MAPK-dependent release of GDNF from Schwann cells mediates tumor-independent pain in neurofibromatosis 1

Abstract

Neurofibromatosis 1 (NF1) is an inherited tumor-predisposition syndrome in which mutations in the neurofibromin gene NF1 cause various dysfunctions in the nervous system, including pain that substantially diminishes quality of life. Pain can be both tumor-dependent and tumor-independent and is challenging to manage therapeutically. Mice with homozygous loss of Nf1 in Schwann cells exhibit pain before tumors form, and this pain is associated with increased gene expression of glial cell line–derived neurotrophic factor (GDNF). Here, we investigated the cause of increased GDNF production and its downstream target cells in a mouse model of NF1. We found that Schwann cells were the dominant source of GDNF in NF1 mice. Schwann cell–derived GDNF activated the receptor GFRα1 on high-threshold mechanoreceptors and polymodal C-fibers, which mediated mechanical hypersensitivity in mice. Treating the mice with pharmacological inhibitors of mitogen-activated protein kinase (MAPK) signaling reduced pain-like behaviors and the expression of GDNF at the mRNA and protein levels in Schwann cells. The findings provide insight into the signaling pathways that underlie tumor-independent pain in NF1 and identify a targetable pathway for therapeutic intervention.

Introduction

Neurofibromatosis 1 (NF1) is an autosomal dominant disorder, usually diagnosed in the first decade of life. It is caused by mutation in the NF1 gene, which encodes neurofibromin, a negative regulator of the RAS signaling pathway [1-3]. NF1 is a multisystem disorder characterized by a predisposition to benign nerve-associated tumors, called cutaneous and plexiform neurofibromas. Other clinical features include café-au-lait macules, cognitive and learning disabilities, skeletal abnormalities, and neuropathic pain [4-6]. Schwann cells (SCs) and SC-precursors have been shown to play a primary role in neurofibroma formation in this disease [7, 8].

Although pain in NF1 was first documented as early as 1948, research into its mechanisms and management remains limited. Conversations with NF1 patients consistently reveal the urgent need for effective pain relief, as current treatment options often provide only temporary respite [5, 9]. Despite trying various therapies, many patients experience recurring pain, which is further complicated by tumor predisposition, adding another layer to the already complex biological nature of pain in NF1 [9-11].

The progression from benign tumors to malignant peripheral nerve sheath tumors (MPNSTs) in NF1 is well-documented and contributes substantially to the disease burden [1, 5, 12], but the associated pain is challenging to manage and can severely limit the ability of individuals with NF1 to carry out even routine daily activities [13, 14]. Although tumor-associated pain is common in NF1, individuals also report substantial pain in areas unaffected by tumors [13] [15]. Skin biopsy samples from tumor-free areas of NF1 patients and animal models of NF1 show exuberant cutaneous axons covered by SCs—termed micro-neuromas—near hair follicles, sweat ducts, and other mechanosensory end organ structures; these were suggested to serve as precursors to cutaneous neurofibromas (cNFs) and were closely associated with typical termination sites for nociceptive fibers, causing pain [16, 17]. Further research into the etiology and origin of non-tumor-related pain in NF1 is essential to develop effective treatment strategies.

Emerging evidence suggests that SCs play a role in modulating the function of primary afferent neurons [18-20], and SCs contribute to the detection and transmission of noxious stimuli from the periphery [18, 20, 21]. Hence, we hypothesized that analyzing how Schwann cells modulate somatosensory and nociceptive processing could lead to the identification of potential therapeutic targets for treating pain in NF1.

A rodent model that closely replicates many of the peripheral nerve manifestations (including tumor development and disruption of Remak bundles) in human NF1 has been developed [7, 22]. In these DhhCre;Nf1^f/f [7] mice, pain-related behaviors are detected in lower limbs at 4–5 months of age, before tumors are formed, suggesting that pain precedes detectable tumor development in NF1 [20]. We previously found that Nf1 deletion in SCs, but not neurons, induces pain-related behaviors, and implicated increased GDNF signaling from SCs to neurons in causing these behaviors [20]. GDNF signaling through the kinase ERK and the transcription factor Runx1 has been implicated in pain associated with cancer and other conditions [23, 24]. Indeed, therapies targeting the GDNF pathway have been explored as treatments for sciatica, highlighting its role in pain modulation and suggesting that targeting GDNF might be a way to relieve in NF1 [25]. In this study, we investigated the role of GDNF, and the potential pathways involved in tumor-independent pain in the DhhCre;Nf1^f/f mouse model of NF1. Because previous work implicated GDNF in NF1-related, tumor-independent hypersensitivity, we sought to determine the mechanisms by which this factor may contribute to this phenomenon in mice at an age that do not display lumbar tumors [7]. We found that activation of the kinase MEK in Nf1-mutant SCs induces the production of GDNF, which activates neuronal GFRα1 receptors, causing afferent sensitization and pain-like behaviors. Blocking MEK decreased GDNF and alleviated pain behaviors in the NF1 model mice. These preclinical results support clinical testing of MEK inhibitors (MEKi) to treat pain in individuals with NF1.

Results

SC-specific deletion of Gdnf reduces mechanically evoked nociceptive behaviors.

Non-neuronal cells release chemokines, cytokines, and growth factors that can influence nociception [26, 27], and our previous report implicated increased GDNF signaling from SCs as a potential modulator of pain in NF1 prior to tumor formation [20]. To test whether GDNF is sufficient to cause pain-like phenotypes, we first administered bilateral intraplantar injections of GDNF and assessed mechanical pain using a modified mechanical conflict-avoidance (MCA) assay. In this modified assay, mice that were acclimated to an empty, non-bedding chamber could cross a chamber of varying-level spikes to reach a familiar-bedding chamber; increased time spent in the initial (non-bedding) chamber was interpreted as avoidance of the spike chamber and greater sensitivity to the mechanical stimulus. GDNF injection in WT C57Bl/6 mice (in equal numbers of male and female animals) led to the avoidance of noxious mechanical stimuli within 30 minutes compared to saline injected controls (Fig. 1A). This mechanical avoidance persisted for approximately 24 hours and resolved by seven days (fig. S1A).

Figure 1: GDNF from Schwann cells, not macrophages, regulates hypersensitivity in NF1 model mice.

(A) Mechanical avoidance, assessed by MCA in equal numbers of female/male WT C57Bl/6 mice at baseline and 30 min after intraplantar injection of -vehicle (saline)or GDNF (in saline) with their tendency to cross the spikes at 0mm and 0.5mm.. Data are time spent in the non-bedding chamber as a measure of spikes avoidance from n=6 mice per group. *p<0.05 by two-way repeated measures ANOVA with Tukey’s test. (B) Representative images and analysis of GDNF staining in L3 DRG tissue from Nf1^fl/flGdnf^fl/fl (Control), DhhCreNf1^fl/fl Gdnf^fl/fl, and DhhCreNf1^fl/flGdnf^fl/+ mice from three biological replicates. Scale bar, 50μm. Data are mean ± SEM from three non-consecutive sections per slide (L3 DRG) each from three independent biological samples. **p< 0.01, *p< 0.05 by one-way ANOVA with Tukey’s post hoc test,. (C) Mechanical avoidance in male and female DhhCreNf1^fl/flGdnf^fl/fl, DhhCreNf1^fl/flGdnf^fl/+, and Nf1^fl/flGdnf^fl/fl mice. Data are mean ± SEM from n= 35, 10, and 9 mice, respectively, per group. ***p<0.001, **p<0.01, and *p<0.05 by two-way ANOVA with Tukey’s post hoc test. (D) Representative image of for macrophages in L3 DRGs showing GDNF is expressed in Schwann cells from male and female C57Bl/6 mice (n=3)(E) Representative images of IBA1-positive macrophages (green) in the sciatic nerve of clodronate liposome injected male and female C57Bl/6 mice (top) compared to PBS liposome-injected controls (bottom). Scale bar, 50μm. Data are mean ± SEM from nine non-consecutive sections from three independent biological samples. **p<0.01 by unpaired t-test. (F and G) Mechanical hypersensitivity, assessed as described above on the DhhCreNf1^fl/fl, by group before (F, Baseline) and after (G) injection with either clodronate or PBS liposomes. Data are mean ± SEM from 7–11 mice per group. *p<0.05 by two-way ANOVA with Tukey’s post hoc test.

We previously found that Gdnf mRNA and GDNF protein were upregulated in SCs in the NF1 mouse model [20]. We therefore tested whether SCs are a major cellular source for GDNF-related hypersensitivity in NF1. To do this, we deleted Gdnf from SCs in DhhCre;Nf1^fl/fl mice (DhhCre;Nf1^fl/fl;Gdnf^fl/fl). Genotyping (fig. S1B) confirmed the knockout at the gene level, and immunohistochemistry of DRGs (Fig. 1B) confirmed the suppression of GDNF protein production in knockout mice.. Mice lacking Gdnf and Nf1 in SCs spent significantly less time in the non-bedding chambers in the MCA assay (Fig. 1C, and fig. S2), indicating a notable reduction in noxious mechanical avoidance compared to the wildtype knockout and control groups. To confirm that these results were not due to alterations in sensory neuron innervation of the skin, we performed innervation density analyses and found that deletion of Gdnf from SCs did not alter the abundance of PGP9.5+ axons found in the skin (fig. S3). This data supports the idea that SCs are a major contributor to GDNF-dependent pain in the NF1 model.

Macrophages are also known to produce GDNF [28] and are a dominant cell type in nerves in which SCs lack Nf1. Therefore, we tested whether macrophages also contribute to the observed hypersensitivity in NF1 model mice [29-31]. In slices of L3 DRGs from these mice, the GDNF signal overlapped with that of macrophage marker IBA1, indicating that macrophages in these mice express GDNF (Fig. 1D), Macrophages were substantially reduced after injection with clodronate-containing liposomes [30, 32, 33] (Fig. 1E). However, macrophage depletion did not alter baseline or NF1-induced mechanical avoidance (Fig. 1, F and G). Thus, macrophages are unlikely to be a major contributor to mechanical hypersensitivity in this NF1 model. These results support the hypothesis that SC-specific GDNF is involved in tumor-independent pain-associated behaviors in NF1.

Knocking down neuronal GFRα1 decreases mechanical hypersensitivity in DhhCre;Nf1^fl/fl mice

Given the reduction in pain-associated behavior in mice following the deletion of Gdnf from SCs, we investigated the pathways contributing to this tumor-independent pain. Our previously reported single cell RNA sequencing (scRNA-seq) analysis predicted that that SC-derived GDNF signals to sensory neurons that express the receptor GFRα1 (GDNF family receptor α1) in DhhCre;Nf1^fl/fl mice [34-36]. Because GDNF primarily signals through GFRα1, we chose to knock down this co-receptor in dorsal root ganglion (DRG) neurons using an Adeno-associated virus (AAV)-based approach. We delivered bilateral injections of an AAV containing a short hairpin RNA targeting GFRα1 (AAV9-GFP-U6-GOI-GFRA1-shRNA) or a GFP-only control AAV into the sciatic nerves of mice with NF1 [37, 38].By four to five weeks after injection, in the number of GFRα1-positive neurons in the shRNA-AAV–injected mice was reduced (Fig. 2A). Knockdown of GFRα1 did not induce a compensatory upregulation of other GDNF family receptors (GFRα2, GFRα3 or Ret) following AAV injection (fig. S4). We next conducted behavioral assessments using the mechanical conflict-avoidance (MCA) assay. Prior to AAV injection, all mice with SC-specific deletion of Nf1 displayed avoidance of noxious mechanical stimuli compared to controls (Fig. 2B). Four to five weeks after GFRα1 knockdown in neurons, there was a significant reduction in mechanical avoidance in the DhhCre;Nf1^fl/fl animals (Fig. 2C). These findings are consistent with the hypothesis that SC-derived GDNF signals to GFRα1 on sensory neurons to modulate pain-like behaviors in NF1.

Figure 2: Neuronal GFRα1 levels are associated with decreased mechanical hypersensitivity in the NF1 mouse model. .

A; Representative images for the expression of GFRα1 positive cells in the DRGs (L3) of indicated groups. GFRα1+ cells (red), DAPI (blue); Scale bar=50μm. GFRα1 expressing cells in DhhCreNf1^fl/fl mice injected with AAV GFRα1 (AAV9-GFP-U6-GOI-GFRA1-shRNA; 1-3μL) is significantly decreased compared to AAV GFP (AAV9-CAG-EGFP shRNA; 1-3 μL) injected DhhCre;Nf1^fl/fl mice and uninjected Nf1^fl/fl controls. Equal number of male and female injected at 3.5 months, and DRGs harvested after 4/5 weeks post injections. Data are mean ± SEM from three non-consecutive sections per slide (L3 DRG) each from three independent biological replicates ****p<0.0001 vs controls, 1-way ANOVA with Tukey’s post hoc test. B: The mechanical avoidance was measured by the time spent in non-bedding chambers at 0mm and 0.5mm spike length and was significantly increased in both DhhCreNf1^fl/fl mice injection groups at baseline prior to AAV injections. Data are Mean ± SEM from, n= 9 DhhCreNf1^fl/fl mice injected with AAV GFRα1, n= 8 DhhCreNf1^fl/fl mice injected with GFP,; and n= 12 Nf1^fl/fl mice injected with AAV GFRα1.. **p<0.01, *p<0.05 vs Nf1^fl/fl, 2-way ANOVA with Tukey’s post hoc test). C: After AAV injection, the mechanical avoidance was significantly decreased in DhhCreNf1^fl/fl mice injected with AAV GFRα1, compared to DhhCreNf1^fl/fl mice injected with AAV GFP. Data are Mean ± SEM from, n= 9; DhhCreNf1^fl/fl mice injected with AAV GFRα1, n= 8 DhhCreNf1^fl/fl mice injected with GFP, and n= 12 Nf1^fl/fl mice injected with AAV GFRα1. **p<0.01, *p<0.05 vs Nf1^fl/fl, 2-way ANOVA with Tukey’s post hoc test

To explore the functional changes in peripheral sensory neurons after decreased GFRα1, we employed an ex vivo electrophysiology recording technique [20, 39, 40]. As previously reported [20], high-threshold mechanoreceptors (HTMRs) displayed reduced mechanical thresholds and increased firing to mechanical stimulation of their cutaneous receptive fields (RFs) in control AAV–injected DhhCre;Nf1^fl/fl mice (Fig. 3, B-C), but in the mice with NF1 injected with AAV GFRα1, these phenotypes were restored to control levels (AAV-GFP injected) (Fig. 3, B to E). Similar results were observed in the polymodal C-fiber (CPM) subtype (Figure 3A, 3D-E). No significant changes were observed in conduction velocity, spike amplitude, resting membrane potential, overshoot, or spontaneous activity in any cell type, nor was any sensitization observed in other neuronal subtypes (figs. S4, S5, and S6). Thus, knocking down neuronal GFRα1 reduces NF1-induced afferent sensitization.

Figure 3: Neuronal GFRα1 is associated with reduced sensitization of HTMRs and CPM neurons in DhhCre;Nf1^fl/fl mice.

A: Representative firing patterns of A-fiber HTMRs and polymodal C-fibers (CPM) in DhhCre;Nf1^fl/fl mice, in equal numbers of male and female mice injected with AAV GFRα1 or AAV GFP, compared to Nf1^fl/fl controls. B: HTMRs of the AAV GFRα1 injected mice showed a significant increase in mechanical thresholds compared to NF1 mice injected with AAV GFP injection. Data are Mean ± SEM from n= 23 cells from DhhCre;Nf1^fl/fl injected with AAV GFRα1, n=18 cells from DhhCre;Nf1^fl/fl injected with AAV GFP, and n=23 cells Nf1^fl/fl mice. ***p<0.001 vs. DhhCre;Nf1^fl/fl injected with AAV-GFP, ****p<0.0001*, Nf1^fl/fl vs. DhhCreNf1^fl/fl injected with GFP, 1-way ANOVA with Tukey’s post hoc test;; ). C: Firing rates of HTMRs show decreased firing to mechanical stimuli in DhhCre;Nf1^fl/fl mice injected with AAV GFRα1 when compared with in DhhCre;Nf1^fl/fl mice injected with AAV-GFP. Data are Mean ± SEM from n= 23 cells from DhhCre;Nf1^fl/fl injected with AAV GFRα1, n=18 cells from DhhCre;Nf1^fl/fl injected with AAV GFP, and n=23 cells from Nf1^fl/fl mice. ****p<0.0001 vs DhhCreNf1^fl/fl w GFP and Nf1^fl/fl, 1-way ANOVA with Tukey’s post hoc test; Mean ± SEM). D: CPMs in AAV GFP injected DhhCre;Nf1^fl/fl mice also showed decreased mechanical thresholds compared with controls that is restored by injection of AAV GFRα1. Data are Mean ± SEM from n= 9 cells from DhhCre;Nf1^fl/fl injected with AAV GFRα1, n=11 cells from DhhCre;Nf1^fl/fl injected with AAV GFP, and n=13 cells from Nf1^fl/fl mice. (**p<0.01 Nf1^fl/fl vs. DhhCreNf1^fl/fl w AAV GFP, ***p<0.001 DhhCreNf1fl/fl w GFP vs. DhhCreNf1fl/fl w GFRα1, 1- way ANOVA with Tukey’s post hoc testE: DhhCre;Nf1^fl/fl mice injected with AAV GFRα1 also showed decreased firing to mechanical stimuli in CPMs vs control GFP injected NF1 mice. Data are Mean ± SEM from n= 9 cells from DhhCre;Nf1^fl/fl injected with AAV GFRα1, n=11 cells from DhhCre;Nf1^fl/fl injected with AAV GFP, and n=13 cells from Nf1^fl/fl mice. (***p<0.001 Nf1^fl/fl vs. DhhCreNf1^fl/fl w GFP, and ***p<0.001 DhhCreNf1^fl/fl w GFP vs. DhhCreNf1^fl/fl w GFRα1, 1 way ANOVA with Tukey’s post hoc test;

MAPK signaling regulates the abundance of SC-derived GDNF and decreases pain-related behaviors in NF1 mice

Identifying pathways that result in GDNF expression in SCs could identify new strategies to ameliorate pain. We investigated the role of the MAPK pathway in mediating pain in the NF1 model, because the RAS-MAPK pathway is a well-characterized and extensively studied pathway in the context of NF1 [3, 41-43], and because GDNF is regulated by MAPK signaling in other contexts [44]. We targeted the MAPK pathway using a MEK inhibitor (MEKi; mirdametinib) [45] (Fig. 4A). At a clinically relevant dose, the MEKi decreased the phosphorylation of ERK in the DRG (Fig. 4, B and C). We then confirmed that the SC Nf1 mutant mice exhibited avoidance of aversive mechanical stimuli prior to MEKi administration as assessed using the MCA analysis (Figure 4D). Initiation of repeated oral gavage administrations (5 days per week) caused variability in body weight, fecal output, and mechanical avoidance during week one, but these measures stabilized by week two (fig. S7). By week three, MEKi-treated DhhCre;Nf1^fl/fl mice showed a significant reduction in mechanical avoidance when compared to vehicle-treated DhhCre;Nf1^fl/fl mice (Fig. 4E). MEKi-treated DhhCre;Nf1^fl/fl mice also exhibited a reduction in SC-specific GDNF expression (Fig. 4, F and G). Thus, loss of Nf1-associated MEK signaling increases SC-produced GDNF, which acts upon nearby neurons to increase pain sensation.

Figure 4: MEK inhibitors and tumor-independent hypersensitivity in DhhCre;Nf1^fl/fl mice.

A: Schematic for experimental design. B: Representative western blots from L3 DRGs, three-weeks post MEKi administration (5mg/kg of mirdametinib or vehicle (200mg/mL of DMSO + 400μL of PEG400 + 50μL of Tween20 + 500μL of ddH₂O) in equal numbers of male and female DhhCre;Nf1^fl/fl mice, showing a decrease in phosphorylated ERK (pERK, top panel) with no change in total ERK (middle panel), normalized to GAPDH (bottom panel). C: pERK is decreased in the MEKi treated groups compared to vehicle treated DhhCre;Nf1^fl/fl mice. Data are Mean ± SEM from three non-consecutive sections per slide (L3 DRG) each from three independent biological samples. *p<0.05 vs control, 1-way ANOVA with Tukey’s post hoc test.). D: Mechanical avoidance was significantly increased in both groups of DhhCreNf1^fl/fl mice at baseline prior to MEKi or vehicle administration. Data are Mean ± SEM from n= 11 DhhCreNf1^fl/fl mice injected with MEKi, n= 9 DhhCreNf1^fl/fl mice injected with vehicle, n= 9 Nf1^fl/fl mice injected with MEKi, and n= 9 Nf1^fl/fl mice injected with vehicle..*p<0.05, vs Nf1^fl/fl, 2-way ANOVA with Tukey’s post hoc test.. E: After injections, mechanical avoidance was significantly decreased in DhhCreNf1^fl/fl mice injected with MEKi, compared to DhhCreNf1^fl/fl mice injected with vehicle group three weeks post administration. . Data are Mean ± SEM from n= 11 DhhCreNf1^fl/fl mice injected with MEKi, n= 9 DhhCreNf1^fl/fl mice injected with vehicle, n= 9 Nf1^fl/fl mice injected with MEKi, and n= 9 Nf1^fl/fl mice injected with vehicle. **p<0.01 vs Nf1^fl/fl, 2-way ANOVA with Tukey’s post hoc test.). F: Representative images of GDNF expression in SCs of the L3 DRG, Nf1^fl/fl with MEKi (top panel), DhhCreNf1^fl/fl mice injected with vehicle (middle panel) and DhhCreNf1^fl/fl mice injected with MEKi (bottom panel);GDNF (green), Dapi (blue), s100β (purple and multi color (Merged)); Scale bar=50μm. G: MEKi administration significantly decreased GDNF expression in the DRGs(L3) compared to vehicle treated DhhCreNf1^fl/fl mice and controls. Data are Mean ± SEM from three non-consecutive sections per slide (L3 DRG) each from three independent biological samples for each group. *p<0.05 vs controls, 1-way ANOVA with Tukey’s post hoc test.

Because alterations in both the G-protein–dependent cAMP pathway and the MAPK pathway are found upon deletion of Nf1 in SCs [46-49], we examined the potential involvement of the cAMP pathway using a Gs-coupled Designer Receptor Exclusively Activated by Designer Drug (GsDREADD; LSL-rM3Ds-GFP) [50] that we expressed in SCs (using DhhCre). Upon Cre recombinase expression, the GsDREADD was acutely activated by administering its ligand, compound 21 (C21) [20]. At baseline, GsDREADD mice showed no changes in mechanical avoidance compared to controls; however, activation of the cAMP pathway led to substantial mechanical avoidance within one day after C21 injection (Fig. 5B).

Figure 5: cAMP activation in Schwann cells (SCs) leads to hypersensitivity with no changes in GDNF expression:.

A: Schematic for experimental design. B: Mechanical avoidance, assessed by the time spent in the non-bedding chambers at 0mm and 0.5mm spike heights, was increased in the DhhCre;GsDREADD mice (equal numbers of male and female mice) one day post C21 injection. Data are Mean ± SEM from n=8 GsDREADD controls, n=8 DhhCre;GsDREADD at baseline. 1 day post C21 injection, n=5 DhhCre;GsDREADD. ***p<0.001, **p<0.01 vs control; 2-way RM ANOVA with Tukey’s post hoc test.; Mean ± SEM). C: Representative images of pPKA (green) and GDNF (purple) expression in the L3 DRGs (arrows) from GsDREADD controls (top panel) and DhhCre;GsDREADD mice (bottom panel) from three biological replicates where three non-consecutives images were taken from each independent sample. DAPI (blue) is provided with merged images; Scale bar=50μm. D: A significant increase in the expression of p-PKA in the DRGs of DhhCre;GsDREADD mice injected with C21 is found compared to controls. Data are mean ± SEM from three biological replicates where three non-consecutives images were taken from each independent sample. . **p<0.01, unpaired t-test. E: Quantification of GDNF shows no difference between the indicated conditions. Data are mean ± SEM from three biological replicates where three non-consecutives images were taken from each independent sample.

In the DhhCre;Nf1^fl/fl model, low cAMP levels are present in Nf1-mutant SCs [51]. It seemed plausible that cAMP may also modulate mechanical responsiveness in the NF1 mouse model by regulating Gdnf expression. To test this idea, we first determined if Gs activation could alter GDNF expression in SCs, as was found in cells with increased activation of ERK. Immunocytochemical data on DRGs from these mice showed increased signal for phosphorylated PKA (Fig. 5, C and D), a surrogate for cAMP activation [51], but no changes in GDNF were observed in the DRGs upon Gs activation (Fig. 5, C to E). It is therefore likely that hypersensitivity in NF1 related to enhanced GDNF production from SCs is not modulated by Gs signaling or cAMP activation.

Discussion

Our findings confirm that GDNF released from SCs contributes to tumor-independent pain in a mouse model of NF1 (Fig. 6). Depletion of GFRα1 in neurons, the main co-receptor for GDNF, lead to a reduction in pain-like behaviors, and this decrease in hypersensitivity was associated with an inhibition of HTMR and CPM neuron sensitization. Treatment with a FDA approved, MEKi (Mirdametinib) used to shrink tumors in individuals with plexiform neurofibromas, reduced behavioral hypersensitivity by downregulating GDNF expression in the PNS. In contrast, targeting cAMP using GsDREADDs did not alter GDNF.

Fig. 6. Model of the mechanism.

Our data reveal a potential, targetable source of neuropathic pain in patients with NF1. In mice, deletion of Nf1 in SCs activates MEK-ERK–dependent release of GDNF, which, through its receptor, heightens the sensitivity of associated peripheral sensory neurons to mechanical stimuli.

Our previous study [20] linked SC GDNF with tumor independent pain in a model of NF1 [20]. Here we established that SCs, rather than macrophages, are a critical source of GDNF, contributing to hypersensitivity in NF1. Although macrophages can play roles in pain modulation [29], depleting macrophages in peripheral nerves using clodronate liposomes did not alter tumor-independent pain in the NF1 mouse model used here (DhhCre;Nf1^fl/fl). Depletion of GFRα1, the receptor for GDNF [52, 53] in neurons, via AAV-mediated knockdown resulted in a reduction of pain-like behavior, indicating that GDNF mediates tumor-independent pain through GFRα1 signaling in primary afferents. Thinly myelinated HTMRs and unmyelinated CPMs have been shown to become sensitized and contribute to pain in various painful conditions [54, 55]. In the rodent model of NF1, each of these neuronal subtypes display lower thresholds and at higher firing rates to mechanical stimuli [20]. Knockdown of GFRα1 restored the firing rates of HTMR and CPM neurons to control levels. Although patch clamp electrophysiology is needed to assess the intrinsic excitability of these cells, collectively, these findings indicate that SC derived GDNF contributes to noxious mechanical avoidance in NF1 by modulating the thresholds and firing rates of C-fiber nociceptors and myelinated HTMRs.

NF1 is caused by mutations in the Nf1 gene, which encodes neurofibromin, a negative regulator of the RAS-MAPK signaling pathway [8, 56-58]. Loss of Nf1 in SCs leads to dysregulation of the ERK pathway, promoting abnormal cell growth and tumor development [57, 59]. To date, the only FDA-approved treatments targeting this pathway are selumetinib, approved for pediatric use in 2020, and mirdametinib (approved in 2025) for children over 2 years and adults [60, 61]. Both drugs are allosteric inhibitors of Mek1/2/5, reduce tumor size and blunt tumor related pain [60, 62, 63]. Components of the MAPK pathway have long been investigated as therapeutic targets for managing various types of pain, including neuropathic pain [64-67], but their role in NF1-associated pain has not been explored in tumor independent pain. Therefore, we targeted the MAPK pathway using a MEKi to assess its effect on tumor independent pain. MEKi administration resulted in a reduction of mechanical hypersensitivity in the DhhCre;Nf1^fl/fl mice prior to tumor formation [7, 20]. Ongoing efforts aim to improve the efficacy of MEKi while decreasing toxicity [68, 69].

GDNF signals through a multicomponent receptor complex composed of the Ret receptor tyrosine kinase [70]. Binding of GDNF to GFRα1 through phosphorylation and dimerization with RET allows the activation of RAS/ERK [70]. The GDNF/GFRα1/Ret/ERK signaling pathway contributes to the sensitization of DRG neurons and plays a key role in the pathogenesis of different types of pain [71]. MEKi administration led to a decreased expression of GDNF in SCs with the DRGs. While MEKi treatment has been shown to reduce tumor-related pain in patients [62], we report that MEKi also alleviates non-tumor-related pain. This study also demonstrates for the first time the involvement of SC-derived GDNF signaling downstream of the MAPK pathway in mediating NF1-associated pain. Thus, GDNF receptor signaling offers a new target for therapeutic development that may have less toxicity than MEKis.

In other systems, SC cAMP has been implicated in pain regulation. Here, increased SC cAMP induced pain-like behaviors in animal models but activating this pathway did not alter GDNF expression, suggesting that SC MAPK signaling (rather than cAMP) is more relevant to GDNF-dependent hypersensitivity in NF1 prior to tumor development. The fact that activation of cAMP using DREADDs readily caused mechanical hypersensitivity may suggest that restoring SC cAMP in NF1, which already has underlying pain, may worsen the pain. If so, targeting cAMP may not be a viable treatment strategy for tumor independent pain in NF1. This will need to be investigated in more detail in the future.

SCs and afferent neurons are key contributors to the initiation and maintenance of pain, as well as the transmission of sensory signals from the periphery to the central nervous system [18, 72, 73]. SCs and SCP are well-established as critical players in the initiation of tumor development in NF1 [57, 74]. While replicating the full spectrum of NF1 phenotypes in rodent models remains a challenge, several studies have employed various strategies to investigate pain in the context of NF1, ranging from targeted knockdown of the Nf1 gene via CRISPR in adult models [75], to the use of standard behavioral pain assays [76]. These studies have highlighted the difficulty of studying pain in NF1 mouse models, and how the commonly used assays such as the von Frey and Randall–Selitto tests are unable to capture a NF1-related pain phenotype. Hence we utilized the MCA assay, which robustly revealed the phenotype in this NF1 model [77]. Our current and previous reports indicate that using non-experimenter evoked, choice-based pain assays, along with utilization of a model that involves inducing Nf1 mutations in SCs and SC precursors during embryonic development, which mimics human NF1 development [7, 20], are necessary to reveal these pain phenotypes in mouse models.

Foundational research can uncover the underlying causes, symptoms, and mechanisms of pain and identify potential therapeutic targets. Our findings show that GDNF released from SCs causes tumor-independent pain in NF1. Thus, early intervention targeting GDNF or the upstream, MEK-ERK–dependent mode of GDNF release from SCs or its receptor on peripheral sensory neurons may be a way to alleviate pain and support daily functioning in patients with NF1.

Materials and Methods

Animals

An equal number of male and female mice, 3-5 months of age were utilized in all studies. To delete Nf1 in SCs and SC precursors, we used the DhhCre; Nf1^fl/fl mouse model of NF1 [7, 20]. To delete Gdnf in the NF1 mouse model, the mice were crossed with Gdnf^fl/fl mice (The Jackson Laboratory, #:01409) (DhhCre;Nf1^fl/fl;Gdnf^fl/fl). In other experiments, we used equal numbers of male and female Cre-dependent, Gs-coupled DREADD knock-in mice (Rosa26-LSL-rM3Ds mice) generously donated by Dr. Bryan Roth (University of North Carolina). To activate the DREADD, mice were treated with the agonist compound 21 (C21) at 10 to 20 μg/μL per day for 1 to 7 days [17]. All animals were housed in a barrier facility under a 14:10 hour light–dark cycle in a temperature-controlled environment, with food and water provided ad libitum. All procedures were approved by the Cincinnati Children’s Hospital Institutional Animal Care and Use Committee and adhered to the Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011) under the Association for Assessment and Accreditation of Laboratory Animal Care International–approved practices.

AAV-mediated knockdown

To achieve targeted knockdown of the GDNF receptor, GFRα1, in DRG neurons that innervate the hind paws, adeno-associated virus 9 (AAV9) containing a short hairpin RNA (shRNA) against GFRα1 and/or a GFP reporter at 1-2μL (AAV9-GFP-U6-m-GFRA1-shRNA, titer 6.2 x 1012 GC/mL) (table S1) was bilaterally injected into the sciatic nerves of mice at 3.5 months of age using a quartz microelectrode connected to a picospritzer. Animals were housed for 4–5 weeks after injection prior to further experimental procedures. GFP AAVs without the GFRα1 shRNAs were used as injection controls. GFRα1 knockdown was validated through immunohistochemical staining of DRGs.

GDNF overexpression

Mouse recombinant GDNF was injected into both paws of mice (intraplantar) at 2 μg/mL in saline (table S1) or with vehicle (saline) [78]. Behavioral assays were conducted at baseline, 30 minutes, 24 hours, and 7 days after.

Macrophage depletion

To deplete macrophages, clodronate liposomes or PBS liposomes (table S1) at 2mg/20g were injected intravenously into the DhhCre;Nf1^fl/fl mouse and their littermate controls once daily for two consecutive days. Two days after the last injection, the mechanical conflict-avoidance (MCA) assay was performed. Two days post MCA, sciatic nerves were isolated, and immunohistochemistry (described below) was performed to validate the elimination of macrophages.

MEKi administration

The MEK inhibitor (MEKi) mirdametinib (table S1) or vehicle (a solution of 200 mg/mL DMSO, 400 μL PEG400, 50 μL Tween20, and 500 μL ddH₂O) was administered orally at a dose of 5 mg/kg for five days each week over the course of three weeks [79] to mice and littermate controls. Behavioral assays were conducted on the same cohort of mice, at baseline and again at one week, two weeks, and three weeks after treatment. Tissue samples were collected at the three-week time point for immunohistochemistry (IHC) and Western blot (WB) analyses.

Immunohistochemistry

DRGs from mice at the indicated time points were removed and immersion fixed in 3% paraformaldehyde in 0.1M phosphate buffer (PB) for 30 minutes at room temperature. Fixed DRGs were embedded in OCT and incubated at −80°C. DRG sections were cut on a cryostat at 12μm and mounted onto charged slides. Sections were then fixed for ~15 minutes, blocked with blocking buffer (5% normal goat serum + 5% normal donkey serum + 2.5% fish gelatin + 10% bovine serum albumin + 0.1% Tween-20 in 0.1M PBS) incubated overnight with the specified primary antibodies GDNF (1:250), p-PKA (1:300), s100β (1:1000), or IBA1 (1:500) (table S1), then incubated with appropriate fluorescently conjugated secondary antibodies (1:400, table S1). Slides were coverslipped in DAPI-Fluro-Gel (Electron Microscopy Sciences) and stored in the dark at 4°C until imaged. In another instance, DRGs were embedded in 10% gelatin followed by overnight fixing in 4% paraformaldehyde and immersion in 10% sucrose in 0.1 M PBS overnight. DRG sections were cut on a sliding microtome (Microm HM 430, Thermo Fisher Scientific) at 50 μm and placed in 12-well plates containing 0.1 M PBS, pH 7.3. Sections were blocked for 1.5 hours and incubated overnight with GFRα1 (1:500), primary antibodies (Table S1). Sections were then washed and incubated with appropriate fluorescently conjugated secondary antibodies (1:400) for 1.5 hours at room temperature. Sections were then placed on gelatin-coated slides prior to cover slipping. Labeling was characterized and documented using a Nikon confocal microscope.

For sensory innervation assessments, the hind paw was parafilm-fixed and cut at 20 μm on a Leica HistoCore MULTICUT microtome. After deparaffinizing the sections on the slides with a series of Xylene and ethanol baths, sections were probed with primary antibody PGP9.5 (1:200) (table S1), after 1 hour of blocking as described above Finally, sections were then incubated with appropriate fluorescently conjugated secondary antibodies (1:200, table S1). Positive cells were counted using Element Software 2 from three images taken from three different slides per each of 3 or 4 subjects, along with their respective controls. Similarly, intensity was detected using Element Software at respective wavelengths (pPKA at 488 and GDNF at 640nm) from three slides per each of 3 or 4 subjects. To assess the effects of GDNF deletion from SCs in the NF1 model on sensory axon prevalence in the periphery, innervation density analyses were performed. . Innervation within the skin was quantified using Element Software 2. 12 randomly selected regions were analyzed in three non-consecutive sections per mouse by drawing circular areas across glabrous and hairy sides of the skin, and the mean signal intensity was measured, averaged and compared across experimental conditions.

Behavioral measurements

Behavioral analyses were conducted by experimenters who were blinded to genotype and treatment conditions. Mechanical hypersensitivity was determined by a modified version of the mechanical conflict-avoidance (MCA) test that does not utilize aversive light. Mice were placed in the three-chamber set up: the first chamber had no bedding, the second chamber had a varying height of spikes, and the third chamber contained bedding from the cage in which that mouse was housed. For the experiment, following a 3-minute acclimation period in the MCA apparatus, each mouse was placed in the first chamber for 10 seconds before being allowed to move toward the bedding chamber by crossing a second chamber containing spikes of varying heights. Spike heights ranged from 0 mm to 2 mm, in 0.5 mm increments, resulting in four 3-minute trials separately for each spike height. The time spent in each chamber was recorded, and the percentage of time spent in the non-bedding chamber was averaged within a group and compared across conditions of a given experiment [20].

Ex vivo recording

Our ex vivo hairy skin, saphenous nerve, DRG, spinal cord recording preparation was used to assess the response properties of individual sensory neurons across conditions. Mice were anesthetized using 90 mg/kg ketamine with 10 mg/kg xylazine, then intracardially perfused with oxygenated artificial cerebrospinal fluid (aCSF) equilibrated with 95% O₂ and 5% CO₂. The aCSF composition (in mM) was: 1.9 KCl, 1.2 KH₂PO₄, 1.3 MgSO₄, 2.4 CaCl₂, 26.0 NaHCO₃, and 10.0 D-glucose, supplemented with 253.9 mM sucrose. The solution was maintained at approximately 12°C during perfusion. Then the spinal cord and vertebral column were dissected in continuity with the hairy hind paw skin innervated by the saphenous nerve [20, 40]. Intracellular single-unit recordings were performed on the dorsal root ganglia (L2/3) using a quartz microelectrode containing 1 M potassium acetate. Current was delivered through a suction electrode attached to the side of the nerve to identify the somata with axons residing in the saphenous nerve. After the identification of an electrically active cell, the receptive field (RF) in the skin was identified using the small paint brush, or hot (~51°C) and cold (~1°C) physiological saline. Once the RF was established, an increasing series of Von Frey filaments (0.07g-10g) were used to assess mechanical responsiveness. After the mechanical responsiveness was confirmed, a controlled thermal stimulus using a 3 × 5 mm contact area Peltier element (Yale University Machine Shop) was delivered to the cell’s RF. First, a variable-rate cold stimulus beginning at 31°C, dropping to approximately 2°C to 4°C, holding for 4 to 5 seconds, and slowly returning to 31°C was delivered. After bath temperature was maintained for approximately 4 to 5 seconds, a heat ramp was applied to the RF, which went from 31°C to 52°C in 12 seconds. This heat stimulus was then held at 52°C for 5 seconds and ramped back down to 31°C in 12 seconds. A recovery period of approximately 20–30 seconds, was employed between different modalities of stimulation. All the obtained responses were recorded digitally for offline analysis of thresholds, firing rates, and mean peak instantaneous frequencies to the various stimuli using Spike2 software (Cambridge Electronic Design).

Western Blotting

DRG (L2/3) samples were isolated and freshly grinded to harvest protein using lysis buffer containing 1% sodium dodecyl sulfate, 10 mM Tris-HCL (pH 7.4), and protease inhibitors (1-mg/mL pepstatin, leupeptin, aprotinin, 1 mM sodium orthovanadate, and 100-mg/mL phenylmethylsulfonyl fluoride (table S1). 25-μg samples were boiled in gel loading buffer containing β-mercaptoethanol and sodium lauryl sulfate as a reducing agent and loaded onto 12% precast polyacrylamide gels (table S1) and ran at 120V for 90 minutes. The samples were then transferred to a nitrocellulose membrane for 90 minutes at room temperature, followed by the blocking with 5% non-fat milk in tris buffer saline (TBS) for 2 hours, and incubated from another ~2 hours at room temperature to overnight at 4°C with primary antibodies: ERK1/2 (1:500), pERK (1:500), and GAPDH (1:1000) (table S1). Afterward, the membrane was incubated with secondary antibodies (1:10,000) (table S1) for one hour, with 3 to 4 washes between antibody incubations and imaged on LiCor-Odyssey. The immunoreactive bands were analyzed by densitometry, where rectangles were drawn around the bands to measure signal intensity, using GAPDH as the internal control.

Real-Time PCR

RNA was isolated from the DRGs (L2/3) using the Qiagen RNeasy^® Mini Kit^® (QIAGEN, Valencia, CA, USA) per the manufacturer’s protocol. After purification, 1μg of RNA was transcribed into cDNA using M-MLV Reverse Transcriptase (Promega, Madison, WI, USA). 25ng samples of cDNA were then used in SYBR Green real-time PCR reactions on a Step-One real-time PCR machine (Applied Biosystems). Cycle time (Ct) values for all targets were all normalized to a GAPDH internal control, and fold change was determined as 2^ΔΔCt (Applied Biosystems). Values were converted and reported as a percentage change, where 2-fold change = 100% change. Primers used are listed in the supplement (table S2).

Statistical analysis

All datasets were analyzed using Prism statistical software. Data were first assessed using Shapiro-Wilk test/D’ Agostino and Pearson test for normality and equal variance prior to performing the indicated statistical tests. For behavioral analyses, one- or two-way ANOVA with Tukey’s post hoc or Fisher’s LSD test were performed. For electrophysiological analyses, 1-way ANOVA with Tukey’s post hoc (parametric) tests were performed. For immunohistochemistry, 2-tailed t tests, or a one-, or two-way ANOVA with Tukey’s post hoc test were performed. Data are represented as mean ± SEM. Significance was defined with p < 0.05.

Diagrams:

The summary model figure and other diagrams in the figures were generated using Biorender.com

Supplementary Material

Figures S1–S7

Tables S1–S2