**1. Introduction**

Targeting the RAS/ERK signaling pathway is an e ffective treatment for numerous cancers with hyper-activation of the RAS pathway. The most striking clinical responses to inhibitors of BRAF, MEK, and EGFR have been observed in melanoma and lung cancers where RAS pathway activation is intrinsic. Even though these targeted therapies have resulted in an extension of the overall survival in these inherently aggressive cancers, the clinical response is often transient and complete remission is rare. Resistance to kinase inhibition is a significant clinical challenge and numerous studies have identified multifactorial and heterogeneous mechanisms of resistance to kinase inhibition [1].

Malignant Peripheral Nerve Sheath Tumors (MPNSTs) are aggressive, highly chemoresistant sarcomas arising from Schwann cells that are a leading cause of death in patients with Neurofibromatosis Type 1 (NF1) [2,3]. Neurofibromatosis Type 1 (NF1) is caused by germline mutations in the *NF1* gene and is the most common single-gene disorder, a ffecting 1 in 3000 live births. The *NF1* gene encodes neurofibromin, a GTPase-activating protein that negatively regulates RAS (including HRAS, NRAS, and KRAS), where the loss of NF1 leads to deregulated RAS signaling. Deregulated RAS signaling caused by the loss of neurofibromin is both permissive and instructive for MPNST progression (3–5). Recent clinical trials have focused on targeting members of the RAS signaling pathway or the PI3K/mTOR pathway. To date, these trials have failed to identify consistent therapeutic vulnerabilities in MPNSTs; however, few studies have examined why these therapies failed. These clinical results highlight our limited knowledge of the mechanisms that drive resistance to kinase inhibition in MPNSTs.

In addition to loss of the *NF1* gene, NF1-relatedMPNSTs exhibit highly complex genomic alterations that result in substantial tumor suppressor gene loss and oncogene copy number variations [4,5]. How MPNST genomic alterations a ffect therapy resistance is currently unclear. Recently, we performed a genomic analysis of longitudinally collected MPNST samples. This study revealed the early concomitant presence of *MET*, *HGF*, and *EGFR* amplifications, as well as the site-specific expansion of these loci over time and treatment. These data point to an adaptive mechanism involving RTK signaling for both malignant transformation and clonal selection in MPNSTs [6]. To advance our understanding of the MPNST therapeutic response and resistance to RAS pathway inhibition, we developed diverse preclinical NF1-related MPNST models, including an "MET-addicted" model of NF1-related MPNSTs (NF1-MET), an *Nf1*/*Trp53*-deficient model (NF1-P53), and an NF1 model (P53WT, *Hgf*-amplified) [7–9]. Using these MPNST models, we determined that P53 deficiency significantly exacerbates resistance to MEK inhibition; however, combined MEK and MET inhibition overcame therapy resistance [6]. Importantly, these results demonstrated that NF1-related MPNSTs maintain multiple signaling dependencies beyond RAS, and that genomic determinants, such as P53 and RTK genomic alterations, profoundly influence the therapy response.

Kinome reprogramming is a powerful barrier to a durable treatment response to kinase inhibition [10–12]. These signaling adaptations occur as a result of the compensatory activation of evolutionarily conserved signaling pathways that drive growth and proliferation, especially when central pathways such as RAS/MEK and PI3K/mTOR are blocked by drugs [13]. Kinome adaptation leads to diverse mechanisms of therapy resistance that can be classified as three categories, all of which can occur simultaneously during treatment [1]. The most common resistance mechanism is defined as "pathway reactivation", which can occur through multiple mechanisms that reinforce oncogenic signaling in the face of strong target inhibition. The second common mechanism of resistance is "pathway bypass", which occurs when oncogenic pathways are activated at a downstream convergence point by a parallel pathway, despite e ffective upstream inhibition (e.g., PI3K/AKT/ERK activation during MEK inhibition). The third resistance mechanism is "pathway indi fference", where the cancer cells transition to an alternative survival state that is independent of the targeted oncogenic pathway. Each of these resistance mechanisms have been observed in response to kinase inhibition in cancers with RAS-activation dependencies. E fforts to delineate patterns of therapy resistance have been valuable in understanding the treatment response and the identification of targets for salvage therapy. Even though genetic mechanisms of resistance (e.g., somatic mutations or gene amplifications) have

been a major focus of resistance research, the impact of phosphoproteomic changes in therapeutic resistance has been increasingly acknowledged [1].

The mechanisms that regulate adaptive kinome reprogramming in NF1-deficient cancers are not well-elucidated. In this study, we aimed to define both mechanisms and categories of resistance to standard chemotherapy and targeted kinase inhibition in NF1-related MPNSTs in order to identify 1) novel therapeutic strategies that correlate with the genomic status or 2) salvage therapies that are focused on the emerging 'resistant' tumor populations. Kinome reprogramming can be heavily influenced by genomic alterations (i.e., amplification and mutation) of kinases, the overexpression of other kinases, or ligand activation [13–16]. Our genomic analysis of MPNST progression identified genomic events during human MPNST progression that heighten RTK signaling, AKT/mTOR activation, and cell survival [6]. To understand how these genomic contexts influence the therapy response to kinase inhibition, we conducted a phosphoproteomic analysis using a reverse phase protein array (RPPA) in our established MPNST mouse models. RPPA is a powerful research tool that simultaneously interrogates a broad range of phosphosites across the entire kinome, including activating, inactivating, and alternative sites. Using this comprehensive analytical technique, we confirmed substantial signaling heterogeneity following kinase inhibition in NF1-related MPNSTs leading to therapy resistance, including examples of pathway reactivation, bypass, and indi fference. Our data verifies that a broad range of signaling pathways are activated as a result of e ffective MET, MEK, and MET/MEK blockade, most notably, AXL, NFκB, RAS/RAF/MEK, and AKT/mTOR pathways. Interestingly, the patterns of kinome adaptation were distinct based on the kinase target and the genomic context of the tumor. We also demonstrated that administration of the DNA damaging reagen<sup>t</sup> doxorubicin resulted in a distinct pattern of kinome adaptation that was partially, but not fully, mitigated by MET and MEK inhibitors. Categorizing these patterns of therapy resistance is valuable for inferring patient stratification and the identification of targets for salvage therapy.

#### **2. Materials and Methods**

#### *2.1. Murine MPNST Tumorgrafts and Treatment*

Immediately following the euthanasia of tumor-bearing mice, 15–25 mg portions of each tumor were transplanted into the flank of NSG-SCID mice using a 10 gauge trochar. Tumors were measured twice weekly and euthanized when the tumor size exceeded 2500 mm3. When the tumor volume reached approximately 150 mm3, mice were randomized into treatment groups, treated, and euthanized as independent groups at 4-h, 2-day, or 21-day time points (or until mice reached the euthanasia criteria). Respective doses across all treatment combinations were capmatinib (30 mg/kg twice daily vial oral gavage), trametinib (1 mg/kg daily via oral gavage), and doxorubicin (1 mg/kg once via subcutaneous injection). For the 4-h time point, all animals were euthanized 4 h after a single dose. For the 2-day time point, capmatinib-treated animals received three total doses and were euthanized 4 h after the final treatment, whereas trametinib-treated animals received two total doses before euthanasia. Tumors were immediately harvested and either snap-frozen or formalin-fixed for further analysis. Three representative tumors were assessed by RPPA for each time point, treatment, and genotype group. Capmatinib and trametinib were obtained from Novartis. Doxorubicin was obtained from LC Laboratories. All animal experimentation in this study was approved by the Van Andel Institute's Internal Animal Care and Use Committee (XPA-19-04-001).

#### *2.2. Sample Collection and Preparation for RPPA Downstream Analysis*

Samples were frozen in liquid nitrogen within 20 min upon surgical resection to preserve the integrity of the phosphoproteome. Specimens were then embedded in an optimal cutting temperature compound (Sakura Finetek, Torrance, CA, USA), cut into 8 μm cryo-sections, mounted on uncharged glass slides, and stored at −80 ◦C until microdissected. Each slide was fixed in 70% ethanol (Sigma Aldrich, Darmstadt, Germany), washed in deionized water, stained with hematoxylin (Sigma Aldrich,

Darmstadt, Germany) and blued in Scott's Tap Water substitute (Electron Microscopy Sciences), and dehydrated through an ethanol gradient (70%, 95%, and 100%) and xylene (Sigma Aldrich, Darmstadt, Germany). In order to prevent protein degradation, complete protease inhibitor cocktail tablets (Roche Applied Science, Basel, Switzerland) were added to the ethanol, water, hematoxylin, and Scott's Tap Water substitute [17]. For each sample, an average of 15,000 tumor cells was isolated from the surrounding microenvironment using a Pixcell II LCM system (Arcturus, Mountain View, CA, USA). Microdissected cells were lysed in a 1:1 solution of 2× Tris-Glycine SDS Sample bu ffer (Invitrogen Life Technologies, Carlsbad, CA, USA) and Tissue Protein Extraction Reagent (Pierce, Waltham, MA, USA) supplemented with 2.5% of 2-mercaptoethanol (Sigma Aldrich, Darmstadt, Germany). Cell lysates were boiled for 8 min and stored at −80 ◦C.

#### *2.3. Reverse Phase Protein Microarray Construction and Immunostaining*

Using an Aushon 2470 arrayer (Aushon BioSystems, Billerica, MA, USA) equipped with 185 μm pins, samples and standard curves for internal quality assurance were printed in triplicate onto Oncyte Avid nitrocellulose-coated slides (Grace Bio-labs, Bend, OR, USA), as previously described [17]. A Sypro Ruby Protein Blot Stain (Molecular Probes, Eugene, OR, USA) protocol was used to stain selected arrays to quantify the total amount of protein within each sample.

Before immunostaining, each array was first incubated with Reblot Antibody stripping solution (Chemicon) for 15 min at room temperature, followed by two washes in PBS. To minimize potential nonspecific bindings, arrays were then incubated in I-block solution (Invitrogen Life Technologies, Carlsbad, CA, USA) for 1 h. Each array was tested with a single primary antibody using an automated system (Dako, Santa Clara, CA, USA). The antibody specificity was tested by immunoblotting using a wide panel of cell lysates, as previously described [17,18]. Negative control arrays were incubated with the anti-rabbit secondary antibody only to account for unspecific binding and background noise. A commercially available catalyzed signal amplification system (Dako, Santa Clara, CA, USA) coupled with a biotinylated anti-rabbit secondary antibody (Vector Laboratories) and a streptavidin-conjugated IRDye680 (LI-COR Biosciences, Lincoln, NE, USA) were used for the amplification and detection of the fluorescent signal. Arrays were probed with a total of 99 antibodies targeting protein kinases involved in major cellular functions, and the results of the broad screening were previously published.

Antibody and Sypro Ruby-stained arrays were scanned using a laser-based PowerScanner (TECAN, Mönnedorf, Switzerland). Acquired images were analysed using the MicroVigene software version 5.1 (Vigene Tech, Carlisle, MA, USA). This commercially available software performs spot finding, averages the triplicates, subtracts the background from the negative control slide(s), and normalizes each sample to the corresponding amount of total protein measured by Sypro Ruby staining. Intraand inter-assay reproducibility of the RPPA platform has been previously reported [19,20].

## *2.4. Statistical Methods*

*Tumor growth analysis*: Linear mixed-e ffects models, with random slopes and intercepts, and false discovery rate-adjusted contrasts, were used to estimate and compare tumor growth rates for the di fferent mono and combo therapies. For visualization of the changes in tumor growth, the tumor volume was imputed using the last observation carried forward, until the animal was euthanized. Curves terminated once >50% of mice had been euthanized in the respective treatment group. All analyses were conducted using R v3.2.2 (https://cran.r-project.org/), with an assumed level of significance of α = 0.05.

*Proteomic analysis*: Fold change in expression for each phosphosite was calculated by log2 transformation of the treatment relative to the vehicle mean for that genotype group. Fold change was ranked-ordered by the median of the treatment group for each genotype and for each condition, the top and bottom 15 proteins from the ranked list were plotted in balloon plots. A total of 98 protein sites passed quality control metrics and were used for analysis. Fold change in the expression for each protein was calculated as the protein expression relative to the vehicle mean for that genotype-time

group. Proteins were rank-ordered on the *y*-axis by the median transformed fold change for each treatment-genotype-time group and plotted from highest to lowest fold change. Each column represents a single animal. For the 4-h and 2-day time points, animals were plotted randomly on the x-axis. For the 21-day time point, animals were plotted on the *x*-axis based on tumor size (largest to smallest) at sacrifice, corresponding to tumors 4–6, respectively, in the associated tumor-graft plots. The balloon color indicates the log2 fold change in protein expression. Proteins with a greater than 4-fold increase or decrease in expression relative to the vehicle were plotted as log2 fold change >2 or <sup>&</sup>lt;−2, respectively. The balloon size indicates the absolute protein expression normalized to the total protein input and background. Head-tail balloon-maps were created by plotting the 30 proteins with the highest and lowest fold change in expression for each treatment-genotype-time group. Plots were generated using R v3.6.1. For pospho-AXL and phospho-NFkB correlation analysis, normality was first assessed by a Shapiro–Wilk normality test. Correlations were analyzed by two-sided Spearman's rank correlation rho. For correlation plots, data were fit by stat\_smooth using loess with span = 1 using ggplot2 (v3.2.1). Missing data points (due to a failure to meet RPPA quality control standards) were omitted from the analysis. All analyses were done using R (v3.6.2).
