Targeting the EMT transcription factor TWIST1 overcomes resistance to EGFR inhibitors in EGFR-mutant non-small-cell lung cancer.

Patients with EGFR-mutant non-small-cell lung cancer (NSCLC) have significantly benefited from the use of EGFR tyrosine kinase inhibitors (TKIs). However, long-term efficacy of these therapies is limited due to de novo resistance (~30%) as well as acquired resistance. Epithelial-mesenchymal transition transcription factors (EMT-TFs), have been identified as drivers of EMT-mediated resistance to EGFR TKIs, however, strategies to target EMT-TFs are lacking. As the third generation EGFR TKI, osimertinib, has now been adopted in the first-line setting, the frequency of T790M mutations will significantly decrease in the acquired resistance setting. Previously less common mechanisms of acquired resistance to first generation EGFR TKIs including EMT are now being observed at an increased frequency after osimertinib. Importantly, there are no other FDA approved targeted therapies after progression on osimertinib. Here, we investigated a novel strategy to overcome EGFR TKI resistance through targeting the EMT-TF, TWIST1, in EGFR-mutant NSCLC. We demonstrated that genetic silencing of TWIST1 or treatment with the TWIST1 inhibitor, harmine, resulted in growth inhibition and apoptosis in EGFR-mutant NSCLC. TWIST1 overexpression resulted in erlotinib and osimertinib resistance in EGFR-mutant NSCLC cells. Conversely, genetic and pharmacological inhibition of TWIST1 in EGFR TKI-resistant EGFR-mutant cells increased sensitivity to EGFR TKIs. TWIST1-mediated EGFR TKI resistance was due in part to TWIST1 suppression of transcription of the pro-apoptotic BH3-only gene, BCL2L11 (BIM), by directly binding to BCL2L11 intronic regions and promoter. As such, pan-BCL2 inhibitor treatment overcame TWIST1-mediated EGFR TKI resistance and were more effective in the setting of TWIST1 overexpression. Finally, in a mouse model of autochthonous EGFR-mutant lung cancer, Twist1 overexpression resulted in erlotinib resistance and suppression of erlotinib-induced apoptosis. These studies establish TWIST1 as a driver of resistance to EGFR TKIs and provide rationale for use of TWIST1 inhibitors or BCL2 inhibitors as means to overcome EMT-mediated resistance to EGFR TKIs.

INTRODUCTION

Lung cancer remains the leading cause of cancer related mortality in the United States and worldwide. Despite a 15% five-year survival rate, there have been improvements in the treatment of subsets of non-small cell lung cancer (NSCLC) patients with known targetable molecular drivers such as mutations in EGFR, BRAF and MET, and translocations involving ALK, ROS1, RET and NTRK1/2 (1–3). Previous studies have demonstrated that patients with EGFR-mutant tumors (~15%) can have marked responses to EGFR tyrosine kinase inhibitors (TKIs). While approximately 70% of patients demonstrate responses to such therapies, long-term efficacy of these therapies is limited due to the inevitability of acquired resistance and frequent de-novo resistance (~30%) (4–6). Efforts to identify drivers of acquired resistance to first generation EGFR TKIs have revealed multiple mechanisms of resistance including T790M gatekeeper EGFR mutations (~49%), MET amplification (~5%), conversion to small-cell lung cancer (~14%), and PIK3CA mutations (~5%) (7).

In as many as 20% of patients, resistance to EGFR TKIs including third generation inhibitors, such as osimertinib, is associated with an epithelial-mesenchymal transition (EMT) phenotype (7–10). EMT is a reversible process of transdifferentiation in which epithelial cells lose their polarity and cell-cell interactions and adopt a mesenchymal phenotype (11, 12). This process is associated with a variety of pro-tumorigenic functions such as with increased invasion, metastasis, and suppression of failsafe programs of apoptosis and senescence (11, 12). Interestingly, the presence of an EMT or mesenchymal phenotype is associated with both de-novo as well as acquired resistance to EGFR TKIs (7, 13–15). Previous studies have demonstrated that upregulation of AXL, TFG-β, and IGF1R signaling axes are drivers of EMT-mediated acquired resistance to EGFR TKIs (8, 16–18). Recent studies have implicated EMT transcription factors (EMT-TFs), which are drivers of global transcriptional changes that lead to EMT, in resistance to targeted therapies in EGFR-mutant NSCLC (19). Specifically, upregulation of the EMT-TFs, SNAI2 and ZEB1, have been shown to can confer resistance to EGFR TKIs (20–22). However, the mechanism(s) by which these EMT-TFs mediate resistance and therapeutic strategies to target these EMT-TFs have been lacking.

We have previously demonstrated that the EMT-TF, TWIST1, is required for oncogene-driven NSCLC (23). In multiple oncogene-driver dependent settings, including tumors with EGFR mutations, TWIST1 functions to suppress oncogene-induced senescence and apoptosis (23–25). In addition to suppressing failsafe programs, TWIST1 has also been shown to promote EMT, metastasis, and therapeutic resistance (26–29). We have also identified a first-in-class inhibitor of TWIST1, harmine that has significant anti-tumor activity in oncogene driver dependent NSCLC including EGFR-mutant NSCLC (25). In the current study, we demonstrated genetic and pharmacological inhibition of TWIST1 resulted in growth inhibition and apoptosis in EGFR-mutant NSCLC cell lines, including cells with acquired resistance T790M mutations. We also identified TWIST1 as a driver of resistance to EGFR TKIs in EGFR TKI naïve EGFR-mutant NSCLC cell lines as well as in EGFR TKI acquired resistant cell lines with T790M mutations. We further demonstrated that TWIST1 is able to mediate resistance in a transgenic mouse model of autochthonous EGFR-mutant lung cancer. We have identified that one mechanism by which TWIST1 mediates resistance is through suppression of EGFR TKI-induced apoptosis by directly binding to the promoter and intronic regions of the pro-apoptotic BH3-only gene, BCL2L11 (BIM) and repressing BIM transcription. Additionally, we demonstrated that TWIST1-mediated EGFR TKI resistance can be overcome with either a BCL-2/BCL-XL inhibitor, or the TWIST1 inhibitor harmine, suggesting that targeting TWIST1 in the clinic may be a viable option to overcome EMT-mediated resistance to EGFR TKIs.

RESULTS

Genetic or pharmacologic Inhibition of TWIST1 results in growth inhibition and apoptosis in EGFR-mutant NSCLC

We previously observed that TWIST1 expression is required for tumorigenesis in oncogene-driven NSCLC as inhibition of TWIST1, in oncogene driver dependent NSCLC cell lines, results in activation of latent senescence and/or apoptotic programs (23–25). To more comprehensively test the role of TWIST1 in EGFR-mutant lung cancers, we first assessed relative expression of TWIST1 in a panel of EGFR-mutant NSCLC cell lines with TKI sensitizing mutations such as the L858R mutation and Exon 19 deletions and EGFR-mutant NSCLC cell lines with the EGFR TKI acquired resistance T790M mutation (Figure 1a). TWIST1 was broadly expressed across both EGFR TKI sensitive and resistant cell lines (Figure 1a). Of note, we identified an EGFR-mutant cell line, H1650, previously demonstrated to have de-novo resistant to EGFR TKIs (30) had increased levels of TWIST1 mRNA and protein. Next, we infected the panel of EGFR-mutant lines with shRNAs targeting TWIST1 or with scrambled control shRNA. Genetic inhibition of TWIST1 inhibits growth in the majority of lines screened (Figure 1b). We have previously identified and characterized a novel TWIST1 inhibitor, harmine, that had anti-cancer activity in oncogene driver defined NSCLCs, inhibited multiple TWIST1-dependent functions, and induced degradation of TWIST1 (25). Similar to our previous findings in a limited number of EGFR-mutant cell lines, harmine markedly inhibited growth across the large panel of EGFR-mutant NSCLC cell lines, including EGFR TKI resistant lines, similar to the effects seen following silencing of TWIST1 (Figure 1b). While we have previously observed that genetic and pharmacological inhibition of TWIST1 primarily results in oncogene-induced senescence (OIS) (23–25), there was a subset of cell lines that appeared more dependent on TWIST1 expression for survival and underwent apoptosis following inhibition of TWIST1 (23, 25). We identified a subset of EGFR-mutant cell lines (H1975 and PC9) that underwent apoptosis following knockdown of TWIST1 and harmine treatment (Figure 2a-b, Supplementary Figure 1). PC9 cells have an EGFR TKI sensitizing EGFR exon 19 deletion (ΔE746-A750) and H1975 cells have both EGFR TKI sensitizing L858R mutation and an acquired resistance T790M mutation, suggesting that targeting TWIST1 may be an effective therapeutic target for EGFR-mutant disease in both the EGFR TKI naïve and EGFR TKI acquired resistance setting. Of note, genetic and pharmacologic inhibition of TWIST1 (Figure 1b) was also effective in the setting of T790M independent resistance such as in the EGFR TKI resistant cell line H1650 (30).

Figure 1:

TWIST1 is required for EGFR-mutant NSCLC.

(A) LEFT: Quantitative RT-PCR demonstrating baseline TWIST1 mRNA (TOP) and protein (BOTTOM) levels in a panel of erlotinib sensitive and resistant EGFR-mutant NSCLC cells. RT-PCR was normalized to Hcc2935 mRNA levels. Data represent mean ±SD (n=3 technical replicates). RIGHT: Chart demonstrating EGFR-mutations in the NSCLC cell lines utilized. EGFR TKI sensitive cells are colored in green and EGFR TKI resistant cells are colored in red. (B) LEFT: Cell-Titer Glo assays demonstrating that knockdown of TWIST1 results in growth inhibition in a panel of EGFR-mutant NSCLC cell lines. Cells were infected with shScram or shRNA targeting TWIST1 (shTWIST1 #1, #2) for 6 days. Viability data was normalized to shScram control. Data represent mean ± SD (n=4 technical replicates). RIGHT: MTS assays demonstrating that harmine has activity in a panel of EGFR-mutant NSCLC cells. Cells were treated with harmine for 72 hours. Data represent mean ± SD (n=4 technical replicates).

Figure 2:

Inhibition of TWIST1 results in apoptosis in a subset of EGFR-mutant NSCLC cell lines

(A) Western blot demonstrating that knockdown of TWIST1 induces apoptosis in PC9 cells with EGFR TKI sensitizing EGFR exon 19 deletion (ΔE746-A750) and H1975 cells with both EGFR TKI sensitizing L858R mutation and an acquired resistance T790M mutation. Cells were infected with shScram and shRNA targeting TWIST1 (shTWIST1 #1–3) for 72 hours (PC9) or 6 days (H1975) and harvested for Western blot. (B) UPPER: Western blots demonstrating that harmine treatment results in PARP cleaved in PC9 and H1975 cells. Cells were treated with harmine for 48 hours and harvested for Western blot analysis. LOWER: Active Caspase 3–7 staining demonstrating induction of apoptosis in PC9 and H1975 following 48 hours of harmine treatment. Data represents mean ± SD (n=3 biological replicates). **, p<.01, 2-tailed Student’s t-test.

TWIST1 is necessary and sufficient for EGFR TKI resistance in a subset of EGFR-mutant NSCLC cell lines

Recent evidence has suggested that EMT-TFs mediate resistance to EGFR targeted therapy in lung cancer (20–22, 31). TWIST1 has been implicated in chemoresistance in lung cancer and other cancer types (27, 32–34). Given the requirement of TWIST1 for EGFR-mutant NSCLC and its role in suppressing OIS and apoptosis in NSCLC, we investigated whether enforced TWIST1 expression would be sufficient to cause resistance to EGFR TKIs, using a panel of doxycycline inducible TWIST1 overexpressing EGFR-mutant NSCLC cell lines. TWIST1 overexpression in these lines was sufficient to cause resistance to both 1st and 3rd generation EGFR TKIs (Figure 3 and Supplementary Figure 2). Additionally, we observed that TWIST1-mediated resistance was associated with suppression of EGFR TKI-induced apoptosis in cells with and without the T790M EGFR gatekeeper mutation (Supplementary Figure 2).

Figure 3:

TWIST1 overexpression is sufficient to mediate resistance to EGFR TKIs.

(A) UPPER: MTS or Cell-Titer Glo assays demonstrating that TWIST1 overexpression results in decreased response to erlotinib. H3255 TRE3G-TWIST1 (UPPER LEFT), 11–18 TRE3G-TWIST1 (UPPER RIGHT), and PC9 TRE3G-TWIST1 (LOWER) were pre-treated with doxycycline for 72 hours and then treated with doxycycline and erlotinib for 72 hours. Data represent mean ± SD (n=4 technical replicates). *, P<.05, **, P<.01, 2-way ANOVA, followed by Tukey’s Test. (B) MTS assay demonstrate that TWIST1 overexpression decreases response to osimertinib. H1975 TRE3G-TWIST1 were pre-treated with doxycycline for 72 hours prior to a 72 hour treatment with osimertinib. Data represent mean ± SD (n=4 technical replicates). *, P<.05, **, P<.01, 2-way ANOVA, followed by Tukey’s Test.

To investigate the requirement of TWIST1 for erlotinib resistance, we assessed whether genetic or pharmacologic inhibition of TWIST1 in EGFR-mutant TKI resistant cell lines could restore sensitivity to EGFR TKIs. As demonstrated in Figure 1a, we identified that the EGFR-mutant cell line, H1650 had increased levels of TWIST1 mRNA and protein. Interestingly, this cell line demonstrates de-novo resistance to EGFR TKIs (30). We found that genetic silencing of TWIST1 increases sensitivity of this cell line to erlotinib (Figure 4a). We observed a similar increase of sensitivity to erlotinib when used in combination with our small molecule TWIST1 inhibitor, harmine (Figure 4b). This increase in erlotinib sensitivity corresponded to increased apoptosis and expression of BIM, a pro-apoptotic BH3 protein shown to be critically important for response to EGFR TKIs (35). As such, BIM expression increased with decreased TWIST1 expression following harmine and erlotinib co-treatment (Figure 4b). Of note, the combination of harmine and erlotinib results in an approximate 2–2.2 fold increase in BIM expression, which suggests that upregulation of BIM may be a mechanism by which inhibition of TWIST1 increases EGFR TKI sensitivity (Supplementary Figure 3a). We next investigated the role of TWIST1 in mediating resistance in an EGFR-mutant NSCLC cell line (HCC827R2) with acquired resistance to erlotinib (36). Although TWIST1 was not increased in the resistant cell line compared to the parental cell line, we observed that this cell line maintained a requirement for TWIST1 expression and that targeting TWIST1 in these cells increased sensitivity to erlotinib (Supplementary Figure 3b-c). These observations indicate that inhibiting TWIST1 may be a viable target in the erlotinib resistance settings in which TWIST1 is expressed.

Figure 4:

Inhibition of TWIST1 is sufficient to overcome EGFR TKI resistance.

(A) LEFT: Western blot demonstrating shRNA targeting TWIST1 decreases TWIST1 levels. The erlotinib resistant cell line, H1650 was infected with the indicated shRNA and harvested six days following infection for Western analysis. RIGHT: MTS assay demonstrating that knockdown of TWIST1 in H1650 cells can re-sensitize cells to erlotinib. H1650 cells which harbor both EGFR and PTEN mutations, were infected with the indicated shRNAs for 48 hours and subsequently treated with erlotinib for 72 hours. Data represent mean ±SD (n=4 technical replicates). *, P<.05, **, P<.01, 2-way ANOVA, followed by Tukey’s Test. (B) LEFT: Western blot demonstrating that the combination of harmine and erlotinib results in increased apoptosis as measured by PARP cleavage as well as BIM expression, and decreased TWIST1 expression. H1650 cells were treated with the indicated doses of harmine and erlotinib for 48 hours and harvested for Western analysis. RIGHT: MTS assay demonstrating that harmine treatment increases H1650 cell sensitivity to erlotinib. Cells were treated with the indicated doses of harmine and erlotinib for 48 hours. Data represent mean ±SD (n=4 technical replicates). *, P<.05, **, P<.01, 2-way ANOVA, followed by Tukey’s Test.

TWIST1 suppresses BIM expression

Previous studies have established that response to oncogene targeted therapies requires BIM expression and loss of BIM expression is associated with EGFR TKI resistance in patients (35–39). BIM expression is regulated both transcriptionally and post-translationally (40). We investigated whether TWIST1 could regulate BIM expression in EGFR-mutant NSCLC cell lines. We found that knockdown of TWIST1 resulted in increased BCL2L11 (BIM gene) mRNA and protein expression of BIM (Figure 5a). In the cell lines in which TWIST1 was sufficient to mediate erlotinib resistance, we demonstrated that TWIST1 overexpression resulted in suppression of the mRNA and protein expression of BIM (Figure 5b-c). To evaluate whether TWIST1 decreased BIM expression through a post-translational mechanism, we performed a pulse-chase experiment and demonstrated that TWIST1 did not decrease BIM half-life, suggesting that TWIST1 negative regulation of BIM expression is primarily at the mRNA level (Supplementary Figure 4). To explore whether TWIST1 was directly repressing the transcription of BCL2L11, we performed TWIST1 ChIP on the promoter region and intron 1 which contained multiple E-box binding sites (CANNTG), the putative consensus binding site for TWIST1. We also performed TWIST1 ChIP on a potential TWIST1 binding site contained within the BCL2L11 genomic region in intron 12 previously identified as a putative binding sequence from a global TWIST1 ChIP analysis (41). We identified that TWIST1 bound to one site upstream of the transcriptional start site (BS1) and a site in intron 12 (BS5) (Figure 5d). These studies establish BIM as a novel target gene of TWIST1. While others have previously established the requirement of BIM for response to EGFR TKIs (35, 37–39), we confirmed that in H1975 cells that BIM expression was required for response to osimertinib (Figure 5e). As BIM appeared to be required for EGFR TKI induced apoptosis we examined whether inhibition of anti-apoptotic BCL2 family members with the BCL-2/BCL-XL inhibitor (ABT-737) would be effective in TWIST1 overexpressing EGFR-mutant NSCLC. We observed that BCL2/BCLXL inhibitor (ABT-737) was able to overcome TWIST1-mediated resistance to osimertinib in H1975 TWIST1 overexpression cells (TWIST1-ON) but did not affect osimertinib sensitivity in the absence of TWIST1 (TWIST-OFF) (Figure 5f). These data suggest that TWIST1-mediated resistance may be overcome through use of BH3 mimetics and that these therapies may be more effective in TWIST1 overexpressing EGFR-mutant NSCLC cells.

Figure 5:

TWIST1 suppresses BIM expression.

(A) LEFT: Quantitative RT-PCR (qRT-PCR) demonstrating increased BCL2L11 mRNA levels following knockdown of TWIST1. PC9 cells were infected with the indicated shRNA for 24 hours. Data represent mean ±SD (n=3 technical replicates). RIGHT: Western blot demonstrating that knockdown of TWIST1 increased BIM protein levels. PC9 cells were infected with the indicated shRNA for 72 hours. (B) qRT-PCR demonstrating that TWIST1 overexpression decreased BCL2L11 mRNA levels. PC9 TRE3G-TWIST1 cells and H1975 TRE3G-TWIST1 were treated with doxycycline for 24 hours. Data represent mean ±SD (n=3 technical replicates). (C) Western blot demonstrating that TWIST1 overexpression decreased BIM protein levels. PC9 TRE3G-TWIST1 cells and H1975 TRE3G-TWIST1 were treated with doxycycline for 72 hours. (D) ChIP assay demonstrating TWIST1 binding to promoter and intronic regions of BCL2L11. UPPER: Model demonstrating E-box sites within the BCL2L11 promoter, Intron 1, and Intron 12 that were interrogated for TWIST1 binding. LOWER: qRT-PCR demonstrating that TWIST1 is enriched at multiple sites within the BCL2L11 promoter and intronic regions. Data represent mean ±SD (n=3 technical replicates). *, P<.05, **, P<.01. 2-tailed Student’s t-test. (E) LEFT: Western blot demonstrating that shRNA targeting BIM decreased BIM expression. RIGHT: MTS assay demonstrating decreased response to osimertinib following knockdown of BIM in H1975 cells. H1975 cells that stably express shScram or shBIM were treated with osimertinib for 72 hours. Data represent mean ±SD (n=4 technical replicates). *, P<.05, **, P<.01, 2-way ANOVA, followed by Tukey’s Test. (F) MTS assay demonstrating that TWIST1-mediated resistance to osimertinib can be overcome with ABT-737. H1975 TRE3G-TWIST1 cells were pre-treated with doxycycline for 72 hours and then co-treated with osimertinib and ABT-737 (1µM) ± doxycycline for 72 hours. Data represent mean ±SD (n=4 technical replicates). **, P<.01, 2-way ANOVA, followed by Tukey’s Test.

Creation and Characterization of an Autochthonous EGFR-Mutant Twist1 Overexpressing Lung Tumor Mouse Model

We previously demonstrated that Twist1 could cooperate with mutant Kras for lung tumorigenesis and that genetic or pharmacologic inhibition of Twist1 in this model inhibited growth of these lung tumors (23–25). To investigate whether EMT and Twist1 could impart erlotinib resistance to EGFR-mutant NSCLCs in vivo we made use of transgenic EGFR and Twist1 inducible mouse models (24, 42). Both of these strains are well established doxycycline inducible lung specific transgenic mouse models: CCSP-rtTA/tetO-EGFR (CE), expressing human EGFR and CCSP-rtTA/Twist1-tetO7-luc (CT), expressing mouse Twist1. We crossed these two lines to create triple transgenic mice, CCSP-rtTA/tetO-EGFR/Twist1-tetO7-luc (CET) (Supplementary Fig. 5a). Cohorts of CE and CET mice, aged 4–8 weeks, were administered doxycycline in the drinking water to turn on the transgenes. After 4 weeks, a point by which CE mice were reported to develop lung tumors (42), mice were sacrificed and necropsies performed. Upon comparison of H&E lung sections from CE and CET mice by a veterinary pathologist, both genotypes resulted in similarly diffuse adenocarcinoma growth in both lungs, as had been previously published for the CE model (42), but CET tumors were more anaplastic and had larger, more irregular nuclei (Supplementary Fig. 5b). We had previously shown that Twist1 expression accelerates mutant Kras tumorigenesis (24), but after 4 weeks on doxycycline, tumor burden was similar between CE and CET mice as shown by pathologic assessment of lung tumor burden (0 meaning no hyperplasia and 5 meaning >75% of the lung was affected) and microCT (Supplementary Figure 5b-c). Thus, TWIST1 expression did not appear to have a primary effect on tumor proliferation rate, but rather resulted in a more aggressive or anaplastic appearance in the CET tumors.

To further characterize the novel CET mouse model, we looked at levels of epithelial and mesenchymal markers. We immunostained lung sections from both CE and CET mice with antibodies for E-cadherin, an epithelial marker, and vimentin, a mesenchymal marker. There was no distinguishable difference in levels of either marker between CE and CET mice (Supplementary Figure 5d). We also did not observe any increased metastasis in the CET mice. In other contexts, Twist1 has been shown to impact the proliferation rate of tumor cells as well as apoptosis levels (23, 43). We next examined the levels of proliferation through Ki-67 staining and apoptosis with cleaved caspase 3 staining. The overexpression of Twist1 in CET mice in fact modestly decreased proliferation rates, as measured by Ki-67 IHC, in comparison to CE mice (Supplementary Figure 5e). There was no significant effect on apoptosis with Twist1 expression (Supplementary Figure 5f).

Twist1 expression induces erlotinib resistance in vivo.

After characterizing the novel CET mouse model in the absence of drug treatment, we investigated whether Twist1 expression could induce resistance to the EGFR TKI erlotinib in vivo. As previously described, upon administration of erlotinib to CE mice, most lung tumors regress, with a distribution of objective responses including disease stabilization, partial response, and complete response (42). In order to compare CE and CET mice tumor responses and overall survival, all mice were put on doxycycline, to turn on transgene expression and allowed to develop tumors for 3 weeks. Both CE and CET mice had similar levels of tumor burden prior to the start of treatment. At that time point, treatment day 0, all mice were scanned by microCT and this scan was used as the baseline. The mice were treated for 3 weeks with erlotinib and scanned by microCT each week (Figure 6a). When baseline scans were compared to scans from after 3 weeks of erlotinib treatment, tumor regression was clearly visible in CE mice, while CET mice showed an increase in tumor burden (Figure 6b). All scans were assessed and tumor burden graded on a scale of 0 (no tumor visible) to 5 (lungs completely filled with tumor). Based on the tumor burden change from the beginning to the end of treatment, a majority of CE mice demonstrated no disease progression with erlotinib, with no progression including complete and partial responses as well as stable disease. Conversely, over half of the CET mice had tumor progression over the three weeks of treatment (Figure 6c). When examining erlotinib treatment responses based on degree of lung tumor regression, two-thirds of CE mice showed lung tumor regression, while only a quarter of CET mouse lung tumors regressed (Figure 6c). After the 3 weeks of treatment, mice were monitored for weight loss, lethargy and other signs indicating a need for euthanasia. CET mice median overall survival time from the beginning of treatment was 6.8 weeks, while CE mice lived a median of 8.7 weeks (Figure 6d). Importantly, we have demonstrated that Twist1 expression does not lead to an increased tumor burden in the EGFR-mutant background so an increased tumor burden cannot explain this decrease in overall survival (Figure 6d). These data support that expression of Twist1 in CET mice induces resistance to erlotinib as shown by increased lung tumor burden by microCT and decreased overall survival time following treatment with erlotinib.

Figure 6:

Twist1 overexpression in vivo is sufficient to cause erlotinib resistance.

(A) Treatment schema for CE and CET mice erlotinib treatment. Mice were started on doxycycline, inducing EGFR and Twist1 transgene expression, at 8 weeks of age and allowed to develop tumors for 3 weeks prior to erlotinib treatment. Mice were scanned at the beginning of treatment, week 11, and each week thereafter until the end of treatment. Mice are treated with 50 mg/kg erlotinib by oral gavage 6 days a week for 3 weeks (weeks 11–14). (B) Representative CT images from baseline and after 3 weeks of erlotinib treatment for CE and CET mice. CE mice show a decrease in tumor burden at the end of treatment compared to day 0. CET mice show a drastic increase in tumor burden despite 3 weeks of treatment. (C) Tumor burden, as visualized by CT image, was graded on a scale of 0 (no tumor) to 5 (lungs filled with tumor) at day 0 and the end of treatment. No progression was considered a complete or partial response as well as stable disease. Only 1 CE mouse demonstrated disease progression, while over half of the CET mice progressed despite erlotinib treatment. Regression was a decrease in tumor burden grade at 3 weeks compared to baseline. Two-thirds of CE mice regressed, while only one quarter of CET mice showed regression. (D) Kaplan-Meier overall survival from beginning of treatment. Median survival for CE mice was 8.7 weeks, for CET mice was 6.8 weeks. Difference in survival was statistically significant using the Mantel Cox test, P=0.0073.

To confirm the differences seen by microCT, a cohort of CE and CET mice were treated with erlotinib for 1 week followed by euthanasia for macroscopic and histologic tumor assessment. While partial and complete responses were seen in CE mice, only partial and no responses occurred in the CET mice (Figure 7a). Tumor burden as assessed on H&E slides by a veterinary pathologist between CE and CET mice treated with erlotinib demonstrated an early trend towards CET mice having greater tumor burden at 1 week (Figure 7a).

Figure 7:

Characterization of TWIST1-mediated erlotinib resistance in vivo.

(A) LEFT: H&E images showing comparison of responses seen in CE and CET mice after 7 days of erlotinib treatment. Black bars equal 500 (top) and 50 (bottom) μm. RIGHT: Pathology scores indicating tumor burden as percent of total lung affected. (B) Similar levels of E-cadherin and vimentin staining in CE and CET mice with and without erlotinib treatment, with CET mice expressing Twist1. (C) LEFT: Representative images of Ki-67 and cleaved caspase 3 staining and quantification (RIGHT) of staining showing a decrease in proliferation to similar levels with erlotinib treatment in both CE and CET mice and a decrease in apoptosis in CET compared to CE mice following erlotinib treatment. Differences were statistically significant using Student t-test, * p<0.05, ** p<0.005, *** p<0.0005.

We then examined the mechanism of Twist1-mediated resistance. Since Twist1 is one of the key mediators of EMT, the tumor cells could be undergoing this phenotypic change. However, staining for E-cadherin and vimentin showed no change with Twist1 expression, with or without erlotinib treatment (Figure 7b). Additionally, there was no significant difference between proliferation levels in CE and CET mice following erlotinib treatment (Figure 7c). Interestingly, when the amount of apoptosis was assessed through staining for cleaved caspase 3, the levels of apoptosis were decreased in CET erlotinib treated lung tumors compared to CE erlotinib treated lung tumors (Figure 7c). These data suggest that while the level of proliferation and EMT status is unchanged following erlotinib treatment, Twist1 expression inhibits apoptosis in EGFR-mutant lung tumors following erlotinib treatment. Of note, we also examined Twist1 expression in residual tumors in CE mice following erlotinib treatment. Interestingly, we demonstrated that despite not expressing Twist1 at baseline, Twist1 was upregulated following erlotinib treatment in a subset of residual CE tumors (Supplementary Figure 6a). In addition, in CE tumors with Twist1 upregulation post-erlotinib treatment, cleaved caspase-3 was not expressed (Supplementary Figure 6b). While this is correlative data, it is consistent with our evidence that Twist1 is a mediatory of EGFR TKI resistance by suppressing EGFR TKI-induced apoptosis.

DISCUSSION

We have previously demonstrated that TWIST1 expression is required for oncogene-driven tumorigenesis and that loss of TWIST1 expression results in activation of latent senescence and/or apoptotic programs. Here, we demonstrated that both genetic silencing and pharmacological inhibition of TWIST1 results in growth inhibition in a large panel of EGFR-mutant cell lines. Additionally, we identified that in a subset of EGFR-mutant cell lines inhibition of TWIST1 results in induction of apoptosis. Of note, targeting TWIST1 resulted in growth inhibition in cells with EGFR TKI sensitizing mutations and acquired resistance T790M mutations, suggesting that targeting TWIST1 may be a viable option in EGFR-mutant NSCLC both in the treatment naïve and acquired resistance settings.

Recently, others have demonstrated that EMT-TFs, specifically ZEB1 and SLUG, can contribute to resistance to EGFR TKIs (18–20). Hwang et. al have recently shown that TWIST1 overexpression is sufficient to cause EGFR TKI resistance in a single erlotinib sensitive cell line in long term assays and that VGF regulates TWIST1 (35). Here, we significantly expand upon these studies by demonstrating that TWIST1 overexpression is sufficient to cause resistance to EGFR TKIs, in multiple EGFR-mutant cell lines with and without T790M mutations. We also establish that Twist1 overexpression promotes erlotinib resistance in vivo, using a mouse model of autochthonous EGFR-mutant Twist1 overexpressing lung cancer. In both EGFR-mutant NSCLC cell lines and our mouse model of EGFR-mutant lung cancer, Twist1 overexpression was associated with suppression of EGFR TKI-induced apoptosis.

Importantly, as the third-generation EGFR TKI, osimertinib, has now been adopted in the first line setting (44), the frequency of T790M mutations will likely significantly decrease in the acquired resistance setting (44, 45). Previously uncommon mechanisms of resistance have already been observed at increased frequency after osimertinib including MET and HER2 amplifications, KRAS mutations, additional second site EGFR mutations, EMT and SCLC transformation (44–51). Of note, there are no other FDA approved targeted agents for after progression on osimertinib (45). Thus, there is clearly a need for the development of novel targeted agents to prevent and overcome EGFR TKI resistance.

Our study is the first to establish that TWIST1 expression is required for resistance in EGFR-mutant cells that demonstrate de-novo or acquired resistance to EGFR TKIs. Importantly, our study demonstrates that therapeutic targeting of an EMT-TF, is able to restore sensitivity to erlotinib in EGFR-mutant NSCLC cells that are resistant to EGFR TKIs. Our findings suggest that use of small molecule compounds that inhibit TWIST1 may be a viable option to overcome de-novo and acquired resistance to EGFR TKIs in lung cancer. Harmine is an active β-carbolin alkaloid found in the herb Peganum harmala used in traditional medicine in Central Asia and the Middle East (52). However, in mouse model systems and in humans, the harmine therapeutic efficacy may be limited due to neurotoxic side effects, such as tremors (52, 53). We have identified analogues of harmine that are potentially more potent inhibitors of TWIST1 without the neurotoxicity associated with harmine and are currently performing further preclinical evaluation of these compounds.

Others have previously demonstrated that BIM expression is required for response to EGFR TKIs (35, 37–39). Additionally, BIM polymorphisms which result in decreased expression of functional BIM protein, are associated with resistance to EGFR TKIs (32, 54, 55). Here, we establish that TWIST1 suppresses BIM expression through direct binding at both the promoter and intronic regions. Overall, these data suggest that one of the mechanisms by which TWIST1 mediates EGFR TKI resistance is through inhibition of EGFR TKI-induced apoptosis by suppression of BIM expression. Interestingly, we demonstrated that TWIST1-mediated resistance can be overcome with use of BCL2/BCLxL inhibitors. BCL2/BCLxL inhibitors, such as ABT-263 are in clinical trials, and our data suggest that use of these inhibitors may provide rapid means to overcome TWIST1-mediated resistance in the clinic. While we established that one mechanism by which TWIST1 can mediate resistance is through suppression of apoptosis, TWIST1 has been previously shown to suppress senescence in both oncogene-driven NSCLC and breast cancer (23, 24, 56). We are currently exploring whether TWIST1 mediated suppression of senescence is potentially another mechanism by which TWIST1 promotes EGFR TKI resistance. Of note, a recent study demonstrated that TWIST1 can mediate resistance to 3rd generation EGFR TKIs through upregulation of the EMT-TF, ZEB1 (9). This study established that ZEB1 can also directly suppress BCL2L11 transcription (9). This study and our current study suggests that there are potentially multiple mechanisms by which TWIST1 can promote EGFR TKI resistance and multiple mechanisms by which TWIST1 can suppress BIM expression.

In summary, we demonstrated that genetic and pharmacological inhibition of TWIST1 results in growth inhibition in EGFR-mutant NSCLC. In a subset of cell lines, including cell lines with acquired resistance T790M mutations, inhibition of TWIST1 is associated with the induction of apoptosis. Additionally, we established that TWIST1 is both sufficient and, in some lines, required for EGFR TKI resistance in EGFR-mutant NSCLC both in vitro and in vivo. We demonstrated that one of the mechanisms by which TWIST1 mediates resistance is through suppression of apoptosis via suppression of BIM expression. We also demonstrated that use of a TWIST1 inhibitor, harmine, was able to overcome both de-novo and acquired resistance to EGFR TKIs. Of note, targeting TWIST1 may be associated with minimal side effect because it is rarely expressed post-natally (57, 58). Our data suggests that targeting TWIST1 may be option to overcome EGFR TKI resistance in EGFR-mutant NSCLCs both in the de-novo and acquired resistance settings.

METHODS

Cell lines and Reagents

PC9, H1975, H1650, Hcc4006, Hcc4011, Hcc2935, Hcc827, H3255, and HEK 293T were acquired from the American Type Culture Collection (ATCC) and were cultured in the recommended ATCC media. Hcc827R2 and 11–18 cells were obtained from Dr. Christine Lovly (Vanderbilt University) and cultured in the recommended media. The identity of the aforementioned cell lines was verified by autosomal STR (short tandem repeat) profiling done at University of Arizona Genetics Core (UAGC). Mycoplasma testing was performed every six months using MycoAlert Detection Kit (Lonza). Osimertinib and erlotinib were purchased from Selleck Chemicals (Houston, TX). Harmine was purchased from Sigma-Aldrich (St. Louis, MS). ABT-737 was purchased from ApexBio Technology (Houston, TX)

CellEvent™ Caspase-3/7 Green Flow Cytometry

Cells were seeded at appropriate density in 25-cm2 plates and incubated for 24 hours. Following incubation, cells were treated with harmine at 0, 20, 40µM for 48 hours. Apoptosis was analyzed as previously described (25).

Quantitative RT-PCR

RNA isolation, cDNA generation, PowerUp™ SYBR® Green Master Mix (Perkin Elmer Applied Biosystems) and TaqMan® Universal PCR Master Mix (Perkin Elmer Applied Biosystems) were utilized as previously described (25). List of primers is provided in Supplementary Table 1–2.

Cell proliferation assays

For all viability experiments, cells were seeded at an appropriate density in 96 well plates and incubated for 24 hours. Cells were subsequently treated with a range of doses of the appropriate inhibitor for 72 hours. Viability was determined using the CellTiter96® Aqueous One Solution Cell Proliferation Assay kit (Promega) or Cell-Titer Glo 2.0 Assay (Promega) according to manufacturer’s protocol. Data was analyzed as previously described (25). To ensure consistent and reproducible results, experiments were performed at least twice.

Western blot and antibodies

Following appropriate treatment, cells were harvested and lysed and subsequent protein was quantified and western blotting was performed as previously described (23). All information on antibodies is included in Supplementary Table 3.

Chromatin Immunoprecipitation

H1975 TRE3G-TWIST1 cells were seeded in 15cm dishes and incubated for 24 hours. Cells were treated with 50ng/ml of doxycycline. Following 24 hours of doxycycline treatment, cells were harvested and Chromatin Immunoprecipitation (ChIP) assays were performed utilizing the SimpleChIP Enzymatic Chromatin IP Kit (Cell Signaling Technology) according to manufacturer’s recommendations. ChIP primers that were used are included in Supplementary Table 4. For ChIP, 2µg of ChIP-grade TWIST1 antibody (Abcam, Ab5087) and 2µg Mouse IgG, Whole Molecule Control (Thermo Scientific, 31903) were used.

Lentiviral cDNA and shRNA production

HEK 293T cells were seeded at a density of 4 X 106 in 25-cm2 flasks. Following a 24 hour incubation period, cells were transfected to generate lentivirus as previously described (23, 59). A complete list of the constructs used is provided in Supplementary Tables 5–7 and sequences are available upon request.

Transgenic mice

Animals were housed in a pathogen free facility and all studies were approved by The Johns Hopkins University IACUC. Mice were housed in groups of no more than five per cage in facilities with controlled temperature and humidity with regulated light and dark cycles. Animals had free access to food and water.

Inducible EGFR and Twist1/EGFR transgenic mice in the FVB/N inbred background were of the genotype: CCSP-rtTA/tetO-EGFR (CE) or CCSP-rtTA/tetO-EGFR/Twist1-tetO-luc (CET). The tetO-EGFR mice were obtained from Dr. Katerina Politi (Yale University). All the mice were weaned at 3–4 weeks of age and then placed on doxycycline (DOX) drinking water at 4–8 weeks of age as previously described (24, 42). After three weeks of DOX treatment, CE and CET mice were randomized to vehicle and erlotinib treatment groups and stratifying by similar levels of tumor burden with micro-CT. Mice without tumor burden were excluded. These criteria were pre-established. Tumor burden was assessed by micro-CT imaging and quantitated as previously described (24). Evaluation of treatment response by microCT was blinded otherwise the identity of the animals were known to investigators.

Erlotinib was purchased from Selleckchem (Houston, TX). For in vivo experiments, erlotinib was dissolved into a slurry in 0.5% methylcellulose. The mice received 50 mg/kg erlotinib or vehicle via oral gavage 6 days a week for 3 weeks.

Histology and immunohistochemistry

Tissues were fixed and subsequent histology and immunohistochemistry was performed as previously described (60). For immunohistochemistry, the primary antibodies were used at the following concentrations: Twist1 at 1:200, vimentin and e-cadherin at 1:400; cleaved caspase 3 at 1:500, and Ki-67 at 1:2000.

Statistical analysis

Student t-test, ANOVA with Tukey’s multiple comparison testing, and Mantel-Cox testing was performed where indicated. For transgenic animal studies, we used cohorts of >12 animals each. This was based on sample size calculations assuming experimental condition will result in animals that have a mean survival that is 45% longer than control treated mice with a power of 80% to detect a difference using the Kaplan-Meier long-rank test.