A molecular basis for neurofibroma-associated skeletal manifestations in NF1.

PURPOSE: Plexiform neurofibromas (pNF) develop in children with neurofibromatosis type 1 (NF1) and can be associated with several skeletal comorbidities. Preclinical mouse studies revealed Nf1 deficiency in osteoprogenitor cells disrupts, in a MEK-dependent manner, pyrophosphate (PPi) homeostasis and skeletal mineralization. The etiology of NF-associated skeletal manifestations remains unknown.
METHODS: We used mouse models of NF1 neurofibromas to assess bone mineralization of skeletal structures adjacent to tumors. Expression of genes involved in pyrophosphate homeostasis was assessed in mouse and human NF tumors and Schwann cell cultures. We used dual-energy X-ray absorptiometry (DXA) to assess tumor-associated changes in bone mineral density (BMD) in an individual with NF1 following treatment with the MEK inhibitor selumetinib.
RESULTS: We detected increased nonmineralized bone surfaces adjacent to tumors in mouse models of NF1 neurofibromas. Expression of Enpp1, a PPi-generating ectophosphatase, and ANKH, a PPi transporter, was increased in mouse and human neurofibroma-derived tissues and Schwann cells, respectively. In one patient, tumor-associated reductions in BMD were partially rescued following therapy with selumetinib.
CONCLUSION: Results indicate that NF-associated skeletal pathologies in NF1 are associated with dysregulated pyrophosphate homeostasis in adjacent NF tumors and suggest that treatment of NFs with MEK inhibitors may improve skeletal manifestations of the disease.

INTRODUCTION

Neurofibromatosis type 1 (NF1) is a cancer-predisposing genetic syndrome caused by inherited or sporadic mutations in the NF1 gene[1]. Twenty to 50% of individual with NF1 present with plexiform neurofibromas (NF)[2,3], which are congenital tumors that grow over lifetime and are thus more apparent in older individuals. These Schwann cell-derived tumors enlarge most prominently during the first decade of life[4,5] and, like many other NF1 manifestations, arise from somatic mutations of the wildtype NF1 allele or from loss of NF1 heterozygosity. Loss of neurofibromin, encoded by NF1, allows for persistent RAS activity and thereby leads to activation of the RAS/MEK/ERK pathway. Recently, the MEK inhibitor Selumetinib was shown to significantly reduce NF burden in children with NF1[6,7].

As many as 30% of individuals with NF1 develop skeletal manifestations, including osteopenia, tibial dysplasia and pseudarthrosis, or scoliosis [8]. While somatic loss of NF1 in bone cells was detected in tibial pseudarthrosis[9-11] and dystrophic scoliosis cases[12], and while the different types of spinal deformities (with dystrophic features or not) in NF1 are well described[13-15], NF-associated skeletal manifestations have been less investigated. Individuals with NF1 presenting with paraspinal neurofibromas have six-fold higher odds of developing scoliosis compared to patients without paraspinal tumors, and up to 63% of NF1 individuals with dystrophic scoliosis have evidence of paraspinal tumors on MRI[16-18]. Although “erosive” defect of bone from contiguous neurogenic tumors were suspected, no direct evidence beyond X-rays and BMD are available to explain how tumors may alter adjacent skeletal elements[19,20]. Figure 1A,B illustrates one such patient with rapidly progressing dystrophic scoliosis associated with vertebral body and rib deformities immediately adjacent to a plexiform neurofibroma. Plexiform tumors affecting the extremities may also present with progressive skeletal deformity, bone loss and length discrepancy. Figure 1C,D depicts a patient with unilateral leg NF and progressive deformity of the pelvis, acetabulum and proximal femur associated with the tumor. Subtrochanteric femoral diameter of the tumor-laden leg in this patient decreased over 7 years (Figure 1E). Thus, skeletal co-morbidities exist at multiple sites among individuals with NF1 and NFs.

Figure 1.

Plexiform neurofibroma-associated NF1 skeletal manifestations.

(A) Standing X-ray (top) and MRI (bottom) of an adolescent female patient with NF1 NF-associated dystrophic scoliosis. The scoliotic curve progressed from a Cobb angle of 40° to 79° in one year. (B) Three-dimensional CT spine reconstructions of the same patient. White arrows indicate vertebral body (top) and rib head (bottom) deformity on the concave side adjacent to tumor that are not evident on the convex side or further from the tumor (black arrows). (C) Longitudinal X-ray imaging of a child with bone loss associated with a unilateral NF extending from the left pelvis throughout the entire extremity. White arrows show deformity of the pelvis, acetabulum and proximal femur. (D) MRI imaging at presentation depicts the location and involvement of the NF. (E) Quantification of tumor-associated femur deformity by serial measurements of the sub-trochanteric femur diameter. Affected: Tumor-laden femur; Unaffected: contralateral side. X-rays with measurement sites are shown in Supplementary Figure 2.

Results from conditional mouse models of Nf1 deletion in the bone mesenchymal lineage and expression profiling of patient-derived bone stromal cells supported a model whereby loss of NF1 leads to activation of genes promoting the accumulation of inorganic pyrophosphate (PPi), a strong inhibitor of hydroxyapatite formation and bone mineralization [21]. We have shown that upregulation of PPi pathway-related genes in mouse osteoprogenitor cells following deletion of Nf1, including Enpp1 (an ectophosphatase generating PPi), Ank (a channel transporting PPi in the extracellular milieu) and Spp1 (Osteopontin, an extracellular protein inhibiting hydroxyapatite formation), is MEK-dependent [21]. In this study, we investigated whether the expression of genes involved in PPi homeostasis was altered in NFs and whether skeletal changes were associated with adjacent neurofibromas. Furthermore, we present an individual with NF1 presenting with reduced bone mineral density (BMD) associated with a unilateral NF who saw marked improvement in BMD concomitant with response of the tumor to MEKi therapy.

METHODS

Mouse models.

Control and tamoxifen-inducible Plp-cre;Nf1 mice [22-24] received Tamoxifen (75mg/kg) i.p. twice daily for 3 days at 2-months of age. Periostin-cre;Nf1 mice [24-26] and Dhh-cre;Nf1 mice [24,27] were generated as described before. No randomization was used for animal studies. Mice were euthanized between 8–10 months of age. Both genders were included in the measurements. All mouse procedures were approved by appropriate institutional IACUC boards.

Histological measurements.

Tissues were immersion-fixed in PBS-based 4% paraformaldehyde and embedded undecalcified in methyl methacrylate before staining by Von Kossa/Van Gieson to stain mineralized bone and non-mineralized collagenous bone matrix, respectively. Histomorphometric measurements were performed using the Bioquant Image Analysis System (R&M Biometrics, Nashville, TN), blinded to genotype and following ASBMR guidelines [28,29]. Osteoid parameters were measured directly by manual tracing from images taken at 20X magnification.

Immunocytochemistry.

Sections were deplastified before rehydration, followed by IHC using an antibody against S-100 (clone 15E2E2, Sigma, MAB079–1), a HRP-coupled goat anti-mouse IgG (Abcam, ab6789) and the Dako Liquid DAB+ Substrate Chromogen System (Agilent Technologies, K346811–2).

Human Schwann cell gene expression analysis.

Enrichment and culture of NF1−/− Schwann cells from NFs was done as described previously [30]. Normal human embryonic Schwann cells (ScienCell Research Laboratories, Carlsbad, CA) were cultured as NF1−/− Schwann cells. RNA-Seq libraries were constructed from 1μg total RNA after ribosomal RNA depletion (Ribo-Zero GOLD and Illumina TruSeq RNA Sample Prep V2 Kit, Illumina, San Diego, CA). Data were processed using RTA v.1.18.64 and CASAVA v.1.8.2. software. Data for cultured Schwann cells from NFs of individuals with NF1 (N=23) and cultured normal human embryonic Schwann cell controls (N=7) were further processed using standard Tuxedo pipeline [31]. Resulting gene expression datasets were then Log2 transformed. Genes with median Log2 FPKM score below 0.5 in both groups were considered not expressed and removed from further analysis.

Human and mouse gene expression microarray data.

We re-analyzed published human and mouse (GEO accession: GSE14038, GSE41747) microarray datasets, including human nerve, human NF tumor, mouse nerve, and mouse neurofibroma samples [32,33]. Raw CEL files were pre-processed and gene expression levels were normalized (Bioconductor/affy package’s Robust Multi-array Average method) [34]. Differentially expressed genes were predicted using Bioconductor/limma package [35].

RNAseq analysis.

RNA was isolated from trigeminal nerve tissues from Periostin-cre;Nf1 mice (RNeasy Plus Mini Kit, Qiagen) with DNase I treatment (Qiagen) for 15min. 400ng total RNA was used as input for RNA-seq library construction (Kapa Stranded mRNA kit). TruSeq adapter sequences were used for indexing. Library were PCR-amplified 12 cycles. Equimolar amounts of each library were run as 12-plexed 1.65pM pools, single-indexed, on a NextSeq 500—75 cycle, high output V2 kit to produce single-end reads. QC-passed reads were aligned to the mouse reference genome (mm10) using STAR 2.4.2a [36] and reads were translated to transcriptome coordinates using Salmon 0.60 [37]. Isoform data were collated to single gene IDs using the R package biomaRt [38], and abundance estimates were upper quartile normalized using R. Abundance values for genes were used as input for DESeq2 to identify differentially-expressed genes [39] using an adjusted p-value cutoff of <0.05.

Whole-body Dual-energy X-ray analysis.

Children aged 5–16 enrolled at the NIH on a phase 2 clinical trial of Selumetinib (SPRINT, NCT01362803) underwent a baseline DXA of whole body, proximal femur and lumbar spine using a Hologic Discovery A, using standard positioning technique. The study was approved by the Institutional Review Board at the National Cancer Institute, and all participants provided written informed consent. Patients found to have an abnormal result in any of these body regions (Z-score ≤ −2.0) had repeat imaging performed after 1 year on treatment. Height and age-adjusted Z-scores were calculated for whole body, total body less head (a.k.a. “subtotal”), lumbar spine, total hip, and femoral neck using data from the Bone Mineral Density in Childhood Study. One subject out of 11 was excluded due to hip surgery during the study period.

Femoral diameter measurements.

Femur diameter was measured directly from X-ray films taken as standard-of-care. Locations of diameter measurement are shown in Supplementary Figure 2.

Statistics.

All gene expression data and distributions are represented as boxplots. Pair-wise comparison were performed using standard two-sided t-tests in R.

Ethics Statement.

The study was approved by the Institutional Review Board at the National Cancer Institute, and all participants provided written informed consent.

RESULTS

We first evaluated differences in skeletal mineralization adjacent to Nf1-deficient neurofibroma tumors using the inducible Plp-Cre;Nf1 mouse model, where the Nf1 allele is conditionally deleted in myelinating Schwann cells at 2 months of age under the control of the Myelin proteolipid protein (Plp) promoter [23]. Histological examination of vertebral bones from mutant mice euthanized at 9–11 months of age, at a time where large paraspinal plexiform neurofibromas associated with peripheral nerves and dorsal root ganglia are readily detectable, showed vertebral processes surrounded by S100+ NFs that appeared completely unmineralized (Supplementary Figure 1A). We also observed cases of tumors infiltrating the pelvic bones, where non-mineralized bone matrix (osteoid) and S100+ tumors were co-localized (Supplementary Figure 1B). Trabecular osteoid surfaces and volumes were increased in the vertebrae of Plp-creERT;Nf1flox/flox mice adjacent to S100+ neurofibromas, although to a lesser extent than in vertebral processes immediately adjacent to tumors. In mutant mice, this increase in cancellous osteoid parameters was not evident in vertebrae adjacent to normal nerves (Figure 2A,B) and bone volume in lumbar vertebrae not associated with tumors was normal (BV/TV: 18.06 ±5.8 in WT versus 18.30±4.8 in cKO, n= 6–9). In a second mouse model in which the Cre recombinase is driven by the Desert Hedgehog (Dhh) promoter to conditionally ablate the Nf1 allele in embryonic Schwann cell precursors (Dhh-cre;Nf1 )[27], increased vertebral osteoid volume and surface adjacent to neurofibroma tumors were also observed (Supplementary Figure 1C). These findings demonstrate, in two independent Schwann cell-specific Nf1 conditional knockout mouse models, the existence of altered mineralization closely associated with neurofibromas.

Figure 2.

Osteoidosis and increased expression of PPi-related genes associated with murine neurofibromas.

(A) Representative images of Von Kossa/Van Gieson-stained (top) and S100+ nerves (bottom) thoracic vertebral sections from control and Plp-Cre;Nf1 mice. N: Normal nerve; T: plexiform tumor adjacent to vertebral bone. Osteoid (non-mineralized bone) is stained pink (white arrows) and mineralized bone is stained black. Bar: 50um. (B) Osteoid Surfaces/Bone Surfaces (OS/BS), Osteoid Volume/Bone Volume (OV/BV) and Osteoid Width (O. Wi.) in vertebral cancellous bone regions adjacent to normal nerves or NF tumors in control and Plp-creERT;Nf1 (cKO) mice (WT: n=6 vertebral bodies from 3 mice), cKO without adjacent tumor (“-“, n=9 vertebral bodies from 4 mice) and adjacent to tumor (“+”, n=7 vertebral bodies from 4 mice). *, p<0.05, one-way ANOVA and post hoc pair-wise comparison with Bonferroni adjustment. (C,D) Normalized expression levels in neurofibromas from (C) Dhh-cre;Nf1 (n=15) and control (normal sciatic nerve, n=15) mice (Microarray) and (D) from Periostin-cre;Nf1 (n=6) and control (Trigeminal nerve, n=6) mice (RNAseq). *: p<0.05; **: p<0.01 by two-sided t-test.

To determine whether dysregulation of PPi homeostasis potentially contributed to neurofibroma-associated mineralization defects, and because the measurement of tissue PPi level was technically not feasible, we measured expression of genes promoting PPi generation and extracellular transport in tumors from multiple mouse models. Microarray analysis of neurofibromas from Dhh-cre;Nf1 mice compared to WT sciatic nerves (GSE41747)[32] identified increased expression of Enpp1 in neurofibroma tumors, whereas expression of Ank was reduced (Figure 2C). Transcriptome profiling of trigeminal nerve tumors from Periostin-cre;Nf1 mice[25] and control nerves confirmed these results in an independent mouse model of Nf1-deficient neurofibromas (Figure 2D), in which Cre-mediated Nf1 ablation occurs in the neural crest-derived Schwann cell lineage after embryonic day E11.

We then tested whether activation of the PPi pathway was also evident in human NF1 NFs. Transcriptome profiling of low-passage Schwann cell cultures from NFs of individuals with NF1 showed increased expression of ANKH compared to control embryonic Schwann cells (Figure 3A), but no difference in ENPP1 expression (not shown). Independent microarray expression analysis of human NF tissues (GSE14038, GSE41747)[33] further confirmed higher ANKH expression compared to normal sciatic nerves (Figure 3B). Although not statistically significant, expression of ENPP1 trended higher in NF1 NFs compared to control (Figure 3B).

Figure 3.

Activation of genes regulating PPI in human plexiform neurofibromas.

(A) Gene expression in cultured Schwann cells (SCs) from NFs of individuals with NF1 (n=24) versus cultured embryonic Schwann cell controls (n=7)(RNAseq). (B) Gene expression in human NF1 NF tissue (n=13) versus normal sciatic nerve control tissue (n=3)(Microarray). *: p<0.05; ***: p<0.001 by two-sided t-test.

Results presented here from both conditional mouse models and patient-derived tumors all concur to support the notion that activation of genes regulating PPi homeostasis in NF1 neurofibromas is associated with decreased bone mineralization and likely contributes to skeletal co-morbidities in children with NF1. The response of NF1 NFs to MEK inhibitor (MEKi) therapy[6] and the MEK dependency of PPi pathway activation[21] thus implied that treatment of NF1 plexiform tumors with MEK inhibitors may have beneficial effects on bone mineralization. To investigate this model further, baseline and post-therapy whole-body Dual-Energy X-ray Absorptiometry scans of ten subjects from the National Cancer Institute phase 1/2 clinical trial (NCT01362803) of Selumetinib for inoperable NFs were examined. Of these, one subject had a unilateral NF of the leg that allowed for comparison of BMD to the contralateral leg without tumor. In this patient, BMD of the tumor-laden leg was dramatically reduced compared to the contralateral leg without tumor; this difference was not observed among participants without leg NF (Figure 4A, baseline). Following 1-year of MEKi therapy, the subject’s NF showed a partial response with volumetric reduction of 23% (1267 mL vs 973 mL), and the difference in BMD between tumor-laden and contralateral leg improved from −15% to −8% (Figure 4A, post-MEKi). Whereas 1-year of MEKi therapy did not affect total hip and femoral neck adjusted Z-scores in subjects without leg NFs, it markedly improved these parameters in the affected leg of this patient (Figure 4B,C).

Figure 4.

Case presentation of improved tumor-associated BMD following MEKi therapy.

(A) Difference in BMD between left and right legs is shown for 9 individuals without leg NF (grey circles) and one patient with a left-leg NF (blue diamond) at baseline (left) and following 1-year MEKi therapy (right). MRI shows left-sided tumor involvement and volumetric reduction following therapy. (B, C) Height adjusted total hip (B) and femoral neck (C) Z-scores at baseline and following 1-year MEKi therapy.

DISCUSSION

MEK-dependent PPi pathway activation in NFs suggests tumor-associated skeletal co-morbidities in NF1 result, at least in part, from a local inhibition of bone mineralization. The rate of bone accrual and mineralization is high during childhood/adolescence and coincides with the initial active growth of NFs and, importantly, their response to MEKi therapy. Therefore, our findings highlight a common therapeutic window where MEKi therapy may reduce both tumor burden and tumor-associated PPi pathway activation, allowing for proper mineralization of the affected bony structures in children. This secondary therapeutic effect was observed in the PN-laden femur/hip of a single available patient with unilateral NF, despite the short-term duration of the treatment and the remaining presence of tumor post-treatment. This effect was not observed in the skeleton of patients without adjacent tumors, thus supporting it is specific to the skeleton associated with regional PN. Because of the preliminary nature of our findings in a single human patient, longitudinal evaluation of additional patients will be necessary to confirm efficacy of MEK inhibitors to treat the skeletal manifestations associated with neurofibroma tumors, as well as other potential effects at other bone sites and on bone growth. This PPi-dependent mechanism will also require confirmation of increased PPi generation in plexiform tumors and does not exclude other mechanisms such as alterations in blood flow or nutrient utilization.

The model proposed here may apply to other neurofibroma-associated skeletal manifestations of NF1, including craniofacial abnormalities associated with orbital NFs, dystrophic scoliosis, and, potentially, any skeletal structure adjacent to NFs. The potential clinical utility of MEKi may thus extend beyond the treatment of NFs to positively impact the many neurofibroma-associated skeletal co-morbidities in children with NF1, thus supporting the need for integrated, comprehensive and collaborative care clinics for patients with NF1. Lastly and importantly, results from this study highlight the opportunity for ongoing and future clinical trials testing MEK inhibitors as therapeutics for NF1 NFs to investigate skeletal co-morbidities, which may inform the design of more effective clinical management strategies for patients with NF1.