Targeting FBW7 as a Strategy to Overcome Resistance to Targeted Therapy in Non–Small Cell Lung Cancer Free

Activating mutations in EGFR and rearrangements in anaplastic lymphoma kinase (ALK) serve as predictive biomarkers for clinical benefits in non–small cell lung cancer (NSCLC) patients who receive EGFR and ALK inhibitors (1–3). Therapeutic strategies using tyrosine kinase inhibitors (TKI) to inhibit EGFR and ALK signaling have become first-line therapies for these patients. Gefitinib and crizotinib are two paradigms of first-generation TKIs with robust clinical responses and improved outcomes (4–7). Unfortunately, almost all patients relapsed on TKIs within 1 to 2 years of treatment. Several mechanisms that drive resistance have been identified. The most common molecular alteration is the emergence of resistance mutations. For example, the EGFR T790M “gatekeeper” mutation mediates gefitinib resistance in more than 50% of cases (8, 9), and the C1156Y and L1196M mutations within the ALK tyrosine kinase domain drive resistance to crizotinib in vitro and in vivo (10). In addition, amplification or activation of bypass kinases, such as c-Met, ErbB2 and c-kit, have also been proposed (11). By means of both mechanisms, cancer cells evade apoptosis and become resistant to targeted therapy by maintaining PI3K/Akt and MEK/ERK signaling. However, the underlying mechanism of resistance in over 20% of patients remains elusive. Thus, the identification of biomarkers that predict patient responses to TKIs would greatly aid clinicians in the discovery of novel therapeutic strategies.

The SCF–E3 complex, consisting of Skp-1, Cul-1, F-box protein, and Rbx-1, represents the largest E3 ubiquitin ligase family that regulates protein ubiquitination and degradation in eukaryotes. It has been extensively documented that the substrate specificity targeted by the SCF-E3 complex is determined by each F-box protein through the WD40 domain or LRR domain (12). There are more than 70 F-box genes in the human genome, and three of them (Skp2, β-TrCP, and FBW7) are well characterized. Skp2 and β-TrCP are often overexpressed in cancers and considered as potent oncogenes (13, 14). In contrast, the heterozygous inactivation of FBW7 has been identified in many cancers, with an overall frequency of 6% (15–17). The importance of FBW7 is underscored by the observation that Fbw7^+/− mice are more susceptible to radiation-induced carcinogenesis than Fbw7^+/+ mice, suggesting FBW7 has a tumor-suppressive function (18–20). However, knowledge regarding the exact role of FBW7 has not been reported in EGFR-mutated and ALK-rearranged cancers, and its correlation with targeted therapy resistance is largely unknown.

Myeloid cell leukemia sequence 1 (MCL-1) is a member of the BCL-2 family that negatively manipulates the intrinsic apoptotic pathway by modulating mitochondrial integrity. Unlike other BCL-2 members, MCL-1 is very unstable and regulated through a host of transcriptional, posttranscriptional, and posttranslational mechanisms that allow for rapid and tight control of MCL-1 abundance. MCL-1 is frequently amplified in cancers and often associated with poor response to chemotherapy (21). Genetic silencing of MCL-1 leads to apoptosis in a broad spectrum of cancer cells, arguing that it is a promising therapeutic candidate for precision medicine. Unfortunately, the development of cancer-selective MCL-1 inhibitors has proven to be technically challenging, and several compounds under early stages of investigation have failed to show single-agent efficacy (22). There is an urgent need to develop alternative approaches to target MCL-1 for cancer research and treatment.

Here, we report previously undescribed PI3K/Akt and MEK/ERK signaling–independent machinery that drives resistance to TKIs. Depletion of FBW7 leads to TKIs resistance through stabilization of the MCL-1 protein. We also suggest a model of how phosphorylation of MCL-1 protein Ser159 governs its subcellular localization and degradation. Collectively, our study reports a novel biological function of FBW7 E3 ligase in regulating sensitivity to targeted therapy. Screening FBW7 expression may help to predict responsiveness, and targeting FBW7 helps to overcome resistance to targeted therapy.

The PC-9, HCC827, HCC827 GR5, H3122, H3255, H1975, H1299, and H1650 NSCLC cell lines have been extensively described. All the cell lines were obtained between 2012 and 2015 and validated by short tandem repeat analysis, tested for mycoplasma contamination within the last 6 months, and used at passage numbers <10. PC-9, HCC827, HCC827 GR5, H3122, H1975, H1299, and H1650 cells were cultured in RPMI1640 (Corning) with 10% FBS (Gibco) and 1% penicillin/streptomycin (Invitrogen). H3255 cells were cultured in ACL-4 medium (Invitrogen) supplied with 10% FBS as described previously. Mouse embryonic fibroblasts (MEF) from Fbw7^flox/flox mice were prepared following the standard protocol and cultured in DMEM (Invitrogen) with 10% FBS and 1% penicillin/streptomycin.

All the compounds were purchased from Selleck Chemicals, except for gefitinib (AstraZeneca), d-Luciferin (Promega), cycloheximide (Sigma), puromycin (Sigma), LY294002 (Calbiochem), and leptomycin B (Tocris). Each compound was dissolved in DMSO for cell culture experiments.

shRNAs against FBW7, Cul-1, and Skp1 were obtained from the Biological Resource Center of NIBS (Supplementary Table S1). shRNA lentivirus was produced by transfecting HEK293 cells with pLKO.1 shRNAs together with packaging plasmids psPAX2 and pMD2.G using Lipofectamine 2000 (Invitrogen). Culture supernatants containing lentivirus particles were collected at 48 and 72 hours after transfection and stored at −80°C. Cells were infected with the lentivirus supernatants and selected with 2 μg/mL puromycin.

Overexpression plasmids for mutant EGFR (delE746_A750), Myc-FBW7α, Myc-FBW7β, Myc-FBW7γ, His-Ub, His-Ub K48 only, His-Ub K63 only, HA-Ub, Flag-MCL-1, GST-MCL-1, HA-JNK, HA-GSK3β, HA-Erk1, HA-Myr-Akt, and HA-DN-Akt were preserved in our in-house construct bank. A QuikChange Site-Directed Mutagenesis Kit (Stratagene) was used to generate Myc-FBW7ΔF, Myc-FBW7ΔWD40, Myc-FBW7 R465H, Myc-FBW7 R505L, Flag-MCL-1 K5R, Flag-MCL-1 K40R, Flag MCL-1 K136R, Flag MCL-1 K194/197R, Flag-MCL-1 K279R, Flag-MCL-1 K308R, Flag-MCL-1 ΔPEST, Flag-MCL-1 ΔNLS, Flag-MCL-1 S159A, Flag-MCL-1 T163A, Flag-MCL-1 S159A/T163A, HA-GSK3β S9A, and HA-GSK3β K85M mutants. The PCR primer sequences for mutagenesis are available in Supplementary Table S1. All the constructs used in our study have been thoroughly sequenced.

Cells were seeded in triplicate at 2 × 10⁴ cells per well in 96-well plates. After waiting 12 hours to allow attachment, cells were transfected with siRNA pools targeting each gene using Lipofectamine 2000 according to the manufacturer's recommendations. Forty-eight hours after transfection, cell viability was determined by the MTT assay (Sigma) as described previously. siRNA sequences are shown in Supplementary Table S2.

Total RNA was extracted with TRIzol reagent (Invitrogen). First-strand cDNA was prepared from 1 μg of total RNA, and qPCR was performed following the standard protocol (Takara). qPCR primers sequences are shown in Supplementary Table S3. The relative amount of each gene was normalized to the amount of β-actin and calculated using the 2^−ΔΔC_t method.

HEK293 cells were transfected with the indicated plasmids for 48 hours and lysed in lysis buffer (Pierce) with a protease/phosphatase inhibitor cocktail (Roche). The cell extracts were then incubated with Ni-NTA agarose (Qiagen) for 3 hours, eluted with TI buffer (25 mmol/L Tris-HCl and 20 mmol/L imidazole, pH 6.8), and subjected to Western blot analysis.

Subcellular protein fractionation was performed following the manufacturer's instructions (Thermo Scientific). Tubulin and Lamin A/C were used as equal loading control for cytoplasmic and nuclear proteins, respectively.

Cell apoptosis was measured using an Annexin V-FITC/PI Kit (Millipore). Flow cytometry was performed using a Becton Dickinson FACS-420 flow cytometer.

After the indicated treatment, cells were harvested and lysed. The cell extracts were incubated with MCL-1 antibody, Flag tag antibody, or Myc tag antibody with gentle rotation overnight at 4°C. The immunocomplex was pelleted with Protein A/G agarose beads (Santa Cruz Biotechnology), resuspended, and boiled with 2× SDS loading buffer and subjected to Western blot analysis. For Myc-Flag tandem pull down, cell extracts were incubated with high-affinity Myc resin (BioTools) overnight at 4°C and eluted with Myc tag peptide (BioTools). A second round of immunoprecipitation was performed using anti-Flag resin (BioTools).

Equal amount of proteins (10–30 μg) were separated by SDS-PAGE gel and transferred onto a nitrocellulose membrane (Millipore). The membrane was blocked with 5% non-fat milk and incubated with primary antibodies overnight at 4°C (Supplementary Table S4). Protein bands were detected with horseradish peroxidase (HRP)-conjugated secondary antibodies and visualized by enhanced chemiluminescence (Millipore).

Recombinant GST-MCL-1 protein was purified from BL21-competent cells. The recombinant protein was incubated with GSK3β kinase (New England BioLabs), ERK1 kinase (Promega), or both for 3 hours at 30°C in 20 μL of kinase reaction buffer. The protein mixture was diluted in lysis buffer, and phosphorylation of Ser159 was determined by Western blot analysis. For the peptide pull-down experiment, His-tagged MCL-1 peptides (AugctBioTech) were bound to Ni-NTA agarose and incubated with HEK293 cell lysate expressing Myc-tagged FBW7. The agarose was eluted with TI buffer and subjected to Western blot analysis.

Approximately 2 × 10⁶ cells stably carrying luciferase were resuspended in 100 μL of Matrigel (BD Biosciences) and subcutaneously injected into the flanks of 6-week-old athymic nude mice (Laboratory Animal Center of FMMU, Xi'an, China). When massive tumor formation was observed, the mice were then divided into two groups that received either vehicle as a control or 100 mg/kg gefitinib (n = 6 per group) for 15 days. Each mouse was intragastrically administered 150 μL of a vehicle control (PBS) or an equal volume of gefitinib solution every day for 2 weeks. Tumor growth was monitored and recorded every three days. At the end of experiment, the mice were intraperitoneally injected with d-Luciferin for in vivo bioluminescence imaging. The mice were then humanely sacrificed, and tumors were carefully isolated and processed for histologic studies. All the animal experiments were conducted in compliance with institutional guidelines and approved by the Animal Care and Use Committee of FMMU (Xi'an, China).

IHC staining was performed on representative tissue sections from formalin-fixed and paraffin-embedded tissue blocks. IHC was performed with anti-FBW7, anti-phospho-EGFR, anti-phospho-Akt, anti-phospho-Erk, anti-Ki67, and anti-MCL-1 antibodies (Supplementary Table S4). The slides were then incubated with HRP-conjugated secondary antibodies and visualized using a DakoEnVision Detection Kit (Dako).

For immunofluorescence staining, cells were cultured in a confocal dish and treated as indicated. The cells were fixed with formalin, incubated with anti-MCL-1 or anti-Flag antibodies, and FITC-conjugated secondary antibodies, and counter stained with DAPI. Fluorescence signals were detected using a Nikon A1 Confocal Microscope (Nikon).

Each experiment was repeated at least three times. Values are expressed as the means ± SD. A paired t test and one-way ANOVA were used to determine statistical significance between different groups. Statistical significance was set at P < 0.05 (*).

As FBW7 deficiency or loss-of-function confers resistance to anti-tubulin chemotherapeutics, we evaluated whether FBW7 defects also contribute to TKIs resistance. FBW7 expression was silenced by shRNAs that targeted its coding sequence (shFBW7.1) or its 3′-untranslated region (shFBW7.2) in gefitinib-sensitive PC-9 cells. Both shRNAs efficiently inhibited FBW7 protein expression without affecting the activation of EGFR signaling (Supplementary Fig. S1A). However, PC-9 shFBW7 cells were resistant to gefitinib treatment compared with PC-9 shGFP cells (Fig. 1A). To evaluate the biological consequences of FBW7 deficiency under more physiologic conditions, we transformed immortalized Fbw7^flox/flox/ER-Cre MEFs with the oncogenic EML4-ALK fusion gene. Efficient deletion of the Fbw7 alleles and loss of Fbw7 protein expression have been confirmed in 4-hydroxytamoxifen (4-OHT)-treated MEFs (Supplementary Fig. S1B). The Fbw7 KO MEFs were resistant to the ALK TKI crizotinib, whereas the Fbw7 WT MEFs were much more sensitive to such killing (Fig. 1B; Supplementary Fig. S1C). Similar findings were also observed in mutant EGFR–transformed MEFs (Supplementary Fig. S1D). In addition, the Fbw7 agonist oridonin synergized with crizotinib to promote Fbw7 WT MEFs apoptosis, whereas the Fbw7 KO MEFs remained resistant to the combinational treatment (Fig. 1C; Supplementary Fig. S1E).

Numerous studies have demonstrated that despite the presence of the indicated TKIs, resistant cancer cells have constitutively activated PI3K/Akt and MEK/ERK signaling, downstream of the original receptor tyrosine kinases (8, 10, 11, 23, 24). We therefore speculated that FBW7 depletion possibly leads to targeted therapy resistance by restoring PI3K/Akt and MEK/ERK signaling. To our surprise, similar potent suppression of EGFR, Akt, and ERK phosphorylation following gefitinib treatment was observed in PC-9 shGFP and PC-9 shFBW7 cells (Fig. 1D). Accordingly, crizotinib abolished ALK signaling in both H3122 shFBW7 and H3122 shGFP cells (Fig. 1E). These results were further confirmed by in vivo bioluminescence imaging and IHC staining. As early as 6 days after treatment, the shGFP xenograft recipients began to display a reduction in tumor burden, whereas the shFBW7 xenograft recipients did not. At the end of the 2-week TKI treatment, the tumor burden of shGFP recipients was reduced to 35% of the pretreated size, while all the mice bearing FBW7-deficient xenografts succumbed to lung cancer (Fig. 1F). Consistent with the results of the cellular assay, gefitinib administration efficiently abolished phosphorylation of EGFR and downstream signaling in both xenografts. However, TKI treatment could not inhibit the expression of the proliferative marker Ki67 in shFbw7 xenografts, as determined by IHC (Supplementary Fig. S1F). These findings suggested that the well-documented, sustained PI3K/Akt and MEK/ERK dogma is dispensable for FBW7 depletion–induced TKIs resistance, arguing the need for further understanding of resistance mechanisms in the setting of FBW7 deficiency.

To investigate how FBW7 deficiency reduces sensitivity to TKIs, we studied whether FBW7 exerts these effects by modulating the abundance of its targeted proteins. Expression vectors encoding FBW7 were introduced into PC-9 shFBW7.2 cells and the rescue experiment was performed. Indeed, reexpression of wild-type FBW7 (FBW7 WT) restored the sensitivity to gefitinib, whereas the FBW7 ΔF-box (FBW7 ΔF) catalytic inactive mutant that cannot form the SCF–FBW7 complex compromised this effect (Fig. 2A), suggesting the E3 ligase activity of FBW7 is required for restoring TKI sensitivity. In further support of this notion, knockdown of endogenous Cul-1 and Skp1, two essential core components of the SCF–FBW7 complex, imitated FBW7 silencing and rendered PC-9 cells resistant to gefitinib. Likewise, Cul-1 and Skp1 deletion in H3122 cells significantly decreased the apoptotic response to crizotinib (Fig. 2B). Moreover, overexpression of FBW7 R465H or R505L, two dominant negative mutants that lack the ability to bind substrates, failed to restore gefitinib sensitivity in PC-9 shFBW7.2 cells (Supplementary Fig. S2A).

FBW7 as an E3 ligase has commonly been known to target oncoproteins, such as c-Myc, cyclin E, c-Jun, and MCL-1 (15, 25–29), for ubiquitination and proteasome-dependent degradation. It is conceivable that FBW7 deficiency reduces sensitivity to TKIs by diminishing the degradation of distinct substrates. Using siRNAs to knockdown FBW7 substrates individually, we found that downregulation of each substrate readily reduced PC-9 shFBW7.2 cell viability (Supplementary Fig. S2B). Knockdown of c-Myc, cyclin E and c-Jun did not have a significant impact on cell apoptosis within 24 hours, whereas MCL-1 silencing resulted in robust apoptosis at this early time point (Supplementary Fig. S2C). The endogenous MCL-1 protein was found to coimmunoprecipitate with endogenous FBW7 and core components of the SCF–FBW7 complex in PC-9 and H3122 cells. This interaction was enhanced when treated with the indicated TKIs (Fig. 2C), highlighting a physiologic interaction between FBW7 and MCL-1. Meanwhile, endogenous MCL-1 protein underwent remarkable ubiquitination in the presence of TKIs (Fig. 2C). To determine the MCL-1 polyubiquitin linkage by FBW7, HEK293 cells were transfected with His-tagged Ub and Flag-tagged MCL-1 with or without Myc-tagged FBW7 plasmids. Ni-NTA pull-down assay and anti-Flag Western blot analysis showed that the ubiquitin-conjugated MCL-1 adducts were detected in cells concurrently expressing Myc-tagged FBW7 (Fig. 2D). To further verify the ubiquitination status, His-tagged Ub mutants were used that contained only a single lysine (K48 only and K63 only) and all other lysines mutated to arginines. MCL-1 binding to K48-linked Ub chains was elevated significantly when FBW7 was simultaneously overexpressed. However, overexpression of FBW7 had little or no effect on MCL-1 binding to nonproteolytic K63-linked Ub (Fig. 2D). These results were further supported by the finding that in PC-9 and H3122 cells, TKIs profoundly promoted MCL-1 protein undergoing K48-linked ubiquitination, which was dramatically diminished upon FBW7 depletion (Supplementary Fig. S2D), highlighting that FBW7 mediates the K48-linked Ub chain attached to MCL-1 protein for ubiquitination and degradation.

The MCL-1 protein contains a total of 13 lysine residues, and N-terminal lysines K5, K40, K136, K194, and K197 have been shown to be targeted by FBW7 (29). Sequence alignment analysis of MCL-1 indicated that residues K279 and K308 on the C-terminal tail are also highly conserved from yeast to human, but their function in MCL-1 ubiquitination has not been elucidated (Supplementary Fig. S2E). We generated a panel of MCL-1 KR mutants and wished to determine the exact ubiquitination sites following TKI treatment. Interestingly, mutations at K40, K136, and K194/197 residues markedly abolished TKI-induced MCL-1 protein ubiquitination, whereas mutations at K279 and K308 residues partially impaired the ubiquitination. Moreover, amino acid substitution at residue K5 seemed to have minimal impact on MCL-1 protein ubiquitination, arguing that K40, K136, and K194/197 residues are major FBW7-targeting sites and that K279 and K308 residues are minor targeting sites of TKI-induced ubiquitination of MCL-1 protein (Fig. 2E). Taken together, these data demonstrate that the integrity of FBW7 maintains sensitivity to targeted therapy. TKIs promote MCL-1 protein binding to FBW7 and undergo K48 linkage ubiquitination and proteolysis.

Proteins undergoing rapid turnover often contain PEST sequences, which stands for the sequences enriched for proline (P), glutamic acid (E), serine (S), and threonine (T; ref. 30). The PEST region is generally believed to serve as a phosphodegron that provides docking sites for the recruitment of F-box protein. The computational online PEST analysis program revealed a conserved PEST region in MCL-1 (Supplementary Fig. S3A). To investigate the function of the PEST region in MCL-1 protein turnover and stability, we expressed the ΔPEST-truncated mutant in HEK293 cells and measured MCL-1 protein half-life. In contrast to MCL-1 WT, the MCL-1 ΔPEST mutant became much more stable, and the half-life time was significantly protracted (Fig. 3A). To map whether the PEST region is required for interacting with FBW7, an immunoprecipitation experiment was performed, and the result showed a dramatic reduction in FBW7 binding upon PEST region removal (Fig. 3B), indicating that the PEST region is essential and efficient for its interaction with FBW7.

Another important aspect of substrate binding lies in the notion that the substrate needs to be phosphorylated within the CDC4 phosphodegron (CPD) motif before binding to FBW7 (31, 32). In support of this notion, MCL-1 dephosphorylated by λ-phosphatase failed to coprecipitate with FBW7 (Supplementary Fig. S3B). To determine which residues within the PEST region are phosphorylated, we performed a Myc-Flag tandem pull-down assay. Only phosphorylated serine (pS), but not phosphorylated tyrosine (pY) or threonine (pT), efficiently bound FBW7 (Supplementary Fig. S3C). The PEST sequence alignment from different species indicated two regions that resembled putative CPDs were highly conserved. The first CPD, designated CPD-1, is located in human MCL-1 between codons 121-125 in the N-terminal region of PEST and contains a canonical JNK consensus motif (33). CPD-2 is situated between codons 159 and 163 in the C-terminal region and contains GSK3β (34) and MEK/ERK (35) phosphorylation sites. However, MCL-1 protein degradation was only promoted by the overexpression of GSK3β but not by kinases JNK and ERK. Consistently, overexpression of the constitutively activated GSK3β S9A counterpart further reduced MCL-1 protein abundance (Fig. 3C). In further support of this notion, inhibition of GSK3β kinase activity by EGF or GSK3β inhibitor stabilized the MCL-1 protein, whereas MCL-1 degradation was recovered by overexpression of GSK3β S9A (Fig. 3D; Supplementary Fig. S3D). An immunoprecipitation assay with the MCL-1 S159A mutant revealed that GSK3β-mediated Ser159 phosphorylation directed its interaction with FBW7 (Supplementary Fig. S3E). To further corroborate these findings, His-tagged phospho-peptides corresponding to the two putative CPDs were synthesized and incubated with cell lysates containing Myc-tagged FBW7. Similar results were obtained from the Ni-NTA pull-down assay in which FBW7 directly bound to Ser159-phosphorylated peptide, but not phosphorylated Ser121 and other nonphosphorylated peptides (Supplementary Fig. S3F). Moreover, increasing doses of GSK3β inhibitor decreased the binding affinity of MCL-1, and the S159A mutation severely impaired its association with FBW7 (Fig. 3E).

Although phosphorylation of MCL-1 at Thr163 by MEK/ERK did not lead to direct binding to FBW7, inactivation of this phosphorylation site abrogated MCL-1 and FBW7 interaction (Fig. 3F). This apparent discrepancy suggested that MCL-1 phosphorylation at Thr163 presumably provides a priming site for GSK3β-mediated Ser159 phosphorylation. To investigate the role of MEK/ERK on Ser159 phosphorylation, we treated GST-MCL-1 protein in kinase reaction buffer with ERK1 and GSK3β kinases individually or in combination. As shown in Supplementary Fig. S3G, GSK3β could not phosphorylate Ser159 in the absence of ERK1, whereas MCL-1 was robustly phosphorylated at Ser159 when both kinases were present. Consistently, the MCL-1 T163A mutant that inactivates MEK/ERK–dependent phosphorylation was not able to bind to FBW7, even in the presence of the GSK3β S9A mutant (Fig. 3F), highlighting MEK/ERK in cooperation with GSK3β leads to sequential phosphorylation of MCL-1. Degradation of MCL-1 protein by FBW7 occurs when both residues are phosphorylated. In particular, Ser159 phosphorylation directs the binding to FBW7.

FBW7 has three isoforms that are localized in the nucleus (α), cytoplasm (β), and nucleolus (γ), respectively. Both nuclear isoforms interacted with MCL-1, whereas the MCL-1 and FBW7β interaction was not identified. Meanwhile, the MCL-1 and FBW7γ interaction was reproducibly lower than that of FBW7α (Fig. 4A), indicating that both nucleus isoforms regulate MCL-1 protein abundance, with the FBW7α isoform (hereafter referred to as FBW7) being the most important and efficient. While endogenous MCL-1 protein primarily localized to the cytoplasm, the degradation of MCL-1 by FBW7 should be attributed to either the nuclear translocation of MCL-1 or the nuclear export of FBW7. Cell fractionation experiments showed that a brief treatment of indicated TKIs resulted in accumulation of MCL-1 protein in the nuclear fraction of PC-9 and H3122 cells (Supplementary Fig. S4A). Confocal laser scanning further confirmed that the inhibition of receptor tyrosine kinases led to MCL-1 protein being shuttled from the cytoplasm to the nucleus (Fig. 4B). To test the possibility that MCL-1 protein degradation might also be a consequence of FBW7 nuclear export, we pretreated the cells with a nuclear exportin inhibitor, leptomycinB (LMB), which failed to prevent the reduction of MCL-1 protein abundance (Supplementary Fig. S4B). Furthermore, both PC-9 and H3122 cells express endogenous FBW7 protein that is localized exclusively in the nucleus, regardless of the TKI treatment (Supplementary Fig. S4A), arguing that only the nuclear translocation regulatory machinery is responsible for MCL-1 protein degradation.

To expand the insight into MCL-1 translocation, we generated Flag-tagged truncated mutants that lack putative nuclear localization signal (ΔNLS) but retain the FBW7 degron. PC-9 cells were treated with DMSO or 1 μmol/L gefitinib for an additional 3 hours and fixed with paraformaldehyde 24 hours after transfection. The localization of ectopically expressed MCL-1 was examined by fluorescence microscopy. The result showed that Flag-tagged MCL-1 WT and ΔNLS truncations displayed a diffused distribution over the entire cytoplasm. In contrast, the localization of MCL-1 WT changed into a nucleus-enriched pattern in gefitinib-containing medium, while the ΔNLS truncations constitutively localized in the cytoplasm regardless of gefitinib treatment (Supplementary Fig. S4C). Examination of PC-9 cell fractions also revealed that TKIs induced a rapid movement of MCL-1 to the nucleus, whereas the MCL-1 ΔNLS truncations did not (Fig. 4C). Furthermore, the proportion of apoptosis coincided with MCL-1 nuclear translocation. PC-9 cells expressing MCL-1 WT underwent substantial apoptosis in response to gefitinib, whereas the MCL-1 ΔNLS–expressing cells were insensitive to such killing (Supplementary Fig. S4D).

It was noteworthy that even though the MCL-1 ΔNLS truncations retained all phosphorylation sites and the region required for FBW7 binding, its inability to translocate to the nucleus abolished the interaction with FBW7. As shown by immunoprecipitation, the absence of FBS in HEK293 cells promoted MCL-1 WT binding to FBW7, while MCL-1 lacking NLS completely lost this association (Fig. 4D). In agreement with this finding, the MCL-1 ΔNLS truncated protein became stabilized, and the half-life was significantly extended (Supplementary Fig. S4E).

Having established that MCL-1 phosphorylation at Ser159 directs the interaction with FBW7 and that MCL-1 protein nuclear translocation is a required step for subsequent degradation, we next investigated the possibility that GSK3β-mediated MCL-1 Ser159 phosphorylation governs its subcellular localization. HEK293 cells were transfected with indicated GSK3β plasmids, and endogenous MCL-1 protein was visualized by immunofluorescence staining. As shown in Supplementary Fig. S5A, endogenous MCL-1 protein was enriched in the nucleus when coexpressed with the GSK3β S9A mutant. In contrast, coexpression of the kinase-dead GSK3β K85M mutant resulted in redistribution of the MCL-1 protein to the cytoplasm, suggesting that the kinase activity of GSK3β is required for MCL-1 protein shuttling from the cytoplasm to the nucleus. To directly visualize the dynamic subcellular localization, MCL-1 ORF was fused to C-terminal EGFP in frame and transiently expressed in HEK293 cells. Consistent with our previous observation, two distinct localization patterns were observed. The FBS deprivation clearly induced a striking accumulation of MCL-1 protein in the nucleus, whereas the S159A mutation that abolished GSK3β-mediated phosphorylation dramatically reduced MCL-1 protein nucleus translocation (Fig. 5A), indicating that phosphorylation at the Ser159 residue by GSK3β enables the MCL-1 protein to enter the nucleus.

On the basis of these data, the dual requirement of GSK3β for MCL-1 protein nuclear translocation and subsequent interaction with FBW7 implied its critical role in MCL-1 protein homeostasis. Because GSK3β is inactivated by PI3K/Akt–dependent phosphorylation, MCL-1 protein subcellular localization and degradation is linked to growth factor–dependent signaling. This finding might not be surprising, as inhibition of EGFR and ALK signaling in PC-9 and H3122 cells resulted in MCL-1 protein nuclear translocation and degradation (Fig. 4B; Supplementary Figs. S4A and S5B). Additional evidence implicating Ser159 as a critical residue in determining MCL-1 protein subcellular localization and degradation was obtained from NSCLC H1650 cells, which harbored an EGFR del E746_A750–sensitizing mutation and a PTEN homozygous deletion. H1650 cells were de novo resistant to EGFR TKIs and gefitinib treatment had minimal effects on the suppression of PI3K/Akt/GSK3β signaling and MCL-1 degradation (Fig. 5B). MCL-1 protein Ser159 phosphorylation and nuclear translocation were also impaired in H1650 cells despite MEK/ERK signaling being efficiently inhibited. Conversely, the PI3K/Akt inhibitors profoundly increased MCL-1 protein Ser159 phosphorylation and nuclear translocation (Supplementary Fig. S5C), indicating that selective inhibition of PI3K/Akt signaling immediately upstream of GSK3β is the fundamental regulatory machinery responsible for MCL-1 protein subcellular localization and degradation.

Finally, we determined the biological consequences of overexpression of the constitutively activated isoform of Akt (Myr-Akt) or its dominant negative counterpart (DN-Akt) on the expression of endogenous MCL-1 protein. Transfection with Myr-Akt rescued the reduced MCL-1 protein levels under FBS-deprived conditions, whereas the DN-Akt did not. The combination of PI3K/Akt inhibitors dephosphorylated Ser9 of GSK3β and preserved its kinase activity, thereby promoting MCL-1 protein degradation (Fig. 5C).

The association between MCL-1 protein degraded by FBW7 and sensitivity to targeted therapy was further confirmed in NSCLC H3255 cells (EGFR L858R) and H1975 cells (EGFR L858R/T790M). Gefitinib profoundly blocked EGFR signaling and reduced MCL-1 expression posttranscriptionally in H3255 cells (Fig. 6A; Supplementary Fig. S6A). However, H1975 cells harboring the T790M mutation were resistant to gefitinib even though this cell line has the same L858R-sensitizing mutation. The second-generation irreversible EGFR TKIs afatinib was able to inhibit EGFR signaling and reduce MCL-1 protein levels at higher concentrations, whereas the third-generation EGFR TKIs AZD9291, targeting both sensitizing and T790M-resistant mutant forms of EGFR, was much more efficient in blocking EGFR signaling and reducing MCL-1 protein levels (Fig. 6B). Afatinib and AZD9291 also had no effect on MCL-1 mRNA expression, suggesting that TKIs prominently reduced MCL-1 expression posttranscriptionally (Supplementary Fig. S6B). We hypothesized that the greater potency of the TKIs/FBW7 agonist combination may indicate increased cell death. Thus, we compared the efficacy of TKIs, oridonin, and their combination in gefitinib-resistant H1299 cells (low endogenous FBW7). Gefitinib and oridonin did not show single-agent efficacy on apoptosis induction or MCL-1 protein reduction, while their combination efficiently promoted apoptosis and reduced MCL-1 protein abundance (Fig. 6C). To translate our findings in the clinical setting, Western blot analysis of human NSCLC specimen also revealed a negative correlation between MCL-1 and FBW7 abundance at protein level. As a control for the BCL-2 family, BCL-2 and BCL-xL protein expression levels were not inversely associated with FBW7 protein expression levels (Supplementary Fig. S6C). The correlation between FBW7 and TKIs sensitivity was further confirmed in paired NSCLC specimens in which the EGFR-mutated treatment-naïve samples had high endogenous FBW7. In contrast, repeated biopsy samples obtained from the same patients who relapsed on EGFR TKIs revealed a significant reduction in endogenous FBW7 together with high MCL-1 protein abundance (Fig. 6D).

Because cancer cells often require elevated MCL-1 levels for survival and proliferation, targeting MCL-1 proteolysis would universally suppress tumor outgrowth. Indeed, inhibition of ErbB2 signaling by trastuzumab dramatically decreased MCL-1 protein levels in ErbB2-amplified SKBR3 breast cancer cells, whereas it failed to do so in trastuzumab-resistant MDA-231 breast cancer cells (Supplementary Fig. S6D). Overexpression of MCL-1 WT in several genetically defined cancer cell lines did not cause a targeted therapy–resistant phenotype. However, overexpression of the degradation-defective MCL-1 S159A mutant resulted in targeted therapy resistance (Supplementary Fig. S6E). These data suggested that the loss of FBW7 E3 ligase or the defect in MCL-1 protein degradation universally contributed to resistance to targeted therapy. Targeting FBW7-mediated MCL-1 degradation is a promising strategy to overcome resistance to targeted therapy.

Our study revealed an unexpected resistance mechanism in which loss of FBW7 E3 ligase resulted in accumulation of antiapoptotic protein MCL-1, thereby enabling cancer cells to escape apoptosis following EGFR and ALK signaling inhibition. Importantly, we provided preclinical evidence demonstrating that targeting MCL-1 proteolysis universally benefits therapeutic outcomes for currently available targeted therapy (Supplementary Fig. S7).

An increasing number of genetically defined cancer patients are gaining rapid and robust benefits from targeted therapy, but resistance to targeted therapeutic agents has become the major challenge of their sustainable efficacy. It is generally believed that constitutive activation of PI3K/Akt and MEK/ERK signaling cascades are the common nodes driving resistance to multiple targeted therapeutics (8, 10, 11, 23, 24). Conversely, resistance to TKIs driven by FBW7 deficiency does not follow this dogma and underlies a challenging resistance mechanism. Induction of functional FBW7 restored TKIs sensitivity, and oridonin treatment recapitulated the potency of FBW7 overexpression; these results suggest that molecules downstream of PI3K/Akt and MEK/ERK and those targeted by FBW7 are likely to mediate this unrecognized machinery. Using several in vitro and in vivo models, we reported sensitivity to targeted therapy and FBW7 converges at the antiapoptotic protein MCL-1. As intratumor heterogeneity exists, more than one resistance mechanism may be present simultaneously in an individual patient. Thus, targeting downstream signaling, such as MCL-1, may provide a feasible approach to overcome resistance to targeted therapy.

Although MCL-1 is one of the top ten amplified oncogenes in tumors, the results released by the TCGA database show that there are no superior benefits in PFS and OS in lung cancer patients with or without MCL-1 mRNA alterations (Supplementary Fig. S8A). In contrast, high expression of MCL-1 protein has been linked to poor prognosis and pathogenesis of refractory cancers, suggesting that the manipulation of MCL-1 at the protein level, rather than the mRNA level, serves to inhibit cancer cell proliferation and facilitate apoptosis. MCL-1 is a labile protein, and several groups recently reported that it is targeted by MULE, β-TrCP, and FBW7 E3 ligases for ubiquitination and degradation (29, 36–38). Unfortunately, MULE and β-TrCP are not typically correlated with lung cancer prognosis (Supplementary Fig. S8B and S8C). The prognostic significance of FBW7 has been assessed by Yokobori and colleagues (39), intriguingly, they observed that the patients with low FBW7 expression levels presented with more progressive cancer and significantly shorter survival. In our study on EGFR-mutant lung cancer patients who relapsed on TKIs, FBW7 expression was significantly suppressed. Therapeutic induction of FBW7 by oridonin did not elicit single-agent efficacy in TKI-resistant H1299 cells; however, H1299 cells were sensitive to concomitant treatment with TKIs and oridonin. In total, these results provide therapeutic insight into FBW7 agonists could be advantageous in FBW7 low–expressing resistant cancers.

The other approach to treat FBW7 low–expressing cancer is to directly target its downstream substrates. Indeed, the mTOR inhibitor has shown single-agent efficacy in lung adenocarcinoma patients harboring the FBW7 R465H mutation (40). Accordingly, MCL-1 protein is the major survival output of tyrosine kinases; it stands out as the core regulator of intrinsic apoptosis in genetically defined NSCLC cells (Supplementary Fig. S9A and S9B). Therefore, manipulating MCL-1 protein abundance may help to promote apoptosis and overcome resistance. Consistently, gefitinib and PHA-665752 failed to elicit single-agent efficacy in inhibiting MCL-1 protein expression in c-Met amplified gefitinib-resistant HCC827 GR5 cells, whereas their combination efficiently suppressed the MCL-1 protein (Supplementary Fig. S9C). Unlike K63-linked ubiquitination of the XRCC4 protein during irradiation and nonhomologous end-joining repair (41), the MCL-1 protein undergoes K48-linked ubiquitination by FBW7 following the inhibition of tyrosine kinases signaling. This finding is consistent with the model put forth by our study, in which TKIs inhibit PI3K/Akt signaling, thereby enabling GSK3β to phosphorylate MCL-1 Ser159 and leading to MCL-1 protein ubiquitination. We further demonstrated that Ser159 phosphorylation is primed by MEK/ERK–dependent phosphorylation of Thr163. This result indicates that MCL-1 protein abundance is determined by dual PI3k/Akt/GSK3β- and MEK/ERK–dependent regulatory machinery and that the crosstalk between these two signaling cascades precisely maintains the MCL-1 protein at a suitable level to keep cancer cells alive or dead. Of note, we observed that GSK3β-dependent phosphorylation also governs the subcellular localization of MCL-1 protein. Impeding MCL-1 protein nuclear localization protected the EGFR-mutant and ALK-rearranged NSCLC cells from TKI-induced MCL-1 protein degradation and apoptosis. On the other hand, MCL-1 protein is found to be present at low levels in the nucleus in treatment-naïve cells, and this nucleus localization of the MCL-1 may possess a nonapoptotic biological function, such as the maintenance of genomic stability and the DNA damage response (42, 43). To our surprise, treatment with targeted therapeutic agents triggered an apoptotic response in MCL-1 WT–overexpressing cells to a similar magnitude observed in EV control cells. Indeed, overexpression of MCL-1 WT has been found to delay, but not prevent, cells from undergoing apoptosis in hematopoietic cells and CHO cells (44, 45). These data suggest that MCL-1 WT only enhances short-term cell survival and that the elevation of MCL-1 protein in refractory and relapsed cancers may not be merely a reflection of MCL-1 allele amplification. Loss of FBW7, disruption of MCL-1 protein phosphorylation, and nuclear translocation are also contributing factors to the stabilization of this antiapoptotic protein. Thus, we hypothesized that the TKI-resistant patients may be sensitive to MCL-1 inhibition. Screening endogenous FBW7 expression in treatment-naïve patient biopsies may help to identify patients who will not benefit from targeted therapy as persistently and substantially.

Collectively, our study reports a novel resistance mechanism and provides compelling evidence that MCL-1 ubiquitination and degradation by FBW7 is a multi-step process involving phosphorylation and nucleus translocation. Targeting FBW7-mediated MCL-1 protein ubiquitination and degradation is a promising strategy to kill cancer cells and overcome resistance. Deciphering pathways regulating FBW7 expression should lead to the discovery of new therapeutic targets to increase cancer cell sensitivity to targeted therapy.

No potential conflicts of interest were disclosed.

Conception and design: L. Yao, J. Zhang, J. Zhang

Development of methodology: M. Ye, J. Zhang, J. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Ye, Y. Zhang, X. Zhang, J.-B. Zhang, P. Jing, L. Cao, N. Li, J. Zhang, J. Zhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Ye, Y. Zhang, X. Zhang, J.-B. Zhang, P. Jing, N. Li, J. Zhang

Writing, review, and/or revision of the manuscript: M. Ye, J. Zhang, J. Zhang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Li, J. Zhang, J. Zhang

Study supervision: L. Yao, J. Zhang, J. Zhang

We would like to thank Jeffrey Engelman (Massachusetts General Hospital Cancer Center, Boston, MA) for supplying H3122, H3255, HCC827, and HCC827 GR5 cells, Hiroyuki Mano (The University of Tokyo, Tokyo, Japan) for supplying EML4-ALK constructs, Ping Wang (Tongji University, China) for supplying the homozygous Fbw7^flox/flox mice. We also express gratitude to Hiroyuki Inuzuka (Beth Israel Deaconess Medical Center, Boston, MA) and Chia-Hsin Chan (Stony Brook University, Stony Brook, NY) for critical comments and technical advice.

This work was sponsored by grants National Natural Science Foundation of China (#81272518 and #81472192 to J. Zhang).

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.