Increased Synthesis of MCL-1 Protein Underlies Initial Survival of EGFR-Mutant Lung Cancer to EGFR Inhibitors and Provides a Novel Drug Target

Metastatic EGFR-mutant non–small cell lung cancers (NSCLC) have high response rates to EGFR inhibitors (EGFRi), in excess of 60% (1). Treatment with EGFRi has vastly improved the care of these patients, and advances in specificity of EGFRi toward mutated EGFR appear to improve these response rates even further (2, 3).

However, responses to EGFRi are transient—∼12 to 18 months usually—as acquired resistance to these inhibitors continues to be the main barrier to long-lasting responses (4). Acquisition of resistance to the EGFRi gefitinib is often caused by a second mutation in EGFR, T790M, which hinders the ability of gefitinib to inhibit EGFR (4). Studying EGFR-mutant cancer cell lines, we recently demonstrated that T790M can exist—even at hard-to-detect frequencies—prior to initiation of EGFRi treatment, or alternatively, appear de novo (5).

In the latter scenario, it remains largely unknown how clones from EGFR-mutant lung cancers treated with EGFRi and have no preexisting T790M mutations but eventually acquire them or other secondary bypass track mutations like MET amplification (6), survive initial drug therapy. To this point, our knowledge of the events leading to the survival of a subpopulation of cells [deemed “persister cells” or “drug-tolerant cells” (DTC)] remains limited. A better understanding of DTC survival capacity, and mechanisms supporting that capacity, could inform upfront drug treatment to kill DTCs, delaying or possibly even eliminating the onset of acquired resistance. In fact, important studies have previously demonstrated distinct molecular characteristics of DTCs within subpopulations of cancer cells, including engagement of the insulin growth factor (IGFR) pathway and epigenetic modifications (7, 8), presenting survival advantages to different drugs.

Although these other studies have focused mostly on signaling cascades and epigenomic alterations, we and others have demonstrated that deficiencies in apoptosis can underlie intrinsic resistance to EGFRi (9–13). This begs the question as to whether a deficient apoptotic subpopulation may transiently enrich following initiation of EGFRi therapy, which could survive in a relatively dormant state until the eventual acquisition of a secondary T790M mutation/other bypass track mutations. Furthermore, if these DTCs could be pharmaceutically targeted with an appropriate drug within the current coterie of targeted therapies, they could be eliminated and theoretically delay or prevent the acquisition of resistance.

The cell lines used in this study were the EGFR-mutant NSCLC cell lines PC9, HCC4006, and HCC827 and have been rigorously characterized (5, 9, 13). These cell lines were cultured in RPMI 1640 (Lonza) with 10% fetal bovine serum plus 1% penicillin and streptomycin (Gibco). All the cell lines were routinely tested for mycoplasma. In order to make the DTC lines, cells were seeded at a density of 50% confluence in 100 mm³ dishes. Cells were treated with 50 nmol/L of gefitinib for 6 days, then the media were changed, and cells were incubated for 3 days without gefitinib (Fig. 1C). Stable cell lines (GFP-IRES and GFP-IRES-MCL-1) were produced by transduction with pLENTI-GFP-IRES and pLENTI-GFP-IRES-MCL-1 as previously published (14).

Gefitinib, dinaciclib, and osimertinib (AZD9291) were from AbMole Biosciences. A-1210477 was kindly provided by AbbVie. S63845 was purchased from Chemietek. mTOR inhibitors (AZD2014 and MLN0128) were from Selleckchem and AbMole, respectively. All drugs were dissolved in DMSO at the stock concentration of 10 mmol/L for in vitro experiments.

Lysates were separated on Nu PAGE 4% to 12% Bis-Tris midi protein gel (Invitrogen). Polyvinylidene fluoride (PVDF) membranes were probed with antibodies against phospho-EGFR Tyr1068 (D7A5), pAKT Ser473 (D9E), MCL-1 (D35A5), pMCL-1 Ser159/Thr163 (4579S), pERK1/2 Thr202/Thr204 (D13.14.4E), NOXA (D8L7U), BCL-XL (C54H6), BCL-2 (D55G8), cleaved PARP Asp214 (D64E10), pRpb1 CTD Ser2/5 (D1G3K), pS6 Ribosomal Protein Ser240/244 (D68F8), mTOR (7C10), total 4E-BP1 (53H11), p4E-BP1 Thr37/46 (236B4), p4E-BP1 Ser65 (D9G1Q), c-MYC (D84C12), pp70S6K Thr389 (108D2), cyclin D1 (2922S), and eIF4E (C46H6) from Cell Signaling and eIF4G (A10) and GAPDH (6C5) from Santa Cruz Biotechnology.

The pFR_HCV_xb was a gift from Phil Sharp (Addgene plasmid #11510; ref. 15). We transiently transfected the vector by lipofectamine 2000 (Thermo Fisher Scientific) for 48 hours. After 48 hours, the cells were lysed by 1X passive lysis buffer for 20 minutes, and luciferase activity was measured using the Dual Luciferase Reporter Assay System (Promega, E1960).

For apoptosis analysis, cells were plated to be ∼50% confluent in 100 mm³ dishes. The next day, cells were subjected to 3 cycles of treatment with 50 nmol/L of gefitinib. Each cycle consisted of treatment with drug for 3 days, assaying for apoptosis at 72 hours, followed by 2 days of no treatment. Cy^TM5 Annexin V and propidium iodide (PI) staining solution were purchased from BD Pharmingen. After each cycle, FACS was performed to measure the presence of apoptotic cells on a Guava easyCyte (Millipore; Fig. 1A and B). For cell-cycle analysis (Fig. 4A), the indicated cells were exposed to drug for 24 hours and then assayed as previously described (13).

mTOR siRNAs (si mTOR) were purchased from Santa Cruz Biotechnology (cat. #SC-35409) and Thermo Fisher Scientific (cat. #AM16708; assay ID: 145119 and #AM16708; assay ID: 145120)). Scramble (si Sc) control siRNA was from Qiagen. HiPerFect transfection reagent (Qiagen) was used to produce the knockdown cells with 50 nmol/L of mTOR siRNAs following the manufacturer's protocol.

For the MCL-1 assay, we mixed either 10% of GFP control cells or 10% of GFP-MCL-1 expressing cells with 90% of the parent cell line. The mixed cells were seeded at a low density (3 × 10³ per well) in two 96-well black plates (plates 1 and 2). One of the plates (designated “P1”) measured the percentage of GFP among the cell population at day 1 after seeding. The second plate (designated “P2”) was treated as outlined in Fig. 1C, and then cells were observed under a high-throughput cell content imager (Image Xpress 6 Micro XLS, Molecule Devices) 9 days later to determine if the population of GFP-MCL-1 cells increased compared with control cells (Fig. 3B and C). DAPI (4′,6-diamidino-2-phenylindole, 1:5,000) was used for counterstaining (SC-3598, Santa Cruz Biotechnology). The fluorescence intensity of GFP within living cells after drug treatment in each well using the multiwavelength cell-scoring module was quantified on the imager using MetaXpress High-Content Image Acquisition and Analysis software.

Cells (3 × 10³ per well) were seeded in 96-well black plates. The next day, we made various concentrations of drugs as indicated in the cell viability graphs through serial dilutions. We used the CellTiter-Glo luminescent cell viability assay (G7173, Promega) according to the manufacturer's protocol, with the exception that we added half of the reagent to each well that is recommended.

PC9 and HCC4006 cells were seeded in 6-well plates at a density of 3 × 10⁵ cells/well. The next day, cells were treated with DMSO, cisplatin (10 μmol/L), or gemcitabine (500 nmol/L) for 24 h and 72 hours. After chemotherapy, the cells were replenished with drug-free complete media in order to recover for 24 and 72 hours. Cells were then harvested and lysates were prepared.

PC9 and HCC4006 cells were seeded in a 6-well plate at a density of 3 × 10⁵ cells/well. The next day, cells were exposed to 5 gray (Gy) radiation. A nonradiated plate was simultaneously maintained as a control. After radiation treatment, the media were changed in both plates and the cells were incubated for an additional 24 or 72 hours before harvesting.

Patients with EGFR-mutant lung cancer were treated with gefitinib. Following confirmation of partial response to the treatment, residual tumors were surgically resected. The stainings were conducted using paired formalin-fixed, paraffin-embedded tissues before and after gefitinib treatment. Immunostaining with anti-MCL-1 (S-19, ×100 dilution, Santa Cruz Biotechnology) was followed by standard DAKO autostainer. The study was approved by Institutional Review Board at Aichi Cancer Center and patients provided written informed consent. The analysis was conducted in accordance with Declaration of Helsinki.

For the experiments in Fig. 4D and E, we injected 3 × 10⁶ PC9-GFP or PC9 GFP-MCL-1 cells with a 1:1 ratio of matrigel matrix basement membrane HC (Corning) subcutaneously in the rear flank of Nod/SCID gamma (NSG) mice (6–8 weeks old, female). Drug treatment was started when the tumor reached ∼150 to 170 mm³, and mice were randomized (n = 4–6 mice per group). Tumors and body weight were measured 3 times per week using a digital caliper and electronic scale for the duration of the experiment. Tumor volumes were assessed by using length (L) × width (W)² × 0.52. Gefitinib was administered 4 days per week by oral gavage at a concentration of 50 mg/kg/body weight in 1% Tween 80 in sterile water.

For the experiment in Fig. 5G, we injected 3 × 10⁶ PC9 cells (left) or 5 × 10⁶ HCC827 cells (right) subcutaneously in both flanks of NSG mice (6–8 weeks old, female, n = 3–4 mice per group). When tumor reached ∼100 to 170 mm³, we administered gefitinib (50 mg/kg/body weight), S63845 (25 mg/kg/body weight), or combination treatment. The gefitinib group received treatment by oral gavage daily for 3 consecutive days. The S63845 group received treatment for 3 consecutive days by tail-vein injection in 25 mmol/L HCl, 20% hydroxypropyl-beta-cyclodextrin (HPBCD). For the gefitinib and S63845 combination cohort, we administered S63845 first, then gefitinib 2 hours later for 3 consecutive days. Tumors were measured every day using a digital caliper, and body weight was measured 3 times per week. For the experiment in Fig. 6C, we injected 3 × 10⁶ PC9 cells subcutaneously in NSG mice (6–8 weeks old, female, n = 5–7 mice per group). Drug treatment was started when the tumor reached ∼150 to 170 mm³. Tumors and body weight were measured 3 times per week using a digital caliper and electronic scale for the duration of the experiment. Dinaciclib was administered by intraperitoneal (i.p.) injection at 20 mg/kg/body weight twice per week with 20% HPBCD. The combination group was administered both gefitinib and dinaciclib in sequence on days when dinaciclib was administered. The animal experiment was approved by the Virginia Commonwealth University Institutional Animal Care and Use Committee (IACUC protocol #AD10001048).

Total RNA was isolated with Quick-RNA Miniprep kit (Zymo Research). cDNA was synthesized from 700 ng of total RNA by SuperScript III Reverse Transcriptase (Invitrogen) with Oligo (dT) primer (Ambion). RT-qPCR analysis was performed in triplicate with SYBR Green master mix (Life Technologies) on a 7500 Fast Real-time PCR system (Thermo Fisher Scientific) according to the manufacturer's protocol. The primer information is as follows: (i) MCL-1 (F) 5′-GGGCAGGATTGTGACTCTCATT-3′, (R) 5′-GATGCAGCTTTCTTGGTTTATGG-3′, (ii) c-MYC (F) 5′-AATGAAAAGGCCCCCAAGGTAGTTATCC-3′, (R) 5′-GTCGTTTCCGCAACAAGTCCTCTTC-3′, (iii) cyclin D1 (F) 5′-TGAACTACCTGGACCGCT-3′, (R) 5′-GCCTCTGGCATTTTGGAG-3′, ß-actin (F) 5′-GGCATGGGTCAGAAGGATT-3′, (R) 5′-AGGATGCCTCTCTTGCTCTG-3′.

For Figs. 1A, B and E, 2D, 4C, 5A–G, 6B and C, P values were determined by unpaired Student t test (***, P < 0.001; **, P < 0.01; *, P < 0.05).

We hypothesized that cells persisting after EGFRi exposure were deficient in EGFRi-induced apoptosis, and as such, could constitute a DTC population that could be pharmaceutically targeted. To test this hypothesis, we treated EGFR-mutant PC9 and HCC827 NSCLCs with cyclical gefitinib. After each cycle, FACS was performed to measure the presence of apoptotic cells. We found that both EGFR-mutant lung cancer cell lines were, as expected (9, 14), vulnerable to gefitinib-induced apoptosis over a 72-hour period (i.e., cycle 1; Fig. 1A and B). However, reexposure to gefitinib over an additional 72-hour period was inadequate to induce robust apoptosis in PC9 cells, and by the third cycle HCC827 cells also lost their ability to undergo apoptosis (Fig. 1A and B). These data indicate a loss of apoptotic potential in early-surviving EGFR-mutant lung cancer cells to EGFRi.

To further assess the characteristics of early-formed DTCs, we exposed EGFR-mutant NSCLCs for 6 days with gefitinib, followed by replenishment with drug-free complete media for 3 days. The replenishment period served two purposes: (i) to allow baseline signaling/expression to be studied in the DTCs, and (ii) to be able to gather sufficient cellular material for study. The initial 6-day treatment led to a cell confluency that empirically was ∼5% of the starting cell population; the subsequent 3 days without drug exposure allowed for modest regrowth of the cells (Fig. 1C). We collected these cells as well as the parental cells and performed either RNA-seq to look at global gene-expression changes or Western blots to analyze select critical proteins in the DTCs and parentals. In particular, because BCL-2 family members govern the cell death response to stress such as kinase inhibition, we focused on the major BCL-2 family effector molecules (reviewed in ref. 16) in each of the EGFR-mutant lung cancer cell lines tested. Strikingly, we found expression of MCL-1 was sharply increased in the bulk cell population of DTCs (Fig. 1D, left). Osimertinib (AZD9291; refs. 17, 18) is a third-generation EGFRi that targets EGFR-mutant NSCLC, including those with T790M mutations, and may be used as first-line treatment (19). We also made DTCs to osimertinib as we did with gefitinib, as depicted in Fig. 1C. Again, we found sharp upregulation of MCL-1 in the osimertinib-DTCs (Fig. 1D, right). In both sets of DTCs, other than BCL-2 in the PC9 osimertinib-DTCs, other antiapoptotic BCL-2 family members were not upregulated, and, with the exception of NOXA in the HCC827 gefitinib-DTCs, expression of the key proapoptotic BCL-2 family members BIM and NOXA was not markedly diminished (Fig. 1D, left and right).

Although MCL-1 was upregulated over 8.2-fold by Western blot, qPCR analysis showed a modest (1.9-fold) increase, indicating posttranscriptional induction (Fig. 1E). To determine if this phenomenon was also active in patients' specimens, we identified two EGFR-mutant lung cancer patients which we had paired samples before and after initial gefitinib treatment. The rebiopsy was taken prior to therapy resistance, but following a confirmed partial response to getfitinib, therefore mimicking the in vitro DTC cells. Consistent with the in vitro data, expression of MCL-1 was strongly expressed in the post gefitinib-treated tumors compared with the gefitinib-naïve tumors (Fig. 1F). Altogether, these data indicate a role for MCL-1 in EGFR-mutant DTC survival.

To better assess how MCL-1 was regulated in DTCs, we performed RNA-seq in the PC9 gefitinib-DTCs and PC9 parental (i.e., drug-naïve) cells. (The GEO accession number for these studies is GSE117610.) Consistent with the qPCR analysis, the RNA-seq data reflected only a small increase in MCL-1 mRNA expression in DTCs versus parental cells (Supplementary Table S1). MCL-1 is notably controlled by mTORC1/eIF4E-mediated cap-dependent translation (20–23). Strikingly, mTOR and the cap-dependent initiation factors eIF4G1, eIF4G3, and eIF3A (24) were four of the most significantly upregulated genes in the DTCs (Supplementary Table S1). In cap-dependent translation, mTOR directly phosphorylates the eIF4E-binding protein 1 (4E-BP1) at residues 37/46, which primes 4E-BP1 for further phosphorylation at 65, all of which contribute to disassociation of 4E-BP1 from eIF4E, promoting the EIF4 complex and cap-dependent translation (22, 23, 25). Consistently, the phosphorylated form of 4E-BP1 was sharply upregulated in the DTCs (higher migrating band; ref. 26), as indicated by Western blots probed with an antibody raised against total 4E-BP1 and phospho-specific residues (Fig. 2A; Supplementary Fig. S1A).

We therefore hypothesized that DTCs had a greater capacity for the biosynthesis of MCL-1 protein. As only some cancers downregulate MCL-1 in response to cap-dependent translation disruption (21), we first asked whether MCL-1 was under cap-dependent regulation in EGFR-mutant lung cancer cells. We have previously demonstrated gefitinib treatment leads to downregulation of mTORC1/p4E-BP1/MCL-1 in EGFR-mutant lung cancers (14) and confirmed osimertinib did the same (Supplementary Fig. S1B). We more directly probed the mTOR pathway by treating the PC9, HCC4006 and HCC827 cells with specific, pharmaceutical inhibitors of TORC1/2. Here, we found that phosphorylation of 4E-BP1 was significantly inhibited with both TORC1/2 inhibitors, AZD2014 and MLN0128 (Fig. 2B). Consistently, MCL-1 levels were decreased in both AZD2014 and MLN0128-treated cells (Fig. 2B). Furthermore, knockdown of mTOR with two different siRNAs also led to a sharp reduction in p4E-BP1 and MCL-1 (Fig. 2C; Supplementary Fig. S1C). The combination of MLN0128 with gefitinib diminished cell viability more than either drug alone (Supplementary Fig. S1D) and increased apoptosis compared with either agent alone (Supplementary Fig. S1E) in parental EGFR-mutant cell lines. In addition, the combination of TORC1 inhibition and EGFR inhibition prevented the outgrowth of DTCs (Supplementary Fig. S1F).

We next investigated whether gefitinib/osimertinib surviving DTCs maintain MCL-1 expression under mTORC1 regulation. Affirming this pathway remained intact in the DTCs, phosphorylated 4E-BP1 and MCL-1 were inhibited in the DTCs following either AZD2014 or MLN0128 treatment, compared with control DTCs (Supplementary Fig. S1G), and similar to the parental cells (Fig. 2B). To ensure T790M was not contributing to the DTC state, which could theoretically have enriched over the 6-day treatment to confer resistance to gefitinib, we treated gefitinib-DTCs with osimertinib and found the DTCs were similarly resistant to osimertinib as they were to gefitinib, in stark contrast to the sensitivity of the parental cells (Supplementary Fig. S1H). These data demonstrate the mTORC1/p4E-BP1/MCL-1 pathway is under the control of the EGFR pathway in EGFR-mutant lung cancers, and the mTORC1/p4E-BP1/MCL-1 pathway is hyperactive in DTCs.

We next probed further the involvement of enhanced protein translation downstream of TORC1. Beyond MCL-1, other key oncogenic proteins with short mRNA half-lives, prominently c-MYC and cyclin D1, are tightly regulated by TORC1-mediated cap-dependent translation (27–29). Consistent with an enrichment of TORC1-mediated cap-dependent translation, both c-MYC and cyclin D1 were sharply elevated at the protein level in PC9 gefitinib-DTCs (Fig. 2A), with little change at the RNA level (Fig. 2D); upregulation of c-MYC and cyclin D1 were confirmed in the HCC827 gefitinib-DTCs (Supplementary Fig. S1A). We next measured translation rates directly in DTCs, and compared them to parental cells. We used a dual luciferase reporter assay that determined both rates of cap-dependent and cap-independent (IRES-dependent) translation, with the latter serving as a control for transfection efficiency (15, 30, 31). In this assay, Firefly luciferase activity associates with cap-dependent translation while Renilla luciferase activity associates with IRES-dependent translational rates, and the ratio is reflective of cap-dependent rates. We detected a sharp increase in cap-dependent translation in the DTCs compared with the parental cells (∼6-fold; Fig. 2E). These data altogether indicate an increased cap-dependent translation in DTCs.

MCL-1 functions primarily as a binding partner to proapoptotic BIM and BAK, by binding to and neutralizing them (16). Indeed, we and others have demonstrated BIM mediates apoptosis in EGFR-mutant lung cancers treated with EGFRi (14, 32). In order to determine whether the increased MCL-1 was functioning at this level, we immunoprecipitated MCL-1 in the parental and DTC cells. As demonstrated in Fig. 2F, MCL-1 was bound to more BIM and BAK in the DTCs compared with the parental cells, indicating the increased MCL-1 was neutralizing the two proapoptotic proteins. In addition, phosphorylated MCL-1 at S159/T163 is concomitantly upregulated in DTCs, further evidencing altered translation as causative of increased MCL-1, and not dephosphorylation-driven stabilization (ref. 33; Fig. 2G). We next performed binding assays with the 5′-7 methylguanosine (m⁷G) cap structure, which is responsible for interacting with mRNAs undergoing cap-dependent translation (Fig. 2G). In fact, sharply increased levels of cellular eIF component proteins (eIF3A, eIF4G1, and eIF4G3) led to increased m⁷G cap:eIF4E:eIF4G binding complexes (∼30-fold increase) in DTCs compared with parental cells (Fig. 2G and H), indicating increased cap-dependent translation. In total, these data indicate DTCs upregulate the TORC1–EIF4 pathway, leading to an increased rate and reliance toward translation of MCL-1. Furthermore, this MCL-1 binds BIM and BAK, which is important given that “free” BIM mediates EGFRi-mediated apoptosis in EGFR-mutant lung cancers (14, 34, 35).

As Western blot analyses can only clue us to an increase in MCL-1 expression within the whole-cell population, we sought to more clearly determine whether a subpopulation of cells that were high in MCL-1 expression could enrich within the total population to survive as DTCs. To address this question, we designed a GFP-tagging assay on a high-throughput cell content imager to determine whether preexisting, high MCL-1–expressing cells would enrich following initial gefitinib treatment, therefore recapitulating the increase in MCL-1 expression seen in the pooled cell population experiment (Fig. 3A and B) and the increase in MCL-1 in the gefitinib-treated patient specimens following partial responses (Fig. 1F). We utilized a bicistronic MCL-1-IRES vector tagged with GFP (GFP-IRES-MCL-1; ref. 36). We performed this experiment with a starting cell population of approximately 10% of GFP and 10% of GFP-MCL-1 cells (Fig. 3A and B), and treated the cells as we did for the pooled cell population experiments (Fig. 1C). At the end of treatment at day 7 followed by no drug for 3 days, the population of GFP-MCL-1 cells at day 10 went from ∼10% to ∼60% in PC9 cells, and ∼10% to ∼30% in HCC4006 cells (Fig. 3C). In contrast, the GFP control cells remained at similar levels (i.e., 10%) at day 10 (Fig. 3B and C). We also seeded untreated GFP and GFP- MCL-1 as a control (Supplementary Fig. S2). As expected, both GFP and MCL-1–expressing cells with gefitinib treatment undergo key signaling shutdown, as reflected by cell-cycle arrest (refs. 5, 9, 14; Fig. 4A). In contrast, the MCL-1 expressing cells are protected from gefitinib-induced apoptosis (Fig. 4B), which translates to large differences in viability across multiple doses of gefitinib in EGFR-mutant NSCLCs (Fig. 4C). Altogether, these data indicate cells expressing high amounts of MCL-1 within an EGFR-mutant lung cancer population of cells are refractory to EGFRi-induced apoptosis, and high MCL-1 expression is sufficient for EGFR-mutant cancer cells to persist following gefitinib treatment.

We next determined whether exogenous MCL-1 would protect EGFR-mutant NSCLCs from gefitinib treatment in vivo. For these experiments, we injected PC9 tumors expressing exogenous MCL-1 or GFP-expressing control plasmids into immunocompromised mice. Consistent with the in vitro studies (Fig. 4B and C), the control tumors shrunk significantly (Fig. 4D), whereas the PC9 tumors expressing MCL-1 failed to shrink in response to gefitinib (Fig. 4E). These data indicate MCL-1 expression is sufficient to maintain the viability of PC9 tumors (Fig. 4E and F).

MCL-1 is an experimentally proven drug target in several cancers (37–40) and is an important oncogene downstream of TORC1 (21, 25, 36, 41). Recently, MCL-1–specific inhibitors have been developed. We first utilized the specific MCL-1 inhibitor A-1210477 (39, 42) and treated the EGFR-mutant PC9 and HCC4006 cells with gefitinib in the presence or absence of A-1210477. We found that the combination of A-1210477 and gefitinib led to markedly fewer viable cells than either drug alone (Fig. 5A) and increased apoptosis compared with either agent alone (Fig. 5B). Importantly, the addition of A-1210477 to gefitinib in EGFR-mutant lung cancer cells was sufficient to prevent the survival of most DTCs (Fig. 5C). Using a second MCL-1 inhibitor, with in vivo capabilities, S63845 (43), we reproduced these results in these three assays (Fig. 5D–F). Lastly, lentiviral transduced shMCL-1 stable cells (Supplementary Fig. S3) greatly sensitized EGFR-mutant cells to increasing doses of either gefitinib or osimertinib, compared with scramble cells (Supplementary Fig. S3A). These data altogether indicate MCL-1 inhibitors are sufficient to prevent the survival of most EGFR-mutant lung cancer DTCs through apoptosis, and suggest targeting MCL-1 in DTCs is a potent and rational drug strategy.

We next investigated whether the addition of the MCL-1i and EGFRi was effective in vivo at improving initial responses to single-agent EGFRi. We treated EGFR-mutant tumors with gefitinib for 3 days, S63845 for 3 days, or the combination for 3 days, and compared these treatments to the control group. Measurements of PC9 tumors at 8 days (Fig. 5G, left), similar to 6 (Supplementary Fig. S4A) and 7 days (Supplementary Fig. S4B), demonstrated treatment with the combination of S63845 (25 mg/kg/body weight) and gefitinib (50 mg/kg/body weight) led to a robust shrinking of tumors, significantly exceeding what single-agent gefitinib was capable of (Fig. 5G). This was also demonstrated in HCC827 tumors at 11 (Fig. 5G, right), 9 (Supplementary Fig. S5A), and 10 days (Supplementary Fig. S5B). Importantly, the combination was well tolerated (Supplementary Figs. S4C and S5C). These data indicate newer MCL-1 mimetics can combine with EGFRi to substantially improve initial responses to EGFRi in vivo.

Several pure MCL-1 inhibitors such as S63845 and AMG 176 are now being clinically developed (39, 43), with Amgen's AMG 176 in clinical trials (for example, clinical trial number NCT02675452). Although it is not yet known whether these direct antagonizers of MCL-1 will demonstrate efficacy or tolerability in humans, the cyclin-dependent kinase (CDK) inhibitor dinaciclib has emerged as an effective MCL-1 inhibitor (44–46) that demonstrates tolerability and activity in clinical trials (47). We therefore assessed dinaciclib as a sensitizer to gefitinib in EGFR-mutant lung cancers to determine whether this strategy would also lead to the elimination of MCL-1 and suppression of DTCs. Indeed, the combination led to potent downregulation of MCL-1 and the induction of apoptosis in EGFR-mutant lung cancer cells to reasonably low doses of gefitinib (Fig. 6A). MCL-1 expression was sufficient to block the ability of the dinaciclib/EGFRi combination to induce apoptosis (Fig. 6B), translating to protection of total cell viability. In vivo, PC9 xenografted tumors were treated with either single-agent gefitinib, single-agent dinaciclib, or the combination at the same doses. Single-agent dinaciclib had little effect on tumors, whereas, as expected, single-agent gefitinib was able to block tumor growth and modestly shrink tumors (Fig. 4D); however, the combination was sufficient to eliminate 60% of palpable tumors (Fig. 6C). These data further corroborate MCL-1 as a key early target in EGFR-mutant lung cancers (Fig. 6D) and demonstrate a second class of drug that may be used in combination with EGFRi to enhance early tumor responses.

We next asked whether MCL-1 was also upregulated in other anticancer therapies, particularly ones that induced marked cell death, in EGFR-mutant NSCLC. EGFR-mutant NSCLC lung cancer cell lines were exposed to 10 μmol/L of cisplatin and 500 nmol/L of gemcitabine for 24 hours (Supplementary Fig. S6A, left), or treated for 72 hours and left to recover for an additional 72 hours without drug to enrich DTCs (Supplementary Fig. S6A, right). Although treatment with cisplatin led to the downregulation of MCL-1 and cell death, as evidenced by cleavage of PARP (Supplementary Fig. S6A, left), the surviving fraction of cells (i.e., DTCs) had higher MCL-1 levels (Supplementary Fig. S6A, right). Similarly, MCL-1 levels were also higher in the gemcitabine-surviving PC9 cells ((Supplementary Fig. S6A, right). We also treated the PC9 and HCC4006 cells with radiation and harvested the cells 24 hours and 72 hours later, respectively (Supplementary Fig. S6B). In the cells exposed to radiation, MCL-1 levels were initially decreased (Supplementary Fig. S6B, left). However, as the amount of cell death markedly increased in both the PC9 and HCC4006 cells over 72 hours, MCL-1 levels also increased in the surviving populations treated with radiation, and in fact became markedly higher in the surviving HCC4006 cells compared with untreated cells. These data altogether indicate that MCL-1 enrichment among surviving cells following therapeutic cellular insults appears widespread in EGFR-mutant NSCLC.

In this study, we demonstrate that (i) a subpopulation of EGFR-mutant NSCLCs can endure initial EGFRi therapy, including both first-generation and third-generation EGFRi, via prosurvival MCL-1, (ii) the increase in MCL-1 expression seen in the bulk population can be recapitulated by the enrichment of a small population of cells expressing exogenous MCL-1, (iii) the increase in MCL-1 occurs through an enrichment of TORC1–eIF4E-mediated cap-dependent translation of MCL-1, (iv) MCL-1 is functional and binds to BIM to neutralize it, and (iv) pharmaceutical inhibition of MCL-1 can largely eliminate EGFR-mutant DTCs in vitro and improve initial responses in vivo.

The inability of EGFRi to induce apoptosis has been recently delineated as an important mechanism of upfront resistance to EGFRi (9–12, 48–50). Although these studies, including ours, have focused on a clearly important role for the proapoptotic BIM, MCL-1 has become an increasingly relevant drug target and accordingly the development of MCL-1–specific inhibitors has excelled (39, 43, 51). Of note, MCL-1 is an intimate partner of BIM, serving as a key BIM neutralizer (17, 25). As further support of an important role of MCL-1 in EGFR-mutant lung cancer survival, and consistent with MCL-1 acting as a key survival factor in these cancers (Figs. 3 and 4), it was recently reported that MCL-1 levels increase in EGFRi-acquired resistant patients (52). In addition, recent studies have demonstrated a key role for MCL-1 in the survival of subsets of not only NSCLC (40), but also breast (42, 53), neuroblastoma (25), and blood cancers (43). Although the protein levels of MCL-1 in DTCs were sharply upregulated (Fig. 1D), the transcript levels of MCL-1 from the same DTCs did not markedly increase (Fig. 1E). This was also true of the other two oncogenic proteins with markedly short half-lives—cyclin D1 and c-MYC (29). Therefore, DTCs with high MCL-1 expression were largely a result of protein changes, with subsequent studies indicating this change was occurring at the level of translational output. To put these data into context, while a disconnect between transcriptional outputs and translational outputs is well established in mammalian cells, interestingly, activation of the EGFR pathway reportedly confers a larger disassociation between these outputs (54). Additionally, biosynthesis of much of the translational machinery itself is governed by translation (55), creating a self-regulating loop, at least somewhat liberated from transcriptional constraints. Biosynthesis of proteins in mammalian cells are initiated through two major pathways: cap-dependent translation, where recognition of a modified guanosine to the 5′ end of an mRNA by eIF4E is required (56), and a process independent of the eIF4 complex that relies on internal ribosome entry sites (IRES). In cap-dependent translation, the mTORC1 complex phosphorylates 4E-BP1, resulting in 4E-BP1 sequestration away from eIF4E, promoting the binding of eIF4E to eIF4G and assembly of the eIF4F complex: in cancers in which this pathway is perpetually active, dependence develops on short-lived oncogenes, such as MCL-1 (20, 23). In addition, phosphorylation of 4E-BP1 by the mTORC1 complex is likely rate-limiting for cap-dependent translation in cancers (57). mTOR itself was highly upregulated (Supplementary Table S1; Fig. 2A), and 4E-BP1 was markedly more phosphorylated, in the EGFR-mutant DTCs (as evidenced by higher migrating bands and/or phospho-specific antibodies; Fig. 2A; Supplementary Fig. S1A). When we measured cap-dependent translation directly, it was increased ∼6-fold in the DTCs (Fig. 2E), and MCL-1 was functional (Fig. 2F).

It is reasonably well established that cells, in response to different stresses, often downregulate global protein synthesis (58, 59), the bulk of which is orchestrated by cap-dependent mRNA translation, therefore making our data on DTCs counterintuitive. However, eIF4, mTOR and pp70S6K signaling are among the highest upregulated pathways in DTCs (Supplementary Table S2), corroborating the laboratory experiments in this study (Figs. 1 and 2). It remains unclear if the cap-dependent translational machinery preferentially translates MCL-1 and other proteins in DTCs, or the identity of the underlying signal to do so, if one such exists. Future studies will be aimed at delineating such things.

In this study, we found EGFR-mutant NSCLC DTCs survive initial EGFRi therapy by an orchestrated evasion of apoptosis caused by a sharp increase in MCL-1 mRNA translation. These data are reminiscent of a previous study that demonstrated coincubation with the BCL-2/xL inhibitor ABT-737 sensitized early-surviving EGFR-mutant lung cancers to EGFRi (60). Our data pointing to a dependence on MCL-1 translation is not altogether surprising because MCL-1 is highly regulated at the level of translation (61). Interestingly, while FBW7 depletion can lead to high levels of MCL-1 in EGFR-mutant NSCLCs (52), in our DTC model these cells did not have depleted FBW7 levels (Supplementary Table S3), indicating alternative strategies EGFR-mutant lung cancer cells may use to increase MCL-1 levels. In addition, Wu and colleagues recently found the p23-activated kinase (PAK1) could activate the PI3K pathway and upregulate MCL-1, contributing to EGFRi resistance (62); the same group also found the focal-adhesion–related protein paxillin could alter MCL-1 phosphorylation and expression, again contributing to EGFRi resistance (63). This study also further highlighted the important relationship between BIM and MCL-1 in EGFR-mutant lung cancers (63). The findings in these studies and ours make a compelling case that different cellular processes can commonly lead to what seems to be a critical signaling event in EGFRi resistance: MCL-1 upregulation. Of note, we also found evidence of increased KDM5A expression (Supplementary Table S1), verifying past reports on EGFR-mutant DTCs (7, 8).

There are now multiple MCL-1–specific inhibitors being developed. As one of these drugs, AMG 176 (e.g., NCT02675452) is currently in clinical trial testing, the use of these drugs in combination with drugs such as EGFRi could be eventually tested. Importantly, we demonstrated in EGFR-mutant cell line xenografts that S63845 markedly sensitized these tumors to early gefitinib treatments (Fig. 5G).

We also demonstrate that dinaciclib, which has an established clinical tolerability profile and could be tested right away (47, 64), can serve as a surrogate MCL-1 inhibitor to thwart the emergence of DTCs. Although dinaciclib also inhibits CDK1, CDK2, and CDK5, it has emerged as a potent MCL-1 inhibitor in vivo due to the short half-life of MCL-1 mRNA that is sensitive to inhibition of CDK9 (46, 65). Although dinaciclib has pleiotropic effects over long-term treatments as a result of inhibiting the other CDKs, the emerging picture in people is much different. In a recent pharmacodynamics study of dinaciclib, it was demonstrated that its exposure time peaks at 2 hours in humans (64). The same study demonstrated dinaciclib effectively inhibited MCL-1 at 4-hour exposure, which returned to baseline levels at 24 hours. Importantly, dinaciclib failed to inhibit pRB at both of those time points, indicating it poorly inhibits CDK1/2 in humans (64). These data suggest dinaciclib primarily acts through inhibitory effects on CDK9 in humans, without the exposure time to affect CDK1/2. This would indeed suggest one of dinaciclib's primary modes of activity is via downregulation of MCL-1 via CDK9, and therefore would serve as a rational copartner with EGFRi to eliminate DTCs and improve efficacies of these drugs. Our in vivo data further support the combination of EGFRi and dinaciclib as a viable combination therapy to eliminate DTCs and enhance initial tumor responses (Fig. 6C).

Bhola and colleagues have recently identified a subpopulation of acute myeloid leukemia cells that are resistant to apoptosis (66). Impressively, the most apoptosis-resistant subpopulations are superior to predict response to chemotherapy than the overall apoptosis-primed state of the entire population, demonstrating a critical role of this subpopulation in chemotherapy response. This study also points to a phenomenon in which DTCs emerge by circumventing apoptosis; in EGFR-mutant NSCLCs treated with EGFRi, the result is a surviving fraction of cells that may survive long enough to acquire secondary and bypass track mutations and regrow in the presence of drug (Fig. 6D). Eliminating these cells early could thwart the development of acquired resistance in EGFR-mutant lung cancers and, as such, provide important insights into new pharmaceutical strategies that could improve responses to EGFRi in lung cancer.

J.D. Leverson and A.J. Souers hold ownership interest (including patents) in AbbVie. A.N. Hata reports receiving commercial research grants from Amgen, Novartis, and Relay Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K.-A. Song, A.N. Hata, H. Ebi, A.C. Faber

Development of methodology: J. Ham, A.C. Faber

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Hosono, S. Jacob, Y. Murakami, N.U. Patel, B. Hu, K.M. Powell, C.M. Coon, Y. Oya, J.E. Koblinski, Y. Yatabe

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.-A. Song, C. Turner, T.L. Lochmann, Y. Murakami, B.E. Windle, J.D. Leverson, A.J. Souers, S. Boikos, Y. Yatabe, H. Ebi, A.C. Faber

Writing, review, and/or revision of the manuscript: K.-A. Song, C. Turner, T.L. Lochmann, J.E. Koblinski, H. Harada, J.D. Leverson, A.J. Souers, A.N. Hata, S. Boikos, H. Ebi, A.C. Faber

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.-A. Song, C. Turner, J. Ham, H. Harada

Study supervision: K.-A. Song, A.C. Faber

Other (management of In vivo work): K.-A. Song

We thank Katherine Borden (University of Montreal) and Cristian Bellodi (Lund University) for helpful discussions. This work was supported by an NCI K22-CA175276 Career Development Award (A.C. Faber). A.C. Faber is supported by the George and Lavinia Blick Research Fund and is a Harrison Endowed Scholar in Cancer Research. Services and products in support of the research project were generated by the VCU Massey Cancer Center Mouse Model Shared Resource, supported, in part, with funding from NIH-NCI Cancer Center Support Grant P30CA016059. H. Ebi is supported by Grants-in-Aid for Scientific Research (16K07164) and Fund for the Promotion of Joint International Research (15KK0303) from Japan Society for the Promotion of Science. The pFR_HCV_xb (Addgene plasmid #11510) was a gift from Phil Sharp.

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.