Nf1-Mutant Tumors Undergo Transcriptome and Kinome Remodeling after Inhibition of either mTOR or MEK

Neurofibromatosis I (NF1) syndrome is a tumor predisposition syndrome that is caused by a germline mutation in the NF1 gene. Tumorigenesis in the setting of germline NF1 heterozygosity is characterized by loss of the wild-type NF1 allele, resulting in nullizygosity. This central event in NF1-mediated tumorigenesis leads to hyperactivation of RAS signaling pathways (1).

Downstream of RAS, activated RAF–MEK–ERK and PI3K–AKT–mTOR pathways are pharmacologically inhibited for therapy, the most clinically validated targets being MEK or mTOR. Unfortunately, single target inhibition leads to compensatory feedback and subsequently the development of drug resistance. For example, MEK inhibition leads to an increase in AKT and mTOR activity by decreasing IRS-1 phosphorylation in a variety of tumor types, including lung, pancreas, colon, bladder, and prostate (1, 2), whereas in breast cancer, MEK inhibition is associated with enhanced EGFR/HER2 signaling and activation of AKT (3). Such differences in kinase signaling profile and therapeutic response have been linked to organ or context dependence (4).

Although Neurofibromatosis type 1 is classified as a subtype of RASopathies, a group of clinical syndromes caused by hyperactivation of the RAS/MAPK pathway, the NF1 syndrome is notable for the phenotypic variation observed in affected individuals (5, 6). This clinical variation suggests significant heterogeneity in the mechanism by which NF1 loss promotes multiorgan disease features.

Kinase inhibition is an important therapeutic strategy in the management of neoplasms in NF1. MAPK pathway inhibition suppresses the growth of malignant peripheral nerve sheath tumors (MPNST; refs. 7, 8) as well as the development of hematopoietic malignancy in Nf1-mutant mice (9). The MEK inhibitor (MEKi) selumetinib was recently identified as the first active therapy for plexiform neurofibromas in NF1, and the FDA granted breakthrough therapy designation for this therapy (10). Combined mTORC1 and MEK inhibition is proposed to be superior to inhibition of either of these effectors alone in the treatment of MPNSTs (11). However, questions remain concerning optimal strategies against RAS pathway activation in many NF1-mutant malignancies, with novel therapeutic approaches yielding mixed responses in clinical trials (12). One potential explanation for the apparent mixed response in NF1-mutant tumors may be that NF1 loss, compared with mutant hyperactive RAS, is a weak driver of tumorigenesis. As such, cooperating events that promote tumorigenesis may also form the basis for heterogeneous clinical responses. The signaling framework also likely influences the mechanisms underlying acquired drug resistance, which can evolve from secondary mutations as well as alternate mechanisms of kinase pathway activation.

We sought to determine whether neoplasms linked by a common pathogenetic mechanism (loss of heterozygosity of Nf1, as occurs in NF1 syndrome) demonstrate shared or divergent signaling features with functional and therapeutically significant consequences. In the context of a germline-mediated genetic syndrome such as NF1, in which multiple neoplasms may arise but are driven by a common initial genetic lesion, identifying shared biochemical and genetic alterations may point to therapeutic strategies with broader efficacy. Conversely, elucidating the mechanisms of divergent therapeutic responses should inform and predict patterns of treatment failure and the development of drug resistance.

In prior work, we generated mouse models of Nf1-driven tumorigenesis in which we mutagenized Nf1 heterozygous mice with radiation, recapitulating the susceptibility of patients with NF1 to radiation-induced cancers (13, 14). These mouse models produced multiple diverse Nf1-mutant tumor histologies (sarcomas and carcinomas) and associated cell lines that were universally characterized by Nf1-nullizygosity (14, 15). Here, we derived MEKi- or mTORi-resistant Nf1-mutant tumor cell lines by continuous exposure to the MEKi PD0325901 (PD901) or the mTOR inhibitor RAD001, then compared transcriptome and kinome profiles of these cells to parental cells. These studies identify stably acquired molecular signatures in the setting of drug resistance and suggest potential treatment strategies in the Nf1 malignancy context.

Cell lines were established from tumors arising in irradiated Nf1^+/− F1 mice (16). Early passage stocks were established for all cell lines, and experimental cells were drawn from these stocks approximately every 20 to 30 passages. Mycoplasma testing was done approximately yearly. Inhibitors were purchased from Selleck Chemicals [PD0325901 S1036, everolimus (RAD001) S1120, dactolisib (BEZ235) S1009, PD184352 (CI-1040) S1020, torkinib (PP242) S2218, PIK-75 HCl S1205, tozasertib (VX680) S1048, barasertil (AZD1152) S1147, alisertib (MLN8237) S1133, binimetinib (MEK162) S7007, trametinib (GSK1120212) S2673, rapamycin (Sirolimus) S1039]. Drug-resistant cell lines 989 PD_R, 989 RAD_R, 881 PD_R, and 881 RAD_R were derived from each original parental cell line by continuous exposure to PD0325901 or RAD001. Cell lines 989 and 881 were exposed to increasing concentrations (from 25 to 500 nmol/L) of PD0325901 or RAD001 for 6 months. Resistant cell lines were continuously maintained in medium supplemented with inhibitor. For drug washout experiments, cells were washed twice with PBS, inhibitors were withdrawn for 12 days, and further analyses were performed.

Genomic DNA was extracted from parental and resistant cell lines and amplified by PCR utilizing custom primers (Supplementary Table S2) targeting exons 2 and/or 3 of murine Hras, Nras, and Kras.

After 24 hours of serum starvation, 50%–60% confluent cells were stimulated for 0, 5, 30, and 60 minutes with medium supplied with 10% FBS, harvested, and protein lysates were collected. RAS-GTP levels were assessed using the Ras Pulldown Activation Assay Biochem Kit according to the manufacturer's protocol (Cytoskeleton).

Cell proliferation was assessed by a colorimetric MTS assay according to the manufacturer's instructions (Cell Titer 96 Aqueous One Solution Cell Proliferation Assay, Promega). Cells seeded in 96-well plates were treated with different inhibitors, and cell viability was measured after 72 hours.

Parental and resistant cell lines were grown to a confluence of 70%. Parental cell lines were treated with 500 nmol/L PD0325901 or RAD001 for 24 hours. Washed cell lines were lysed in multiplexed inhibitor beads (MIB) lysis buffer [50 mmol/L HEPES, 150 mmol/L NaCl, 0.5% Triton X-100, 1 mmol/L EDTA, 1 mmol/L EGTA, 10 mmol/L NaF, 2.5 mmol/L NaVO4, cOmplete protease inhibitor Cocktail (Roche), 1× Phosphatase Inhibitor Cocktail 2 (Sigma-P57726), and 1× Phosphatase Inhibitor Cocktail 3 (Sigma-P0044)]. Samples were sonicated, clarified by centrifugation, filtered, and stored at −80°C. RNA and genomic DNA were also extracted from parental and resistant cell lines by using the AllPrep DNA/RNA Mini Kit (Qiagen). RNA and DNA quality was verified using Agilent RNA 6000 Nano Kit and Agilent High Sensitivity DNA Kit (Agilent Technology). DNA quantity was assessed by the Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific). Biological duplicates were analyzed for each sample.

MIB chromatography and quantitative mass spectrometry (MS) methodology was performed as described previously (17) to examine the kinomes of PD901- or RAD001-resistant Nf1 cell lines and matched parentals. label-free quantification (LFQ) intensity is the normalized abundance across all runs and relative to all identified proteins (typically in the 10e5–10e9 range). One hundred ninety-four kinases divided in 11 kinase families were selected by filtering “Razor + unique peptides” number, which was set ≥ 2. Log2LFQs were calculated and used for further analyses and comparisons. Averages of Log2LFQ duplicates were calculated. Log2LFQ averages of parental cell lines (989 and 881) were subtracted from the Log2LFQ averages of drug-resistant or drug-sensitive counterparts (989 PD_R, 989 RAD_R, 881 PD_R, 989 + PD, 989 + RAD, 881 + PD, and 881 + RAD). Heat maps were created using Morpheus, and the color scale represents Z-score. Cytoscape software integrated with Genemania was used to visualize large-scale integration of molecular interaction network data. Kinase lists were imputed in Genemania to determine physical interactions that were then integrated in the Cytoscape for visualizing kinase–kinase network. Log2LFQ differences >1 or < −1 were included in visualizing the protein interactions.

Total RNA (1 μm) was used as input for Illumina's TruSeq stranded mRNA library preparation protocol. The libraries were made according to the manufacturer's instructions. Library quality and quantity were assessed using the Agilent 2200 TapeStation instrument prior to sequencing. Paired-end 75 bp reads were generated using Illumina NextSeq500 with the high output 150-cycle flowcell. For pathway enrichment analysis, we computed the up/downregulated genes in RAD_R versus P and PD_R versus P in the 989 PD_R and 989 RAD_R cell lines. Specifically, consistently differentially expressed genes in each cell line were analyzed in a generalized linear model to compute the differential gene-expression test (P < 0.001 as a significant cutoff).

RAW sequencing data (fastq files) were mapped to the mouse mm10 genome by using STAR pipeline (18). On average, 65.4% reads were uniquely mapped to the exon regions and were used to compute the counts mapped to each gene. DESeq2 was applied to identify differentially expressed genes among the conditions, with FDR<0.05 and log fold change >1.0 or < −1.0, as the significance cutoff. MDS plots and heat maps were plotted using the log(RPKM+1) normalized gene expression levels. All the sample-wise clustering analysis utilized the hierarchical clustering method (hclust function) by R with the default setting.

Whole-exome sequencing was performed on two biological replicates of each cell line. Whole-Exome Sequence Analysis Genomic DNA was extracted from cells using standard, previously described methods (19). All sequence data including read alignment (to mm10 build) quality and performance metrics after processing, somatic mutation detection, and variant annotation were performed as previously described (20) using the mm10 build of the mouse genome.

The NF1 gene product is the RAS-GTPase activating protein neurofibromin, which negatively regulates RAS. Loss of neurofibromin leads to RAS activation, commonly involving two main signaling pathways downstream of RAS, RAF/MEK/ERK and PI3K/AKT/mTOR. We compared RAS signaling profiles in 12 Nf1-null mouse tumor cell lines derived from radiation-induced tumors arising in Nf1 heterozygous mice (14, 21). Tumor histologies included mammary carcinomas, squamous cell carcinomas, and soft-tissue sarcomas (13, 14). These malignancies were previously analyzed for genetic variants, indicating that these malignancies demonstrate loss of wild-type Nf1 and Trp53 in cis via loss of a large segment of chromosome 11 (13). Although these tumor cell lines share genetic loss of Nf1 as an initiating feature, Western blotting analysis of PI3K–mTOR and RAF–MEK revealed differences in basal RAS signaling among our cell lines when grown in the presence of serum, as evidenced by variable phosphorylation of RAS effectors (Fig. 1A).

We then assessed Ras activity in the context of serum starvation followed by serum stimulation. After 24 hours of serum starvation, cells were stimulated for 0, 5, 30, and 60 minutes with medium supplied with 10% FBS, and then protein lysates were collected. All tumor cell lines demonstrated increased RAS-GTP at 5 minutes post-stimulation with variable kinetics of attenuation (Fig. 1B; Supplementary Fig. S1A). In contrast to the basal Ras pathway signaling, stimulation after serum starvation was uniformly associated with sustained activation of RAS-GTP and downstream effectors such as phosphorylated AKT (S473), S6 (Ser235/236), and p44/42 MAPK (ERK1/2) (Thr202/Tyr204), although even a subset of these (881, 917, 9223, and 989) demonstrated relatively more activation of RAS-GTP and downstream effectors. (Fig. 1B; Supplementary Fig. S1A). These comparisons between signaling under either basal or starvation-stimulation conditions reveal that genetic Nf1-deficient tumor cell lines share intact core Ras pathway activation when acutely stimulated but under basal growth conditions diverge to nonuniform Ras pathway activation.

The variability in RAS pathway activation led us to ask whether these Nf1-mutant tumors harbored-mutant RAS, the presence of which could produce constitutive activation and amplify downstream signaling above that achieved with wild-type RAS. NF1 and RAS are uncommonly comutated in cancer, but NF1 and other RASopathy genes have been found to be comutated in melanomas (6). To identify activating Ras mutations we sequenced exons 2 and 3 of Kras, Nras, and HRas in our 12 Nf1-mutant mouse tumor cell lines.

Ras mutations infrequently co-occurred with Nf1 loss in our mouse model of Nf1-driven tumorigenesis. Of the 12 lines sequenced, one cell line (cell line 169) harbored a somatic mutation in codon 12 of Hras (Supplementary Fig. S2A and S2B). This point mutation involves a recognized hotspot residue that is implicated in binding nucleotide phosphates (16, 22, 23), substituting a guanine for an adenine (GGA>GAA), replacing a glycine for a negatively charged glutamic acid (G12E) in the protein. Notably Ras-GTP activation and downstream signaling after serum stimulation is similar between the cell line 169 and other Nf1-mutant cell lines with wild-type Ras (Fig. 1B), suggesting that this Ras mutation fails to amplify Ras signaling beyond that achieved by Nf1 loss alone. These data indicate that acquired Ras mutations are uncommon in our genetic model of Nf1-driven tumorigenesis and thus do not explain Ras signaling heterogeneity in this system.

The apparent heterogeneity in Ras signaling among the panel of cell lines indicates that Nf1 loss does not produce a common shared Ras signaling signature and suggests underlying heterogeneity in kinome function. Rather than focusing the analysis on cell lines functioning at extremes of signaling, we focused analysis on a subset of cells to investigate genetic, transcriptomic and kinome differences, these being 881 and 989 lines because they are both mouse sarcoma cell lines that share similar kinetics of Ras signaling after starvation followed by stimulation (Fig. 1B), showing peaking of pAKT and pERK at 5 minutes post-stimulation.

Transcriptome and kinome profiling of the 989 and 881 lines demonstrated that although these lines share Nf1 loss and similar basal and stimulated PI3K/AKT and MAPK pathway activation, kinome-wide differences distinguish these lines (Fig. 1C). The five kinases in each cell line exhibiting the greatest MIB binding (Supplementary Fig. S1B) were also transcriptionally upregulated (Fig. 1D). Cytoscape software integrated with Genemania was used to visualize protein interaction networks constituted by the kinomes of the 881 and 989 cell lines (Supplementary Fig. S3A and S3B). Network analysis revealed that Aurora Kinase (AURKA) was among the kinases demonstrating increased MIB binding common between these two cell lines (Supplementary Fig. S3A).

Aurora kinases (AURK) are involved in mitotic control, and inhibitors of AURKs have been shown to have therapeutic potential against cancers (24). To test whether AURKA identified by kinome analysis is functionally relevant to growth in our tumor cell lines, we assessed pharmacologic inhibition of AURKA in the parental cell lines 881 and 989. We assessed the AURKA inhibitor MLN827, and for comparative purposes the AURKB inhibitor AZD1152 and the pan-AURK inhibitor VX680, each decreasing phosphorylation levels of their respective targets (Supplementary Fig. S3C and S3D). AURK inhibition reduced growth of both the 989 and 881 cell lines, although AURKA inhibition with MLN827 produced the greatest negative effect on cell viability, validating the identification of AURKA as a functionally relevant kinase that supports growth by both 989 and 881 Nf1-mutant cell lines.

The development of drug resistance by tumors is an important clinical problem that can involve different mechanisms, such as new mutations in the drug target, activation of redundant biological feedback loops, or overexpression of proteins that compensate for the loss of the drug target. Notably the 989 and 881 cell lines share similar phosphorylation response profiles to PD901, CI1040 (an ATP noncompetitive MEKi), PP242 (an ATP-competitive mTOR inhibitor), PIK75 (PI3K inhibitor), RAD001, and BEZ235 (dual PI3K/mTOR inhibitor) (Supplementary Fig. S4) and we sought to determine whether the similar biochemical responses present in these two lines would extend to the drug-resistant setting.

The best studied therapeutic approaches in NF1-mutant tumors include pharmacologic inhibition of MEK and mTOR (10). To model the emergence of resistance to chronic MEKi or mTORi, we cultured parental Nf1 989 and 881 cell lines in the MEKi PD0325901 (PD901, an ATP noncompetitive MEKi) or the mTORi RAD001 (a rapalog mTORi). Drug-resistant derivatives were established by exposing parental 989 and 881 lines to increasing concentrations (from 25 up to 500 nmol/L) of either PD901 or RAD001 continuously in the media. Resistant cell lines were pooled cells that were continuously maintained in medium supplemented with inhibitor. Two sets of resistant cell lines were established from each chronic drug exposure, cultured in either 100 nmol/L (PD_R1 and RAD_R1) or 500 nmol/L (PD_R and RAD_R). 989 and 881 resistant derivatives are referred to as 989 PD_R, 881 PD_R, 989 RAD_R, and 881 RAD_R (MEKi- and mTORi-resistant, respectively). Culturing cells in 500 nmol/L PD901 produced acquired drug resistance in both lines (Fig. 2A; Supplementary Fig. S5A). In contrast, extended chronic exposure to escalating concentrations of RAD001 led to the development of a relative resistance to RAD001 compared with the parental line, rather than the complete resistance developed after PD901 exposure.

The apparent similarities in Ras signaling between the parental 989 and 881 were lost in the setting of acquired drug resistance. The biochemical changes in Ras signaling varied between cell lines; compared with matched parental cells, levels of pAKT and pS6 markedly increased in 989 PD_R cells compared with parental, whereas 881 PD_R cells lacked this response (Fig. 2B; Supplementary Fig. S5B). Although basal pAKT and pERK levels decreased in 989 RAD_R cells compared with parental, the 881 RAD_R cells demonstrated an increase in these effectors compared with parental 881 cells (Fig. 2B; Supplementary Fig. S5B).

To compare Ras activation in response to stimulation by serum, cells were serum starved for 24 hours and then stimulated for 0, 5, 30, and 60 minutes with serum containing media, after which protein lysates were collected. RAS-GTP levels were assessed by GTPase pulldown. Acquired resistance to MAPK or mTOR inhibition was associated with altered basal Ras pathway activation as well as in response to inhibitor exposure (Fig. 2E; Supplementary Fig. S5). Notably, for both cell lines phosphorylated S6 and ERK, readouts of biochemical response to RAD and PD901, respectively, remain suppressed in drug-resistant derivatives, indicating that reactivation of these pathways does not occur.

Combinatorial kinome inhibition is a strategy used to improve treatment efficacy, and we assessed the sensitivity of PD_R cells to RAD001 and RAD_R cells to PD901. Both 989 PD_R and 881 PD_R cell lines show sensitivity to RAD treatment alone (Fig. 2D and E; Supplementary Fig. S5C). PD901 decreased cell viability in both the 989 RAD_R and 881 RAD_R cell lines (Fig. 2D; Supplementary Fig. S5D). Finally, combined PD901 and RAD001 produced the greatest suppression of cell viability in both sets of drug-resistant derivatives, pointing to the combined importance of each of these pathways in promoting proliferation (Fig. 2E; Supplementary Fig. S5C).

The mechanisms responsible for acquired drug resistance may affect processes mediating sensitivity to other drugs, and this can be a point of differentiating between cell lines and types of acquired drug resistance. The issue of cross-resistance, whereby resistance to one kinase inhibitor confers resistance to other inhibitors, is a clinically significant issue and challenging to predict. We analyzed drug-resistant derivatives for the acquisition of resistance to other kinase inhibitors. AURKA is a functionally relevant kinase that supports growth by both parental 989 and 881 Nf1-mutant cell lines (Supplementary Fig. S3); we assessed whether acquired resistance to either MEKi or mTORi also conferred resistance to AURKA inhibition. Both RAD_R cell lines and the 881 PD_R cell line retained sensitivity to all three Aurora inhibitors (with AZD1152 producing the weakest response), demonstrating decreased cell growth and phosphorylation of AURKA, AURKB, and H3 (Supplementary Fig. S6A and S6B). The 989 PD_R cells were the exception among the drug-resistant derivatives, demonstrating resistance to Aurora inhibitor AZD1152 alone as well as when combined with PD901 (Supplementary Fig. S6A). The 989 PD_R tumor cells serve as an example of how tumor cells that acquire resistance to one inhibitor (MEKi) can acquire resistance to inhibitors of other molecular classes (AURK), although given that this did not also occur in the 881 PD_R cells, we can surmise that this cross-resistance is cell specific rather than a generalized feature of resistance to MEKi.

Drug-resistant derivatives were also challenged with alternative MEK and mTOR inhibitors (MEK162, trametinib, BEZ235, and rapamycin) already approved for clinical use or tested in different clinical trials. Parental and PD-resistant 881 cells were unaffected by the MEK162 (Supplementary Fig. S7); however, the RAD-resistant 881 RAD_R acquired sensitivity to MEK162 (Supplementary Fig. S7). This acquired sensitivity was not similarly observed in the 989 lines (parental and drug-resistant derivatives), all of which, similar to the parental 881 line, remained resistant to MEK162. BEZ235, a dual PI3K and mTOR inhibitor markedly suppressed the growth of all drug-resistant cell lines (Supplementary Fig. S7; ref. 25), indicating that Nf1-mutant tumor cells acquiring resistance to either MEKi or mTORi continue to be functionally dependent on PI3K.

We compared the transcriptomes of 989 cells treated with either acute or chronic MEKi and mTORi (989 PD_R and 989 RAD_R; Fig. 3A) to parental 989 cells using unsupervised hierarchical clustering (Fig. 3B). This analysis indicated that parental and acutely treated parental cells (989 + PD9901 and 989 + RAD001) retained transcriptomic similarity as a group, whereas drug-resistant cell lines clustered separately from parental treatment groups and controls (Fig. 3B; Supplementary Fig. S8A). Overall, acutely treated cells (with either PD901 or RAD001) demonstrated significantly fewer differentially expressed genes as compared with 989 PD_R and 989 RAD_R cell lines (Fig. 3C).

Given the significant changes in expression associated with acquired drug resistance, we performed pathway enrichment analysis to identify preferentially affected biological processes. We first computed the pathways enriched by differentially expressed genes (DEG; P < 0.01). We then analyzed DEGs in a generalized linear model (P < 0.001 as a significant cutoff) and found a very significant overlap between the DEG lists of RAD_R versus P and PD_R versus P, P < 1e−16 for both upregulated and downregulated genes in the two resistant lines (Supplementary Table S3). This analysis identified pathway enrichment shared between the drug-resistant cell lines; significantly increased gene expression in extracellular matrix organization pathway occurred in both RAD and PD-resistant cell lines. Genes significantly downregulated in both resistant cell lines commonly functioned in pathways modulating metabolism of noncoding RNA and mRNA processing.

To determine how acute versus chronic drug exposure differs in effects on the kinome, we used MIBs coupled with mass spectrometry (MIB/MS) to quantitatively measure dynamic changes in kinase activity. For this analysis, we compared parental 989 tumor cells that were acutely treated with either PD901 or RAD001 (for 24 hours) and drug-resistant 989 tumor cells that were continuously grown in either of these drugs (989 PD_R or 989 RAD_R). Figure 3D shows these four experimental groups that are each compared with control untreated parental cells: the drug-resistant cell lines 989 PD-R, 989 RAD-R, and the parental cell lines treated with either drug acutely. The heat map depicts the subtraction of the average kinase Log2LFQ (194 total) of two biological replicates of parental cell lines from the average Log2LFQs of two biological replicates of resistant or acutely treated cells. Thus, untreated parental controls were analyzed in parallel with drug-treated cells, and the data are displayed as the difference from untreated parental controls.

Kinome analysis distinguishes drug-resistant cells from acutely treated cells (Fig. 3D). One hundred ninety-four kinases were assayed in resistant and acutely treated 989 cells and compared with untreated parental 989 cells; the differences in MIB binding between drug-treated cells and parental controls are visualized as a heat map in which kinases are organized into 11 kinase families (Fig. 3D). Strikingly, the kinome profiles of the drug-resistant 989 cells segregated from acutely treated cells, irrespective of the drug target, supporting the concept that Nf1-mutant tumors undergo broader remodeling of kinome activity in response to targeted kinase inhibition.

Integrated kinome and RNA-seq analyses of PD901- and RAD001-resistant Nf1 cell lines (compared with parental) indicate that changes in the functional kinome generally correlate with mRNA expression changes (Fig. 4A). Kinome data were organized into molecular interaction networks for PD_R and RAD_R cell lines (S8B-G).

Common, shared changes in specific kinases were identified in drug-resistant cells (Fig. 4B–E), with the MIB binding of some kinases correlating with changes in transcription (Fig. 4A). Shared kinases with increased MIB binding were diverse, and apparently not functionally linked to each other or to MAPK and PIK3 signaling pathways, thus representing potentially new mechanisms of therapeutic relevance. Shared kinases included those involved in metabolism, such as PFKM (26), as well as CDK18 (27) and CHEK1 (28), which are involved in DNA repair (Fig. 4B).

Interestingly, drug resistance was associated with reduced MIB binding of three functional classes of kinases (Fig. 4C). Kinases involved in metabolism demonstrated significantly reduced binding; these included HK2, which catalyzes the first essential step of glucose metabolism, the conversion of the substrate glucose into glucose-6-phosphate (29) and the mitochondrial enzyme PCK2, which catalyzes the conversion of oxaloacetate to phosphoenolpyruvate in the presence of guanosine triphosphate (GTP; refs. 30, 31). Growth factor receptors EGFR and PDGFRA demonstrated reduced MIB binding in the setting of drug resistance (Fig. 4C). Three protein kinases involved in the PIP2 pathway were downregulated in both 989 PD_R and 989 RAD_R cell lines, these being two PRKC isoforms, PRKCA and PRKCD, and one of their downstream effectors PRKD2 (Fig. 4C). The calcium-activated PRKCA kinase can promote cell growth by phosphorylating and activating RAF1, which mediates the activation of the MAPK/ERK signaling cascade (32).

Notably, few kinases were commonly altered in MEKi-resistant cell lines (989 PD_R and 881 PD_R; Fig. 4D and E). EPHB2 exhibited increased MIB binding in both PD901-resistant lines (989 PD_R and 881 PD_R; Fig. 4D) and also in the 989 RAD_R line, making it the only kinase demonstrating increased MIB binding in all three cell lines analyzed. EPHB2 (Ephrin type-B receptor 2) is a member of the largest family of receptor tyrosine kinases in the mammalian genome and demonstrated increased MIB binding in both. EPHB2 is capable of activating MAPK pathway inhibition and has been described to modulate multiple cancer cell behaviors, such as invasion, dedifferentiation, and metastasis (33, 34). Consistent with pharmacologic inhibition, reduced MIB binding was observed for MAP2K1/MEK1 in cells chronically treated with MEKi (Fig. 4E).

The kinome profiles of resistant phenotypes revealed a limited number of targetable nodes; however, the drug resistance can also be driven by the genetic events such as acquired somatic variants. To investigate this, we performed whole-exome sequencing analysis of the parental 989 cell line, its matched drug-resistant derivatives, and germline (tail) control to identify somatic single-nucleotide variants and compare the mutational profile among these lines. Somatic single-nucleotide variants (SNV) were quantified in each cell line.

Acquired drug resistance was associated with unique somatic variants in the exomes of resistant lines that were not present in the parental lines. Both 989 PD_R– and 989 RAD_R–resistant cell lines harbored more somatic SNVs than the parental line from which they were derived (Fig. 5A). The degree of genetic similarity between parental and drug-resistant derivatives can be inferred from the percentage of somatic variants that are shared or private; the numbers of shared and private somatic SNVs among 989, 989 PD_R, and 989 RAD_R are shown in Fig. 5A and B. One hundred thirty somatic SNVs were shared among all three cell lines. 989 PD_R and 989 RAD_R cell lines shared 39 SNVs, suggesting that these SNVs are not drug specific. Expressing the private and shared somatic SNVs as percentages of each cell line's total somatic SNVs (Fig. 5B), we find that private mutations in 989 PD_R and 989 RAD_R comprised 54.33% and 33.95% of the total somatic SNVs, respectively.

These analyses indicate that acquired drug resistance is associated with increased genetic diversity, evidenced by minor retention of parental somatic SNVs resulting in fewer than 5% of somatic SNVs in either drug-resistant line being shared exclusively with the parental line. Interestingly, a greater proportion of somatic SNVs was shared only by the drug-resistant lines, suggesting possibly common resistance variants (Fig. 5A and B).

Molecular pathways involved with somatic mutations in drug-resistant cell lines included pathways regulating release of intracellular calcium and PKC activation. Plcd3 hydrolyzes the PIP2 to generate the second messenger molecules diacylglycerol and inositol 1,4,5-trisphosphate (IP3) (35), and was mutated in the 989 PD_R cell line. Ptprq, which regulates the phosphorylation state of AKT1 by suppressing the PIP3 level (36), was also mutated in the 989 PD_R cell line. Dgkq, which phosphorylates diacylglycerol to generate phosphatidic acid (PA; ref. 37), was mutated in both 989-resistant cell lines (Supplementary Table S1).

The 989 PD_R and 989 RAD_R cell lines shared a mutation in the gene Nfrkb (Fig. 5C). NFRKB is a member of the chromatin remodeling INO80 complex and binds to the DNA consensus sequence 5′-GGGGAATCTCC-3′ (38). WES and RNA-seq each identified the same somatic mutation in the Nfrkb gene in both 989-resistant cell lines (Fig. 5C). The NFRKB protein sequence is highly conserved in multiple species (39); the point mutation acquired by drug-resistant 989 cell lines substitutes a cytosine for a guanine (CCG>CGG) in position 223, replacing a proline for an alanine (P75A). Due to the large size of the protein, the full-length NFRKB structure has not yet been determined; however, mutations throughout the NFRKB protein have been identified in human cancers (COSMIC database, https://cancer.sanger.ac.uk/cosmic/), and the overall distribution of somatic SNVs is suggestive of NFRKB functioning as a tumor suppressor.

Managing drug-resistant cancers is a practical challenge, and effective strategies are likely defined by the mechanisms by which drug resistance develops and whether these mechanisms are reversible. Drug-resistant tumors can develop an addiction-like state that renders them sensitive to drug withdrawal (40), a characteristic that may support discontinuation of a drug. We sought to test whether drug-resistant Nf1-null tumor lines demonstrate drug addiction evidenced by sensitivity to drug withdrawal. Drug-resistant 989 and 881 lines were grown in drug-free media (named washout, “WO”) for 1 week (989 PD_R WO, 989 RAD_R WO, 881 PD_R WO, and 881 RAD_R WO), and then cell viability was compared with parental and drug-resistant lines (Fig. 6A and B).

After drug washout, both 989 RAD_R WO and 881 RAD_R WO demonstrated significantly reduced growth as compared with the lines from which they were derived, 989 RAD_R and 881 RAD_R (Fig. 6B). In contrast, 989 PD_R WO and 881 PD_R WO maintained growth and retained their resistance to PD901 (Fig. 6A). Phosphorylated ERK1/2 in 989 PD_R WO cells after drug washout failed to recover to parental levels (Fig. 6C), suggesting that the development of resistance to MEKi is associated with stable impaired MAPK signaling. Notably, withdrawing drug from 989 RAD_R cells (989 RAD_R WO) resulted in impaired growth compared with parental control cells despite the recovery of mTOR signaling, as evidenced by phospho S6 and phospho Akt (Fig. 6D). Re-exposure of both 989 RAD_R WO and 881 RAD_R WO cells to RAD was associated with complete loss of phosphorylated S6, recapitulating the original RAD-resistant 989 and 881 lines. These data contrast chronic MEKi with chronic mTORi with respect to drug addiction and reconstituting inhibition observed in drug-naïve cells; chronic mTORi was associated with drug dependence in both Nf1-mutant cell lines and reduced growth upon drug withdrawal, whereas this response was absent in PD-resistant Nf1-mutant cell lines tested in parallel.

Our analysis of acquired drug-resistant cell lines demonstrated that drug exposure is associated with extensive transcriptomic changes that distinguish drug-resistant cells from parental cells (Fig. 3C). To investigate the stability of these transcriptomic alterations, we then asked whether drug withdrawal might reveal stable versus reversible transcriptomic changes after chronic drug exposure. We compared the transcriptomes of WO cells to the drug-resistant cells from which they were derived and the parental cells from which all drug-exposed cells were derived. Unsupervised hierarchical clustering analysis distinguishes 989 PD_R WO and 881 PD_R WO from the 989 PD_R and 881 PD_R lines (Fig. 6E; Supplementary Fig. S8C), suggesting that drug withdrawal itself triggers a transcriptomic response. Notably, hierarchical clustering analysis restricted to only kinome transcripts (Fig. 6E) differentiates 989 PD_R and 989 PD_R WO from drug-resistant and parental cells, suggesting that the kinome robustly reflects cellular adaptations in response to MEKi. In neither cell line does drug washout restore the kinome to the parental state (Fig. 6E), providing evidence that kinome-wide consequences of MEKi or mTORi are relatively irreversible. Collectively, these findings indicate separate fates associated with prolonged drug exposure, with MEKi being associated with irreversible drug resistance, whereas mTORi failed to produce complete and durable drug resistance (Fig. 6F).

Tumors harboring NF1 mutations are classified as a subtype of RASopathies. Although Nf1 loss is the shared genetic alteration leading to neoplastic progression in NF1, our data indicate that loss of heterozygosity of Nf1 as an initial genetic event does not confer a common functional signature. We interrogated a unique set of murine Nf1-mutant tumors to assess the biochemical and functional spectrum in this genetic class, and in so doing identified basal Ras pathway activation and variable responses to kinase inhibition. These data demonstrate basal heterogeneity in Ras signaling in the context of dysregulated but largely wild-type Ras signaling. Furthermore, in the absence of a strong driver such as oncogenic Ras, Nf1-null tumors, in which Ras pathway activation is heterogeneous, may be uniquely positioned to utilize diverse mechanisms to circumvent drug therapy.

Using multiple approaches to investigate the genetic and functional similarities among Nf1-mutant tumors, we uncovered shared vulnerabilities; for example, kinome profiling of parental Nf1-null tumor lines identified overlap with respect to kinase affinity (Supplementary Fig. S3). Among the activated kinases common to the Nf1-null cell lines was Aurora Kinase A (AURKA). Interestingly, AURKA is not a canonical Ras pathway effector, but represented a common therapeutic target, as the AURKA inhibitor MLN8237 decreased phospho-AURKA levels and reduced cell growth. Importantly, AURKA is not known to have a prominent function in NF1-mutant cancers and thus its identification represents a novel opportunity. AURKA inhibition has been used to overcome drug resistance in several cancers (41, 42), and targeting AURKA in Nf1-null neoplasms might similarly serve as a strategy to prevent or delay drug resistance.

Our study examines the biochemical consequences of chronic mTOR inhibition and demonstrates that the signaling consequences of prolonged inhibition differ from the acute effects. RAD001 inhibits mTORC1, which can trigger a negative feedback loop resulting in AKT S473 phosphorylation by mTORC2 (43). However, prolonged rapamycin treatment over hours to days) has been described to inhibit mTORC2 assembly and subsequently decrease phosphorylation of Akt S472 (44). We found that Nf1-mutant tumor cells chronically exposed to RAD001 demonstrate downregulated phosphorylation of both ERK and AKT compared with their matched parental lines (Fig. 2C). Similarly, increasing concentrations of rapamycin treatment (100 and 1,000 nmol/L) for 24 hours in cancer cells produces a dose-dependent reduction in phospho-ERK (45) in an mTORC2-dependent manner. The reduced phospho-ERK observed in RAD001-resistant 989 cells likely reflects the same phenomenon, similar to AKT.

Tumors evolve and can develop complex genomic alterations in response to therapy (46). These post-therapy alterations can have important therapeutic implications; however, neither post-treatment nor drug-resistant NF1-mutant tumors have been characterized to date. To study the consequences of inhibitor-mediated selective pressure leading to a therapy-resistant tumor cell population, we modeled the emergence of drug-resistant Nf1 tumors in response to chronic MEKi or mTORi. This approach enabled systematic comparisons between experimentally derived drug-resistant cells and matched original parental lines. These analyses revealed functional remodeling of the kinome, as evidenced by significant changes in MIB binding and transcript abundance developing in the setting of both MEKi and mTORi. Importantly, comparing MIB binding between resistant cell lines revealed few kinases commonly increased or decreased between pairs of drug-resistant cells, but the greatest overlap or shared kinases between 989 PD_R and 989 RAD_R (Fig. 4B and C).

MEKi- and mTORi-resistant tumors were genetically distinct; somatic SNVs were predominantly unique acquired mutations, with minor representation of parental SNVs. Also notable was the absence of RAS mutations in drug-resistant cell lines, indicating that Nf1-driven tumorigenesis and responses to therapy occur in the context of wild-type Ras. The propensity of drug-resistant tumors to acquire a significant proportion of private mutations, thereby becoming “less related” to the original parental tumor, resists attempts to neatly classify drug-resistant disease on a genetic basis. Given this challenge and suggested heterogeneity in genetic alterations post-therapy, our findings highlight the need for studies directed to this issue. Further analyses of greater numbers and types of after treatment tumors, experimental and clinical, will be needed to elucidate the most common or dominant mutational motifs.

Drug resistance can be associated with addiction, whereby drug withdrawal results in loss of essential growth signals and subsequent tumor regression. Interestingly, RAD-resistant cell lines demonstrated evidence of drug addiction, whereas PD-resistant cells did not. Upon drug withdrawal, RAD-resistant lines reverted to their parental mTOR signaling and reexposure to RAD reconstituted inhibition of mTOR. In contrast, following drug cessation, PD_R WO cell lines maintained similar functional and biochemical mechanisms to the PD_R cells. Our data show that drug addiction develops in a drug-specific manner, a finding that supports therapeutic strategies tailored to specific drug resistance, for example, a drug holiday for mTORi-resistant cells and mTORi for MEKi-resistant cells.

In summary, tumorigenesis initiated by Nf1 loss is mediated in the context of wild-type Ras and produces biochemically heterogeneous cancers. This diversity is apparent in the basal signaling, response to therapeutics, and subsequent acquired drug resistance. Nf1-mutant tumors treated with MEKi and mTORi, the two most commonly utilized therapeutic approaches against hyperactive Ras signaling in NF1 neoplasms, acquire genetic and transcriptomic alterations distinguishing them from parental tumors. The reversibility of these changes is defined in a drug-dependent manner and therapeutic strategies in the posttreatment context must be informed by data specific to the clinical scenario.

D.W. Clapp reports grants from NIH during the conduct of the study. G.L. Johnson reports grants from NIH (DHART SPORE where I was head of the omics core) during the conduct of the study and grants from NCI (Breast SPORE), NIH (U24 grant for ODG program), and NCI (U01 for systems biology of breast cancer signaling) outside the submitted work. J.L. Nakamura reports grants from NIH/NCI and Department of Defense and other from Hagar Family Foundation (philanthropic gift) during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

D. Pucciarelli: Formal analysis, validation, investigation, writing–original draft, writing–review, and editing. S.P. Angus: Formal analysis, methodology, and writing–original draft. B. Huang: Formal analysis, methodology, and writing–original draft. C. Zhang: Formal analysis, supervision, methodology, writing–original draft, writing–review, and editing. H.J. Nakaoka: Validation. G. Krishnamurthi: Investigation. S. Bandyopadhyay: Conceptualization. D.W. Clapp: Conceptualization. K. Shannon: Conceptualization and writing– original draft. G.L. Johnson: Conceptualization. J.L. Nakamura: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, project administration, writing–review, and editing.

We thank Benjamin Braun and his lab for helpful discussions. J.L. Nakamura was supported by a gift from the Hagar Family Foundation and the Department of Defense (W81XWH1810287). J.L. Nakamura, K. Shannon, D.W. Clapp, G.L. Johnson, S.P. Angus, and C. Zhang were supported by U54 CA196519 (NIH/NCI).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.