Oncogenic Receptor Tyrosine Kinases Directly Phosphorylate Focal Adhesion Kinase (FAK) as a Resistance Mechanism to FAK-Kinase Inhibitors

Focal adhesion kinase (FAK) is considered a major cancer drug target due to its vast overexpression in 80% of all solid tumors both at the protein and mRNA level (1–3). In addition, it has been shown to play a critical role in multiple aspects of tumor progression, such as proliferation, survival, invasion, metastasis, angiogenesis, cancer stem cell maintenance, and recently, immune cell suppression (4–11). Multiple studies using transgenic mouse models have shown a defect in tumor growth, invasion, and metastasis with specific FAK deletion (12–15). In further support, FAK overexpression has been clinically correlated with higher tumor stage, metastasis, and in some tumors (breast, lung, endometrium, liver, gastrointestinal system, and esophagus) is associated with poor prognosis (16).

FAK functions as a dual kinase and scaffolding protein where it forms complexes with various oncogenic proteins such as receptor tyrosine kinases (RTK), integrins, and tumor suppressor proteins (17–20). Upon extracellular stimuli, FAK is activated by autophosphorylation of critical residue tyrosine 397 (Y397) through the FAK-kinase domain (21). Phosphorylated Y397 subsequently serves as a canonical SH2-domain docking site for SRC-family kinases, PI3K, and GRB7 (22–24). These signaling molecules then lead to the full activation of FAK through phosphorylation of effector residues (Y861 and Y925) as well as the activation of downstream oncogenic pathways ERK and AKT (25–27).

In an attempt to inhibit FAK in cancer, numerous groups have developed FAK-kinase inhibitors which bind to the ATP-binding pocket of the FAK-kinase domain and block catalysis (28–31). Although very potent (low nanomolar) and specific inhibitors of FAK-kinase activity have been discovered, limited efficacy has been observed in phase I/II clinical trials with lack of mechanistic knowledge explaining this phenomenon (32–34). Furthermore, there are no current molecular markers available that successfully predict the response of FAK-kinase inhibitors in patients, nor are there biomarkers that would predict resistance to FAK-kinase inhibition. As there are 18 reported clinical trials involving FAK-kinase inhibitors (35), the ability to predict both response and resistance to FAK-kinase inhibitors is critical to clinical development.

Along with the canonical FAK activation pathway through autophosphorylation, it is also known that FAK can be activated independently of FAK catalytic activity (19). PDGF- and EGF-stimulated FAK phosphorylation and cell migration occurs independently of FAK-kinase activity in a genetically modified FAK kinase-dead (K454R) model system. However, the tyrosine 397 site via the Y397F mutation was shown to be indispensable for FAK-dependent migration. Despite these data, it is still unknown how Y397 is activated by RTKs and whether multiple oncogenic RTKs can directly phosphorylate FAK at Y397 as a resistance mechanism to FAK-kinase inhibitors.

In this report, we have investigated whether multiple oncogenic RTKs could re-activate FAK and therefore cause resistance to FAK-kinase inhibitors. We found that multiple classes of RTKs directly phosphorylated FAK at Y397 independently of FAK catalytic activity. FAK-kinase inhibitor treatment induced rapid RTK activation, leading to FAK reactivation as well as MAPK/AKT activation. In addition, the initial presence of HER2 activity predicted resistance to FAK-kinase inhibitors in vitro. Together, these studies have identified a novel drug resistance mechanism to FAK-kinase inhibitors through transphosphorylation by RTKs. We call this “oncogenic protection” of FAK due to the variety of oncogenic RTKs that can phosphorylate Y397 to protect a cancer cell's ability to maintain signaling through SH2 domain pathways. Furthermore, we believe that these results have identified a fragile point in FAK signaling pathways that can be exploited with precision-targeted therapeutics directed at FAK Y397 opposed to simply targeting the intrinsic FAK-kinase activity.

Cell lines MDA-MB-231, MDA-MB-453, and MCF7 (courtesy of Katerina Gurova, Roswell Park Cancer Institute), FAK^−/− MEFs (courtesy of Duško Ilić, Kumamoto University School of Medicine), SYF^−/− MEFs (courtesy of Irwin Gelman, Roswell Park Cancer Institute), H292 (courtesy of Pamela Hershberger, Roswell Park Cancer Institute), and MCF7/HER2 tet-off cells (courtesy of Andrei Bakin, Roswell Park Cancer Institute) were cultured at 37°C with 5% CO₂. H292 cells were cultured in RPMI1640 (Gibco, Life Technologies), 10% HI-FBS (Gibco, Life Technologies), 1% penicillin/streptomycin (Gibco, Life Technologies), MEM nonessential amino acids (Cellgro, Mediatech, Inc.), sodium pyruvate (Cellgro, Mediatech, Inc.), and HEPES (Gibco, Life Technologies), and 0.2% Normocin (InvivoGen). Remaining cell lines were cultured in DMEM (Gibco, Life Technologies), 10% HI-FBS (Gibco, Life Technologies), 1% penicillin/streptomycin (Gibco, Life Technologies), and 0.2% normocin (InvivoGen). Cells lines were not submitted for cell authentication services; however, they were instead validated for proper HER2, EGFR, and FAK expression via Western blotting and were tested for Mycoplasma infection via the MycoAlert Mycoplasma Detection Kit (Lonza).

HER2 and HER3 retrovirus was prepared from 293T-Phoenix cells transfected with either pLXSN-HER2 or pLXSN-HER3 vector (courtesy of Dr. H. Shelton Earp III, UNC at Chapel Hill) similarly as described previously (36). Cultured FAK^−/− MEFs were transduced with retrovirus and selected for HER2^High/HER3^High expression using FACS (Roswell Park Flow Cytometry Core). Antibodies used for selection were Alexa Fluor 488 anti-human erbB2/HER-2 (cat # 324410, BioLegend) and APC-conjugated anti-hErbB3/Her3 (cat # FAB3481A, R&D Systems). Stable cells were subsequently transiently transfected with various HA-FAK constructs (WT, Y397F, K454R) using LipoD293 (SignaGen Laboratories) and processed for Western blotting analysis.

His-tagged avian FAK-FERM (AA 31-405) in modified pET vector (provided by Dr. Michael Eck, Harvard University) was expressed in BL21 (DE3) E. coli (Life Technologies) and purified on Ni-NTA resin (Thermo Scientific) using buffers as described previously (37). His-tagged human FAK-CD (AA 677-1052) was cloned into pET15b vector and similarly purified on Ni-NTA resin (Thermo Scientific). GST-FAK-NT (AA 1-415), GST-FAK-KD (AA 416-676), GST-FAK-CD (AA 677-1052), GST-FAK-NT1 (AA 1-126), GST-FAK-NT2 (AA 127-243), and GST-FAK-NT3 (244-415) protein in pGEX-4T1 vector were expressed in BL21 (DE3) E. coli and purified on Glutathione Sepharose 4B resin (GE Healthcare) as described previously (20, 38). GST-HER2-ICD (AA 676-1255), GST-HER2-ICD1 (AA 676-801), GST-HER2-ICD2 (AA 802-1029), and GST-HER2-ICD3 (AA 1030-1255) constructs were designed into the pGEX-4T1 vector, sequenced by the Roswell Park Sequencing Core, expressed in BL21 (DE3) E. coli, and purified on Glutathione Sepharose 4B resin (GE Healthcare). HER2-ECD protein (cat#BMS362) was purchased from eBioscience. Recombinant HER2-ICD (cat# PV3366), EGFR (cat# PV3872), Tie2 (cat# PV3628), EphA2 (cat# PV3688), and FGFR4 (cat# P3054) were purchased from Life Technologies.

Purified FAK-FERM (100 ng) or FAK-CD (100 ng) were incubated with purified HER2 (100 ng) or additional RTK for 30 minutes in the presence or absence of ATP (10 μmol/L) in standard tyrosine kinase buffer: 20 mmol/L HEPES (pH = 7.3), 5 mmol/L MgCl₂, 5 mmol/L MnCl₂, 2 mmol/L DTT, 0.1 mg/mL BSA, and 0.1 mmol/L Na₃VO₄. Proteins were resolved on 4%–20% gradient gels and probed for phosphorylated FAK (Y397), FAK (Y925), HER2 (Y1248), FAK-FERM, FAK-CD, and HER2/RTK using standard Western blotting techniques.

Purified GST-FAK or GST constructs were incubated in NP40 buffer plus 0.1% BSA with purified HER2-ECD and HER2-ICD. GST or GST-FAK constructs were pulled down using Glutathione Sepharose 4B (GE Healthcare Life Sciences) and washed three times with NP40 buffer. Proteins were eluted off of beads in 2× Laemmli buffer (Bio-Rad) and boiled. Samples were resolved on 4%–20% gradient gels and probed for HER2-ECD or HER2-ICD using standard Western blotting techniques and antibodies as described below. Secondary gels were run and stained with SimplyBlue SafeStain to confirm protein loading.

Cell lysates were collected using NP40 lysis buffer containing protease (Roche) and phosphatase inhibitors (Roche) and were incubated on profiler slides according to manufacturer's instructions (Cell Signaling Technology; map in Supplementary Methods). Slides were imaged via chemiluminescence on film and dot intensities were analyzed using ImageJ densitometry software and graphed in GraphPad Prism 6. Phosphorylation (relative to 0 hours) was quantified for each protein by subtracting the negative control dots within each panel, respectively. Subsequently, values were divided by signal obtained at 0 hour to obtain phosphorylation levels relative to 0 hour.

Cells were plated at a density of 1,000 cells per well in a 1:50 solution of Matrigel:complete DMEM culture media on top of a base matrix composed of 1:1 Matrigel:DMEM media. Doxycycline (1 μg/mL) was added for MCF7-HER2 Tet-Off cells. Defactinib was added at various concentrations the day after initial plating. Cell proliferation was evaluated after 5 days using CellTiter AQueous One Solution Cell Proliferation Assay (Promega) according to manufacturer's instructions. Viability was plotted relative to DMSO control and IC₅₀s were calculated using dose–response – Inhibition nonlinear regression algorithm [log(inhibitor) vs. response] in GraphPad Prism 6.

Protein Data Bank files 2AL6 (crystal structure of FAK FERM domain), 3RCD (crystal structure of HER2 kinase domain), 4RIW (crystal structure of EGFR kinase domain), and 2J0J (crystal structure of FAK FERM-Kinase domains) were downloaded and utilized for all modeling experiments. “Active” surface residues determined from CPORT studies were utilized as input residues for restraints to drive the HADDOCK docking process (39). Restraints were only set to “active” residues that corresponded to FAK FERM F1 lobe and HER2 Kinase N-lobe subdomains that were found to be required for HER2-FAK binding in experimental assays. EGFR restraints were solely based from CPORT results. The HADDOCK docking protocol was performed similarly as described previously (40). The top-ranking HADDOCK cluster (based on HADDOCK score and Z-score) was selected and visualized using PyMOL software (Incentive version). Images were ray-traced and saved for publication quality purposes.

Comparisons between two groups were made using a Student t test (GraphPad Prism 6). Data were considered significant when P < 0.05. Two-way ANOVA and Tukey multiple comparison test were used to calculate significance when comparing multiple groups within the same experiment (GraphPad Prism 6).

FAK has been shown to be primarily activated at Y397 by autophosphorylation through the FAK-kinase domain. Whereas some data suggest that RTKs can signal through FAK independently of FAK-kinase activity (19), it is still unclear how RTKs activate FAK and whether this alternative pathway plays a role in cancer cell resistance to FAK-kinase inhibitors. Thus, we hypothesized that oncogenic RTKs could directly transphosphorylate FAK Y397 as a drug resistance mechanism to bypass FAK-kinase inhibition. To test this, in vitro studies were carried out in HER2⁺ MDA-MB-453 and SkBr3 breast cancer cells as well as EGFR⁺ H292 and A549 lung cancer cells (Fig. 1 and Supplementary Figs. S1 and S2; Supplementary Table S1) treated with three different FAK-kinase inhibitors (defactinib, PF-228, and PF-271), two of which have been in clinical trials (28, 29, 32, 34, 41, 42). We used 10 μmol/L of FAK-kinase inhibitor treatment to ensure full inactivation of the FAK enzyme in a variety of cell lines (Supplementary Figs. S3 and S4). HER2 activation by Heregulin (HER2-activating ligand) reactivated FAK Y397 phosphorylation after treatment with all three FAK-kinase inhibitors in both MDA-MB-453 and SkBr3 cell lines, suggesting that FAK Y397 phosphorylation could be maintained independent of its intrinsic kinase activity. While the effects of EGF stimulation on FAK reactivation were not as robust as HRG stimulation, EGF stimulation slightly reactivated FAK pY397 after treatment with PF-228 in H292 cells. Nonetheless, growth factor stimulation activated both AKT and ERK pathways in the presence of all three FAK-kinase inhibitors in all four cell lines, indicating that FAK-kinase activity was dispensable for downstream pathway activation. Conversely, FAK depletion using FAK-null MEFs or FAK siRNA in MDA-MB-453 cells partially diminished HER2-dependent activation of AKT/ERK, indicating that FAK total protein, however not kinase activity, could regulate downstream AKT/ERK pathways (Supplementary Fig. S5). We also noticed that even in the absence of HRG or EGF stimulation, all three FAK-kinase inhibitors stimulated phosphorylation of HER2/AKT in MDA-MB-453 cells as well as HER2/AKT/ERK in SkBr3 cells and EGFR/AKT/ERK in A549 cells (Fig. 1). These data indicated a novel drug resistance pathway to FAK-kinase inhibitors through the reactivation of FAK Y397 and downstream AKT/ERK pathways by oncogenic RTKs.

To verify that the compensatory increases in HER2, AKT, and ERK after FAK-kinase inhibitor treatment were in fact due to FAK-kinase inhibition and not off-target kinase effects of the 10 μmol/L dose, we revisited our dose-titration experiments using clinical-stage FAK inhibitor, defactinib (Supplementary Figs. S3 and S4). In MDA-MB-453 and SkBr3 cells, once FAK pY397 levels were decreased by defactinib treatment (0.3–1.0 μmol/L and 0.004 μmol/L respectively), compensatory increases in pHER2, pAKT, and pERK were also observed. In MDA-MB-453 cells, maximal FAK inhibition (10–30 μmol/L) caused maximal compensatory increases in pAKT and pERK. Intriguingly, at higher doses of defactinib (3 μmol/L) in SkBr3 cells, pHER2 and pAKT levels were maximally increased and FAK pY397 levels were paradoxically restored. In H292 and A549 cells, compensatory increases in pEGFR and pERK were observed after low-dose defactinib treatment (0.001–1 μmol/L). We also noticed that 0.01–0.3 μmol/L defactinib treatment in H292 cells induced paradoxical increases in FAK pY397 that were in alignment with increases in pEGFR, suggesting that FAK-kinase inhibition led to FAK hyperphosphorylation by EGFR. To further confirm the FAK specificity of these compensatory increases, we performed FAK siRNA studies in SkBr3 cells using all three FAK-kinase inhibitors (Supplementary Fig. S6). FAK knockdown by siRNA itself induced the phosphorylation of AKT, ERK, and the reactivation of FAK pY397. In addition, FAK siRNA reduced the relative increase in pAKT and pERK after FAK-kinase inhibitor treatment. In all, these data supported that the compensatory increases in pHER2, pAKT, and pERK upon FAK-kinase inhibition were indeed due to the specific inhibition of FAK.

Next, we used genetic studies in FAK-null MEFs to confirm the phenomenon of RTK-driven Y397 phosphorylation (Fig. 2A and B, and Supplementary Fig. S7). Both WT-FAK and K454R-FAK (kinase-dead) were found to be rephosphorylated at Y397 upon stimulation of both HER2/HER3-stably expressing and EGFR⁺ FAK-null MEFs. In addition, studies in SRC/YES/FYN (SYF)-null MEFs confirmed FAK Y397 phosphorylation by HER2 independently of known SYF adaptor proteins (Supplementary Fig. S8). These data demonstrated that Y397 can still be phosphorylated by RTKs independently of both FAK and SRC activity, consistent with the RTK-dependent drug resistance we observed with all three FAK-kinase inhibitors.

Because HER2 reactivated kinase-dead FAK phosphorylation at Y397 in MEFs, we then hypothesized that HER2 could maintain FAK-dependent biological functions as well. To test this, we performed transwell migration and invasion assays utilizing HER2/HER3 stably expressing FAK-null MEFs that were transiently transfected with WT-FAK, K454R-FAK, or Y397F-FAK. As shown in Fig. 2C, HRG-stimulated migration was enhanced by not only WT-FAK, but K454R-FAK as well. Conversely, only mutation of Y397 (Y397F-FAK) was found directly to completely block HRG-stimulated migration. Intriguingly, basal cell migration (-HRG) was fully inhibited by K454R mutation, suggesting that basal migration, but not HER2-stimulated migration, is regulated by FAK-kinase activity. These findings were confirmed in cell invasion assays as well (Fig. 2D). Kinase-dead FAK (K454R) moderately reduced HRG-stimulated invasion compared with WT-FAK; however, only Y397F-FAK completely abrogated FAK-enhanced cell invasion. These data showed that HER2 can maintain FAK-dependent biological functions (migration and invasion) under kinase inhibition as a result of the FAK transphosphorylation phenomenon.

Because FAK was reactivated by RTKs independently of both FAK and SRC activity, we hypothesized that RTKs were directly binding to FAK to promote phosphorylation. Although HER2 has been previously shown to colocalize and form a complex with FAK in cells, it is still unknown whether HER2 and FAK form a direct protein–protein interaction (43, 44). To further confirm the mechanism of direct FAK activation by HER2, we performed a series of GST pull-down assays with purified recombinant proteins. GST-FAK constructs were cloned according to FAK-NT (N-terminus), -KD (kinase domain), or -CD (C-terminal domain) regions and fusion proteins were incubated with either HER2-ICD (intracellular domain) or -ECD (extracellular domain) proteins (Fig. 3A and Supplementary Fig. S9A). Significant binding was detected between GST-FAK-NT and HER2-ICD, but minimal binding was observed for FAK-KD and -CD regions. As expected, no binding was detected between GST-FAK constructs and HER2-ECD protein.

To further define the binding region between HER2 and FAK, we performed similar assays with GST-FAK-NT1, NT2, and NT3 proteins cloned according to the F1, F2, and F3 lobes of the FAK N-terminal FERM domain (Fig. 3B and Supplementary Fig. S9B). The FAK NT1 (or F1 lobe) region, but not NT2 or NT3, was found to bind to HER2-ICD. In addition, we tested various HER2 segments with GST-HER2-ICD, ICD1, ICD2, and ICD3 proteins cloned according to whole ICD, kinase domain N-lobe, kinase domain C-lobe, and C-terminal tail regions of the HER2 ICD (Fig. 3B and Supplementary Fig. S9C). Both the HER2-ICD and ICD1 (or kinase N-lobe) regions, but not ICD2 or ICD3, were found to bind to FAK-NT. These data confirmed the direct interaction of FAK and HER2 between the FAK FERM F1 lobe and HER2 kinase N-lobe.

We then examined the X-ray crystal structure of the FAK-FERM domain (PDB 2AL6) and observed that Y397, located within the flexible FERM-Kinase linker region, binds back onto the structured FERM F1 lobe where HER2 interacts (Supplementary Fig. S10). As Fl lobe binding could provide proximity of HER2 to Y397, we utilized HADDOCK protein–protein docking to develop a structural model for HER2-FAK binding and putative Y397 transphosphorylation (40). FAK FERM (PDB 2AL6) and HER2 kinase domain (PDB 3RCD) crystal structures were computationally analyzed for active surface residues likely to be involved in protein–protein interactions using the program CPORT (39). Subsequently, CPORT-predicted residues that overlapped with experimentally validated amino acids from GST pull-down assays were set as restraints to drive the HER2-FAK docking model. Intriguingly, HADDOCK produced a HER2-FAK structural model similar to a prototypical kinase–substrate interaction (Fig. 3C). As observed in Fig. 3D, the FAK FERM F1 lobe makes contact with the HER2 N-lobe to orient the flexible Y397 FERM-linker segment close to the HER2 substrate–binding region and ATP-binding pocket. HADDOCK docking was also performed similarly with the EGFR kinase domain (PDB 4RIW) to approximate a model between EGFR and FAK (Supplementary Fig. S11). In agreement with the HER2 model, the FAK FERM domain was predicted to bind EGFR through the F1 lobe, facilitating orientation of Y397 into the EGFR ATP-binding pocket. These data provided structural rationale for the HER2–FAK interaction, where binding of HER2 (and potentially other RTKs) to the F1 lobe promotes proximity to Y397, allowing HER2 to directly phosphorylate Y397.

Direct HER2-FAK binding assays as well as our HADDOCK docking model suggested that HER2, and potentially other RTKs, could directly activate FAK at Y397 (Fig. 3). To evaluate direct phosphorylation and experimentally validate our HADDOCK model, we tested whether HER2 could directly phosphorylate FAK using an in vitro kinase assay with purified HER2-ICD, FAK-FERM, and FAK-CD proteins (Fig. 4A and B). In agreement with our HADDOCK model, HER2 directly phosphorylated FAK-FERM specifically at Y397. However, no phosphorylation was detected at FAK-CD Y925, confirming the specificity of the kinase reaction. Intriguingly, multiple RTKs (EGFR, Tie2, EphA2, FGFR4) were also found to directly phosphorylate FAK at Y397 (Fig. 4C), indicating a redundant FAK activation pathway by multiple RTKs. These data supported a direct Y397 phosphorylation mechanism by HER2 and other RTKs, where HER2 directs interacts with the FAK-FERM F1 lobe to phosphorylate Y397.

As mentioned previously, even in the absence of growth factor stimulation, all three FAK-kinase inhibitors stimulated the phosphorylation of RTK/AKT/ERK pathways as a drug resistance mechanism (Fig. 1). Given our data of FAK reactivation by RTKs, we hypothesized that cancer cells might induce novel RTK signatures (rapid vs. long-term depending on initial RTK expression) in response to FAK-kinase inhibitors as a mechanism to maintain FAK and AKT/ERK signaling. To test this, we performed time course experiments in RTK^High cell lines (MDA-MB-453, SkBr3, H292) as well as RTK^Low cell lines (MDA-MB-231, MDA-MB-468) and evaluated changes in phosphorylation patterns. We focused on defactinib due to its current involvement in multiple ongoing clinical trials and its recently failed phase II clinical trial (34). In addition, we used a dose of 1 μmol/L, which is in alignment with patient serum concentrations of defactinib achieved with the recommended phase II dose of 425 mg twice daily (42). In MDA-MB-453 cells, defactinib treatment induced rapid activation of HER2, EGFR, AKT, and ERK within 15 minutes, with maximal activation of HER2 and EGFR coming at 1 and 4 hours, respectively (Fig. 5A and Supplementary Fig. S12A). When FAK-kinase inhibition was lost and pY397 levels were restored at 48–72 hours, pHER2 and pEGFR levels also returned to basal levels.

In addition, we utilized RTK arrays to assess a broader kinome response to FAK-kinase inhibition (Fig. 5 and Supplementary Figs. S13, S14; Table S2). In H292 cells, we found consistent rapid activation of RTK pathways as observed in MDA-MB-453 cells, with increases in EGFR, HER2, ERK, and AKT from 15 minutes to 4 hours after treatment. Immunoblots were used to confirm our findings of EGFR, AKT, and ERK activation after defactinib treatment (Fig. 5B and Supplementary Fig. S12B). Intriguingly, S6 ribosomal protein (S6 RP), a marker of increased translation of mRNA transcripts encoding for proteins regulating cell-cycle progression, was rapidly activated at 15–30 minutes in two of three RTK^High cell lines. These data showed that RTK^High cancer cells are involved in a rapid feedback loop, where FAK-kinase inhibition induces corresponding RTK activation to compensate for loss of FAK Y397 phosphorylation. Conversely, in RTK^Low cell lines, the RTK reprogramming in response to defactinib treatment occurred at longer timepoints (48 and 72 hours). In MDA-MB-231 triple-negative (ER/PR/HER2⁻) breast cancer cells, defactinib induced the expression of HER2 and EGFR as well as the activation of other RTKs such as FGFR4 and EphA2 at 72 hours. Immunoblots were used to confirm to total protein increases in both EGFR and HER2 after defactinib treatment (Fig. 5C and Supplementary Fig. S12C). These data demonstrated that RTK^Low cancer cells require longer time points to undergo kinome changes, most likely due to the requirement to express new RTKs. In addition, these data showed that the selective pressure of FAK-kinase inhibition is able to drive triple-negative breast cancer cells to express HER2.

On the basis of our findings of rapid versus long-term RTK reprogramming in RTK^High versus RTK^Low cells, we predicted that cancer cells driven by oncogenic RTKs would be more resistant to FAK inhibition than those that are not driven by RTKs. We utilized 3D Matrigel-on-top growth assays to measure of the efficacy of FAK-kinase inhibitors, as the 3D-environment of these assays were shown to recapitulate the cellular requirement for FAK (45, 46). Indeed, HER2-addicted breast cancer cell lines MDA-MB-453 (IC₅₀ ND) and SkBr3 (IC₅₀ > 10 μmol/L) showed no changes in viability in response to defactinib treatment in 3D growth assays, whereas defactinib treatment decreased the viability of HER2⁻ cell line MDA-MB-231 (IC₅₀ = 0.281 μmol/L) in a dose-responsive manner (Fig. 6A). We confirmed these findings using the MCF7-HER2 Tet-Off system, where HER2 expression was modulated by the simple removal or addition of doxycycline in this isogenic system (Fig. 6B). HER2^High (−Dox) cells showed minimal response to defactinib (IC₅₀ = 1.58 μmol/L), whereas HER2^Low (+Dox) cells were quite responsive to defactinib treatment (IC₅₀ = 0.052 μmol/L). Together, these data showed that HER2 expression levels can distinguish cancer cells that may or may not respond to FAK-kinase inhibitors and demonstrated that RTKs such as HER2 can drive resistance to FAK-kinase inhibitors.

The data described in this report have identified a novel mechanism of drug resistance that may explain some of the failures of FAK-kinase inhibitors in clinical trials. Here, we have shown that oncogenic RTKs, such as EGFR or HER2, directly phosphorylated FAK at Y397 to rescue kinase-inhibited FAK in cancer cells. In addition, HER2 formed a direct protein–protein interaction complex with FAK nearby this critical residue Y397. We also have shown that RTK^High cell lines displayed rapid resistance to FAK-kinase inhibitors due to upregulation of RTK signaling pathways while RTK^Low cells were initially sensitive to the inhibitors. However, RTK^Low cell lines, such as MDA-MB-231, started to express HER2 and EGFR protein after longer incubations with defactinib, explaining a possible acquired resistance mechanism to FAK-kinase inhibitors. An overview of our findings is summarized in Supplementary Fig. S15. In all, we have identified a common resistance mechanism to FAK-kinase inhibitors, whereby RTKs can bypass FAK-kinase inhibition to rephosphorylate FAK Y397.

In our analysis, we noticed that RTK^High cell lines showed a trend whereby RTK activation led to increased FAK pY397, but we also observed that different RTK^High cell lines had different capacities to rephosphorylate FAK at Y397. In addition, EGFR⁺ cell line, A549, showed minimal reduction of pY397 by FAK-kinase inhibitors and compensatory increases in pEGFR, suggesting an innate resistance to these inhibitors. Also observed was a temporal difference between cell lines in their ability to induce compensatory RTK reprogramming after defactinib treatment. To explain these results, we propose that the ability of RTKs to transphosphorylate FAK at Y397 is very cell line-dependent, where multiple variables (i.e., RTK levels, mutation status, adaptor protein levels) can affect both the levels and kinetics of Y397 phosphorylation. Future experiments will be designed to evaluate other cellular factors that may drive Y397 phosphorylation and FAK-kinase inhibitor resistance.

Of additional notice was the strong compensatory activation of HER2/AKT/ERK pathways as a drug resistance mechanism to FAK-kinase inhibitors. We acknowledge that high-dose kinase inhibitor treatment has the potential for off-target kinase effects. Nonetheless, our data with low concentrations of defactinib and FAK-specific siRNA suggest that the compensatory RTK reprogramming after FAK-kinase inhibitor treatment was FAK-specific. Kinase inhibitor off-target kinase effects are generally inhibitory (due to the conserved ATP-binding region) and not agonistic in nature. In addition, the chemical structures of all three FAK-kinase inhibitors (defactinib, PF-271, PF-228) are different and therefore display different off-target kinase effects as observed (28, 29, 41). However, all three FAK-kinase inhibitors still activated HER2/EGFR and AKT/ERK in a similar manner, supporting the notion that these compensatory increases are FAK-specific. Furthermore, as mentioned previously, the recommended phase II dose of defactinib is 425 mg twice daily, and at steady-state dosing (day 15) the average serum C_max was greater than 1.0 μg/mL, or 1.95 μmol/L, indicating that our experimental doses are not only appropriate for cell culture studies, but are also in line with circulating levels of drug in patients (42). Therefore, we believe our data describe a clinically relevant drug resistance mechanism to FAK-kinase inhibitors at concentrations utilized in current clinical trials.

Surprisingly, we found that multiple RTKs from different family members could directly phosphorylate FAK FERM at Y397. In addition, we found a wide array of upregulated RTKs in response to the FAK-kinase inhibitor in widespread clinical trials, defactinib. These findings suggest there are redundant transphosphorylation mechanisms where cancer cells can hijack multiple different RTK pathways (depending on cellular context and genomic accessibility) to maintain FAK Y397 pathway activation during FAK-kinase inhibition. Also, FAK depletion by siRNA in SkBr3 cells induced the hyperactivation of FAK Y397 despite lower levels of total FAK, implying cancer cell dependency on FAK Y397 phosphorylation. Therefore, we propose FAK Y397 phosphorylation as a fragile point in cancer cells, required to maintain cancer cell motility and proliferation. We have termed our observation “oncogenic protection of FAK” because of the ability of different oncogenic RTKs to phosphorylate and protect this fragile Y397 site. In fact, several other groups have shown the requirement of FAK for tumor initiation and progression (12, 13). In addition, another group has implicated the FERM domain in binding of FAK to RTKs in cells, but did not show a direct interaction (19). We have demonstrated a direct interaction of FAK and HER2 and further hypothesize that multiple RTKs bind to the FAK-F1 lobe as a means to directly phosphorylate Y397. Although our direct kinase and cellular data support the redundant phosphorylation of FAK Y397 by multiple RTKs, we acknowledge the known promiscuity issue with in vitro kinase assays (47). Future cellular evaluation of novel RTK-FAK signaling complexes as well as in vivo drug resistance models will be helpful as we seek to use precision RTK biomarkers of a patient's tumor to predict their response/resistance to FAK-kinase inhibitors.

Several recent reports have characterized a novel drug resistance mechanism to kinase inhibitors involving rapid dynamic kinase signaling called kinome reprograming (48–50). These groups have identified global changes in cancer cell kinome expression and phosphorylation events after kinase inhibitor treatment. Interestingly, MEK inhibitor and HER2 inhibitor treatment induced global upregulation of redundant intracellular signaling nodes, RTKs, and MAPK/AKT signaling pathways. Our data describe a novel and previously uncharacterized form of compensatory RTK reprogramming to FAK-kinase inhibitors through the rephosphorylation of FAK Y397 but are in agreement with these studies where other kinase inhibitors induced global changes in kinase expression/activity. It remains to be determined whether FAK-kinase inhibition induces both whole kinome as well as nonkinase forms of drug resistance.

At the time this manuscript was prepared, there were 18 reported clinical trials involving FAK inhibitors on clinicaltrials.gov (35). Although FAK is considered a promising drug target for cancer therapy, questions still remain on which function of FAK (kinase vs. scaffold) is optimal to target and under which molecular context (17). Several preclinical studies have shown that molecular markers, such as Merlin^Low expression, can be utilized to stratify tumor subtypes which may be responsive to FAK-kinase inhibitors (46). Unfortunately, clinical trial results of the COMMAND (Control of Mesothelioma with MAiNtenance Defactinib) trial have shown limited efficacy and no difference in control versus defactinib-treated patients with Merlin^Low tumors (34). As such, it is important to identify new molecular markers, such as HER2 and other RTKs that may have the ability to predict which population of patients will or will not respond to FAK-kinase inhibitor therapy.

Finally, the data presented here have several direct clinical implications regarding FAK-kinase inhibitors. First, we show that expression of RTKs in cancer cells, such as HER2, can cause resistance to FAK-kinase inhibitors due to rephosphorylation of Y397. This suggests an opportunity to retrospectively analyze tumor samples from patients who have received FAK-kinase inhibitors to determine the levels of specific RTKs such as HER2 and EGFR and correlate them with clinical response. Second, our data suggest that RTK^Low cancers will be initially sensitive to FAK-kinase inhibitors and that RTK expression can be induced as a mechanism of acquired resistance. This suggests the need to regularly monitor tumors for changes in their kinome that could predict acquired resistance to the FAK-kinase inhibitor. Third, these studies provide a molecular rationale for combination therapy of FAK-kinase inhibitors with RTK inhibitors such as lapatinib or erlotinib. However, we predict eventual resistance to combination therapy, due to the observed upregulation of diverse RTKs in RTK reprogramming assays. As an alternative, the discovery and development of FAK-scaffold inhibitors that directly inhibit FAK Y397 may provide a more durable response to FAK-based therapy (51) by blocking the fragile Y397 site to globally prevent its phosphorylation. As we further understand the dynamic nature that surrounds FAK and its complex interactome, we believe that more precision-targeted therapeutics will be elucidated.

W.G. Cance has immediate family members with ownership interest (including licensed intellectual property) in FAKnostics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.A. Marlowe, F.L. Lenzo, S.A. Figel, W.G. Cance

Development of methodology: T.A. Marlowe, F.L. Lenzo, S.A. Figel, W.G. Cance

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.A. Marlowe, F.L. Lenzo, S.A. Figel

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.A. Marlowe, F.L. Lenzo, S.A. Figel, W.G. Cance

Writing, review, and/or revision of the manuscript: T.A. Marlowe, F.L. Lenzo, S.A. Figel, W.G. Cance

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.A. Marlowe, F.L. Lenzo, S.A. Figel, A.T. Grapes

Study supervision: W.G. Cance

We would like to thank Drs. Vita Golubovskaya, Elena Kurenova, and Shelton Earp for scientific discussions regarding the research topic, Mr. Thomas Mathers for pharmaceutical industry advice, the Roswell Park Flow Cytometry Core for performing FACS analysis, and Mr. George Clinton for creative influences.

This work was funded by the NCI (R01CA065910) awarded to W.G. Cance, a Ruth L. Kirschstein National Research Service Award (NRSA) Institutional Research Training Grant (T32CA009072) awarded to T.A. Marlowe, and a NCI Cancer Center Support Grant for Roswell Park Cancer Institute (P30CA016056).

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.