Activity-Based Proteomics Reveals Heterogeneous Kinome and ATP-Binding Proteome Responses to MEK Inhibition in KRAS Mutant Lung Cancer

One way cancer cells can escape from targeted agents is through their ability to evade drug effects by rapidly rewiring signaling networks. Many protein classes, such as kinases and metabolic enzymes, are regulated by ATP binding and hydrolysis. We hypothesized that a system-level profiling of drug-induced alterations in ATP-binding proteomes could offer novel insights into adaptive responses. Here, we mapped global ATP-binding proteomes perturbed by two clinical MEK inhibitors, AZD6244 and MEK162, in KRAS mutant lung cancer cells as a model system harnessing a desthiobiotin-ATP probe coupled with LC-MS/MS. We observed strikingly unique ATP-binding proteome responses to MEK inhibition, which revealed heterogeneous drug-induced pathway signatures in each cell line. We also identified diverse kinome responses, indicating each cell adapts to MEK inhibition in unique ways. Despite the heterogeneity of kinome responses, decreased probe labeling of mitotic kinases and an increase of kinases linked to autophagy were identified to be common responses. Taken together, our study revealed a diversity of adaptive ATP-binding proteome and kinome responses to MEK inhibition in KRAS mutant lung cancer cells, and our study further demonstrated the utility of our approach to identify potential candidates of targetable ATP-binding enzymes involved in adaptive resistance and to develop rational drug combinations.


Introduction
A cancer cell's ability to rapidly adapt to targeted reagents is a major hurdle to the success of targeted therapy for cancer treatment. Examples include the compensatory activation of PI3K/Akt following mTOR inhibition driven by upstream receptor tyrosine kinase activation [1][2][3][4] and feedback activation of B-and C-RAF upon MEK inhibition, especially in the context of RAS-driven cancers [5,6]. This adaptive resistance is complicated by the diversity of kinases and enzymes expressed in cancer mutant lung cancer cell lines harboring diverse p53 and LKB1 co-mutations and differential histology (adenocarcinoma and squamous cell carcinoma) to address heterogeneous adaptive responses. Using these two MEK inhibitors allows for filtering and focus on "on target" effects and not just idiosyncratic drug targets.

Results and Discussion
To address diverse adaptive responses to MEK inhibition in the context of KRAS mutant lung cancer, we employed five KRAS mutant lung cancer cell lines with differential LKB1/p53 mutation status and histology; four lung adenocarcinoma cell lines including A427, A549 (p53 wild type/LKB1 mutant), Calu-1, and Calu-6 (p53 mutant/LKB1 wild type); and a lung squamous cell carcinoma cell line, H157 (p53 mutant/LKB1 mutant). We then assessed how MEK inhibitors remodel their ATP-binding proteomes. Cells were treated with 1 µM of MEK inhibitors (AZD6244 or MEK162) or vehicle control (DMSO), and then ATP-binding proteins were labeled with the desthiobiotin-ATP probe and trypsin digested. We found 1 µM to be a clinically relevant [17] and a 24-h time point was chosen based on a previous study on kinome-level response to a MEK inhibitor [9]. The probe-labeled peptides were enriched by streptavidin beads, followed by LC-MS/MS analysis. The workflow of our study is illustrated in Figure 1.  After filtering out low-confidence peptides, we identified 5800 peptides that were associated with 1925 protein groups (Table S1). Principal component analysis (PCA) shows clustering based on cell lines and not for treatment type, suggesting that, despite shared KRAS mutant status, the ATP-binding proteomes and responses to MEKis are cell type specific with the exception of similarity between Calu-1 and H157 in the first two principal components ( Figure 2A). Interestingly, the X and Y axes of the PCA plot, corresponding to the first two principal components of the data, are not associated with drug effect on cell viability, which showed a similar reduction across the cell lines (50%-60% reduction at 1 µM; data not shown) nor with co-mutating tumor suppressor (p53 and LKB1) status. This suggests that these are not major factors dictating the behavior of the ATP-binding proteome. However, given the small number of cell lines employed in this study, larger studies with more power are necessary to investigate this further. From these 1925 protein groups, we were able to identify 174 protein kinases or 225 kinases in total (including nine lipid kinases and 42 other generic/small molecule kinases). These results are comparable to previously-reported identifications using this technology (188 protein kinases [22], 136 protein kinases [23], and 41 protein kinases [24]). Although kinases were a primary focus, we were able to identify hundreds of other proteins in a variety of different classes ( Figure 2B). We were also able to detect some proteins that have not been reported to bind ATP. These proteins likely received an ATP-probe by being adjacent to a substrate of, or in a complex with, an ATP-binding protein or by promiscuous binding of the ATP-probe during the labeling step of the experimental workflow. After filtering out low-confidence peptides, we identified 5800 peptides that were associated with 1925 protein groups (Table S1). Principal component analysis (PCA) shows clustering based on cell lines and not for treatment type, suggesting that, despite shared KRAS mutant status, the ATP-binding proteomes and responses to MEKis are cell type specific with the exception of similarity between Calu-1 and H157 in the first two principal components ( Figure 2A). Interestingly, the X and Y axes of the PCA plot, corresponding to the first two principal components of the data, are not associated with drug effect on cell viability, which showed a similar reduction across the cell lines (50%-60% reduction at 1 µM; data not shown) nor with co-mutating tumor suppressor (p53 and LKB1) status. This suggests that these are not major factors dictating the behavior of the ATP-binding proteome. However, given the small number of cell lines employed in this study, larger studies with more power are necessary to investigate this further. From these 1925 protein groups, we were able to identify 174 protein kinases or 225 kinases in total (including nine lipid kinases and 42 other generic/small molecule kinases). These results are comparable to previously-reported identifications using this technology (188 protein kinases [22], 136 protein kinases [23], and 41 protein kinases [24]). Although kinases were a primary focus, we were able to identify hundreds of other proteins in a variety of different classes ( Figure 2B). We were also able to detect some proteins that have not been reported to bind ATP. These proteins likely received an ATP-probe by being adjacent to a substrate of, or in a complex with, an ATP-binding protein or by promiscuous binding of the ATP-probe during the labeling step of the experimental workflow.  Next, we set out to examine how MEK inhibitors remodel ATP-binding proteomes in each individual cell line. We first averaged the log 2 fold-changes of drug/control across all cell lines. This allowed us to observe if there were any differences between the effects of the two drugs. Figure 2C shows a strong, positive correlation between the two inhibitors, suggesting that these drugs are behaving in a similar manner. Next, we defined altered peptides as those whose labeling levels (calculated as log 2 (drug/control)) were changed by at least one standard deviation away from the average by both MEK inhibitors (Table S2). In order to gain a comprehensive view of the altered ATP-binding proteome, individual proteins corresponding to these altered peptides were subjected to GeneGO MetaCore pathway enrichment analysis. We observed strikingly diverse enriched pathways; most of the enriched pathways were observed in only one cell line, which highlights a heterogeneous response to MEK inhibition ( Figure 2D and Table S3). Representative pathways enriched from each cell line are shown in Table 1. Despite the heterogeneity, cytoskeleton remodeling pathways were highly enriched across all cell lines. We observed enrichment of glycolysis/gluconeogenesis pathways in A427, A549, and Calu-1 cells, suggesting MEK inhibition leads to altered glucose metabolism in these cells. Previous studies indicated that pharmacological inhibition of BRAF or MEK suppresses glycolysis in the context of melanoma cells harboring activating mutation of BRAF [25,26], warranting future studies to examine whether MEK inhibition leads to metabolic perturbation and, if so, to determine its clinical implication on lung cancer. Next, we focused on specific kinome responses to MEK inhibitors in each cell line given the importance of kinase signaling in regulating cell growth and survival [27]. In total, 225 kinases, which were quantified from the five cell lines, showed quite similar overall expression patterns between cell lines (Figure 3a and Table S4); however, each cell line showed unique drug-induced altered kinase list (Figure 3b and Table S5). This observation prompted us to hypothesize that each cell line could show a distinct kinome response to MEK inhibition, and the trend of changes in kinases is illustrated in kinome trees (Figure 4). Despite this heterogeneity, we observed a number of peptides that were consistently altered by both MEK inhibitors in more than two cell lines, which suggests common adaptive responses could exist (Table 2).
Given the importance of tyrosine kinases in cell growth and survival, we first focused on altered tyrosine kinases (TK group) in each cell line. Despite the well-known role of drug-induced activation of receptor tyrosine kinases in drug resistance, we could not observe significantly-altered receptor tyrosine kinases from our results. However, we found that the non-receptor tyrosine kinase JAK1 is upregulated in A549 and Calu-1. It has been reported that MEK inhibitors induce the JAK-STAT pathway promoting cancer cell invasiveness in melanoma [28], suggesting that MEK inhibitors could induce a similar phenotype in KRAS mutant lung cancer. We also observed downregulation of focal adhesion kinase FAK1 in Calu-1 and Calu-6, suggesting that MEK inhibition could lead to altered integrin signaling and cell motility. Second, MEK inhibitors upregulated stress-activated protein kinase (SAPK) signaling, including MKK6 (A427, A549, Calu-6) and MKK3 (Calu-1), both of which are upstream of p38 MAPK. Third, MEK inhibitors downregulated mitotic kinases PLK1 and its upstream Aurora A kinase (AURKA). Notably, this effect was observed in most of the cell lines, suggesting potential crosstalk between MAPK signaling and cell cycle kinases in the context of KRAS mutant lung cancer. Several preclinical studies have already indicated KRAS mutant cancers are specifically vulnerable to inhibition of mitotic kinases [14,29,30]. However, significant clinical responses of mitotic kinase inhibitors are still lacking in lung cancer [31,32]. It is, thus, possible that combined inhibition of MAPK and mitotic kinases could be synergistic in the context of KRAS mutant lung cancer. Finally, MEK inhibitors upregulated autophagy kinases ULK1 (A549 and Calu-1), ULK3 (A427), and AMPK (A427 and Calu-1). Autophagy is a self-digestion process that is generally activated by nutrient deprivation, but it is also known to be induced by therapeutic stresses in cancer cells, contributing to drug resistance [33,34]. These observations are consistent with previous studies indicating RAF and MEK inhibitors could induce cytoprotective autophagy, leading to drug resistance in the context of BRAF mutant melanoma [35,36] and KRAS mutant lung cancer [37]. Our results reveal kinases that are potentially responsible for autophagy induction in KRAS mutant lung cancer cells treated with MEK inhibitors, further offering potential rational drug combinations.
The desthiobiotin-ATP probe employed in this study was originally developed for drug target profiling to assess the specificity of kinase inhibitors [15,22], but we employed it here to assess global ATP-binding proteome/kinome response to clinical MEK inhibitors in the biological context of various KRAS mutant lung cancer cell lines. Gygi and his colleagues reported that desthiobiotin-ATP probe did not specifically enrich active forms of kinases [38], thus, the changes are likely to be associated with total protein level, rather than activity, at least in the context of kinases. The importance of drug-induced transcriptome and kinome reprograming and its implication in drug resistance is increasingly being recognized [9,10]. In light of these studies, the value added from our study is heterogeneity of drug-induced kinome and ATP binding protein expressions in KRAS mutant lung cancer. Our study also uncovered novel adaptive responses, suggesting that MEK inhibition could lead to metabolic alteration, abnormal mitosis, and induction of cytoprotective autophagy. Our results could be integrated with phosphoproteomics datasets to address how the drug-induced kinome changes lead to modulation of the phosphoproteome. Further, synthetic lethal kinome RNAi screening [39][40][41][42] or pharmacologic vulnerability screens [43] could be combined to assess translational potential of our results.   [44]. Red and green nodes signify increased and decreased kinases, respectively. Illustrations reproduced courtesy of Cell Signaling Technology. Table 2. Altered kinase peptides by both MEK inhibitors in more than two cell lines.

Gene Symbol
Position Cell Line Direction of Change    [44]. Red and green nodes signify increased and decreased kinases, respectively. Illustrations reproduced courtesy of Cell Signaling Technology. Table 2. Altered kinase peptides by both MEK inhibitors in more than two cell lines.

Gene Symbol Position
Cell Line Direction of Change Figure 4. Kinome response to MEK inhibition. Altered kinases from each cell line are mapped on kinome trees using a Web-based kinome render tool [44]. Red and green nodes signify increased and decreased kinases, respectively. Illustrations reproduced courtesy of Cell Signaling Technology.

Cell Lines and Drugs
Cells were maintained in RPMI 1640 medium supplemented with 10% FBS. Cells were confirmed to be free of mycoplasma using PlasmoTest (Invivogen, San Diego, CA, USA). AZD6244 and MEK162 were purchased from ChemiTek (Indianapolis, IN, USA). Drugs were reconstituted with DMSO at 50 mM concentration, and aliquots were stored at´80˝C.

Drug Treatment and ATP Probe Labeling
Cells at 80% confluence were treated with DMSO (vehicle control) or 1 µM MEK inhibitors (AZD6244 or MEK162) for 24 h; cell pellets were then harvested in ice-cold PBS and stored at 80˝C until probe labeling. Cell lysates were prepared and labeled according to the manufacturer's recommendations for the Pierce Kinase Enrichment Kits and ActivX Probes (Thermo Scientific, Rockford, IL, USA)). Briefly, cell pellets were re-suspended in 600 µL of Pierce IP lysis buffer and sonicated. The lysates were cleared by centrifugation at 16,000 g for 10 min at 4˝C and desalted by Zeba Spin Desalting Columns (Thermo Scientific, Rockford, IL, USA). The concentration of protein was measured using a Bradford assay, and a total of 1 mg was used for ATP probe labeling. MnCl 2 was added to the lysate to a final concentration of 20 mM for 10 min, and then the desthiobiotin-ATP probe was added at a final concentration of 5 mM for 15 min. All reactions were performed at room temperature.

MS Sample Preparation and LC-MS/MS Analysis
The probe-labeled lysates were denatured in 5 M urea and 5 mM DDT at 65˝C for 30 min and then reduced in 40 mM iodoacetamide at room temperature for 30 min. The lysates were desalted by Zeba Spin Desalting Columns and then digested with 20 µg/mL of trypsin for 2 h at 37˝C. Desthiobiotinylated peptides were captured by 50 µL slurry of high-capacity streptavidin beads for 1 h. The beads were washed with lysis buffer, PBS buffer, and LC-MS grade water in sequence with four washes for each buffer, and then the peptides were eluted by 50% acetonitrile in 0.1% formic acid (TFA) water. The peptides were lyophilized in a vacuum concentrator and re-suspended in 20 µL of injection buffer containing 2% acetonitrile and 0.1% TFA. The LC-MS/MS analysis was performed as previously described [45].

Data Analyses
The peptides identified in the mass spectrometry data were analyzed and quantified using MaxQuant Version 1.2.2.5 (http://www.maxquant.org) [21]. For the search parameters, desthiobiotin-ATP labeling was selected, enzyme specificity was set to fully tryptic cleavage, allowing a 10 ppm mass error in the main search, a maximum number of three modifications per peptide, and four maximum missed cleavages. Fragment ion tolerance was set to 0.6 Da. Protein and peptide false discovery rates were set to 0.1 and 0.05, respectively, and the minimum peptide length was set to six amino acids. MS/MS data were searched against the UniProt human database combined with common contaminants and concatenated with the reversed versions of all sequences using the Andromeda search engine integrated into MaxQuant. The protein kinases were annotated using GeneGO (Boston, MA, USA) and intensity of each peptide corresponding to protein kinases were used to calculate the inhibition of each drug on the relevant peptide compared with DMSO control group.
Peptide site intensities were extracted from MaxQuant output and input into the Libaffy software package [46]. Libaffy features access to iterative rank-order normalization (IRON), which we have previously used in proteomics applications [47,48]. Peptide site intensities were normalized using IRON before being input into R/RStudio with corresponding metadata [49,50]. Peptides were filtered to remove posterior error probability >0.1, reverse sequences, non-human peptides, and for peptides that were identified but had no corresponding intensities. Intensities were then log 2 -transformed. Concordance of log 2 ratios between control and treatment pairs was measured using Pearson's correlation (herein referred to as R). GeneGO MetaCore was used to annotate individual proteins from the identified protein groups [51] and used for pathway searches on all proteins in identified protein groups. A protein group was included in the pathway search if there was at least one standard deviation change in the log 2 ratios of both treatments. Inconsistent log 2 ratios (i.e., the log 2 ratio indicates an increase from AZD6244 but a decrease from MEK162) were not included in the pathway analysis.

Conclusions
Our results demonstrate the utility of the desthiobiotin-ATP probe coupled with LC-MS/MS in assessing adaptive kinome/ATP-binding proteome responses to cancer drugs. This approach could allow mass spectrometry analysis of multiple samples or conditions (e.g., time points) since it does not require a large amount of total protein from each sample and peptide fractionation is unnecessary prior to LC-MS/MS analysis; this approach could be an alternative to kinase inhibitor bead-based kinome profiling and phosphoproteomics. These approaches are complementary and could be combined to identify novel adaptive resistance mechanisms.
Supplementary Materials: The following are available online at www.mdpi.com/2227-7382/4/2/16/s1: Table S1. Full list of ATP-binding proteins identified and quantitated on individual cell lines. Table S2. Full list of altered ATP-binding proteins on individual cell lines. Table S3. Full list of enriched pathways from altered ATP-binding proteins on individual cell lines. Table S4. Full list of kinases identified and quantitated on individual cell lines.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

IRON
Iterative rank-order normalization LC-MS/MS Liquid chromatography/tandem mass spectrometry NSCLC Non-small cell lung cancer PCA Principal component analysis SAPK Stress-activated protein kinase