Cabozantinib for neurofibromatosis type 1-related plexiform neurofibromas: a phase 2 trial

Abstract

Neurofibromatosis Type 1 (NF1) plexiform neurofibromas (PN) are progressive, multicellular neoplasms that cause morbidity and predispose to sarcoma. Treatment of Nf1^flox/flox;PostnCre mice with cabozantinib, an inhibitor of multiple tyrosine kinases, caused a reduction in PN size and number and differential modulation of kinases in cell lineages that drive PN growth. Based on these findings, the Neurofibromatosis Clinical Trials Consortium conducted a phase II, open-label, non-randomized Simon two-stage study to assess the safety, efficacy and biologic activity of cabozantinib in patients ≥16 years of age with NF1 and progressive or symptomatic, inoperable PN (NCT02101736). The trial met its primary outcome, defined as ≥25% of patients achieving a partial response (PR, defined as ≥20% reduction in target lesion volume as assessed by MRI) after 12 cycles of therapy. Secondary outcomes included adverse events (AE), patient-reported outcomes (PRO) assessing pain and quality of life (QOL), pharmacokinetics, and the levels of circulating endothelial cells and cytokines. Eight of 19 evaluable (42%) trial participants achieved a PR. The median change in tumor volume was 15.2% (range +2.2% to −36.9%) and no patient had disease progression while on treatment. Nine patients required dose reduction or discontinuation of therapy due to AEs; common AEs included gastrointestinal toxicity, hypothyroidism, fatigue and palmar plantar erythrodysesthesia. A total of 11 grade 3 AEs occurred in 8 patients. Patients with PR had a significant reduction in tumor pain intensity and pain interference in daily life, but no change in global QOL scores. These data indicate that cabozantinib is active in NF1-associated PN, resulting in tumor volume reduction and pain improvement.

Fewer than 10% of the roughly 7000 known rare diseases have therapies available or in development¹. One of the most common of these rare disorders is Neurofibromatosis type 1 (NF1), a tumor predisposition syndrome affecting 1/3000 people worldwide². NF1 is caused by loss of function of the NF1 tumor suppressor gene, resulting in abnormal neurofibromin, a GTPase activating protein (GAP) for p21Ras^3–5. Abnormal neurofibromin results in constitutively activated p21Ras, leading to a lifetime of tumors occurring throughout the central and peripheral nervous system. Although the gene and protein are known, development of therapies for NF1, like many rare diseases, is challenging due to disease heterogeneity, as well as a small and often geographically dispersed patient population. To address many of the limitations that impede the development of therapies for this rare disease, the NF1 scientific community has created a broad infrastructure for drug discovery and evaluation^6–9.

Plexiform neurofibromas (PN) are one of the most prevalent and impactful clinical manifestations of NF1. PN are multicellular tumors composed of tumorigenic Schwann cells, fibroblasts, perineural cells, macrophages, mast cells, and secreted collagen¹⁰. These tumors arise within nerves, grow rapidly during childhood, and can lead to motor and sensory dysfunction, pain, and disfigurement in as many as 40% of patients with NF1^11,12. PN can be life-threatening when impinging on vital structures (such as airway and spinal cord)¹³ and are associated with a 10–15% risk of transformation to malignant peripheral nerve sheath tumors (MPNST), an aggressive and often fatal sarcoma¹⁴. Like many histologically benign and slow-growing tumors, conventional cytotoxic chemotherapy has no benefit in PN^6,15. Similarly, radiation therapy is avoided given the risk of promoting the development of secondary malignancy and uncertain efficacy^16,17. Thus, surgical debulking is standard of care for PN; however, given the intrinsic relationship between the tumor and nerve, surgery is often limited or infeasible. Recently, there have been promising results in phase 1 and phase 2 trials of the MEK inhibitor selumetinib in children with symptomatic or progressive PN^18,19. Although these data are encouraging, there is both heterogeneous response of PN to MEK inhibition as well as a need for continuous drug exposure^13,18. Hence, NF1-associated PN remain a tumor with an urgent, unmet medical need.

Extensive preclinical and translational studies demonstrate the critical role of the NF1 microenvironment in PN tumorigenesis. Specifically, mast cells, macrophages and endothelium are all critical for tumor initiation and maintenance in preclinical models^10,20,21. These studies suggest that broad receptor tyrosine kinase (RTK) inhibitors that disrupt interactions between homozygous Nf1 mutant (Nf1^−/−) Schwann cells and the microenvironment may be effective for treatment of PN. To test this hypothesis, we evaluated the multi-RTK inhibitor cabozantinib in a preclinical murine model with demonstrated fidelity for human PN^20,22,23. Cabozantinib is approved for treatment of medullary thyroid carcinoma²⁴, renal cell carcinoma²⁵, and hepatocellular carcinoma²⁶, and exhibits activity against c-Kit, VEGFR2, MET, RET, FLT3, and the TAM family receptors (AXL, TYRO3 and MERTK)^27,28. Amongst these, c-Kit and VEGF signaling have documented roles in tumor initiation and in stimulating neoangiogensis, respectively, within the PN microenvironment^20–22. Further, other cabozantinib targets have been implicated in these processes in other tumors^24–26.

Here we show that cabozantinib significantly reduced PN tumor burden in Nf1 mutant mice. Using multiplexed inhibitor beads (MIB) coupled with mass spectrometry (MS), we uncovered key kinases modulated by cabozantinib in tumor tissue and within individual cell lineages in the tumor microenvironment. These findings prompted a multicenter phase 2 clinical trial for adolescents and adults with NF1 and unresectable, progressive or symptomatic PN via the Congressionally Directed Medical Research Programs Neurofibromatosis Clinical Trials Consortium (NCT02101736) that revealed anti-tumor activity similar to that seen in the mouse model. Collectively, these studies illustrate the value of exploiting systems biology for rare tumors as well as the efficiency of coordinated translational and clinical efforts to advance targeted therapies for a rare disease such as NF1-associated PN.

RESULTS

Cabozantinib abrogates plexiform neurofibroma growth in Nf1 mutant mice

We evaluated the efficacy of cabozantinib in the Nf1^flox/flox;PostnCre model of PN²³. Pharmacokinetic (PK) analysis in plasma and tumor-bearing nerve tissues from experimental mice were assessed by high performance liquid chromatography (HPLC)-MS/MS. Cabozantinib was absorbed rapidly, achieving a maximum concentration in plasma of 6800 ng/mL at 2 hours post exposure, similar to findings in humans²⁹ (Extended Data Figure 1a). The drug penetrated nerve tissue at a ratio of approximately 1:4 to 1:5 (tissue:plasma), reaching a maximal concentration of 636 ng/g in nerve tissue. Tabulated PK parameters and chromatograms from HPLC-MS/MS method development are provided in Extended Data Figure 1b–d.

Nf1^flox/flox;PostnCre mice were initiated on cabozantinib at 4 months of age when they consistently have multiple PN³⁰. Strikingly, after 12 weeks of daily cabozantinib treatment, the number of PN per mouse was reduced by approximately 60% with normalization of nerve microarchitecture (Figure 1a, b, Extended Data Figure 2, 3) relative to the vehicle control cohort in at least as subset of these tumors. Cabozantinib treatment also reduced the mean volume of the proximal peripheral nerve roots by 38% (Figure 1c). Unexpectedly, despite the well-recognized modulation of c-kit by cabozantinib and the known role of c-kit in mast cell chemotaxis for PN²⁰, mast cell infiltration was not significantly changed (Figure 1d). Given its established inhibitory activity against VEGFR2 and AXL, we also examined blood vessels using CD31 immunohistochemistry, revealing a significant reduction in neoangiogenesis in cabozantinib-treated tumors versus vehicle control (Figure 1e).

Figure 1. Cabozantinib reduces plexiform neurofibroma tumor burden in Nf1 mutant mice.

(a) Representative photomicrographs of H&E (low mag, left and high mag, middle) and trichrome (high mag, right) stained sections of vehicle (top) and cabozantinib (bottom) treated nerve tissue demonstrating normalization of nerve microarchitecture and reduced collagen with cabozantinib treatment. The experiment was repeated three times independently with similar results. (b) The number of plexiform neurofibroma tumors in the nerve tree were scored per mouse in vehicle (n=13) and cabozantinib (n= 16) treated animals. *** P-value = 0.0004 (unpaired, two-tailed Student’s t-test). Data are presented as mean values ± SEM. (c) Mean proximal nerve root volume (mm³) in vehicle and cabozantinib treated animals. N=56 nerve roots from n=14 mice were evaluated in the vehicle group and n=64 nerve roots from n=16 mice were evaluated in the cabozantinib treated group. ** P-value = 0.0011 (unpaired, two-tailed Student’s t-test). Data are presented as mean values ± SEM. (d) Left, mast cells per high power field were scored on toluidine blue stained sections of vehicle (n= 4 mice) and cabozantinib (n= 4 mice) treated nerve tissue. 15 randomly selected 40x fields were scored per slide. Right, representative images of toluidine blue staining in vehicle (Veh) and cabozantinib (Cabo) treated mice. Arrowheads indicate mast cells. ns, no statistically significant difference, P = 0.1013 (unpaired, two-tailed Students t-test). Data are presented as mean values ± SEM. (e) Left, blood vessels per high power field were scored on CD31 stained sections of nerve tissue from vehicle (n= 4 mice) and cabozantinib (n=4 mice) treated mice. 20 randomly selected 40x fields we scored per slide. Right, represenative images of CD31 staining in vehicle and cabozanitinib treated mice. Arrowheads indicate blood vessels. Inset, secondary antibody only control. **** P-value < 0.0001 (unpaired, two-tailed Student’s t-test). Data are presented as mean values ± SEM.

Mass spectrometry based kinome profiling of murine PN reveals molecular signatures of tumor response to cabozantinib

To delineate candidate kinases associated with the anti-tumor activity of cabozantinib, we returned to the preclinical GEM model and utilized an innovative chemical proteomic approach that combines multiplexed inhibitor beads (MIB) with mass spectrometry (MS) to comprehensively define the functional signature of cabozantinib target kinases in PN ^31,32. Kinase binding to MIBs is influenced by kinase protein expression level, affinity of a given kinase for the inhibitors used for MIB capture, and by activity, which combined define a functional kinome. At early time points following treatment (1 and 3 days), many kinases had transient increases in MIB binding in response to cabozantinib, likely reflecting an early cellular stress response (Extended Data Figure 4a and b). Following 7 days of sustained cabozantinib treatment, there was a shift in the functional kinome profile predominated by loss of MIB binding of a small number of kinases (Figure 2a) including MET, DDR1, DDR2, and AXL.

Figure 2. MIB/MS profiling reveals cabozantinib target-kinases in plexiform neurofibroma tumors of Nf1 mutant mice.

(a) Volcano plot showing the log₂-fold change MIB binding (LFQ intensity) in sciatic nerve tumor tissue lysate after 7 days of cabozantinib (n= 3 mice) vs vehicle control (n= 4 mice) treatment, plotted against the −log₁₀ Benjamini-Hochberg adjusted P-value of unpaired, two-tailed Student’s t-tests (the dotted line represents FDR = 0.05). Red points highlight known cabozantinib target kinases with the most significant log₂ fold changes in MIB binding. (b) The log₂ fold change in MIB binding in sciatic nerve tumor tissue lysate at the indicated cabozanitinib concentrations compared to DMSO is plotted for two replicates. Lysate was incubated in vitro with DMSO or the indicated concentrations of cabozantanib for 1 hour prior to MIB/MS kinome profiling.

To further validate that the loss of MIB binding of these kinases was due to cabozantinib binding, murine PN tumor lysates were incubated ex vivo with DMSO or with serial increasing concentrations of cabozantinib (10, 100, and 1000 nM) for 1 hour prior to MIB/MS kinome profiling, an approach used to characterize drug target profiles³³. Overwhelmingly, these same kinases exhibited significantly decreased MIB binding in a concentration-dependent fashion with increasing concentrations of cabozantinib (Figure 2b). Other kinases with cabozantinib binding affinity modulating a diverse array of tumorigenic processes were identified at lower levels (Extended Data Figure 5, Supplementary Table 1). Given the MIB proteomics data, the functional decline in neoangiogensis, and knowledge of the established target kinases of cabozantinib, we further interrogated the roles of key target kinases of cabozantinib, particularly AXL in neoplastic Schwann cells and within lineages of the tumor microenvironment (Extended Data Figures 6–7). Interestingly, AXL was found to be increased in PN tissue as compared to normal nerve, as assessed by MIBs (Extended Data Figure 6a), western blot (Extended Data Figure 6b) and immunohistochemistry (Extended Data Figure 6c).

Further, there was a significant decline in AXL in cabozantinib treated tissues (Extended Data Figure 6c–e). Functionally, Gas 6 mediated activation of AXL resulted in an increase in proliferation of Nf1^−/− Schwann cells as well as fibroblast and endothelial progenitor cells (Extended Data Figure 7a,c,e). This activation was inhibited at both the cellular (Extended Data Figure 7a,c,e,g) and biochemical level (Extended Data Figure 7b,d,e) by cabozantinib. No increase in apoptosis was observed at the cellular level (Extended Data Figure 8a) or in tissues (Extended Data Figure 8b) following cabozantinib treatment. Interestingly, a soluble form of AXL (sAXL) generated by proteolytic cleavage is reported to negatively regulate AXL signaling by acting as a decoy receptor for GAS6 ³⁴ (Extended Data Figure 9a). Here we found that sAXL is significantly elevated in the plasma of cabozantinib treated mice as compared to the WT mice or Nf1^flox/flox;PostnCre mice treated with the vehicle control (Extended Data Figure 9b).

Activity of cabozantinib in a phase 2 clinical trial for NF1-related PN

Based on the preclinical data presented above a multicenter, prospective, open label, single arm phase 2 trial of cabozantinib using a two stage Simon Optimal design was conducted via the Neurofibromatosis Clinical Trials Consortium (NCT02101736). Participants were ≥16 years old with NF1 and either progressive or symptomatic PN not amenable to surgery. The primary outcome measure was volumetric response of the target PN determined by central review of magnetic resonance imaging (MRI). Success was defined as ≥25% of participants achieving a partial response (PR) after 12 cycles. Initially 9 subjects were recruited to provide evidence of effectiveness and safety. If at least one of these 9 subjects achieved a PR, the study would enroll up to 24 subjects in order to recruit a minimum of 17 evaluable subjects. The drug was considered of interest if there were at least 3 responses, for a Type I error of 5%, power of 80%, and a null hypothesis success rate of 5% versus a treatment with a 25% response rate. Cabozantinib was given once daily continuously for up to 24 cycles (one cycle = 28 days) starting at 40 mg with planned escalation to 60 mg after 2 cycles (Figure 3a). Dose reductions for toxicity were allowed to a minimum daily dose of 20 mg. MRI volumetric response was assessed after cycle 4, 8, 12, 18, and 24 for those on treatment. PR was defined as ≥20% reduction in tumor volume from baseline on MRI 35. Participants were considered evaluable for toxicity if they received at least one dose of the study drug and for response if they completed ≥2 cycles of therapy and one follow-up MRI. Participants who did not achieve at least 15% reduction in tumor volume by cycle 8 were removed from study in an effort to balance toxicity risk against a hypothesized low likelihood of achieving PR by 12 months. Secondary endpoints included safety and tolerability, patient-reported outcomes (PRO) assessing tumor pain intensity (Numeric Rating Scale-11; NRS-11), pain interference in daily life (Brief Pain Inventory-Pain Interference Scale; BPI-PI), and disease-specific quality of life (PedsQL NF1 module), pharmacokinetics (PK), and circulating endothelial cells and cytokines. Additional eligibility criteria and required study observations are in Supplementary Tables 2 and 3.

Figure 3. Clinical trial schema and CONSORT diagram.

(a) Enrolled participants were started on 40 mg of cabozantinib daily for the first 2 cycles, then escalated to 60 mg daily for the remaining cycles if no dose-limiting toxicities (DLT) occurred. Each cycle was 28 days in duration. MRI evaluations occurred after cycles 4, 8, 12 and then after cycles 18 and 24 for those who stayed on treatment. Evaluable participants were those who completed at least 2 cycles of therapy and had at least one follow-up MRI. Participants who did not show ≥15% volumetric reduction of the target plexiform neurofibroma by cycle 8 were taken off study; those who had ≥20% volumetric reduction of the target plexiform neurofibroma by cycle 12 continued therapy for up to another 12 cycles. (b) CONSORT diagram showing disposition of the study participants. Nine patients were initially enrolled on study. Per the 2-stage study design, once one of the participants achieved a partial response, an additional 14 patients were enrolled. In total, 23 patients were enrolled on study but only 21 participants received at least 1 dose of cabozantinib, and of these 21 participants, only 19 were evaluable for therapy response.

Participants were enrolled from Neurofibromatosis Clinical Trials Consortium sites. Between September 2014 and December 2014, the initial 9 participants were enrolled. Per the study design, enrollment was halted until May 2015, when one of the subjects achieved a PR meeting the minimum criteria for expansion to the second stage. An additional 14 participants were then enrolled through February 2016. Two participants were determined to be ineligible prior to receiving cabozantinib, one withdrew from study during cycle one after experiencing grade 1 anorexia, constipation, fatigue, and nausea, and one subject was determined to be ineligible after receiving study drug (Figure 3b). In total, there were 21 participants evaluable for toxicity and 19 evaluable for response (median age 23 years, range 16–34 years; median baseline tumor volume 557 mL, range 57–2954 mL) (Table 1).

Table 1.

Baseline Characteristics of All Patients Treated with Cabozantinib and Evaluated for Toxicity and Response

	Evaluable for Toxicity (n=21)	Evaluable for Response (n=19)
Age
Median (Range)	22 (16–34 years)	23 (16–34 years)
Gender
Male	14 (66.7%)	13 (68.4%)
Female	7 (33.3%)	6 (31.6%)
Race
White	16 (76.2%)	14 (73.7%)
Black or African-American	1 (4.8%)	1 (5.3%)
Native Hawaiian or Other Pacific Islander	1 (4.8%)	1 (5.3%)
Asian	1 (4.8%)	1 (5.3%)
Other	2 (9.5%)	2 (10.5%)
Tumor Locations
Head (Face)		1 (5.3%)
Combined head and neck		4 (21.1%)
Combined neck and chest		2 (10.5%)
Trunk		7 (36.8%)
Combined trunk and extremity		4 (21.1%)
Extremity		1 (5.3%)
Tumor Size
Median (Range)		557 (57–2954 mL)

Clinical imaging response

Of the 19 participants evaluable for response, 8 (42%) had a PR and 11 had stable disease (SD) (Figure 4a). No patient had disease progression while on study. The median change in tumor volume was −15.7% (range, +2.8% to −38.0%). Only one subject achieved a PR by the end of cycle 4. The majority of patients achieved PR at the end of cycle 12. Maximal tumor response was not achieved until cycle 18 in 4 participants and cycle 24 in 2 participants. Notably, three responders maintained or had further tumor volume reduction in the setting of dose reduction (Figure 4b). No differences were seen in tumor location, baseline tumor size, demographic characteristics between reponders and non-responders (Supplementary Table 4).

Figure 4. Participant responses to cabozantinib.

(a) Waterfall plot per participant of best response to cabozantinib as determined by the percent change from baseline in target plexiform neurofibroma tumor volume. Eight of nineteen participants (42.1%) demonstrated a partial response (PR) with greater than 20% reduction of target plexiform neurofibroma tumor volume from baseline. (b) Duration of therapy for participants with stable disease (SD), indicated by green bars, and those with PR, indicated by blue bars. The length of the bar shows time from first dose administration until drug discontinuation. Red open stars indicate the first dose reduction and red closed stars indicate the second dose reduction. For responders, yellow triangles show the time of initial PR and yellow circles show the time of maximum response. Two patients achieved first and maximum responses after 2 dose reductions.

Clinical safety and tolerability

Two participants discontinued cabozantinib due to dose limiting toxicity (DLT), palmar plantar erythrodysesthesia (PPE) in both cases. Seven participants required dose reductions while on study; for PPE (n=7), skin infection (n=1), and weight loss (n=2). The most common adverse events (AE) on study were diarrhea (n=17), nausea (n=14), asymptomatic hypothyroidism (n=15), fatigue (n=13), and PPE (n=10) (Supplementary Table 5). Eleven grade 3 adverse events occurred in 8 participants, including PPE (n=4), hypertension (n=2), and one event each of grade 3 anorexia, neutropenia, proteinuria, skin infection, and vomiting (Supplementary Table 6). There were no grade 4 or 5 AEs.

Of the 19 participants evaluable for response, 5 completed all planned therapy, 6 were removed from study at cycle 8 because of <15% tumor volume reduction, and 2 were removed for DLT. Six participants chose to stop treatment (2 at cycle 3, 1 at cycle 6, 1 at cycle 7, and 2 at cycle 20) because of low grade AEs perceived to be intolerable (including anorexia, gastrointestinal symptoms, PPE, acne, and change in hair color).

Patient-reported clinical outcome (PRO) assessments of pain and quality of life (QOL)

All 19 evaluable participants completed the PROs evaluating pain and QOL at baseline. Of these, 17 completed the PROs at cycle 4, 14 at cycle 8, and 8 at cycle 12. All eight participants who had a PR completed the PROs through cycle 12. At baseline, 84% of participants reported tumor pain on the NRS-11 (total sample mean=5.0; SD=3.1; range 0–10) with 75% reporting moderate to severe pain (rating of ≥4). Consistent with the MRI data, PRO assessments of pain improved from study entry in the 8 patients who achieved a PR, with a significant reduction in reported tumor pain occurring as early as cycle 4 (LSM estimate=2.7; t=4.42, p=0.0001) that persisted to cycle 12 (responders: LSM estimate=3.00; t=4.81, p<0.0001; Non-responders: LSM estimate=1.04, t=1.52, p=0.139 at cycle 8). Of note, the responders’ mean tumor pain was slightly but not significantly (p=0.23) higher than the non-responders at baseline, and attrition of non-responder participants with high pain ratings prior to cycle 8 limits the comparison of the subgroups over time. Nevertheless, the mean decrease in worst tumor pain intensity in the responder group with no attrition was 3 points from baseline to cycle 12, consistent with clinically meaningful change³⁶. In addition, only the responder group exhibited significant reduction in pain interference at each evaluation timepoint from baseline to cycle 12, as assessed by the mean total BPI- PI scores (LSM estimate=1.48; t=2.74, p=0.0099).

There was no significant change (p<0.01) over time in the PedsQL NF1 Total Functioning mean score (validated for adolescents and adults with NF1³⁷ at any evaluation from baseline to cycle 12 for the total sample (LSM estimate=−2.83; t=−0.85, p=0.40) or for responders versus non-responders. When examining the physical symptoms and psychosocial functioning domains for the total sample, none of the domain scores changed significantly from baseline to cycle 12 except for an improvement in the general pain and hurt scale (LSM estimate= −14.98; t=−3.12, p=0.0036).

Clinical pharmacokinetics

The dose-normalized mean cabozantinib trough level was 16.5 ng/mL/mg at Cycle 1 Day 15 and 20.7 ng/mL/mg at end of Cycle 1 at the 40 mg dose; consistent with results in renal cell and hepatocellular carcinoma studies at 60 mg/day (raw data on file at Exelixis; Supplementary Table 8).

Cytokine and Biomarker Analysis

Based on the preclinical data, we hypothesized that circulating pro-angiogenic markers may serve as biomarkers for response to cabozantinib in patients. We quantified pro-angiogenic circulating hematopoietic stem and progenitor cells (CHSPCs) and measured the levels of 45 cytokines including soluble AXL (sAXL) (Supplementary Table 9), in peripheral blood samples from patients at baseline, early (Cycle 1 Day 15) and later (end of Cycle 2 and end of Cycle 4) treatment cycles. A possible but not statistically significant association was seen between an increase in sAXL and clinical response in eight participants (p=0.08, Extended Data Figure 10a). Parallel changes between early and late time points associated with response were also seen for IP-10, TNFα and IL-1α (Supplementary Tables 10 and 11). Reductions of proangiogenic circulating hematopoietic stem and progenitor cells (CHSPCs) between baseline/early and later treatment cycles have been associated with therapuetic responses in gliomas³⁸. However, when comparing participants with SD to those who achieved PR on cabozantinb, no such association was observed (Extended Data Figure 10b and c). The raw data is provided in the supplementary materials (CHSPC-and-Cytokine-Data.xlsx).

DISCUSSION

Given the number of potential therapeutics to be assessed, paired with challenging clinical trial logistics in a rare tumor predisposition syndrome such as NF1, establishing a strategy linking preclinical and clinical studies is critical for translational innovation and efficiency. GEM models which accurately recapitulate PN growth kinetics and histopathologic characteristics in humans informed mechanisms-based, NF1-associated PN trials, including a recent successful study of MEK inhibition^{20,22,39–41}. Using a validated GEM model, we identified cabozantinib as an efficacious targeted therapy in NF1-associated PN which prompted a phase 2 multicenter trial. We have evidence that AXL is at least one key kinase associated with cabozantinib treatment response. AXL overexpression has been implicated in tumor angiogenesis and therapeutic resistance to targeted agents, including other RTK inhibitors⁴². PN are vascular tumors and concordant with downregulation of AXL, there was significant reduction in neoangiogenesis in cabozantinib treated PN in the GEM model. While further genetic and cell lineage specific experiments will be required to fully delineate the mechanisms by which cabozantinib exerts biological activity in plexiform neurofibroma, the preclinical studies presented here suggest cabozantinib may target both the Nf1 deficient Schwann cells as well as particular components of the tumor micorenvironent including both fibroblasts and collagen production as well as endothelial cells and neoangiogenesis.

Collaboration between the laboratory and clinical scientists in the Neurofibromatosis Clinical Trials Consortium allowed for these preclinical findings to be efficiently translated to patients with NF1 associated PN. The 42% PR rate with cabozantinib in this trial represents the best documented response to date in adolescents and adults with PN, and only the second agent or class of agents with substantial efficacy for PN. Furthermore, PROs indicated clinically meaningful decreases in tumor pain intensity as well as significant decreases in the interference of pain in daily life in the responding patients. The use of prospectively-administered PROs added important context about the clinical significance of a PR. Outside of the MEK inhibitor selumetinib, which had a 74% PR in children with progressive PN¹⁹, other agents tested have either only prolonged PN time to progression (sirolimus⁴³ and pegylated interferon alfa-2b⁴⁴) or had modest response rates limited to small tumors of less than 20 cm³ (imatinib mesylate²²). Based on the activity of cabozantinib in adolescents and adults, the NF Clinical Trials Consortium recently opened a pediatric cohort, and this study is ongoing. There is also an ongoing study of selumetinib for adults with NF1-associated PN (NCT02407405). These investigations will provide the data necessary to interrogate the matrix of tumor age, growth status, and response to either MEK inhibition or cabozantinib in NF1 associated PN⁴⁵. Furthermore, the anticipated outcomes from these clinical trials paired with ongoing laboratory studies will enable future exploration of combination therapies. Interestingly, cabozantinib and another MEK inhibitor (mirdametinib; PD-0325901), have been found to reduce growth in a highly aggressive, malignant peripheral nerve sheath tumor (MPNST) murine model that has compound mutations of Nf1 and Trp53. Lock and colleagues found that MNK/eIF4E is the key cabozantinib target responsible for tumor regression in their studies⁴⁶. While we did not find an elevation of MNK in the benign PN model, which is genetically and biologically distinct from MPNSTs, both studies point to the potential utility of cabozantinib that targets key kinases found across the spectrum of benign and malignant NF1 associated nerve sheath tumors.

Cabozantinib was reasonably tolerated by patients throughout the study with expected AEs and rare severe AEs compared to prior studies of cabozantinib^47–50. A significant portion of patients, however, discontinued cabozantinib due to low grade AEs. NF1 patients with PN represent a unique population as most have lived with PN throughout their life and may have a low-threshold for drug-related AEs²². In addition, efficacy appears to be achieved at lower drug doses in the setting of NF1 than in other tumor indications with both cabozantinib and MEK inhibitors^18,51. These findings highlight the importance of NF1-specific dose selection that allows for biologic effect and clinically meaningful efficacy with minimum toxicity. This is particularly relevant when considering long-term administration of targeted therapies for NF1-PN. Ongoing pre-clinical studies are exploring intermittent dosing schedules of targeted therapies in an effort to identify monotherapy or combination dosing that minimizes toxicity and maximizes durable response.

Limitations of the clinical trial include the trial design which required participants to stop treatment if they did not have at least a 15% reduction in tumor volume after 8 cycles. Given that maximal response may not be achieved until year 2 of treatment, it is possible that this feature of the study design limited the response rate achieved in this study. Of note, this study requirement was eliminated for the pediatric cohort. In addition, since the protocol did not require that off-treatment MRI scans be performed, we do not have information about durability of response once medication is stopped. Although the study included detailed PRO assessments of pain and QOL, no functional evaluations were performed. Furthermore, as this was a single arm study with limited subject numbers, comparisons with other effective agents for PN, such as selumetinib, cannot be made, and would require a larger randomized trial design.

Given the long latency to reach both PR and maximal tumor response by volumetric MRI criteria in an indolent tumor, the development of a biomarker predictive of benefit would help guide clinical decision making in PN. We did observe a trend that increases in sAXL within the first 4 cycles of treatment were associated with PR to cabozantinib. This data is intriguing, given that other studies have demonstrated that increased sAXL may act as a decoy receptor, and in drug treatment experiments, levels of sAXL seem to correlate with efficacy of RTK inhibition^34,52. However, due to the inherent limitation of sample size, the statistical power of this association in our clinical trial is not robust, with an unadjusted p-value of 0.08. If confirmed in a larger study, such a biomarker may support continuation of cabozantinib in patients with low grade toxicities who have a biomarker response or redirect patients without biomarker response to alternate treatments, avoiding prolonged drug exposure if there is low likelihood of benefit. We also observed a statistical difference in select pro-inflammatory cytokines including IP-10, TNF-α, and IL-1α that have complex roles in immune function/dysfunction and tumor development.

In conclusion, we present an integrated preclinical pipeline using an authenticated GEM model and an innovative proteomic approach to guide therapy for a rare tumor in a multi-center clinical consortium trial. The preclinical model was predictive of a durable response rate to cabozantinib in adolescent and adult patients with NF1-associated PN. Although cancer kinome networks are highly complex and perturbations introduced by multi-RTK inhibitors can have dynamic and broad-reaching impacts beyond their postulated targets, comprehensive evaluation of tumor kinase profiles in vivo revealed unanticipated kinases associated with therapeutic response that merit further investigation. As we continue to refine our understanding of PN tumorigenesis and utilize multi-targeted drug therapies, the importance of determining target engagement, adaptive tumor response, and optimal dosing is paramount. The integrated translational approach outlined here establishes a powerful roadmap to facilitate such complex interrogations in the setting of rare tumor syndromes.

Online Methods

METHODS

Clinical Trial Study Population

Inclusion criteria included patients aged 16 years or older with a diagnosis of NF1 and an unresectable PN that was measurable by volumetric analysis³⁵ and either progressive (defined as ≥20% increase in the volume, ≥13% increase in the product of the two longest perpendicular diameters, or ≥6% increase in the longest diameter in the prior year) or causing significant morbidity, such as (but not limited to) head and neck lesions compromising the airway or great vessels, brachial or lumbar plexus lesions causing nerve compression and loss of function, lesions causing major deformity or significant disfigurement, lesions of the extremity causing limb hypertrophy or loss of function, and painful lesions. Histologic confirmation of tumor was not necessary in the presence of consistent clinical and imaging findings. Previous treatment for a PN was allowed but participants must have recovered from the acute toxic effects of all therapy. Other eligibility criteria included (a) adequate performance status (Karnofsky score ≥ 50); (b) normal hematologic, hepatic, renal, and cardiovascular function; and (c) signed informed consent from the patient (≥18 years of age) or their legal guardians (<18 years of age), according to institutional review board guidelines.

Participants were excluded if they (a) had an active optic glioma or other low-grade glioma requiring treatment with chemotherapy or radiation therapy; (b) required treatment for a malignancy in the prior 12 months; (c) had dental braces or prosthesis that interfere with volumetric analysis; (d) were unable to swallow tablets; (e) were post-pubertal males/females who would not agree to use effective contraception, or were pregnant/breast-feeding females; (f) had a known history of HIV or immunodeficiency; (g) had impairment of gastrointestinal function or gastrointestinal disease that significantly alters the absorption of cabozantinib; (h) had concurrent severe and/or uncontrolled medical disease; (i) required therapeutic treatment with anticoagulants or antiplatelet agents with the exception of low dose aspirin, low-dose warfarin, and prophylactic low molecular weight heparin; or (j) had undergone major surgery within 3 months or minor surgery within one month of first dose of cabozantinib. In participants who had received therapy prior to study enrollment, 2 weeks – 6 months must have elapsed, depending on the specific prior therapy. The use of strong CYP3A4 inducers/inhibitors was not allowed while on study.

A complete list of the inclusion and exclusion criteria is provided in Supplementary Table 2. Ethics oversight was provided by both the local IRB’s (Children’s Hospital Boston/Dana-Farber Cancer Institute/Massachusetts General Hospital, Children’s Hospital Los Angeles, The Children’s Hospital of Philadelphia, Children’s National Medical Center, Cincinnati Children’s Hospital Medical Center, Indiana University, Ann & Robert H. Lurie Children’s Hospital of Chicago, National Cancer Institute, New York University, University of Alabama at Birmingham, University of Chicago, University of Utah, The Washington University in St. Louis) and the U.S. Army Medical Research and Material Command (USAMRMC) Human Research Protections Office (HRPO). Each participating center’s local IRB reviewed and approved the study. Once local approval was obtained, this was submitted to the USAMRMC HRPO for review and approval. No study related activities could commence without approval of both the local IRB and the USAMRMC HRPO. In addition, the study was also approved by the FDA.

Study Design

This prospective, multicenter, nonrandomized phase 2 trial was performed by the NF Clinical Trials Consortium. The primary measure of efficacy was volumetric response of the target PN determined by central review of magnetic resonance imaging (MRI), as per the recommendations from REiNS (Response Evaluation in Neurofibromatosis and Schwannomatosis), an international effort to develop standardized outcome measures for clinical trials³⁵. Secondary endpoints included safety and tolerability, patient-reported outcomes (PRO) assessing tumor pain intensity (Numeric Rating Scale-11; NRS-11), pain interference in daily life (Brief Pain Inventory-Pain Interference Scale; BPI-PI), and disease-specific quality of life (PedsQL NF1 module), PK, and circulating endothelial cells and cytokines. Additional planned secondary outcomes not reported in this manuscript included estimating the overall response rate for non-target PN (there were 5 non-target PN identified in a total of 4 participants), a determination of whether patients who respond to cabozantinib maintained that response for one year off therapy (not reported because the protocol did not require that off-treatment MRI scans be performed, so that information was unavailable), assessing the activity of cabozantinib on mast cell activity by mast cell culture and FACS (not performed as samples were limited and other aims were prioritized), and assessing the PedsQL NF1 QOL Module, a disease specific QOL scale, for use in this patient population (a planned future analysis will combine this data with data from other PN trials using the PedsQL NF1 QOL Module). Cabozantinib was given daily on a continuous dosing schedule for up to 24 cycles. Each treatment cycle was 28 days. The starting cabozantinib dose was 40 mg daily with planned dose escalation to 60 mg daily after 2 cycles if tolerated (Figure 3a). For participants who experienced dose-limiting toxicities (DLT), up to two dose reductions were allowed (40 mg and 20 mg). Participants were removed from therapy if they experienced a cabozantinib-related toxicity requiring removal, for progressive disease, or if they had not achieved a partial response (PR; defined as a ≥20% reduction in tumor volume compared with pre-treatment baseline) by the end of 12 cycle. In addition, participants who had not achieved at least 15% reduction in tumor volume at cycle 8 were removed from study for safety reasons, as it was believed that the likelihood of achieving PR by 12 months was minimal. Participants were considered evaluable for toxicity if they received at least one dose of the study drug and evaluable for response if they completed at least two cycles of therapy and had their first follow-up MRI evaluation.

Baseline MRI was required within 4 weeks prior to study entry. Tumor response by MRI was determined after cycle 4, 8, and 12, and then after cycle 18 and 24 for those who continued therapy. Axial and coronal STIR images were obtained to cover the entire PN with 5–10 mm slice thickness. Imaging response was evaluated centrally at the Pediatric Oncology Branch of the National Cancer Institute using volumetric MRI analysis⁵³. Disease progression was defined as a ≥20% increase, and partial response as ≥20% reduction in the volume of the target PN, confirmed by a follow-up MRI. Safety evaluations including laboratory monitoring, electrocardiograms, physical exams, and vital signs were performed at regular intervals as outlined in Supplementary Table 3. Adverse events were graded according to National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE), version 4.0. Participants were considered evaluable for toxicity if they received at least 1 dose of the study drug. Success was defined as ≥25% of participants achieving PR after 12 cycles without significant toxicity.

Patient Reported Outcome Measures

All participants completed the following patient-reported outcomes (PROs) prior to study entry, at cycle 4, 8, and 12, and then after cycle 18 and 24 if they continued therapy.

Pain Intensity and Pain Interference Measures

Numerical Rating Scale-11 (NRS-11):

The NRS-11 is a self-report 11-point numeric scale that assesses pain intensity⁵⁴. It consists of a horizontal line with 0 representing “no pain” at the left end of the line and 10 representing “worst pain you can imagine” at the right end. Participants were asked to circle the one number from 0 to 10 that best described their “most important tumor pain” at its worst during the past week.

Brief Pain Inventory (BPI)-Pain Interference (PI) Scale:

The BPI-PI Scale is a 7-item self-report questionnaire that measures the extent to which pain interferes with daily functioning⁵⁵. Participants were asked to indicate how much pain interfered with various activities (general activity, mood, walking, normal work, relations with other people, sleep, and enjoyment of life) in the past week, with scores ranging from 0 (does not interfere) to 10 (completely interferes). The total score is the mean of all 7 items.

NF1 Disease-Specific Quality of Life Measures

PedsQL™ NF1 Module:

This scale is a self-report measure assessing NF1 disease-specific QOL³⁷. Participants who were ages 21 years and older completed the 74-item self-report adult questionnaire. It is comprised of 16 scales (physical functioning, emotional functioning, social functioning, cognitive functioning, communication, worry, perceived physical appearance, pain and hurt, paresthesias, skin irritation, sensation, movement and balance, fatigue, daily activities, treatment anxiety, and sexual functioning). Participants 16–20 years of age completed the self-report teen version. It is comprised of 16 scales (physical functioning, emotional functioning, social functioning, cognitive functioning, communication, worry, perceived physical appearance, pain and hurt, paresthesias, skin irritation, sensation, movement and balance, fatigue, daily activities, treatment anxiety, and school activities). Items are rated on a 5-point Likert scale (0 – 4) and transformed to a 0 – 100 scale, with higher domain and total mean raw scores indicating better QOL.

Pharmacokinetic Analysis

Blood samples were collected at 6 time-points for cabozantinib plasma concentration measurement: Cycle 1 Day 1 (pre, 4 hours after first dose), Cycle 1 Day 15 (pre, 4 hours after dose), and Cycle 1 Day 28 (pre, 4 hours after dose) and were shipped to Alturas Analytics (Moscow, ID) for analysis. Cabozantinib concentrations in human plasma matrix were quantified using a validated liquid chromatographic tandem mass spectrometry assay with 0.5 ng/mL as the lower limit of quantification.

Cytokine Analysis

Plasma samples were received from all 19 participants who were evaluable for response, but only 14 had multiple usable time points. Samples were stored at −80°C in the Translational Research Integrated Biology Laboratory (TRIB) in Riley Hospital for Children until collection and analysis of all samples. Samples were thawed, centrifuged to remove particulates (16,300 g for 10 min) and analyzed for 44 biomarkers by Luminex xMAP technology. Targets were detected using premixed kits from MilliporeSigma. The biomarkers measured included SCF (using a 1-Plex human Cytokine/Chemokine Magnetic Bead Panel); TGF-B1, TGFB2, TGFB3 (using a TGFβ−3 Plex Magnetic bead panel); sCD40L, EGF, Eotaxin/CCL11, FGF-2, Flt-3 ligand, Fractalkine, G-CSF, GM-CSF, GRO, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-1RA, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12 (p40), IL-12 (p70), IL-13, IL-15, IL-17A, IP-10, MCP-1, MCP-3, MDC (CCL22), MIP-1α, MIP-1β,TGF-α, TNF-α, TNF-β (using a Human Cytokine/Chemokine 37-Plex Magnetic Bead Panel), and VEGF-A, VEGF-C, VEGF-D (using a Human Angiogenesis/Growth Factor 3-Plex Magnetic Bead Panel). Samples were diluted according to manufacturer’s recommendations, assayed in duplicate using a Bio-Plex 200 System with high-throughput fluidics (HTF) multiplex array system (Bio-Rad Laboratories, Hercules, California, US), and analyzed using Bio-Plex Manager software (Bio-Rad Laboratories). Biomarker assays were run by the Multiplex Analysis Core at the Melvin and Bren Simon Cancer Center, Indiana University.

Levels of circulating sAXL were measured in human plasma using the Human AXL DuoSet ELISA kit (R&D Systems, DY154) and accompanying DuoSet ELISA Ancillary Reagent Kit 2 (R&D Systems, DY008) according to the manufacturer’s instructions. Human plasma samples were assayed in duplicate at dilution of 1:200 in Reagent Diluent. Optical density of each well was measured using a VERSAmax Tunable Microplate Reader (Molecular Devices) with SOFmax PRO software. Optical density readings at 540 nm were subtracted from the readings at 450 nm for wavelength correction. Investigators were blind to patients’ clinical response during data collection and analysis.

Circulating Progenitor Cell Analysis

The flow cytometry assay was performed in real time within 24 hours of peripheral blood collection. Peripheral blood collected off-site was shipped overnight. Peripheral blood mononuclear cells were isolated and incubated with Fc blocking reagent and stained as previously described⁵⁶. Briefly, following Fc block, mononuclear cells were stained with anti-human CD31 (FITC, BD), anti-human CD34 (PE, BD), anti-human AC133 (APC, Miltenyi Biotec), anti-human CD14 (PECy5.5, Abcam), anti-human CD45 (APC-AG740, Invitrogen), and the fixable amine reactive viability dye LiveDead (violet, Invitrogen). Fluorescent minus one (FMO) gating was used to properly delineate positive events. Circulating hematopoietic stem/progenitor cells (CHSPCs) have the phenotype: CD31⁺CD34^brightCD45^dimCD14⁻LIVEDEAD⁻. The proangiogenic fraction of the CHSPCs additionally stain positive for Ac133⁺ and have the phenotype: CD31⁺CD34^brightCD45^dim−Ac133⁺CD14⁻LIVEDEAD⁻. Following staining, cells were fixed in 1% paraformaldehyde (Tousimis) and run within 72 hours of fixation on a BD LSRII flow cytometry equipped with a 405nm violet laser, 488nm blue laser, and a 633nm red laser. Data were acquired uncompensated and exported as FCS 3.0 files for analysis using FlowJo Software (version 9.9.6,Tree Star Inc). Investigators were blind to patients’ clinical response during the data aquistition and analysis.

Statistical Analysis

The study used an optimal Simon 2-stage design⁵⁷, with a null hypothesis response rate of 0.05 and an alternative of 0.25, power of 80% and type I error of 0.05. This called for a first stage sample size of 9 with expansion of up to 24 participants (if at least one of the first 9 participants had a PR) to achieve at least 17 participants evaluable for response. The initial participant enrolled on 9/4/2014, and the last participant was enrolled on 2/24/2016. All participants in this stratum have been off treatment since 1/29/2018; however, on the basis of the promising results of these particpants, the study was amended to include a pediatric stratum (enrollment restricted to those age 3–15 years), which enrolled its first participant on 11/21/18 and is in the last stages of enrollment. Descriptive statistics including means, percentage, etc. were used to summarize the pain and QOL data. The primary analyses were done comparing proportions to the null hypothesized proportion response. The changes were assessed in the total group (n=19) as well as between the volumetric responders (n=8) and non-responders (n=11). We analyzed changes over time in the tumor size response and other parameters using a linear mixed model approach with an unstructured correlation matrix and compared differences using Least Square Means (LSM) from SAS PROC MIXED (version 9.4). This method accommodates missing values, although it assumes that data are missing at random, which was technically violated because the protocol required removal from treatment if sufficient responses were not achieved. Nevertheless, the technique was used to summarize the responses over time.

For biomarker analysis of circulating proangiogenic CHSPCs and cytokines, participants were stratified by clinical response (PR vs SD). Due to the limited sample size and missing data across multiple timepoints, cell and cytokine levels measured at C0 and C1D15 were merged as an early time point, and C2 and C4 as a later time point. If two values were observed at either C0/C1D15 or C2/C4, the mean of the two values was utilized to represent the measure at the early (C0/C1D15) or late time (C2/C4) point. If only one data point was measured at either C0/C1D15 or C2/C4, that value was directly used in subsequent computational analysis. Applying these criteria, CHSPC features from 6 SD patients and 7 PR patients, and cytokine features from 6 SD patients and 8 PR patients were used in the analysis (see Supplementary Materials: CHSPC-and- Cytokine-Data.xlsx). To identify associations between CHSPC and cytokines levels with long term clinical response, changes in CHSPC and cytokine levels between in early (C0/C1D15) and late (C2/C4) cycles were computed and evaluated by the Mann Whitney test in participants with SD vs PR.

Preclinical Study

Experimental animals

Genetically engineered Nf1^flox/flox;PostnCre mice on a mixed C57BL/6 × 129/SV background were utilized for in vivo studies, with genotypes confirmed by polymerase chain reaction (PCR) as previously described²³. Animal care and experiments were conducted according to the guidelines established by the Indiana University Animal Care and Use Committee (IACUC), protocol # 11405. The experimental mice were housed in the Laboratory Animal Research Facility (LARC) at Indiana University at an ambient temperature of 22 ± 2 °C, relative humidity of 30 to 70%, and a 12 hour light-dark cycle from 7:00 AM EST.

Drug treatment in experimental mice

Cohorts of 4–5 month old, male and female (age and sex matched) Nf1^flox/flox;PostnCre mice were treated with either 15mg/kg cabozantinib daily or the vehicle, hydroxypropylmethylcellulose (HPMC) as the control administered by oral gavage for a duration of 12 weeks. The number of mice per group was determined by a power analysis conducted at the outset of the study as per the preclinical statistical analysis section. Mice were then euthanized to evaluate the extent of tumor burden as outlined below. Investigators were blinded to the treatment groups during drug treatment and response evaluation.

Pharmacokinetic analysis of cabozantinib

1. Plasma Sample Analysis

Cabozantinib was quantified from plasma using temazepam as the internal standard and High Performance Liquid Chromatography-tandem mass spectrometry (HPLC-MS/MS) (5500 QTRAP^® AB Sciex, Framingham, MA). In brief, cabozantinib and temazepam were separated by a gradient mobile phase (acetonitrile: 0.1% formic acid) with a Restek C8 150X4.6 mm 5 μm column. The mass spectrometer utilized an electrospray ionization probe run in positive mode. The multiple reaction monitoring (MRM) Q1/Q3 (m/z) transitions for cabozantinib and temazepam were 502.3/323.3 and 301.1/254.9 respectively. For the plasma samples, 20μL were transferred to a polypropylene tube, temazepam was added as the internal standard (20μL of 0.01ng/μL), and the extraction was performed by the addition of 0.1M citric acid buffer (pH=3.0) followed by the addition of hexane:ethyl acetate; 50:50; v/v. The samples were then vortexed, centrifuged, the organic layer was transferred to a clean polypropylene tube, and evaporated to dryness. The samples were then reconstituted with mobile phase (50μL) and an aliquot (10μL) injected to the HPLC-MS/MS.

2. Tissue Sample Analysis

Cabozantinib was quantified from tissue samples using a slightly modified method from the plasma sample analysis. Briefly, the tissue was weighed then transferred to a polypropylene tube. Phosphate buffered saline was added to the tissue to bring the total volume to 0.5mL (assumption 1g=1mL). The tissue was homogenized using a TissueRuptor^® with a single use disposable probe. An aliquot (0.4mL) was transferred to a clean polypropylene tube and temazepam was added as the internal standard, (20μL of 0.1ng/μL). The extraction procedure and HPLC-MS/MS conditions were the same as for the plasma samples.

Non-compartmental analysis of the data was performed using pharmacokinetic (PK) Solver add-ins in Excel®. Pharmacokinetic parameters obtained included: C_max (the maximal plasma concentration), t_max (the time of maximal plasma concentration), AUC_0−∞ (area under the plasma concentration time curve from zero to infinity), k_el (the elimination rate constant), and t_1/2 (half-life, t_1/2=0.693/k_el). The AUC_0−∞ was calculated from the AUC_0-t (time zero to the last quantifiable concentration C_last) and the AUC from C_last to infinity (C_last/k_el). The systemic clearance (Cl/F, where F is the bioavailability) was calculated from the dose and AUC_0−∞.The apparent volume of distribution at steady state (Vdss/F) was calculated by the Cl and k_el. Parallel studies were conducted using tumor specimens.

Nerve tree microdissection and measurement of tumor volume

Immediately postmortem, fresh blood and select tissues were harvested and mice were perfused and fixed in 10% neutral buffered formalin. The bodies were decalcified in 5% formic acid. The proximal spinal nerve trees, brachial, and trigeminal nerves were dissected microscopically. The volume of proximal peripheral nerves was determined using calipers to measure the length and width of dissected tumors (or equivalent region in absence of tumor) in maximal dimension. Volume was then approximated using the formula for the volume of a spheroid = 0.52 × (width)² × length.

Histopathology

Paraffin sections were stained with hematoxylin and eosin (H&E) and Masson’s trichrome to examine tumor histomorphology. Infiltrating mast cells were enumerated on toluidine blue stained sections by three independent investigators who were blinded to the treatment group. Ten randomly selected 40x high power fields were scored on each slide. Five micron-thick paraffin sections were deparaffinized, hydrated and transferred to 0.1 M citrate buffer (pH 6.0) for antigen retrieval in a pressure cooker. Sections were incubated in 3% hydrogen peroxide for 10 minutes, rinsed, and blocked in 5% normal goat serum in TBST (TBS buffer + 0.1% tween 20). Sections were incubated overnight at 4°C in primary antibody diluted in blocking buffer; CD31 (ab28364, 1:50, Abcam) and AXL (ab227871, 1:500, Abcam). Sections were then incubated in a biotinylated secondary antibody for 1 hr at room temperature; goat anti-rabbit (B8895, 1:800, Sigma). Vectastain Elite ABC reagent was applied for 30 min at room temperature. Color development with Vectastain DAB was observed under the microscope and the reaction was terminated by rinsing in distilled water. Sections were counterstained with modified Mayer’s Hematoxylin (Vector), dehydrated, cleared and coverslipped. Slide images were acquired either on an Aperio ScanScope CS or a QImaging Retiga 2000R camera on a Nikon Eclipse 80i microscope. CD31 immunohistochemical staining was performed to evaluate neoangiogenesis in cabozantinib and vehicle treated animals. CD31+ staining blood vessels were counted manually on 10 randomly selected high power fields per slide. Investigators were again blinded to the treatment groups. AXL immunohistochemistry was analyzed using the IHC Profiler plugin for ImageJ (version 1.47), and the percentage of high positive pixels was analyzed by ANOVA using Graphpad Prism.

TUNEL Staining

The FragEL™ DNA Fragmentation Detection Kit (Sigma Aldrich #QIA21) was used according to manufacturer’s instructions as follows. Tissue slides and the kit-provided positive control slide were deparaffinized with xylenes and rehydrated through graded alcohols. Tissue was encircled with an ImmEdge hydrophobic marker, covered with distilled water, and placed into a humidified slide tray. Proteinase K was applied at room temperature for 20 minutes to permeablize the tissue. Slides were rinsed in tris-buffered saline (TBS) between each subsequent step unless otherwise noted. Three percent hydrogen peroxide was applied to the slides for 5 minutes at room temperature to quench endogenous peroxidase. Slides were incubated at room temperature for 20 minutes with Klenow Equilibration Buffer. Without rinsing, excess buffer was suctioned from the slide and fresh Klenow Labeling Reaction mixture was made and applied to each tissue section. Parafilm® coverslips were applied to each tissue section to prevent evaporation of the reaction mixture during the incubation. The humidified slide tray was incubated at 37°C for 1.5 hours. The slide chamber was then brought back to room temperature. Coverslips were removed and Stop Solution was added to terminate the reaction. Slides were then incubated in Blocking Buffer for 10 minutes at room temperature. Without rinsing, the Blocking Buffer was suctioned from the slide and the Conjugate solution was applied to the tissue. Slides were incubated at room temperature for 30 minutes. DAB solution was applied to the slides for 10 minutes, and the positive control slide was checked to ensure positive staining was achieved. Slides were incubated in methyl green solution for 5 minutes then quickly dipped in 2 changes of ethanol and 3 changes of xylenes, then coverslipped using glass coverslips and Cytoseal XYL mounting medium. Whole slide images were generated with an Aperio CS2 Scanscope and representative image were generated using HALO Image Analysis Software (version 2.3, Indica Labs).

Preclinical Kinome Analysis

Multiplexed inhibitor bead (MIB) chromatography and mass spectrometry (MS)

MIB and MS was performed on snap-frozen sciatic nerve tissue from tumor bearing mice treated with vehicle or 1 day, 3 days, or 7 days of cabozantinib. Tissue was crushed by mortar and pestle in ice-cold MIB lysis buffer (50mM HEPES, 150mM NaCl, 0.5% Triton X-100, 1mM EDTA, 1mM EGTA, pH 7.5) supplemented with cOmplete protease inhibitor cocktail (Roche) and 1% phosphatase inhibitor cocktails 2 and 3 Sigma). Extracts were sonicated 3 × 10s, clarified by centrifugation, and syringe-filtered (0.22μm) prior to Bradford assay quantitation of concentration. Equal amounts of total protein were gravity-flowed over multiplexed inhibitor bead (MIB) columns in high salt MIB lysis buffer (1M NaCl). The MIB columns consisted of 125μL mixture of five Type I kinase inhibitors: VI-16832, PP58, Purvalanol B, UNC-21474, and BKM-120 were custom-synthesized with hydrocarbon linkers and covalently linked to ECH-Sepharose and Purvalanol B was linked to EAH-Sepharose beads as previously described³². For in vitro MIB competition assays to demonstrate quantitatively the direct binding of cabozantinib to kinases within the tumor, lysates were incubated with beads (125 μL equal amounts of inhibitors above plus CTx0294885) at 4°C with DMSO or 10, 100, or 1000nM cabozantinib for 1 h prior to flowing them over the MIB columns. Columns were washed with 5mL of high salt (1M NaCl), 5mL of low salt (150mM NaCl) MIB lysis buffer, and 0.5mL low-salt lysis buffer with 0.1%SDS. Bound protein was eluted twice with 0.5% SDS, 1% beta-mercaptoethanol, 100mM Tris-HCl, pH6.8 for 15 min at 100°C. Eluate was treated with DTT (5mM) for 25 min at 60C and 20mM iodoacetamide for 30 min in the dark. Following spin concentration using Amicon Ultra-4 (10k cut-off) to ~100 μL, samples were precipitated by methanol/chloroform, dried in a speed-vac and resuspended in 50mM HEPES (pH8.0). Tryptic digests were performed overnight at 37°C, extracted four times with 1mL ethyl acetate to remove detergent, dried in a speed-vac, and peptides further cleaned using C-18 spin columns according to manufacturer’s protocol (Pierce).

Liquid Chromatography and Mass Spectrometry

Peptides were resuspended in 5% ACN and 0.1% formic acid. Approximately 40% of the final peptide suspension was injected onto a Thermo Easy-Spray 75μm × 25cm C-18 column and separated on a 180min gradient (5–40% ACN) using an Easy nLC-1000. The Thermo Q Exactive mass spectrometry ESI parameters were as follows: 3e6 AGC MS1, 80ms MS1 max inject time, 1e5 AGC MS2, 100ms MS2 max inject time, 20 loop count, 1.8 m/z isolation window, 45s dynamic exclusion. Raw files were processed for label-free quantification by MaxQuant (version 1.6.10.43) LFQ using the Uniprot/Swiss-Prot mouse database, fixed carbidomethyl (C) and variable phospho (STY), oxidation (M), and acetyl (Protein N-term) modifications. LFQ intensities for kinases with at least two peptides were imported into Perseus (version 1.6.10.50).

Western blot

Isolated proteins were fractionated using NuPAGE 4–12% Bis-Tris Gels (Invitrogen Cat#NP0322BOX) and electro-transferred to PVDF membranes. Immunoblots were carried out using antibodies specific to pAXL (Cat#5724, 1:1000, Cell Signaling Technology), AXL (Cat#ab215205, 1:1000, Abcam), pAKT (Cat #9271, 1:1000, Cell Signaling Technology), pGSK-3β (Cat#9323, 1:1000, Cell Signaling Technology), pMEK1/2 (Cat#9121, 1:1000, Cell Signaling Technology), and GAPDH (#CST-5174, 1:1000, Cell Signaling Technology). After incubation with appropriate HRP conjugated secondary antibodies (Anti-rabbit (#NA934V, 1:5000, GE Healthcare), Anti-Rat(#AP-136P, 1:5000, EMD), signals were detected using ECL chemoluminescence substrate (ECL Prime, GE Healthcare).

Schwann cell culture

Mouse embryos were harvested at embryonic day 13.5 and dorsa root ganglia (DRG) were isolated under a dissecting microscope. The DRGs were digested in 0.05% trypsin-EDTA and dissociate with syringes. The resulting suspensions were plated on PDL/laminin coated plates in Schwann Cell Media I (SCM-I) consisting of Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 50 U/mL penicillin, 50 μg/mL streptomycin, 2 mM L-glutamine, 1X N2 supplement, and 250 ng/mL nerve growth factor (NGF). The following day, the medium was changed to Schwann Cell Media II (SCM-II), which was identical to SCM-I except for the substitution of 10 ng/mL neuregulin in exchange for NGF and the addition of 2 μM forskolin to suppress fibroblast growth. Cells were used for experiments after approximately 10–14 days of culture.

Apoptosis assays

Human immortalized NF1^−/− Schwann cells (iPNF95.6) were plated in triplicate at a density of 5,000 cells/well in 96 well plates, allowed to adhere overnight, and then treated with either DMSO or increasing concentrations of cabozatninib from 39 nM to 5 μM. Navitoclax 1 μM was also used as a positive control. Caspase 3/7 Glo activity (#G8093, Promega) was measured by luminescence after 24 hours of treatment according to the manufacturer’s instructions using SynergyH4 microplate reader.

Murine embryonic fibroblast culture

Mouse embryos were harvested at 18.5 days post-coitus and dissected under a low-power microscope. After removal of the heads and internal organs, embryos were minced and trypsinized for 20 min. The resultant suspension was seeded onto culture dishes in Iscove’s Modified Dulbecco Medium (IMDM) supplemented with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin, and 1% L-glutamine. Upon reaching confluence, cells were split at 1:2 ratios until a morphologically homogeneous culture was obtained. The cells were then frozen or expanded for further studies. Cells at passage 2–3 were used for experiments.

Endothelial Colony Forming Cell Assays

Umbilical cord blood derived endothelial colony forming cells (ECFCs) were obtained from the Angio BioCore (Indiana University) and passaged once before plating for a proliferation assay via cell counting. ECFCs were plated on rat tail collagen and were allowed to set down overnight before starting 200ng of GAS6 and/or 100nM cabozantinib treatments. ECFCs were trypsinized and counted manually using a glass hemocytometer after 48 hours with an n=3 independent experiments performed.

Mouse sAXL ELISA

Levels of circulating sAXL were measured in mouse plasma using the Mouse AXL DuoSet ELISA kit (R&D Systems, DY854) and accompanying DuoSet ELISA Ancillary Reagent Kit 2 (R&D Systems, DY008) according to the manufacturer’s instructions. Mouse plasma samples were assayed in duplicate at dilution of 1:200 in Reagent Diluent. Optical density of each well was measured using a VERSAmax Tunable Microplate Reader (Molecular Devices) with SOFmax PRO software. Optical density readings at 540 nm were subtracted from the readings at 450 nm for wavelength correction.

Preclinical Statistical Analysis

For the in vivo therapeutics studies in Nf1^flox/flox;PostnCre mice treated with either vehicle vs cabozantinb, a preliminary power analysis was conducted to determine the appropriate sample size, assuming a 5% type I error rate. Our preliminary data suggested a coefficient of variation of 48% for tumor number. Thus, we determined that 16 animals per group would allow detection of a 57% reduction in tumor number with an 81.3% power. In addition, our preliminary data indicated that the coefficient of variation for proximal nerve root size was less than 38%. Thus, 16 animals per group would allow us to detect a 40% difference in tumor size with greater than 82.1% power.

Statistical analyses were performed with GraphPad Prism 8.0 software (GraphPad, La Jolla, CA). Analysis of variance (ANOVA) and Student’s t-tests with post-hoc correction for multiple comparisons were used to evaluate for statistically significant differences between samples as described in detail within the figure legends.

For MIB/MS analysis from in vivo drug treatments, log₂ transformed LFQ intensities were filtered to include only kinases with at least three valid values in at least one treatment group and missing values were imputed from the total matrix in Perseus. Student’s t-tests were performed in Perseus with Benjamini-Hochberg correction for multiple hypothesis testing. For in vitro MIB competition assays done in duplicate, missing log₂ LFQ intensities were imputed for each column separately.

DATA AVAILABILITY

Raw MS files pertaining to Figure 2 and Extended Data Figures 4 and 5 are available in the PRIDE database (https://www.ebi.ac.uk/pride/archive/projects/PXD019138). Source data for all Figures and Extended Data Figure are provided as individual Excel files. Full-length, unprocessed gels and blots are also provided as individual PDF files for each figure contianing all supporting blots and gels with the linked figures noted directly in the file. Log₂ transformed LFQ MIB binding values are provided in Supplementary Table 1, Source Data Figure 2, Source Data Extended Data Figure 4, Source Data Extended Data Figure 5). Raw data for circulating progenitor cell and cytokine analysis associated with Extended Data Figure 10 and Supplementary Tables 10 and 11 are also provided in a Supplementary Excel file (CHSPC-and-Cytokine-Data.xlsx).

Extended Data

Extended Data Fig. 1. Pharmacokinetics of cabozantinib in Nf1^flox/flox;PostnCre mice.

(a) Concentration-time profiles of cabozantinib in plasma and nerve tissue samples from n=3 mice at each time. Plasma samples were measured at 1, 2, 4, 8, and 24 hours, while tissue samples were measured at 4 and 24 hours after a single 15mg/kg oral gavage dose of cabozantinib. The error bars indicate standard deviations. (b) Pharmacokinetic outcomes of cabozantinib in n=3 mice administered a single 15mg/kg oral gavage dose. All values are presented as the arithmetic mean. (c) Chromatography of cabozantinib (top) at the lowest limit of quantification, 3ng/mL, and the internal standard, temazepam (bottom). (d) Chromatography of cabozantinib (top) and the internal standard, temazepam (bottom). The filled peaks are the analyte of interest.

Extended Data Fig. 2. Representative tumors from vehicle and cabozantinib treated Nf1^flox/flox;PostnCre mice.

H&E stained photomicrographs of plexiform neurofibroma tumors in the vehicle (a) and cabozantinib (b) treated cohorts. Magnification are as denoted by the scale bars. The experiment was repeated three times independently with similar results.

Extended Data Fig. 3. Representative plexiform neurofibroma and nerve tissues within the brachial plexus of Nf1^flox/flox;PostnCre mice before after treatment with vehicle vs cabozantinib.

(a) H&E stained photomicrographs of plexiform neurofibromas and nerve tissues within the brachial plexus of mice of 4 month old Nf1^flox/flox;PostnCre mice prior to treatment (top) and after 12 weeks treatment (7 month old) with either vehicle (middle) or cabozantinib (bottom). (b) The number of tumors in the brachial plexus was quantified as shown in the bar blot. Bars represent the standard error of the mean. The number of independent mice evaluated in each group were as follows Nf1^flox/flox;PostnCre prior to treatment (4 months, n=5), vehicle treated (7 months, n =13), cabozantinib treated (7 mo, n=16). *Adjusted P-value = 0.0254, vehicle vs cabozantinib at 7 months (one-way ANOVA with Tukey’s multiple comparisons test).

Extended Data Fig. 4. Kinome analysis of murine plexiform neurofibromas after treatment with 1 day and 3 days of cabozantinib.

Volcano plot showing log2- fold change MIB binding (LFQ intensity) after (a) 1 day (n=3) and (b) 3 days (n=5) of cabozantinib vs vehicle (n=4) control treated sciatic nerve tumor tissue plotted against the -log₁₀ Benjamini-Hochberg adjusted P value of unpaired, two-tailed Student’s t-tests. The dotted line denotes FDR = 0.05. Increased MIB binding was observed of a number of kinases known to modulate an array of cellular processes including proliferation and survival (RPS6KB1), migration and adhesion (PTK2), interferon signaling (JAK1 and TYK2), NF-kappa B signaling (IKBKB), cell cycle and DNA damage response (CDK5 and NEK1), and longterm potentiation and neurotransmitter release (CAMK2A). Kinases with decreased MIB binding after 3 days of cabozantinib treatment including those modulating B lymphocyte development, differentiation and signaling (BTK), cAMP-dependent signaling (PRKACA and PRKACB), and regulation of immune cell chemotaxis and mast cell degranulation (FGR). None of these kinases were cognate targets of cabozantinib and their 4 suppression was transient, likely reflective of an early cellular stress response invoked by cabozantinib.

Extended Data Fig. 5. Top 20 decreased kinases in MIB competition proteomics assays with cabozantinib.

Stacked bar plot for the top 20 decreased kinases representing the summed log₂ fold change in MIB binding for two biological replicates incubated in vitro at varying concentrations of cabozantinib vs DMSO.

Extended Data Fig. 6. Evaluation of AXL as a cabozantinib target in plexiform neurofibromas of Nf1 mutant mice.

(a) MIB binding (LFQ intensity) for AXL in tumor bearing sciatic nerve tissue of Nf1 mutant mice as compared to the WT control. P-value (unpaired, two-tailed Students t-test) <0.0001 as shown. Nerve tissues from n= 3 mice (WT) and n=4 mice (Nf1^flox/flox;PostnCre) were analyzed in one experiment. Data are presented as mean ± SEM. (b) AXL was detected by western blot in sciatic nerve tissue lysates from plexiform neurofibroma bearing Nf1^flox/flox;PostnCre mice and WT control. GAPDH is shown as the loading control. AXL levels normalized to GAPDH were significantly increased in tumor bearing tissue relative to the WT control, P=0.05 by two-tailed, Mann-Whitney test. Nerve tissue lysates from n=3 mice (WT) and n=4 mice (Nf1^flox/flox;PostnCre) were analyzed in two independent experiments. (c) AXL expression was detected by immunohistochemistry in WT (PostnCre negative) and Nf1^flox/flox;PostnCre tumor bearing tissues following 12 weeks of treatment with either vehicle or cabozantinib. The negative control is shown at the inset. Original magnification x20. The experiment was conducted two times independently with similar results. (d) The percentage of AXL high positive pixels normalized to the tissue area was quantified using ImageJ software and the IHC Profiler plugin. The number of independent mice evaluated per treatment condition were as follows, WT (n=3), vehicle Nf1^flox/flox;PostnCre (n=6), and cabozantinib Nf1^flox/flox;PostnCre (n=6). Three independent tissue regions were scored from each mouse. The experiment was conducted once. ***Adjusted P-value = 0.0003 vehicle vs cabozantinib, Nf1^flox/flox;PostnCre. **Adjusted P-value = 0.0059 WT vs vehicle Nf1^flox/flox;PostnCre (one-way ANOVA with Tukey’s multiple comparisons test). Data are presented as mean ± SEM. (e) AXL, GSK3β, pAKT, and pMEK1/2 and GAPDH were detected by western blot in tumor bearing sciatic nerve tissues from Nf1^flox/flox;PostnCre mice treated with either cabozantinib or the vehicle control. The experiment was conducted two times independently with similar results.

Extended Data Fig. 7. Cabozantinib modulates multiple cellular constituents of plexiform neurofibroma by abrogating AXL signaling.

(a) Primary Nf1−/− Schwann cells were stimulated with GAS6 (200 ng/mL) in the presence or absence of cabozantinib (100 nM). Proliferation was assessed by manual cell counting after 48 hours (n=3 replicates per condition). **Adjusted P-value = 0.0091 unstimulated vs GAS6, **Adjusted P-value =0.0053 GAS6 vs GAS6 + cabozantinib (one-way ANOVA with Sidak’s multiple comparisons test). Data are presented as mean ± SEM. (b) pAXL was detected by western blot following stimulation with GAS6 (250 ng/mL) in the presence or absence of cabozantinib (2500 nM). GAPDH is shown as a loading control. The experiment was conducted two times independently with similar results. (c) Murine embryonic fibroblasts were stimulated with GAS6 (200 ng/mL) in the presence or absence of cabozantinib (2000 nM). Proliferation was assessed by manual cell counting after 48 hours (n=6 replicates per condition). **Adjusted P-value = 0.0032 unstimulated vs cabozantinib only, ****Adjusted P-value <0.0001 unstimulated vs GAS6 only, ****Adjusted P-value <0.0001 GAS6 vs GAS6 + cabozantinib (one-way ANOVA with Sidak’s multiple comparisons test). Data are presented as mean ± SEM. (d) pAXL was detected by western blot in primary murine embryonic fibroblasts following stimulation with GAS6 (250 ng/mL) from 0 to 30 minutes in the presence or absence of cabozantinib (2000 nM). GAPDH is shown as a loading control. The experiment was conducted two times independently with similar results. (e) Human umbilical cord blood derived endothelial colony forming cells (ECFCs) were plated and stimulated with GAS6 (200 ng/mL) in the presence or absence of cabozantinib (100 nM). Proliferation was assessed by manual cell counting after 48 hours (n=6 replicates per condition). ***Adjusted P-value = 0.0003 unstimulated vs GAS6, *Adjusted P-value = 0.0321 GAS6 vs GAS6 + cabozantinb, ns = not statistically significant, adjusted P-value = 0.0714 (one-way ANOVA with Sidak’s multiple comparisons test). Data are presented as mean ± SEM. pAXL was detected by western blot in ECFCs following stimulation with GAS6 (200 ng/mL) from 0 to 30 minutes in the presences or absence of cabozantinib (1000 nM). GAPDH is shown as a loading control.

Extended Data Fig. 8. Cabozantinib does not induce apoptosis in Schwann cells or plexiform neurofibroma tumor tissues in vivo.

(a) Cabozantinib does not induce apoptosis by caspase 3/7 glo assay in human NF1^−/− SC across of range of doses from 39 nM to 5 μM. By contrast, a robust increase in caspase 3/7 activity was observed in NF1^−/− Schwann cells treated with Navitoclax (1 μM) as a positive control shown at the right. n=3 independent cell culture wells were analyzed per condition over two independent experiments. ****Adjusted P-value < 0.0001 Navitoclax vs all other conditions (one-way ANOVA with Tukey’s multiple comparisons test). Data are presented as mean ± SD. (b) Representative nerve tissues in cabozantinib (XL184) and vehicle treated mice are negative for TUNEL staining indicating that cabozantinb does not induce apoptosis. Positive and negative controls are shown at the top of the panel. The experiment was conducted two times independently with similar results.

Extended Data Fig. 9. Increases in sAXL following cabozantinib treatment in Nf1^flox/flox;PostnCre mice.

(a) Cartoon schematic depicting release of soluble AXL following binding of Gas6 and proteolytic cleavage. (b) Soluble AXL was measured by ELISA in the plasma of WT (PostnCre negative) (n=8) and Nf1^flox/flox;PostnCre mice following 12 weeks of treatment with either vehicle (n=13) or cabozantinib (n=16). *Adjusted P-value = 0.0167 WT vs cabozantinib Nf1^flox/flox;PostnCre, *Adjusted P-value = 0.0179 vehicle vs cabozantinib Nf1^flox/flox;PostnCre (one-way ANOVA with Tukey’s multiple comparisons test).

Extended Data Fig. 10. sAXL levels and proangiogenic CHSPCs vs clinical response to cabozantinib in trial participants.

(a) A panel of 45 cytokines including soluble AXL (sAXL) were analyzed in the serum of clinical trial participants (n=6 with SD, n=8 with PR). The barplot depicts the change in the mean level of sAXL (ng/mL) between baseline/early (C0/C15) and later cycles (C2/C4) by clinical response of study participants (SD vs PR; P=0.08 by two-tailed, Mann-Whitney test). Grubb’s test with an alpha value of 0.05 did not identify any statistical outliers within the dataset. Whiskers extend from the minima to maxima. The center line represents the median. The box spans the 25^th to 75^th percentiles. (b) The frequency of proangiogenic CHSPCs was analyzed in the peripheral blood in n=13 participants (n=6 with SD, and n=7 with PR) collected at 4 time points: prior to cycle 1/baseline (C0), cycle 1 day 15 (C1D13) end of cycle 2 (C2), and end of cycle 4 (C4). The change in the mean frequency of proangiogenic CHSPCs from early (C0/C1D15) to later (C2/C4) treatment cycles was plotted in participants with SD vs PR (p=0.6282 by two-tailed, Mann Whitney test). Whiskers extend from the minima to maxima. The center line represents the median. The box spans the 25^th to 75^th percentiles. (c) Gating strategy for identification of proangiogenic CHSPCs by flow cytometry.