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
] and feedback activation of B- and C-RAF upon MEK inhibition, especially in the context of RAS-driven cancers [5
]. This adaptive resistance is complicated by the diversity of kinases and enzymes expressed in cancer cells functioning in interconnected networks; thus, systematically examining how cancer cells respond to drugs could offer novel insights into adaptive resistance mechanisms and point toward effective drug combinations [7
Mass spectrometry-based quantitative proteomics has been employed to assess altered signaling networks after drug treatment at a system-wide level. Gary Johnson and his colleagues developed a chemical proteomics method employing multiplexed kinase inhibitor beads followed by mass spectrometry analysis, allowing a system-wide measurement of drug-induced kinase activity/expression. This platform was employed to assess kinome adaptations to kinase inhibitors in multiple cancer models, including triple-negative breast cancer treated with MEK inhibitor selumetinib [9
] and HER2-positive breast cancer treated with HER2 inhibitor lapatinib [10
]. These studies highlighted drug-induced dynamic kinome reprogramming involving reactivation of co-expressed receptor tyrosine kinases in response to kinase inhibitors. The multiplexed kinase inhibitor bead approach also revealed differential kinome expression/activity between parental and leukemia cells with acquired drug resistance to BCR-Abl inhibitor imatinib [11
As an alternative approach that has shown utility in charting rapid dynamic responses, profiling drug-induced global phosphoproteome changes has identified key adaptive changes linked to drug resistance. A tyrosine phosphorylation profiling in DDR2
mutant squamous lung cancer cells treated with its tyrosine kinase inhibitor dasatinib revealed key compensatory receptor tyrosine kinase activations linked to intrinsic drug resistance [12
]. This approach also revealed differential drug-induced tyrosine phosphoproteome responses between naïve and drug-resistant EGFR mutant lung cancer cells to EGFR tyrosine kinase inhibitor [13
]. Global phosphoproteome (phospho-Ser/Thr/Tyr) profiling showed that ablation of TBK1 expression in KRAS mutant lung cancers leads to compensatory activation of a panel of receptor tyrosine kinases including EGFR and MET [14
]. One limitation of phosphoproteomic approaches is the requirement of large amounts of protein lysates (usually ~30–50 mg) and/or peptide fractionation (usually 12 fractions per sample), which restricts the number of samples or conditions to be analyzed in a practical and reasonably economical mass spectrometry experiments. Finally, important information on adaptive responses driven by other ATP-binding enzymes could be missed by focusing solely on phosphoproteomics.
We hypothesized that we could employ another approach to study adaptive resistance and kinase rewiring using a commercially available desthiobiotin-ATP probe (ActivX, Thermo Scientific), which covalently labels conserved lysine residues in or near the ATP-binding pocket of enzymes, including kinases [15
]. Peptides containing the labeled lysine residues are then enriched by streptavidin beads, identified and quantitated by liquid chromatography-tandem mass spectrometry (LC-MS/MS). This approach is an alternative way to overcome the aforementioned disadvantages since it requires relatively small amount of samples (1 mg) and requires no fractionation for LC-MS/MS analysis. It allowed us to test how multiple KRAS mutant lung cancer cell lines differentially respond to MEK inhibition and reveal heterogeneous ATP-binding proteome responses from each individual cell line.
Here, we profiled ATP-binding proteome responses to two clinical MEK inhibitors, AZD6244 and MEK162, in the context of KRAS mutant lung cancer. KRAS mutations occur in nearly 30% of non-small cell lung cancers (NSCLC), yet therapeutic targets for these cancers have not been realized. MAPK signaling has been known to be essential for KRAS-induced lung tumorigenesis [16
], and pharmacological inhibition of this pathway (e.g., MEK inhibitor) has been attempted to treat KRAS-driven lung cancers. However, significant clinical responses are still lacking, in part due to the cancer cells’ ability to re-activate ERK via feedback activation of RAF [5
]. Recent studies indicated that mutational status of tumor suppressors, p53 or LKB1, in KRAS mutant lung cancer could modulate drug responses to MEK inhibitor AZD6244 [19
] and immune checkpoint inhibitors [20
]. This raises the possibility that heterogeneous adaptive responses could exist in KRAS mutant lung cancer depending on the status of co-mutated tumor suppressors, further complicating the development of a rational co-targeting strategy. For this study, we employed multiple KRAS 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.
2. 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 2
A). 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 2
B). 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 log2
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 2
C 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 log2
(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 2
D 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
], 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 3
a and Table S4
); however, each cell line showed unique drug-induced altered kinase list (Figure 3
b 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
]. However, significant clinical responses of mitotic kinase inhibitors are still lacking in lung cancer [31
]. 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
]. 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
] 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
], 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
]. 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
] or pharmacologic vulnerability screens [43
] could be combined to assess translational potential of our results.