completely recapitulates the characteristic body location, skin thickness, pigmentation, and cutaneous restriction of human diffuse cNF shown in Fig. 3C and D.

Next, we performed histologic evaluation of tumor specimens to confirm the skin lesion from the H7;Nf1mut to be of the diffuse cNF type. Hematoxylin and eosin (H&E) staining revealed hypercellularity in the dermis with cells harboring wavy nuclei typical of Schwann cells mixed with collagen fibers (Fig. 3E). X-Gal staining indicated that these LacZ⁺ cells derived from HOXB7 lineage (Fig. 3E, 2nd row plots). Immunofluorescence with Schwann cell–specific marker (S100β, GAP43) and immunohistochemistry with Sirius Red further confirmed these results along with mast cell staining (toluidine blue; Fig. 3E). In addition, accumulation of dermal pigmentation (Fig. 3E, top right plot insert) was observed. Importantly, all these features completely recapitulate human clinical samples of diffuse cNF (Fig. 3F). Furthermore, as it is known that the nerve microenvironment plays an important role in neurofibroma development (17, 18), immunostaining for the neuronal marker beta-III tubulin reveals consistent presence of nerve fibers in both mouse and human cNF tumors (Supplementary Fig. S1). Whereas topical application of tamoxifen in adult CMV-Cre-ERT2;Nf1^f/− gave rise to discrete cNF (10), biallelic Nf1 inactivation in a larger population of the cell of origin (HOXB7) embryonically gave rise to diffuse cNF.

HOXB7 lineage–derived cells do not solely populate nerve endings in the dermis, and hence may not be restricted to generate diffuse cNF upon Nf1 deletion. Our HOXB7 lineage–tracing experiment (Fig. 1) indicated an overlapping DRG and peripheral nerve staining pattern compared with PLP-Cre-ERT2;LacZ, a mouse model used to trace the cell of origin of pNF (16). This suggests that ablation of Nf1 in HOXB7 lineage–derived cells could trigger the development of pNF in addition to cNF, a phenomenon very frequent in patients with NF1. Some H7;Nf1mut mice began to show clinical signs characteristic of pNF development (scruffy fur, limping, or limb paralysis; ref. 16) as early as 5 months (Supplementary Table S1). To quantify pNF tumor burden, we performed a whole spinal cord dissection by carefully preserving peripheral nerves and DRGs in the cohort previously mentioned in Fig. 3. About half of the H7;Nf1mut mice showed characteristic enlargement of DRGs and peripheral nerves (Fig. 4A; Supplementary Table S1). Histologic characterization of both enlarged DRGs and peripheral nerves confirmed the distinguishing characteristics of pNF such as the presence of neoplastic Schwann cells in an abundant collagen matrix (Fig. 4B). Interestingly, a very similar number of mice developed either only diffuse cNF, only pNF, or both diffuse cNF and pNF (Supplementary Table S1). However, the similar mean age at death for the group of mice developing exclusively cNF or exclusively pNF (Supplementary Table S1) suggests that additional factors may play a role in the development of cNF and pNF. It is also possible that given more time, these mice will develop both diffuse cNF and pNF. To our knowledge, the H7;Nf1mut mouse is the first model to recapitulate both human cNF and pNF.

Although H7;Nf1mut mice develop classic diffuse cNF, they do not develop multiple discrete cNFs that cover the entire body surface. Therefore, we investigated potential modifiers that could favor the formation of classic cNF. The Hippo pathway regulates cell growth and has been implicated in a number of cancer types (19). Recent whole-exome sequencing and pathway analysis of human cNFs revealed somatic mutations in 3 genes [RASSF1A, SFN (encoding the protein 14-3-3), and DLG4] that belong to the Hippo pathway, warranting further investigation of the Hippo pathway in the context of patients with NF1 (20). In fact, patients with neurofibromatosis type 2 (NF2), which harbor a mutation in a key upstream regulator of the Hippo pathway, can develop some clinical features of NF1 such as cNFs (21, 22). Therefore, we hypothesized that the Hippo pathway may be a modifier of neurofibromagenesis. To extend the work of Faden and colleagues (20), we performed mutation analysis on a whole-genome sequencing dataset of 33 cNFs collected from 9 individual patients with NF1 (23) and found multiple germline mutations in genes involved in the Hippo pathway (Fig. 5A). Very interestingly, one particular tumor had a very high number of variants across the whole exome, including 30 somatic DNA mutations in the Hippo pathway (Fig. 5B). These results encouraged us to evaluate the activity of the key effectors of the Hippo pathway (YAP and TAZ). We detected a high expression level of YAP and TAZ in the nucleus of both human discrete and diffuse cNF (Fig. 5C). To make sure that the downstream targets of YAP and TAZ are transcribed in cNF, we confirmed that the gene expression levels of AXL and CTGF, two established genes modulated by the Hippo pathway (24), were upregulated compared with normal skin margin of patients with NF1 (Fig. 5D). Thus, the Hippo pathway is dysregulated in at least some human cNF tumors.

Consequently, we decided to test the impact of altering the Hippo pathway on the development of cNF in vivo. Because Lats1/2 regulate the intracellular localization of the key transcriptional effectors of the Hippo pathway (YAP and TAZ) through phosphorylation, ablating Lats1 and/or Lats2 leads to nuclear accumulation of YAP/TAZ and subsequent transcription of a cell growth gene expression program (19). We crossed the H7;Nf1mut mice to Lats1^f/f or Lats2^f/f mice to generate H7;Nf1mut mice harboring a combination of two flox alleles from Lats1 and/or Lats2 (H7;Nf1mut;Lats1^f/f or H7;Nf1mut;Lats2^f/f or H7;Nf1mut;Lats1^f/+;Lats2^f/+) and hereafter called H7;Nf1mut;Lats1/2mut mice. Strikingly, a fraction of H7;Nf1mut;Lats1/2mut mice (10.7%) develop a mixture of classic discrete nodular cNF as seen in patients with NF1 and spreading to almost the whole back skin (Fig. 5E, far left plots). Histologic evaluation of the diffuse (Fig. 5E, top panels) and nodular (Fig. 5E, bottom plots) mouse cNF confirms that the mouse lesions match human cNFs.

The impact of the modifier can also be appreciated in the diffuse cNF variant. The skin is thicker with infiltrating cNF in the H7;Nf1mut;Lats1/2mut mice (Fig. 5F). cNFs also develop in broader areas of different body locations in the H7;Nf1mut;Lats1/2mut mice (Fig. 5G, left plot; Supplementary Table S1). Because none of the control mice (Hoxb7-Cre;Nf1^+/+;Lats1^f/f;Lats2^+/+ or Hoxb7-Cre;Nf1^+/+;Lats1^+/+;Lats2^f/f or Hoxb7-Cre;Nf1^+/+;Lats1^f/+;Lats2^f/+, hereafter called H7;Lats1/2mut mice) develop any type of cNF (Fig. 5G, right plot), it indicates that the Hippo pathway acts as a modifier for cNF tumorigenesis.

In addition, Nf1 heterozygosity has long been considered as an important supportive factor to sustain neurofibroma tumor microenvironment in Krox20-Cre mouse models (9). However, the requirement of Nf1^+/− microenvironment was not fully emulated in another neurofibroma model system by using Dhh-Cre (25), suggesting that Nf1 haploinsufficiency might contribute to neurofibroma development in a context-dependent manner. Therefore, we also reevaluated the contribution of Nf1 heterozygosity by comparing the tumor progression between Nf1^f/f and Nf1^f/− mice with Hoxb7-Cre mice. We found that both Hoxb7-Cre;Nf1^f/− and Hoxb7-Cre;Nf1^f/f from both H7;Nf1mut and H7;Nf1mut;Lats1/2mut mice in Supplementary Table S1 succumbed to neurofibroma development. However, mice in the Nf1^f/− group have thicker skin with infiltrating cNF, and they died from tumor development much faster than those in the Nf1^f/f group (Supplementary Fig. S2). These findings revealed that Nf1 heterozygosity is not absolutely required for neurofibroma development. However, inclusion of Nf1 heterozygosity in nontumor cells significantly enhanced neurofibroma progression, suggesting that germline Nf1 heterozygosity is a modifying factor for neurofibroma development.

Because alteration of the Hippo pathway modified the growth of cNF, we hypothesized that it may also modify the development of pNF. To determine pNF tumor burden, we further performed whole spinal cord dissection as described in Fig. 4 to quantify pNF development in the mice cohort previously mentioned in Fig. 5. Indeed, paraspinal neurofibromas developed at DRGs are substantially larger in H7;Nf1mut;Lats1/2mut compared with H7;Nf1mut (Fig. 6A). Accordingly, histologic evaluation revealed a higher cellularity, chaotic distribution of neuron bodies in H7;Nf1mut;Lats1/2mut DRGs compared with H7;Nf1mut (Fig. 6B). Moreover, the number of paraspinal NFs causing enlarged DRGs was significantly higher in H7;Nf1mut;Lats1/2mut (Fig. 6C). Similarly, the diameter of large peripheral nerve such as the sciatic nerve was increased in H7;Nf1mut;Lats1/2mut mice (Fig. 6D and F). Histologic evaluation revealed that the nerve fiber high disorganization and hypercellularity found in H7;Nf1mut;Lats1/2mut was almost completely absent in H7;Nf1mut and H7;Lats1/2mut sciatic nerves (Fig. 6E). Thus, we conclude that the Hippo pathway acts as a modifier to promote paraspinal and peripheral nerve pNF development.

Finally, we investigated the impact of interfering with the Hippo pathway on overall survival using a Kaplan–Meier plot. At 200 days, about 50% of H7;Nf1mut;Lats1/2mut had already died because of extensive cutaneous tumors or pNF requiring sacrifice, whereas more than 80% and 100% of H7;Nf1mut and H7;Lats1/2mut, respectively, survived (Fig. 6G). Altogether, it indicates that the Hippo pathway is acting as a modifier of the plexiform in addition to cutaneous neurofibromagenesis and suggests that dampening the Hippo pathway could be an attractive therapeutic target.

Recent studies have shown that estrogen from females correlates with the prevalence of optic glioma, another NF1-associated neoplasm (26, 27). Therefore, in this study, we also utilized our survival data set from both H7;Nf1mut and H7;Nf1mut; Lats1/2mut mice in Supplementary Table S1 to further characterize the pNF development in this new model and to compare the progression of neurofibromas in male and female mice. Although we observed that 100% of female mice developed pNF compared with 71% of their male counterparts, both male and female mice developed similar numbers of pNF in both cervical and thoracic dorsal root ganglia. Furthermore, although more male mice survived beyond 400 days than female mice, the Kaplan–Meier survival analysis between male and female mice is not statistically significant (Supplementary Fig. S3).

To probe the molecular mechanistic link between the Hippo pathway deregulation and its effect on neurofibromagenesis which is driven by NF1 loss leading to MAPK pathway activation, we performed immunohistochemical analysis to monitor MAPK pathway activity using phosphorylated ERK (pERK1/2) as a surrogate marker. Indeed, pERK1/2 is increased in H7;Nf1mut;Lats1/2mut when normalized to total ERK and compared with H7;Nf1mut in both cNF and pNF (Fig. 7A and B). As expected, histologic evaluation of skin and DRG from H7;Lats1/2mut revealed minimal change in the normal tissue organization and confirmed the role of the Hippo pathway as a modifier of neurofibroma (Fig. 7A and B, middle plots). We confirmed these results with western blot analysis by quantifying pERK1/2 in two sets of experiments. First, we compared mouse embryonic DRG/Nerve root Schwann cell precursors (the cell of origin for pNF; ref. 16) with or without ablation of one or two copies of Lats2 and observed an increase in the level of pERK1/2 (Fig. 7C). Second, we compared skin and DRG from wild-type mice versus H7;Nf1mut, H7;Lats1/2mut, and H7;Nf1mut;Lats1/2mut mice and observed again an increase in the level of pERK1/2 in tissues from H7;Nf1mut;Lats1/2mut mice (Fig. 7D). Altogether, we reasoned that the Hippo pathway deregulation enhances the MAPK pathway activation by NF1 loss.

Patients with NF1 present with a wide range of phenotypic manifestations, including tumors that develop at different time points during their lifetime. Several NF1 mice faithfully model human NF1-associated tumors such as pNF, astrocytomas, malignant peripheral nerve sheath tumors (MPNST), and pheochromocytomas, both histologically and with regard to NF1 mutation status (28). However, the most prevalent pathologic signs of NF1 occur in the skin in the form of cNFs and pigmentation. Ironically, these characteristic features are currently the most challenging to model in vivo. We discovered that deleting Nf1 in the HOXB7 lineage–derived cells generates both cNF and pNF spontaneously, faithfully recapitulating human neurofibromatosis. In addition to the histologic signatures and NF1 status, this novel NF1 mouse model highlights three more hallmarks of human neurofibroma: (1) neurofibroma exclusively affects the dorsal sensory nerve fibers; (2) diffuse cNF is consistently present with pigmented skin over cNF; and (3) neurofibroma development is susceptible to modifiers.

From a developmental perspective, most of the tissues sensitive to NF1 gene dosage are derived from the neural crest. The neural crest generates an important number of tissues above the shoulder, including the facial muscles and bones (29) as well as the peripheral nervous system (14). Indeed, Nf1 LOH in Schwann cell lineage leads to neurofibroma development (9, 30), Nf1^−/− melanocytes lead to CALM (3) and Lisch nodules (iris hamartoma), Nf1^−/− osteoblasts lead to pseudoarthrosis of the tibia (31), and Nf1^−/− in glial cells leads to astrocytomas (4). Attempts to study the development of neurofibromas through the use of a neural crest–specific Cre driver in these mice have failed because mice do not survive long enough to allow the development of phenotypic manifestations (12). To circumvent this issue, Schwann cell–specific Cre were successfully used (9, 17, 25, 32, 33). Because cNF and pNF have identical histologic features including neoplastic Schwann cells, one can expect to find a mixture of cNF and pNF in these mouse models. Strikingly, none of these models develop cNF (9, 17, 25, 32, 33). This means that although both tumors contain Schwann cells, they may not come from the same cell of origin. In fact, the biology of these tumors is widely different. First, pNF involves internal nerve plexuses and can further progress to a malignant tumor, whereas a malignant form of cNF has rarely if ever been reported. Second, the onset of pNFs can be as early as the embryonic stage of development, whereas the cNFs typically occur around puberty and cNFs are exclusively in the skin. Thus, the presence of at least two cell-of-origin populations within the Schwann cell lineage may explain the different types of neurofibroma due to inactivation of NF1 in two different spatiotemporal locations.

As cells of origin for both cNF and pNF originate from the neural crest, one can then envision three scenarios (34): If the Nf1 LOH occurs before their bifurcation into distinct lineage, then it will lead to both cNF and pNF development (as in Hoxb7-Cre). If Nf1 LOH occurs only in the pNF cell of origin after the bifurcation, only pNFs develop (as in PLP-Cre and Krox20-Cre). On the other hand, if Nf1 LOH occurs only in the cNF cell of origin, then this will lead to cNF development without pNF (as in topical induction with CMV-Cre-ERT2). At first sight, this model does not seem to reflect what is observed in clinic where most if not all patients with NF1 who develop pNF also have cNF. One explanation for this may be the mechanism by which mutations or LOH is introduced in mouse models but occurs spontaneously in humans. In order to develop pNF, biallelic Nf1 inactivation has to occur precisely at a defined embryonic stage, a narrow window of opportunity. In humans, this window spans only a short amount of time during embryonic development, and the “2nd hit” happens stochastically. In contrast, in mice, the time and cell type where the Nf1 biallelic inactivation occurs is controlled and purposely made in the pNF cell of origin. Thus, a minority of patients with NF1 develop pNF, and even if they do, most of them will eventually also develop cNF because the window for developing solely pNF is very small. On the other hand, the cell of origin of cNF is present throughout adulthood, giving ample time to undergo Nf1 inactivation. This may explain why a near-totality of adult patients bears cNF. Thus, our model reconciles the apparent discrepancies between in vivo mouse experiments and human clinical scenarios.

Modeling NF1 skin pigmentation in mice is not trivial because mice and humans have different pigmentation mechanisms (35). Human skin gets pigmented because its epithelial cells uptake the melanin produced by melanocytes through stem cell factor (SCF)–dependent signaling, resulting in interfollicular pigmentation. In mice, the production of SCF in interfollicular epithelial cells drastically drops at about 2 weeks after birth. Therefore, melanocytes still produce melanin but only in the hair follicle. Consequently, murine skin color is dependent on the presence of hair and follows the temporal cycle of hair growth. In addition, it is technically challenging to formally demonstrate the origin of melanocytic cells due to the inherent nature of dark pigmentation overwhelming any colocalization signal. Therefore, and not surprisingly, the mechanism underlying pigmentation defect in the context of NF1 is still a matter of debate and remains elusive.

Nf1^+/− mice exhibit a subtle darkening of the skin (36), and biallelic inactivation of Nf1 using melanocyte-specific Cre mice phenocopies this trait (36, 37). The hyperpigmentation phenotype in these models is spread across the body, and accumulation of pigmented cells takes place in the dermis. Of note, the skin darkening can be easily overlooked by the naked eye. Because Nf1^−/− melanocytes from the TYR and MITF lineage also originate from neural crest cells and HOXB7 expression coincides with the first wave of neural crest cell migration, it is possible that these melanocytic lineage cells are in the Hoxb7 lineage. Consistent with this, we observed more pronounced pigmentation of tails and paws in H7;Nf1mut mice, suggesting that the melanocytes responsible for the global and subtle pigmentation in NF1 may be derived from the neural crest–derived melanocyte within the HOXB7 lineage.

The fact that in diffuse cNF the skin hyperpigmentation surface matches the tumor area in the dermis suggests that Nf1^−/− Schwann cell and Nf1^−/− melanocytes may share the same cell of origin (a neural crest cell progenitor). Indeed, very early in neural crest migration, SOX10⁺ MITF⁺ stem cells (melanocyte precursor) migrate in a dorsolateral pathway to reach the skin dermis (38), whereas PLP⁺ and Krox20⁺ cells (Schwann cell precursors) start to appear at E10.5 (16). However, this hypothesis has never been demonstrated in mice because the Cre mice strains used up to now were not early enough in the neural crest lineage to encompass the melanocyte lineage. Strikingly, we observed that ablating Nf1 in the HOXB7 lineage results in the concomitant generation of pigmented skin on top of cNF. As seen in patients with NF1, the location of the tumor is mostly restricted to the head and neck area. We conclude that the type of cells, their location, and the timing of NF1 loss along the developmental course may dictate skin pigmentation in patients with NF1.

Several lines of evidence indicate that modifiers may explain some of the high phenotypic variability observed within a subgroup of patients with NF1 sharing the same germline mutation. A genome-wide high-resolution array-comparative genomic hybridization strategy identified a SNP in the gene ANRIL that is associated specifically with the genesis of pNF (39). Interestingly, type 1 microdeletion encompassing the NF1 gene is associated with a high number of cNF and an increased risk for MPNST in patients with NF1. In the context of MPNST, the modifier function has been genetically linked to SUZ12 (40), but it is currently unknown if SUZ12 or any other gene that is commonly inactivated with NF1 in the setting of these large gene deletions plays a role in determining the number of cNF. Other interesting candidate genes were identified and need further validation (41, 42).

A robust cNF mouse model recapitulating human cNF represents an important milestone. H7;Nf1mut mice develop a robust diffuse subtype of cNF rather than the discrete cNF found in many patients with NF1. Therefore, we were looking for factors beyond the cell type of origin and the timing of the Nf1 biallelic inactivation that could modify the development of cNF (body location, time of onset, or gross morphology) to be more representative of discrete human cNF. In the context of mouse pNF, the Nf1^+/− microenvironment is known to promote the development of pNF (9). In fact, and depending on the context, the Nf1^+/− microenvironment may be absolutely required (9), without significant effect (17), or anywhere between these two extremes. In the current study, we did not find any significant difference in the percentage of mice developing cNF between Nf1^+/+ and Nf1^+/− backgrounds, although mice with the Nf1^+/− background develop both cNF and pNF and succumbed to death earlier than the Nf1^+/+ background. Therefore, we looked for alternative modifiers of neurofibromagenesis.

The Hippo pathway, which regulates cell growth and has been implicated in a number of cancer types (19), may be altered in a subset of cNF based on whole-exome sequencing analysis (20). We reported here a high expression level of the key effectors of the Hippo pathway (YAP and TAZ) in the nucleus of both human discrete and diffuse cNF. Our data support the clinical fact that individuals with mutation in the Hippo regulator NF2 develop neurofibromatosis type 2, where cNF can grow (21, 22). Moderately altering the Hippo pathway in addition to Nf1 deletion modifies the course of neurofibroma development. H7;Nf1mut;Lats1/2mut mice generate higher numbers of enlarged DRGs and peripheral nerves as well as much thicker skin lesions where cNFs developed. There is also a trend toward a faster tumor onset compared with H7;Nf1mut mice and a broader body location for tumor distribution. Strikingly, nodular cNFs similar to the classic discrete cNF found in a patient with NF1 were observed in a subset of H7;Nf1mut;Lats1/2mut mice that were not found in control littermates (H7;Nf1mut or H7;Lats1/2mut mice). In contrast to diffuse cNF, these tumor nodules were not restricted to a particular body location, a characteristic shared with the human scenario. Furthermore, moderately affecting the Hippo pathway alone (H7;Lats1/2mut mice) does not produce any skin or nervous system tumor even after 1 year of age. It is also possible that we underestimated the number of mice developing discrete cNF, as the high tumor burden of diffuse cNF and pNF required early sacrifice of mice. Mechanistically, there are many ways NF1 could interact with the Hippo pathway. Because it is known that NF1 plays an inhibitory role in the MAPK pathway in cNF, and it is also known that there is cross-talk reported between the MAPK and the Hippo pathway in some cancer types (43–45), one possibility could be that the Hippo pathway is modulated indirectly due to NF1 loss of function–induced increased MAPK signaling. Alternatively, the Hippo pathway could be tempered in neurofibroma through genetic mutation in the Hippo pathway (ref. 20; Fig. 5A). This suggests that dampening the Hippo pathway may serve as part of a comprehensive treatment approach for cNF and pNF, specifically using new inhibitors in development for human clinical trials that target the Hippo pathway (46).

In conclusion, we discovered that HOXB7 serves as a lineage marker to trace the developmental origin of cNF neoplastic cells. Ablating Nf1 in the HOXB7 lineage faithfully recapitulates both human cutaneous and plexiform neurofibroma. In addition, we discovered that the Hippo pathway acts as a modifier to promote neurofibroma tumorigenesis. This novel mouse model opens the doors for deciphering the evolution of cNF to identify effective therapies, where none exist today.

The Nf1 knockout (8), Nf1 flox (9), Hoxb7-Cre (47), Lats1 flox (48), Lats2 flox (48), ROSA26-YFP (49), ROSA26-lacZ, and athymic nude mice (Foxn1^−/−) are available from Jackson Laboratory. Genotyping was performed by PCR as previously reported (8, 9, 47–49). All mice were housed in the Animal Care Facility at The University of Texas Southwestern Medical Center, and all procedures were approved by the Institutional Animal Care and Use Committee at The University of Texas Southwestern Medical Center and conformed to NIH guidelines. Human subjects and cNF sample collection and use were approved by the Institutional Review Board at The University of Texas Southwestern Medical Center and conformed to NIH guidelines. Written informed consent was obtained from patients.

SKP isolation, culture, and sciatic nerve implantation protocols were performed as previously described (50). Briefly, dorsal skin was cut to yield about 1 square inch of tissue and was dissected into small pieces. To isolate SKP cells, the suspension of tissue pieces was centrifuged (30 seconds, 1,200 rpm), supernatant was discarded and resuspended in a solution of PBS (Sigma, D8662; 8 mL) and collagenase I (2 mL) by inversion, and it was moderately shaken (37°C, 40 minutes). Next, tissue pieces were mechanically disrupted by up and down pipetting (10 times) and were centrifuged (10 minutes, 2,000 rpm, 4°C), and supernatant was discarded and resuspended in ice-cold Hank's Balanced Salt Solution (10 mL). The procedure was repeated twice starting back at the mechanical pipetting disruption step. Finally, the resuspension was filtered on a 70-μm sterile cell strainer, and viable cells were counted (trypan blue). To enrich for SKPs, 1 × 10⁶ cells per 10 cm dish were seeded on ultra-low culture dish (Corning; 3262). After 3 to 4 days, half of the supernatant (5 mL) was discarded, and fresh SKP complete medium (5 mL) was added.

To make SKP base medium, 0.22 μm heparin 0.2% (1 mL), glucose 30% (1 mL), NaHCO₃ 7.5% (750 μL), HEPES 1 mol/L (250 μL), glutamine (5 mL), sodium pyruvate (5 mL), penicillin–streptomycin (5 mL), and N2 supplement (Gibco; 1750248; 5 mL) were mixed into 500 mL of DMEM/F12 and filtered at 0.22 μm. To make complete SKP medium, basic FGF (Sigma; F0291; 0.16 ng), EGF (Gibco; 13247-051; 0.8 ng), amphoterin B (160 μL), and B27 (800 μL) were added into SKP base media (40 mL) for use within 2 weeks.

Tissues were fixed in 10% formalin-buffered solution for at least 48 hours. Then tissues were paraffin embedded (Rushabh Instruments LLC), sectioned at 5 μm using a microtome (Leica RM2135), and allowed to dry on glass slide at room temperature. Tissue slides were deparaffinized, were progressively rehydrated, and were stained with hematoxylin (2 minutes) followed by high definition (10 seconds), bluing agent (10 seconds), and eosin (H&E staining) or a solution of Sirius Red (0.5 g of Direct Red80 dissolved in 500 mL of saturated picric acid) for 1 hour (collagen staining) or a solution of toluidine blue [0.05 g of toluidine blue O dissolved in 70% ethanol (5 mL) and freshly further diluted in 1% sodium chloride (45 mL)] for 2 minutes (toluidine blue staining). Finally, tissue slides were progressively dehydrated and coverslipped.

Tissues were briefly incubated in 4% paraformaldehyde (10 minutes, room temperature), rinsed twice with PBS 1X followed by incubation into X-Gal staining solution (X-Gal 1 mg/mL, KFerro 1X, KFerri 1X, MgCl₂ 50 mmol/L) for 24 to 48 hours, and finally fixed with 10% formalin-buffered solution.

Tissue slides were blocked with 10% donkey serum in PBS (1 hour) and were incubated with primary antibodies [Rabbit anti-YAP (Cell Signaling Technology; AB-2650491); Rabbit anti-TAZ (abclonal; AB-2721146); Rabbit anti-HOXB7 (Novus; AB-2721144); Rabbit anti-S100β (Dako; AB-10013383); Rabbit anti-GAP43 (Abcam; AB-2247459); Rabbit anti–βIII-tubulin (Sigma-Aldrich; AB-262133)] diluted in 3% donkey serum (16 hours, 4°C). The next day, tissue slides were rinsed in PBS (3 × 5 minutes), incubated with secondary antibodies coupled to biotin and diluted in 3% donkey serum (1 hour), rinsed again in PBS (3 × 5 minutes), incubated with a premixture of avidin and biotin (following Vecta Stain Elite ABC kit procedure; 30 minutes), rinsed again in PBS (3 × 5 minutes), and were visualized by adding the DAB substrate (following Vecta Stain Elite ABC kit procedure). Finally, reaction was quenched in distilled water, and tissue slides were counterstained with hematoxylin, dehydrated, and cover slipped. A brown precipitate was deposited on positive cells.

Tissue slides were blocked with 10% donkey serum in PBS (1 hour) and were incubated with primary antibodies [Rabbit anti-S100β (Novus; AB-2184423); Rabbit anti-GAP43 (Abcam; AB-2247459); and Chicken anti-GFP (Aves Labs; AB-10000240)] diluted in 3% donkey serum (16 hours, 4°C). The next day, tissue slides were rinsed in PBS (3 × 5 minutes), incubated with secondary antibodies coupled to a fluorophore from Jackson Immunoresearch (Cy3-AffiniPure Donkey anti-Rabbit; AlexaFluor488, Donkey anti-Goat), diluted in 3% donkey serum (1 hour), and rinsed again in PBS (3 × 5 minutes). Slides were visualized under a fluorescence microscope.

Skin lesion body location was classified as follows: Head and neck is defined as diffuse cNF located at least above the shoulders; upper back tumor is defined as diffuse cNF located at least between the shoulders, and the upper 50% of the dorsal skin and bottom back is defined as diffuse cNF where tumor is located at least in the lower 50% of the dorsal skin. The number of mice with diffuse cNF in these three body locations was expressed as percentage over the total number of mice for each genotype. After euthanasia, a skin incision along the spinal cord was performed, and the skin was lifted up to evaluate potential involvement of the subcutaneous region. Finally, skin tumors of interest were excised, rinsed with PBS 1X, and either quickly immersed in 10% formalin-buffered solution or submitted to X-Gal staining procedure.

Whole spinal cord dissection was performed as previously reported (33). To perform mouse anesthesia, a mixture of ketamine (10 mg/mL) and xylazine (1 mg/mL) solution (100 μL per 25 g of mouse) was administered intraperitoneally. After 15 to 20 minutes, mouse was placed face up in a surgical field, and the chest area was sprayed with 70% ethanol. The left thoracic cage was removed, a catheter was installed in the heart left ventricle, and the mouse was perfused intracardially with 4% paraformaldehyde. Then, the mouse was prepared for microscopic dissection by removing gross tissue (cervical decapitation, whole skin removal, and all internal organs). Next, muscle and other tissue were carefully removed, and bones from the vertebrate column were broken one by one under dissection microscope to end up with intact spinal cord and peripheral nerves. Finally, the whole spinal cord and peripheral nerves were rinsed with PBS 1X and either immersed in 10% formalin-buffered solution or submitted to X-Gal staining procedure.

To quantify the thickness of skin samples, digital H&E images were reviewed, and a line was drawn between epidermis and the tumor area (perpendicular to the epidermis layer) using the cellsenstandard software. Then, the longest line (in mm) was used as a data point to generate the scatter plot in Fig. 5 (one data point per mouse).

To quantify the sciatic nerve diameter, sciatic nerve from dissected whole spinal cord was reviewed, and the point at which L3, L4, and L5 merge was selected as the measurement location (51). Then, the left and right nerve diameters were measured using a Vernier caliper and were used to generate the scatter plot in Fig. 6F (two data points per mouse).

To quantify the number of paraspinal neurofibromas, all DRGs from dissected whole spinal cord were measured using a Vernier caliper. DRGs with a diameter of at least 1 mm were considered neurofibromas (52). The numbers of neurofibromas were counted and were used to generate the scatter plot in Fig. 6C (one data point per mouse).

To estimate the fraction of SKPs derived from HOXB7 lineage, SKPs from Hoxb7-Cre;Nf1^f/f;YFP were passaged, and final resuspension was done in sterile PBS. SKPs were analyzed on a FACSAria instrument (BD Biosciences) equipped with a 488 nm solid-state laser where green fluorescence was detected using 490 LP and 510/20 filters. The fraction of YFP-positive SKPs was estimated using FACSDiva software. Analyses were repeated on SKP cultures derived from at least 3 mice.

To estimate the fraction of cells in neurofibroma derived from HOXB7 lineage, SKPs from Hoxb7-Cre;Nf1^f/f;YFP were implanted into the sciatic nerve of nude mice. Resulting neurofibromas were analyzed by immunohistochemistry using anti-GFP antibodies for the presence of GFP-positive cells. Total cell number (nuclei) and GFP-positive cells were counted in at least 3 random fields from at least 3 different neurofibroma-bearing mice.

RNA was extracted from 30 mg of tissue using the TRIzol reagent followed by on-column DNAse digestion as described elsewhere (53). Reverse transcription step was performed using iScript Select cDNA synthesis kit as per the manufacturer's recommended procedure using a random priming strategy. Real-time PCR assay validation for AXL (5′-TGGAAGGCCAGCTCAACCAG-3′ and 5′-TGCAGACCGCTTCACTCAGG-3′) and CTGF (5′-ACCTGTGCCTGCCATTACAA-3′ and 5′-GCTTCATGCCATGTCTCCGT-3′) was performed as described elsewhere (54) and normalized with GAPDH (5′-AGGGCTGCTTTTAACTCTGGT-3′ and 5′-CCCCACTTGATTTTGGAGGGA-3′).

To investigate Hippo pathway mutations in a clinical dataset, somatic and germline variant calls were acquired from the CTF Cutaneous Neurofibroma Data Resource: http://doi.org/10.7303/syn4984604 (23). This database includes summarized germline mutation data from 9 patients and somatic mutation data from 33 patient-matched cNF samples. All analysis was performed in R 3.5 (https://r-project.org/). Hippo pathway genes were identified by taking the union of genes listed in the literature and in the Kyoto Encyclopedia of Genes and Genomes database (hsa04390; ref. 55). Somatic and germline mutations in the Hippo pathway were summarized and plotted using the R package GeneVisR (56). The scripts used to summarize the variants and generate this figure are available here: http://doi.org/10.5281/zenodo.1273759.

All data are displayed as the mean ± SEM unless specified otherwise. A two-tailed t test and Fisher exact test were applied as appropriate to evaluate statistical significance (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

No potential conflicts of interest were disclosed.

Conception and design: Z. Chen, L.Q. Le

Development of methodology: Z. Chen, C.-P. Liao, L.Q. Le

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Chen, J. Mo, J.-P. Brosseau, Y. Wang, J.M. Cooper, T.J. Carroll, L.Q. Le

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Z. Chen, J.-P. Brosseau, J.M. Cooper, R.J. Allaway, J. Guinney, L.Q. Le

Writing, review, and/or revision of the manuscript: Z. Chen, J.-P. Brosseau, R.J. Allaway, J. Guinney, L.Q. Le

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Mo, J.-P. Brosseau, T. Shipman, Y. Wang, L.Q. Le

Study supervision: S.J.C. Gosline, L.Q. Le

We thank all members of the Le lab for helpful suggestions and discussions. J.-P. Brosseau and C.-P. Liao are recipients of the Young Investigator Award from the Children's Tumor Foundation. C.-P. Liao also received a Career Development Award from the Dermatology Foundation. J.M. Cooper is funded by the Dermatology Research Training Program T32 Grant T32AR065969. L.Q. Le holds a Career Award for Medical Scientists from the Burroughs Wellcome Fund and the Thomas L. Shields, M.D. Professorship in Dermatology. This work was supported by funding from the NCI of the NIH grant number R01 CA166593, the U.S. Department of Defense grant number W81XWH-17-1-0148, the Giorgio Foundation, the Neurofibromatosis Therapeutic Acceleration Program, and the NF1 Research Consortium Fund to L.Q. Le.