A genetic cell context-dependent role for ZEB1 in lung cancer

The Zinc-finger E-box-binding Homeobox-1 (ZEB1) is a transcription factor that promotes epithelial–mesenchymal transition (EMT) and acts as an oncogene in KRAS-mutated lung cancer models. Here we report that ZEB1 exerts the opposite effect in EGFR-mutated lung cancer cells, where it suppresses growth by increasing microRNA-200 targets to antagonize ERBB3, a driver of mutant EGFR-dependent cell growth. Among these targets, NOTCH1 represses ERBB3 promoter activity and the expression of ERBB3. Furthermore, we find that EGFR inhibitor treatment, which inhibits the growth of EGFR-mutated cells, induces ZEB1. Despite its growth-inhibiting effect, EGFR inhibitor-induced ZEB1 strongly promotes EMT-dependent resistance to EGFR inhibitors partially through NOTCH1, suggesting a multifunctional role for NOTCH1 in EGFR-mutated cells. These results support a previously unrecognized genetic cell context-dependent role for ZEB1 and suggest that NOTCH1 may be a useful target for treating resistance to EGFR inhibitors, especially EMT-driven resistance.

C ancer development is associated with complex genetic changes, including mutations of oncogenes and tumour suppressors. Some oncogenic mutations occur in a mutually exclusive manner in specific types of cancers, such as the KRAS and EGFR mutations in lung cancer 1,2 . The biologic basis for this phenomenon is not well understood. It is possible that cellular processes regulated by mutant KRAS or EGFR may have contradictory functions that exclude each other. Clinically, KRAS mutations in lung cancer are associated with resistance to epidermal growth factor receptor (EGFR) inhibitors 1,2 .
ZEB1 is a zinc-finger E-box-binding homeobox protein that induces epithelial-mesenchymal transition (EMT), a reversible process with multiple roles in cancer development 3,4 . Recent studies have shown that ZEB1 acts as an oncogene in invasive and metastatic lung cancer cells, in which ZEB1-induced EMT promotes the loss of epithelial cell polarity and adhesion, induces cytoskeleton remodelling and drives growth, migration, invasion and metastasis [5][6][7][8][9][10][11] . The role of ZEB1 in early-stage lung cancer remains poorly explored. A recent report showed that ZEB1 is required for mutant KRAS-driven lung tumour initiation and progression 12 . Notably, most of these studies address the role of ZEB1 by using cells or mouse models that express mutant KRAS, without exploring whether ZEB1 has different roles in lung cancer cells without KRAS mutations.
Herein, we report an unexpected finding that ZEB1 plays an opposite role in EGFR-mutated lung cancer cells by acting as a suppressor of cell growth. We show that ZEB1 transcriptionally represses microRNA-200 (miR-200) to increase several miR-200 target genes that inhibit the receptor tyrosine kinase ERBB3, which is essential for growth of EGFR-mutated lung cancer cells [13][14][15][16][17] . In vivo, ERBB3 is repressed in lung adenocarcinoma tissue with elevated ZEB1 expression. Furthermore, we find that the growth inhibition induced by an EGFR inhibitor, gefitinib, is associated with strong induction of ZEB1, which in turn promotes EMT-dependent resistance to gefitinib. Mechanistically, we show that NOTCH1 is the mediator of ZEB1 that transcriptionally inactivates ERBB3 and promotes EMT and resistance. Thus, the biologic functions of ZEB1 and NOTCH1 are context dependent. Within the context that EGFR is inhibited, ZEB1 and NOTCH1 exert an additional role that may contribute to the survival of a subset of EGFR-mutated cells by promoting EMT. As there are no clinically effective drugs for EGFR inhibitorresistant lung tumours with EMT features, these findings warrant future studies that test the efficacy of combined use of NOTCH1 and EGFR inhibitors in the treatment of such resistant lung tumours.

Results
ZEB1 was repressed in lung adenocarcinomas. Several E-boxbinding transcription repressors, including ZEB, SNAIL and TWIST, are known to promote EMT by repressing epithelial genes (for example, CDH1) and by increasing mesenchymal genes (for example, VIM) [18][19][20] . To explore the role of such EMT-promoting factors in lung carcinogenesis, we mined the Oncomine data sets of lung adenocarcinomas (nine data sets from www.oncomine.org) for alterations in CDH1, VIM, ZEB1, ZEB2, SNAI1, SLUG (also known as SNAI2) and TWIST (Table 1). We found that lung adenocarcinomas exhibited an epithelial CDH1 high /VIM low gene expression profile compared with normal lung tissues, demonstrating higher CDH1 in eight of nine data sets but lower VIM in all nine data sets. In five of six data sets containing ZEB1 expression data, tumours had lower ZEB1, which was associated with the epithelial-like phenotype of the tumours and supported the role of ZEB1 as a driver of EMT. In contrast, expression levels of the remaining factors were not different between tumours and normal tissues (ZEB2, SNAI1 and SLUG), or even higher in tumours (TWIST), implying that these factors may not be the primary regulators of EMT in lung adenocarcinomas.
Although the above observations raised the possibility that ZEB1 may be more related to EMT in lung adenocarcinoma compared with other E-box-binding factors, they also suggested a growth-suppressor role for ZEB1, which contradicts previous reports supporting an oncogenic role of ZEB1 in KRAS-mutated lung cancer [5][6][7][8][9][10][11][12] . To validate these observations, we quantified the expression of ZEB1, CDH1 and VIM for 41 paired samples of lung adenocarcinoma and adjacent normal lung tissues (Supplementary Table 1). We found that the changes in gene expression in tumours versus normal tissues were similar to what we observed in the Oncomine data sets, namely that tumours had an epithelial-like CDH1 high /VIM low phenotype, which was associated with lower ZEB1 expression (Supplementary Table 2). Consistently, immunohistochemical staining results showed that lung adenocarcinoma cells expressed high levels of E-cadherin but low to non-detectable levels of Vimentin and ZEB1 ( Supplementary Fig. 1a). Notably, ZEB1 was repressed in nearly all tumours from never smokers (15 out of 16) (Fig. 1), a significantly higher frequency of repression than that observed in smokers (Fisher's exact text, Po0.05) (Fig. 1). Given that the incidence of certain somatic driver mutations in lung adenocarcinomas differs in smokers and non-smokers, for instance, lung tumours of never smokers have more EGFR mutations but fewer KRAS mutations compared with tumours from smokers 21,22 , we hypothesize that ZEB1 exerts a growth suppressive role that is genetic context dependent.
Sequencing of the 41 lung adenocarcinomas showed that 14 tumours had KRAS mutations, and that 7 tumours had EGFR mutations ( Fig. 1 and Supplementary Table 3). Of the ten tumours with higher ZEB1, five (50%) had KRAS mutations (Fig. 1). This percentage was higher than that of tumours with lower ZEB1 (9 KRAS-mutated tumours in a total of 31 tumours; 29%) ( Fig. 1), but the association between KRAS mutations and ZEB1 expression changes was not statistically significant (Fisher's exact test, P ¼ 0.11). In contrast, all seven EGFR-mutated tumours had lower ZEB1 (Fig. 1) and the levels of ZEB1 were significantly lower in EGFR mutant tumours than that in EGFR wild-type tumours (Student's t-test P ¼ 0.027) or in KRAS mutant tumours (Student's t-test P ¼ 0.013). In support of these findings, most EGFR-mutated lung adenocarcinomas from The Cancer Genome Atlas (TCGA) data set (7 of 9 paired lung adenocarcinoma and normal lung tissues) or from previous reports (12 of 13 paired lung adenocarcinoma and normal lung tissues; reported in refs 23,24) had lower ZEB1 compared with paired normal lung tissues ( Supplementary Fig. 1b,c). In most Oncomine data sets with lower tumour ZEB1 (Table 1), the expression levels of both EGFR and KRAS were marginally higher in tumours compared with paired normal lung tissues (Supplementary Table 4). Collectively, these results suggest that EGFR mutations correlate with the loss of ZEB1 in lung adenocarcinomas.
ZEB1 oppositely regulated KRAS and EGFR mutant cell growth. Next, we expressed ZEB1 in lung adenocarcinoma cells, including H441, 393P, HCC827 and H3255 cells. H441 and 393P are KRAS mutated, and HCC827 and H3255 are EGFR mutated. None of them expressed endogenous ZEB1 ( Supplementary Fig. 2a). We found that ZEB1 promoted KRAS-mutated cell growth ( Fig. 2a) but inhibited EGFR-mutated cell growth (Fig. 2b). Consistently, ZEB1 also promoted KRAS-mutated xenograft tumour growth ( Fig. 2c) but suppressed EGFR-mutated xenograft tumour growth (Fig. 2d). However, ZEB1 induced EMT in both of them (Fig. 2e,f). These results suggest that ZEB1 plays distinct roles in cells with EGFR or KRAS mutations, and that such roles are likely to be independent of EMT.
We realized that additional gene mutations may coexist with KRAS or EGFR mutations in cancer cells and may also affect the function of ZEB1. To address this point, we expressed KRAS G12D or EGFR D722-726 in BEAS2B cells, a lung epithelial cell line. Western blotting of transfectants showed that KRAS G12D slightly increased (by 27%) and EGFR D722-726 slightly decreased (by 16%) ZEB1 expression ( Supplementary Fig. 2b), and both mutants activated the phosphorylation of extracellular regulated, mitogenactivated proteinkinases, indicating that the transfected genes were functional (Fig. 2g). Similar to the findings from cancer cells, ZEB1 also oppositely regulated the growth of BEAS2B cells expressing KRAS G12D or EGFR D722-726 (Fig. 2h,i).
ZEB1 repressed ERBB3 through miR-200c repression. To understand the molecular basis underlying this new function of ZEB1, we performed a phospho-receptor tyrosine kinase array for EGFR-mutated HCC827 cells expressing ZEB1. We found that ZEB1 suppressed phosphorylation of ERBB3 and MET but promoted phosphorylation of RYK and AXL (Fig. 3a). Next, we focused on ERBB3, because a number of studies have described an essential role for ERBB3 in mutant EGFR-dependent lung cancer cell growth [13][14][15][16][17] . For instance, ERBB3 inactivation not only inhibits EGFR-mutated cell growth in vitro (including HCC827 cells), but also suppresses mutant EGFR-driven lung tumour growth in a mouse model 13,14 . Western blotting showed that ZEB1 suppressed the expression of ERBB3 (Fig. 3b). In a panel of EGFR-mutated cells, ERBB3 was higher in cells expressing low ZEB1 (Fig. 3c). Thus, we concluded that ZEB1 suppresses EGFR-mutated cell growth by repressing ERBB3. The clinical relevance of these findings is further supported by an analysis of The Cancer Genome Atlas (TCGA) (www.cancergenome.nih.gov) lung adenocarcinoma data set (n ¼ 510), in which ZEB1 negatively correlated with ERBB3 (Pearson's R ¼ À 0.24, P ¼ 6.94E À 08). We noted that the level of phosphorylated ERBB3 was low in one ZEB1 low cell line, H3255, in which non-phosphorylated ERBB3 was highly expressed (Fig. 3c), suggesting that the expression and phosphorylation of ERBB3 can be regulated through differential mechanisms.
Next, we explored the mechanism whereby ZEB1 repressed ERBB3. As ZEB1 acts as a transcription repressor 18,19 , we first tested whether ZEB1 directly represses ERBB3 transcription. Our results ( Supplementary Fig. 3a) showed that ZEB1 did not inhibit the activity of an ERBB3 promoter reporter (it actually slightly increased the ERBB3 promoter reporter activity), indicating that ZEB1 does not directly repress ERBB3 transcription. Recent findings have shown that transcriptional inactivation of miR-200  Table 1) were quantified by qPCR and normalized to that of L32 (internal control). The changes in ZEB1 expression were expressed as logged (log2) ratios of ZEB1 level in tumours versus normal tissues. Note: arrow indicates KRAS-mutated tumours and * indicates EGFR-mutated tumours.  by ZEB1 is critical for its biologic functions, for example, EMT and invasion 25,26 . As such, we posited that ZEB1 repressed ERBB3 by inactivating miR-200. In support of this idea, ZEB1 strongly decreased miR-200c ( Supplementary Fig. 3b) and reexpression of miR-200c dramatically restored ERBB3 expression and phosphorylation (Fig. 3d), and promoted cell growth ( Supplementary Fig. 3c). Given that microRNAs are negative regulators of gene expression, we postulated that miR-200c represses its target genes to indirectly promote ERBB3 expression and phosphorylation (Fig. 3e). To identify potential targets, we performed RNA sequencing of HCC827 cells expressing ZEB1 (Supplementary Data 1) and found that 87 predicted miR-200c targets were increased by ZEB1 (fold change 41.5 and Po0.05) ( Supplementary Fig. 4a). The 33 0 -untranslational region (3 0 -UTR) of each of these genes has at least one specific binding sequence for miR-200c (www.targetscan.org).
MiR-200c targets suppressed ERBB3. We randomly selected 16 of the 87 predicted miR-200c targets, individually transfected their small interfering RNAs (siRNAs) ( Supplementary Fig. 4b), and performed western blotting. The results showed that knockdown of NOTCH1, but not any of the other 15 genes, strongly promoted both expression and phosphorylation of ERBB3 (Fig. 3f). On the contrary, expression of NICD1 abrogated ERBB3 expression ( Fig. 3g and Supplementary Fig. 4c). Given that NICD1 is a transcription factor 27 , we posited that it transcriptionally represses ERBB3. In support of this idea, knockdown of NOTCH1 increased not only the protein but also the messenger RNA levels of ERBB3 ( Supplementary Fig. 4d,e). Furthermore, g-secretase inhibitors (DAPT and BMS-708163) induced ERBB3, mimicking the effect of NOTCH1 knockdown and the expression of NICD1 completely abolished the effect of g-secretase inhibitors ( Supplementary  Fig. 4c), suggesting that NICD1 is both necessary and sufficient for ERBB3 repression. Notably, NICD1 complexes with transcription co-factors to form a transcription-regulating complex that directly binds to target gene promoters. Indeed, analysis of the ERBB3 promoter (Fig. 4a) revealed a previously untested binding site for RBPJ, a well-characterized co-factor of NICD1 that plays an essential role in NOTCH1-dependent transcriptional regulation 27 . Our immunoprecipitation-western blotting experiment confirmed that NICD1 bound to RBPJ in lung cancer cells (Fig. 4b). By performing chromatin immunoprecipitation (ChIP) assays, we found that both NICD1 and RBPJ bound to this RBPJ site (Fig. 4c,d). Next, we performed luciferase promoter reporter assays and found that NICD1 inhibited the ERBB3 promoter and either mutation of the RBPJ site or knockdown of RBPJ significantly suppressed the effect of NICD1 (Fig. 4e,f). Similar to the effect of NOTCH1 depletion, knockdown of RBPJ also promoted ERBB3 expression (Fig. 4g,h).
To further validate these results, we knocked down ZEB1 and found that miR-200c was increased and NOTCH1 was decreased ( Supplementary Fig. 4f,g), supporting our finding that ZEB1 promoted NOTCH1 through repression of miR-200c. We also expressed miR-200c or NOTCH1 siRNA in multiple lung cancer cells with different oncogenic backgrounds. In nearly all these cells, miR-200c expression decreased NICD1 (Fig. 4i) and knockdown of NOTCH1 promoted both expression and phosphorylation of ERBB3 (Fig. 4j) We further observed that knockdown of several other genes, including BCL2, RANBP9, LFNG, TOB1 and MAF, did not significantly affect ERBB3 expression but did promote its phosphorylation to varying extents (Fig. 3f). To determine whether these ERBB3 phosphorylation-regulating genes, as well as NOTCH1, are direct targets of miR-200c, we cloned the 3 0 -UTRs of BCL2, RANBP9, LFNG, TOB1 and NOTCH1 into a luciferase-conjugated 3 0 -UTR reporter plasmid 5 . We found that miR-200c suppressed all these 3 0 -UTRs (Fig. 5a) and mutation of their miR-200c binding sequences suppressed the effect of miR-200c (Fig. 5b-e). Consistently, knockdown of these genes promoted cell growth (Fig. 5f), mimicking the effect of miR-200c expression ( Supplementary Fig. 3c). We also noted that the effect of miR-200c on NOTCH1 3 0 -UTR luciferase reporter activity was modest. However, miR-200c significantly repressed endogenous NOTCH1 expression in multiple lung cancer cells (Fig. 4i), suggesting that the regulation of NOTCH1  in tumour growth, we generated EGFR-mutated H1975 cell transfectants that stably express short hairpin RNAs against TOB1 (Supplementary Fig. 5a). We found that the knockdown of TOB1 alone moderately (P ¼ 0.074) increased xenograft tumour growth ( Supplementary Fig. 5b,c), suggesting that these ERBB3regulating genes may collaboratively regulate tumour growth in vivo.
In contrast to the effect of above-mentioned genes, knockdown of several other genes, including FHL, KLF9, NACC2 and PRKAR2B, decreased ERBB3 phosphorylation (Fig. 3f). We also cloned the 3 0 -UTRs of FHL and KLF9, but the reporter assay results showed that they were not regulated by miR-200c (Fig. 5g). Collectively, our results suggest that miR-200c acts as an activator of ERBB3 and multiple mechanisms are likely to be involved in the regulation of ERBB3 by miR-200c.
NOTCH1 drove EMT-dependent gefitinib resistance. Both EMT and ERBB3 are implicated in the acquisition of resistance to EGFR inhibitors, for example, gefitinib 13,28 . As ZEB1 strongly regulates both of them (Figs 2 and 3), we suspected that ZEB1 and its mediators (miR-200c and its targets) are involved in this process. Using gefitinib-sensitive HCC827 cells as a model, we found that chronic treatment of gefitinib strongly induced ZEB1 expression (Fig. 6a), accompanied by the loss of miR-200c (Fig. 6b) and an EMT phenotype ( Supplementary  Fig. 6). Despite its growth inhibitory effect (Fig. 2), ZEB1 dramatically drove resistance to gefitinib (Fig. 6c). As the expression of miR-200c partially reversed both EMT (Fig. 6d) and resistance (Fig. 6e), we screened the 16 miR-200c targets to identify resistance-related genes that may have therapeutic potential. The result revealed that knockdown of NOTCH1 partially reversed gefitinib resistance to an   (Fig. 6f). Knockdown of several other genes, including NR5A2, NACC2, LFNG and HNF1B, marginally reversed resistance and knockdown of FHL or MAF further enhanced resistance ( Supplementary  Fig. 7), implying that miR-200c may target multiple genes and pathways with distinct roles in resistance. Meanwhile, we also screened the 16 miR-200c targets for regulation of EMT by performing western blotting for E-cadherin and Vimentin (Fig. 6g). Only knockdown of NOTCH1 partially reversed EMT by increasing E-cadherin without affecting Vimentin, an effect similar to that of miR-200c (Fig. 6d), further suggesting that NOTCH1 promotes EMT to mediate ZEB1-induced resistance.
To examine the effect of pharmacologic inhibition of NOTCH in regulation of cell growth, we treated the gefitinib-resistant HCC827GR cells with the g-secretase inhibitor BMS-708163, which dose dependently inhibited NICD1 expression (Fig. 6h), increased the expression of ERBB3 (Fig. 6h) and promoted cell growth (Fig. 6i). These results were consistent with NOTCH1 inhibiting both ERBB3 and cell growth (Figs 3-5). However, both the parental HCC827 and HCC827GR cells became more sensitive to gefitinib in the presence of BMS-708163  ( Fig. 7a), suggesting that the combination of EGFR and NOTCH inhibitors may be useful for treating EGFR inhibitor resistance. In support of this idea, treating multiple gefitinibresistant lung cancer cells (including HCC4006GR, H2279 and HCC827-ZEB1) with BMS-708163 or DAPT (another g-secretase inhibitor) also reversed their resistance phenotype to various extents (Fig. 7b-d), demonstrating that NOTCH activation exerts a universal biologic function in EGFR inhibitor resistance in lung cancer cells, and that NOTCH inhibitors may have a therapeutic value in treating EGFR inhibitor-resistant lung tumours. It is known that distinct mechanisms can drive the acquired resistance to EGFR inhibitors. For instance, HCC4006 cells acquire resistance through loss of EGFR copy number 29 and HCC827 cells could develop resistance through gefitinibresistant EGFR T790M mutation-dependent and -independent mechanisms 29,30 . To determine whether changes in EGFR mutation effect ZEB1 induction and NOTCH signalling, we examined the expression levels of ZEB1 and NICD1 in multiple gefitinib-resistant cell lines, including HCC827-GR, HCC4006-GR, HCC827-ZEB1, H1975 and H2279 (Fig. 7e and Supplementary Fig. 8). Gefitinib-resistant HCC827-GR and HCC4006-GR cells were generated by continuously treating HCC827 and HCC4006 cells with 1 mM gefitinib for 10-12 weeks until the cells could be grown into confluence and passaged normally. H1975 is known to harbour T790M EGFR mutation and sequencing of the genomic DNA from gefitinib-resistant cell lines revealed that only H1975 cells had T790M mutation (Supplementary Fig. 9). HCC827-ZEB1 cells were stable transfectants of the parental HCC827 cells and RNA sequencing experiment (partly shown in Supplementary  Fig. 4 and Supplementary Data 1) showed that there was no T790M mutation in these cells. Compared with gefitinibsensitive parental cells (HCC827, HCC4006 and H3255) or control cells (HCC827 vector), all gefitinib-resistant lines had higher levels of ZEB1 except for HCC4006-GR cells (Fig. 7e) and all gefitinib-resistant lines also expressed higher NICD1 except for H2279 cells (Fig. 7e), suggesting that ARTICLE changes in EGFR mutation may not affect ZEB1 induction or NOTCH1 signalling.

Discussion
EMT has been an active area of investigation in cancer biology. Its best-characterized role involves the later stage of solid tumour development, when it acts to centrally initiate tumour cell dissemination by destroying the polarized epithelial cell structure and by remodelling the cytoskeleton to mobilize cells [18][19][20] .
Consistently, several families of E-box-binding transcription factors, including the ZEB factors, are known to promote invasion and metastasis by inducing EMT in various cancer cell types [18][19][20] . Furthermore, inactivation of these factors blocks EMT and suppresses invasion and metastasis in models of metastatic cancer 5,6,31-34 , suggesting that these factors and their associated pathways may have therapeutic value. As such, our findings that ZEB1 distinctly regulates (promotes or suppresses) the growth of lung cancer cells harbouring KRAS or EGFR mutations (Fig. 2) is an important new observation. These findings not only reveal that ZEB1 can act as either an oncogene or a growth suppressor, but also suggest that targeting ZEB1 may have opposite outcomes. The distinct effects of ZEB1 were independent of EMT, indicating that both EMT-dependent and EMT-independent mechanisms collectively mediate the biologic functions of ZEB1 and the latter may specifically mediate the growth suppressive action of ZEB1 in EGFR-mutated cells. ERBB3 belongs to the EGFR family of receptor tyrosine kinases, which includes EGFR and ERBB2-4. Unlike other members of this family, ERBB3 does not have an active kinase domain; it forms heterodimers with EGFR or ERBB2/4 to activate downstream signalling pathways, such as the RAS/mitogen-activated protein kinase pathway [35][36][37] . Our results identify multiple miR-200c targets as novel mediators of ZEB1 that antagonize ERBB3 and suppress growth, probably through both direct and indirect mechanisms (Figs 3-5). For instance, TOB1 was reported to bind and to inhibit ERBB2 (refs 38,39), suggesting a possible indirect mechanism whereby TOB1 suppressed ERBB3 phosphorylation through inhibition of ERBB2. Of particular interest, we found that NOTCH1 directly suppressed ERBB3 transcription, thereby establishing a molecular link between ZEB1 and ERBB3 repression. As depletion of NOTCH1 promotes ERBB3 in cells harbouring various gene mutations, such a NOTCH1-ERBB3 axis may be widely present in lung cancer cells and serve as a common signalling mechanism that regulates the growth of ERBB3dependent cells, especially EGFR-mutated cells. However, in KRAS-mutated cells constitutive KRAS activation may bypass ERBB3 to promote growth, allowing the cells to escape growth inhibition induced by ZEB1 through NOTCH1-mediated ERBB3 repression. This may also partly explain why ZEB1 distinctly regulates the growth of cells harbouring mutant KRAS or EGFR.
RBPJ plays a critical role in NOTCH1-mediated transcription regulation 27 . In the absence of NICD1, RBPJ can recruit transcription co-repressors to silence gene expression. When NICD1 is activated and translocates into the nucleus, it can promote target gene transcription by binding to RBPJ, recruiting co-activator MAML1 and converting the initially transcriptioninactive RBPJ complexes into activators 27 . Although our results suggest that NICD1 binds to both RBPJ and ERBB3 promoter, it remains unclear whether NICD1 directly represses ERBB3 transcription through binding to RBPJ. It should be noted that our current results can not preclude the possibility that NICD1 may also repress ERBB3 through other indirect mechanisms, for example, through HES or HEY family of transcription repressors, both of which are well-established NICD1 transcription targets 27 .
Thus, future studies are needed to determine whether multiple mechanisms are involved in NICD1-mediated ERBB3 repression.
Previous studies have shown that ZEB1 promoted the NOTCH ligand Jagged1 expression and activated NOTCH signalling in pancreatic and prostate cancer cells by repressing miR-200 family members, which directly targeted the 3 0 -UTR of Jagged1 (refs 40,41). Our results showed that ZEB1 promoted NOTCH1 but not Jagged1 in lung cancer cells ( Supplementary Fig. 4 and Supplementary Data 1), suggesting that the ZEB1/miR-200 axis may regulate NOTCH signalling through differential mechanisms in different types of human cancer cells.
Tyrosine kinase inhibitors of EGFR, including gefitinib and erlotinib, are clinically effective drugs that treat EGFR-mutated lung cancer. However, their efficacy is transient and drug resistance inevitably develops, leaving patients with few treatment options. How resistance is generated remains incompletely understood. Several genetic mechanisms, including drug-resistant EGFR mutations and MET gene amplification, drive resistance in more than half of the affected patients 28,42 ; these patients may be treatable with newer-generation EGFR inhibitors or MET inhibitors, which are in clinical development. However, for patients without these genetic changes, the mechanisms of resistance remain elusive and effective targeted therapeutics do not exist. Intriguingly, EMT drives resistance in a subset of these patients 28,42,43 . As such, our results that ZEB1 promoted EMT and resistance partially through repression of miR-200c (Fig. 6) are exciting. These results are consistent with recent reports from others showing that miR-200 inhibits EMT and chemoresistance in breast tumour cells, and that EMT inhibition also promotes sensitivity to gemcitabine treatment in pancreatic cancer cells 44,45 . Mechanistically, our data show that NOTCH1 is a direct target of miR-200c and drives EMT and resistance (Figs 5-6), thereby providing a novel and direct molecular link between miR-200 and the regulation of these processes. Further, our studies show that NOTCH1 depletion or treatment with g-secretase inhibitors reverse resistance in multiple gefitinib-resistant lung cancer cells (Figs 6-7), suggesting a strong translational potential of our above mechanistic findings.
Notably, our results show that treatment with g-secretase inhibitors (BMS-708163 and DAPT) partially reversed the resistance phenotype (Fig. 7), an effect similar to NOTCH1 knockdown or miR-200c expression, suggesting that additional pathways may also be implicated in ZEB1-induced EMT and resistance. In addition, our data are consistent with recent reports [46][47][48] showing that ZEB1 or NOTCH1 promotes EMT and EGFR inhibitor resistance in lung cancer cell lines, including HCC4006, PC9 and H1650. Taken together, these findings suggest that NOTCH1 antagonists, with combined use of EGFR inhibitors, may be useful for treating patients with EMT-dependent resistance. Furthermore, a recent report has shown that HER2 inactivation led to the activation of NOTCH1, which in turn promoted mammary tumour dormancy and recurrence 49 , suggesting that NOTCH1 inhibition may have broader applications in the treatment of drug resistance in multiple human cancers.
Nevertheless, it should be noted that the inhibition of NOTCH1 by g-secretase inhibitor treatment or by siRNAs also significantly promoted both ERBB3 expression and lung cancer cell growth in the absence of gefitinib (Figs 3-6). This is different from the observation from breast cancer cells that NOTCH1 inhibition reduced tumour cell growth 49 . Our findings indicate that the inactivation of NOTCH1 alone, without concomitant inhibition of EGFR, may lead to unexpected outcomes in lung cancer by increasing ERBB3 to promote EGFR-mutated tumour growth, providing a caution against indiscriminate use of NOTCH inhibitors. Thus, personalized therapeutics should be developed to direct the use of these drugs in the treatment of EGFR-mutated lung cancer.
peroxidase-conjugated secondary antibody and supersignal ECL substrates (Pierce). Uncut western blotting films were included in Supplementary Fig. 10.
RNA sequencing. Total RNA was prepared using the miRNeasy kit (Qiagen). RNA sequencing was performed by Mayo Clinic gene expression core laboratory. The sequencing data have been deposited to the GEO repository with an accession code GSE81167 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE81167).
Reporter assay. Cells were transfected with reporter plasmids using Lipofectamine 2000. After 48 h, cells were lysed in passive lysis buffer (Promega) and luciferase activity was measured using a Dual-Luciferase Reporter Assay System (Promega).
Human and mouse studies. Human tissues were acquired from Mayo Clinic lung tissue registry with protocols approved by the Mayo Clinic internal review board and human tissue subcommittee. Mice were purchased from Charles Rivers. All protocols were approved by the Mayo Clinic Institutional Animal Care and Use Committee (IACUC). Briefly, cells (one million per mouse) were suspended in PBS and subcutaneously injected into the flanks of 8-to 10-week-old male athymic nude mice (strain 490; n ¼ 6 per cohort for HCC827 vector and HCC827-ZEB1 xenograft tumours; n ¼ 10 per cohort for H1975-sc and H1975-TOB1sh4 xenograft tumours) or male wild-type 129S2/SvPasCrl mice (strain 476; n ¼ 10 per cohort for 393P-vector and 393P-ZEB1 xenograft tumours). Autopsies were performed at 3 weeks after injection.
Statistics. Statistical significance was determined using two-sided Student's t-tests or one-way analysis of variance as indicated, with Po0.05 being considered a statistically significant difference, unless otherwise indicated.
Immunohistochemistry. For immunohistochemistry, the frozen samples were cut into 10-mm sections using a cryostat microtome (Leica). The slides were fixed in cold menthol for 10 min at À 20°C and washed with tris-buffered saline (TBS). Subsequently, the sections were incubated in 3% H 2 O 2 solution in methanol at room temperature for 10 min. After washing, blocking buffer (5% normal goat serum/tris-buffered saline (TBS)) was added and incubated for 30 min at room temperature. The sections were then incubated with primary antibodies 4°C overnight (anti-ZEB1: 1:100 dilution; anti-E-cadherin: 1:400 dilution; and anti-Vimentin: 1:400 dilution). The detection system used was the avidin-biotin complex method. Finally, the sections were rinsed, counterstained with haematoxylin and mounted on glass slides before evaluation. The slides were examined by a pathologist using light microscopy.
Data availability. RNA sequencing data have been deposited to the GEO repository (accession code GSE81167) and relevant data are available from the authors.