Sox9 mediates Notch1-induced mesenchymal features in lung adenocarcinoma.

Sox9 has gained increasing importance both functionally and as a prognostic factor in cancer. We demonstrate a functional role for Sox9 in inducing a mesenchymal phenotype in lung ADC. We show that Sox9 mRNA and protein are overexpressed in lung ADC, particularly those with KRAS mutations. Sox9 expression correlated with the Notch target gene Hes1, and numerous other Notch pathway components. We observed that Sox9 is a potent inducer of lung cancer cell motility and invasion, and a negative regulator of E-cadherin, a key protein that is lost during epithelial-mesenchymal transition (EMT). Moreover, we show that Notch1 signaling directly regulates Sox9 expression through a SOX9 promoter binding site, independently of the TGF-β pathway, and that Sox9 participates in Notch-1 induced cell motility, cell invasion, and loss of E-cadherin expression. Together, the results identify a new functional role for a Notch1-Sox9 signaling axis in lung ADC that may explain the correlation of Sox9 with tumor progression, higher tumor grade, and poor lung cancer survival. In addition to Notch and TGF-β, Sox9 also acts downstream of NF-κB, BMP, EGFR, and Wnt/β-catenin signaling. Thus, Sox9 could potentially act as a hub to mediate cross-talk among key oncogenic pathways in lung ADC. Targeting Sox9 expression or transcriptional activity could potentially reduce resistance to targeted therapy for lung ADC caused by pathway redundancy.


INTRODUCTION
Sox9 is a member of the high-mobility group-box class DNA-binding protein family of transcription factors and plays a critical role in embryonic development, cell fate determination and lineage commitment. Specifically, Sox9 orchestrates chondrogenesis, bone formation and testis development [1,2]. Germline SOX9 mutations cause campomelic dysplasia, a disorder characterized by numerous skeletal abnormalities and XY sex reversal [3]. Infants with mutant SOX9 usually die due to respiratory distress shortly after birth [4]. This clinical finding likely reflects the role of Sox9 in the developing lung. During development, Sox9 is required for branching morphogenesis in the lungs and for controlling alveolar epithelial progenitor cell proliferation [5,6].
Overexpression of Sox9 is found in more than 50% of lung adenocarcinomas (ADCs), the most common histological lung cancer subtype, and is associated with a poor lung ADC survival [7,8]. Over the past 5 years, it has become increasingly clear that Sox9 plays a major role across numerous human cancers. Sox9 overexpression Oncotarget 3637 www.impactjournals.com/oncotarget acts as an oncogene in several cancer types, including breast, colorectal, pancreatic, and prostate cancer and glioma [9][10][11][12][13], although it appears to act as a tumor suppressor in bladder cancer and melanoma [14,15]. In lung cancer, Sox9 overexpression was recently shown to increase cell proliferation and xenograft tumor formation [16] [8]. However, the functional role of Sox9 in lung cancer has not been fully elucidated. Furthermore, the upstream oncogenic molecular pathways regulating Sox9 overexpression in lung ADC have not been completely delineated.
The Notch signaling pathway regulates cell fate decisions in nearly every developing tissue and organ, and maintains adult tissue homeostasis. In humans, the Notch receptor is activated after binding to one of 5 Delta or Jagged ligands that are expressed on neighboring cells. Ligand binding initiates a series of cleavage events, and the final cleavage is carried out by the γ-secretase protease complex which frees the intracellular domain portion of Notch (ICD), allowing Notch ICD to translocate to the nucleus. Nuclear Notch ICD complexes with the transcriptional repressor RBP-Jκ, converting it into a transcriptional activator [17]. Notch ICD-induced effects on gene expression are tightly controlled by numerous post-translational modifications and protein co-factors. Several Notch target genes are nearly universally activated across many cell types and developmental processes, such as Hes and Hey, while many other Notch target genes are highly cell-type and context specific [18,19].
Notch1 is a putative oncogene that is aberrantly over-activated in over one-third of lung ADCs through gain-of-function mutations (10%) or altered expression of Notch receptors and/or suppressors (30%) [20]. Notch1 is required for tumor initiation in the KRAS-driven lung ADC mouse model [21]. Moreover, overexpression of constitutively activated Notch1 directly participates in lung carcinogenesis; it induces lung adenomas in mice which progress to ADCs in cooperation with Myc overexpression [22]. Aberrant Notch1 signaling contributes to acquired resistance to EGFR inhibitors, the most widely used targeted therapy for NSCLC [23]. Functionally, Notch increases lung cancer stem cell (CSC) self-renewal and promotes epithelial-mesenchymal transition (EMT) [23,24], reviewed in [25], although the downstream mechanisms mediating these effects have not been fully defined. Few studies have identified specific target genes of oncogenic Notch1 in lung ADC, but those validated thus far include Survivin and IGF1R [26,27]. Because of the exquisite specificity of cell-type and context specific effects of Notch signaling, it is necessary to explore Notch target genes in lung cancer in order to elucidate the genes mediating Notch-induced EMT within this disease.
Sox9 is a well-studied downstream effector of the Notch pathway in murine tissues, especially during development [28][29][30][31][32][33][34], through RBP-Jκ binding sites located at positions +40 bp and −325 bp relative to the SOX9 transcriptional start site [35]. However, these binding sites are not conserved in humans. Sox9 expression correlates with Notch signaling in humans [36], but there has been a lack of direct evidence that Notch-induced modulation of Sox9 expression is evolutionarily conserved in humans or that Sox9 is a Notch target in lung cancer. Loss of the TGF-β adaptor β2SP induces TGF-β signaling, resulting in both Notch activation and SOX9 expression in esophageal ADC [37]; although, the specific role of Notch in regulating Sox9 expression was not fully delineated. In this study, we show that the Notch1-Sox9 signaling axis is evolutionarily conserved through a novel RBP-Jκ binding site at -10 bp relative to the SOX9 promoter and that Sox9 expression is regulated by Notch1 in lung ADC independent of TGF-β signaling. Further, we demonstrate that Sox9 mediates Notch1-induced cell motility, invasion, morphological changes, and suppression of E-cadherin, all features of a mesenchymal phenotype.

Sox9 expression is elevated in KRAS-mutant lung ADC.
An analysis of publicly available datasets [38][39][40][41][42][43][44][45][46][47] on Oncomine (www.oncomine.com) revealed an upregulation of Sox9 mRNA in the majority of lung ADC datasets. Out of 10 available datasets comprising a total of 1,021 samples, 8 datasets with 762 samples showed statistically significant upregulation of Sox9 in lung ADC tumor tissue in comparison to normal adjacent tissue (Fig. 1A). We screened a panel of 14 lung ADC cell lines, and observed detectable levels of Sox9 protein in 10 cell lines. Low Sox9 expression was found in immortalized bronchial epithelial Beas2b cells ( Supplementary Fig. S1). We further investigated Sox9 expression at the protein level in 50 human lung ADC surgical samples and paired NATs. Sox9 protein levels were significantly higher in tumors than in NAT (Fig. 1B). Sox9 protein was predominantly nuclear and overexpressed in 70% of human ADCs. In contrast, normal adjacent lung tissue was scored negative, though vascular endothelial cells showed weak positive nuclear and cytoplasmic staining (Fig. 1B), and, when present, bronchial epithelial cells showed moderate cytoplasmic staining with focal nuclear staining.
To further investigate Sox9 overexpression in lung ADC, we queried data with annotated mutational information and analyzed Sox9 expression within lung cancer genetic subgroups, including EGFR and KRAS mutations. Sox9 mRNA was significantly higher in lung ADCs with KRAS mutations compared to NAT. EGFR mutant tumors had a trend of higher Sox9 expression than in NAT, but it was not statistically significant (Fig.  1C). These data are consistent with the report that Sox9 overexpression is required for oncogenic Kras-induced pancreatic intraepithelial neoplasia [48]. To investigate if high Sox9 expression is conserved in a GEMM of lung ADC, we stained lung adenoma and ADC sections from the LSL-K-rasG12D lung ADC mouse model for Sox9 and observed Sox9 protein expression to be mostly absent from normal lung and adenoma tissue, but highly expressed in ADC tumors (Fig. 1D). Expression was not uniform in the tumors and was restricted to the more malignant-appearing outgrowths.
Sox9 is a Notch-responsive gene during murine lung carcinogenesis.
Notch1 was recently shown to be required for the LSL-K-rasG12D lung ADC GEMM [21]. We analyzed if Sox9 is a downstream effector of Notch1 during lung carcinogenesis. Notch1 overexpression in the mouse lung alveolar epithelium, driven by the Clara cell specific protein (CCSP) promoter, leads to pulmonary adenomas that progress to ADCs when crossed with conditional Myc overexpression [22]. We tested these conditional GEMM of lung ADC to determine if Sox9 expression is an early event downstream of Notch1 signaling ( Fig. 2A). As previously reported, within 14 days of doxycycline induced N1ICD expression, hyperplastic regions were clearly discernable in stained lung sections (Fig. 2B) [22]. These hyperplastic regions progressed to adenomas with papillary histology in doxycycline-fed Notch1 mice. Immunohistochemical staining for Notch1 demonstrated high expression levels in the hyperplastic and adenomatous regions. Similarly, Sox9 protein was substantially upregulated in the hyperplastic and adenomatous regions (Fig. 2B). Moreover, multifocal ADCs developed in mice conditionally expressing both N1ICD and Myc driven by the CCSP promoter when induced by doxycycline. Notch protein was highly expressed in this model, and Sox9 displayed a similar pattern (Fig. 2B). To test if enhanced Oncotarget 3639 www.impactjournals.com/oncotarget Sox9 transcription was responsible for the increased Sox9 protein levels in these mouse models, we assessed Sox9 mRNA expression and found an 18-fold increase in lung tissue as early as 7 days after Notch1 induction (Fig.  2C). Levels continued to increase over time through to adenoma formation, and also in the ADC samples from the Notch1-Myc mice (Fig. 2C). These data indicate that Sox9 is responsive to Notch signaling as an early event during Notch1-induced lung carcinogenesis in vivo.
To determine if Sox9 expression correlates with Notch signaling in human lung ADC, we performed gene co-expression analyses across several datasets. There was a signficant correlation between Sox9 and Hes1 expression, a common Notch-target gene, in 3 of 3 examined studies ( Fig. 2D and Supplementary Fig. S2). Both Hes1 and Sox9 positively correlated with the catalytic enzyme that cleaves Notch PSEN1, the Notch ligand JAG1, and positive modulators of Notch signaling, DTX2 and LFNG. Hes1 and Sox9 both negatively correlated with MFNG, which inhibits ligand-dependent Notch activation (Fig. 2E  and Supplementary Table S1). (C) SOX9 mRNA levels were assayed in a set of mouse lung tissues from the indicated conditions (n=3 mice each). Mean fold changes with standard deviations, compared to no doxycycline treatment, are shown; *, P < 0.05. (D) Linear regression of normalized SOX9 and HES1 mRNA levels [41]. (E) Heat map of Pearson correlation coefficients of mRNA gene expression across Sox9 and genes that modulate the Notch pathway. Raw data and P values are detailed in Supplementary Table 1.
Sox9 is responsive to Notch1 signaling in human lung adenocarcinoma.
Because Sox9 was a reported target gene of the Notch pathway during murine development and we found a significant correlation between the Notch pathway and Sox9 expression at the mRNA level across microarray datasets of lung ADC samples, we set out to determine if Sox9 is a key gene downstream of Notch in lung ADC. We treated lung ADC cells with EDTA, known to induce Notch activation by chelating calcium [49]. EDTA-induced Notch activation consistently upregulated expression of Hes1 at least 5-fold at both the RNA and protein levels ( Fig. 3A and Supplementary Fig. S3). We observed dosedependent inhibition of Notch by the gamma-secretase inhibitor (GSI) RO4929097 ( Supplementary Fig. S4). The GSI substantially blocked EDTA-induced Hes1 expression (Fig. 3A, left panel). Similarly, Sox9 expression was significantly increased after EDTA-induced Notch activation in 3 of 4 lung ADC cell lines examined, and Sox9 upregulation was significantly eliminated in the presence of the GSI (Fig. 3A, right panel), suggesting that Sox9 is highly responsive to Notch signaling in lung ADC. The intensities of upregulated Sox9 and Hes1 expression induced by Notch signaling were different, as is typical of Notch-target genes [28,50].
We next set out to determine if Sox9 is downstream of Notch1 or Notch3, the two Notch receptors implicated in lung ADC [20,24,26]. We knocked down Notch1 expression in H1299 cells, which had high basal Sox9 protein levels among the screened cell lines, and Sox9 mRNA and protein expression were repressed (Fig. 3B, Supplementary Fig. S5A). Similar results were observed in A549 and H1650 cells, but not in H358 cells (Fig.  3B), suggesting that other upstream pathways might also drive Sox9 expression. Conversely, knockdown of Notch3 failed to decrease Sox9 expression in any of the 4 cell lines tested. However, in two cell lines, Notch3 silencing increased Sox9 expression, concomitant with an   Fig. S5B and C). This was likely due to a compensatory increase in Notch1 secondary to decreased Notch3 signaling, a known Notch1 antagonist [51]. Overexpression of the active form of Notch1 (N1ICD) induced Sox9 expression in lung ADC cell lines and immortalized bronchial epithelial cells that had low basal Sox9 levels (Fig. 3C). These data confirmed that Notch1 signaling influences Sox9 expression in human lung ADC.

Notch1 directly regulates Sox9 gene expression.
Recently, two RBP-Jκ binding sites in the murine SOX9 promoter were identified at approximately -325 bp (S2) and +40 bp (S5) relative to the transcriptional start site (Fig. 4A) [35,52]. Alignment of the mouse and human Sox9 promoters revealed that these mouse RBP-Jκ binding sites are not conserved in humans (Fig. 4A). We queried -950 to +50 bp relative to the human transcriptional start site using the LASAGNA algorithm [53] and identified two novel RBP-Jκ binding motifs with potential high affinity located at positions -81 bp (S3) and -10 bp (S4) (Fig. 4A). To test whether direct binding of N1ICD to these RBP-Jκ binding sites in the human Sox9 promoter is necessary for induction of Sox9 expression in human NSCLC, we used a luciferase reporter under the control of a 1 kb-fragment of the Sox9 promoter, transiently expressed in the Sox9low H1838 cells. Activity of the wild-type reporter was induced 4.5-fold, when co-transfected with N1ICD (Fig.  4C). We mutated the four putative RBP-Jκ binding sites, as well as one additional site (Site 1) proposed by Song et al. [37] to cooperate with the Smad3 SBE (Smad binding element) upstream of Sox9 in esophageal adenocarcinoma (Fig. 4B). Only mutation of Site 4, that we identified by LASAGNA, abolished activation of the luciferase reporter activity by N1ICD, showing that this -10 bp site is, indeed, a functional site responsible for the induction of Sox9 by Notch1 in human lung ADC (Fig. 4C). Consistent with these data, we performed ChIP-QPCR assays on H1838 cells transiently transfected with N1ICD, using primers designed to amplify Site 4, and found that N1ICD physically associated with the RBP-Jκ binding site in the human Sox9 promoter at or near the -10 bp site (Fig. 4D and E). Taken together, these results establish Sox9 as a direct transcriptional Notch1 target in human lung ADC.
Because Smad3 was recently shown to interact with Notch1 in esophageal adenocarcinoma cells after treatment with TGF-β, we tested if N1ICD binding to the Sox9 promoter RBP-Jκ site was dependent on TGF-β signaling in lung ADC. TGF-β increased Smad3 phosphorylation (Fig. 4F) and significantly induced Sox9 reporter luciferase activity in both the wild-type and mutated Site 4 promoters by 2-fold (Fig. 4G). Combined TGF-β treatment with transient overexpression of N1ICD resulted in an additive increase in luciferase activity, compared to either one individually, indicating that Sox9-promoter driven gene expression does not require cooperativity of N1ICD and TGF-β in lung ADC. Further, TGF-β treatment induced an additive increase in the basal luciferase activity even with a mutation at Site 4, confirming that N1ICD does not bind the putative site immediately upstream of the Smad3 binding site in lung ADC. Inhibition of TGF-β signaling by SB431542 abolished TGF-β-induced Smad3 phosphorylation (Fig. 4F) but had no effect on N1ICD-induced luciferase activity (Fig. 4G). These data confirm that Notch1 and TGF-β induce Sox9 expression independently in lung ADC.

Sox9 mediates Notch1-induced cell motility and suppression of epithelial-like cellular morphology.
To determine if Sox9 affects cell motility, we carried out in vitro wound-healing assays with H1299 and A549 cells, which have high Sox9 expression ( Supplementary Fig. S1). Knockdown of either Notch1 or Sox9 significantly reduced migration, as compared to cells transfected with a scrambled siRNA sequence. Overexpression of Sox9 rescued the reduced cell motility caused by loss of Notch1 signaling, in both H1299 ( Fig. 5A and B) and A549 cells ( Supplementary  Fig. S6A). We then tested if silencing Sox9 could block Notch-enhanced migration. Transient overexpression of activated Notch1 increased A549 cell motility, though not significantly. However, silencing of Sox9 expression reversed the increased cell motility caused by activated Notch1, down to similar levels seen with silenced Notch1 (Supplementary Fig. S6B). These data demonstrate that Sox9 at least partly mediates Notch1-induced cell migration.
H1299 cells are characterized by a mesenchymal gene expression signature and spindle-shaped morphology with few cell-cell contacts. Compared with control cells, H1299 cells with knocked down Notch1 or Sox9 expression underwent profound morphologic changes, which included a larger, flattened phenotype and tighter, more numerous cell-cell contacts. When Sox9 was overexpressed in cells with knocked down Notch1, the mesenchymal appearance of the cells was rescued, with characteristic spindle-like shape and loss of tight cell-cell contacts (Fig. 5C).

Sox9 mediates Notch1-induced cell invasion and loss of epithelial marker expression.
Because our data suggested that Sox9 mediates the adoption of a mesenchymal phenotype induced by Notch1, we analyzed cell invasion, a hallmark of mesenchymal cells. Transient Notch1 or Sox9 knockdown significantly reduced invasion of H1299 cells through Matrigel-coated Transwell inserts, and transient overexpression of Sox9 rescued the cells' invasive ability ( Fig. 6A and B).  sequences is shown, with indicated putative RBP-Jκ binding Sites 1 -5 and Smad3 binding element, relative to the SOX9 transcriptional tart site (TSS). Sites 1, 2, and 5 were previously reported (See text), and we identified Sites 3 and 4 through LASAGNA, (B) Wild-type and mutated primer sequences used for site-directed mutagenesis of Sites 1 -5 are shown. (C) Normalized luciferase fold activity (to Renilla luciferase) and standard deviations in cells co-transfected with Notch1 intracellular domain (N1ICD OE) and a luciferase reporter driven by a 1 kb fragment of the wild-type or mutated SOX9 promoter in sites 1 -5, compared to the vector control (PCDNA3); *, P < 0.05. (D) ChIP was performed using either a Notch1 or IgG control antibody on cells transfected with N1ICD or a control vector (PCDNA3), and the SOX9 promoter was amplified by PCR, using primers that were designed to amplify Site 4. (E) Quantitative PCR was performed after ChIP; *, P < 0.05. (F) pSmad3 was analyzed by western blot after treatment with vehicle controls, recombinant human TGF-β or the TGF-β inhibitor, SB431542. (G) normalized luciferase fold activity (to Renilla luciferase) in cells transfected with N1ICD and a luciferase reporter driven by a 1 kb fragment of the wild-type or mutated SOX9 promoter at Site 4, and treatment with TGF-β or SB431542, each compared to its empty vector control, which was set to 1 (Not shown); *, P < 0.05.

Oncotarget 3643 www.impactjournals.com/oncotarget
The cell-cell adhesion molecule E-cadherin is a major determinant of epithelial features. In line with the morphological changes, E-cadherin expression was substantially increased after knockdown of Notch1 expression in H1299 cells. Overexpression of Sox9 substantially, though not completely, blocked the upregulation of E-cadherin expression caused by knockdown of Notch1 (Fig. 6C and Supplementary Fig. S6C). Similarly, in A549 cells, repression of Sox9 upregulated E-cadherin protein expression, even in the presence of activated Notch1 ICD overexpression ( Supplementary  Fig. S6D). Knockdown of Notch1 significantly reduced expression of MMP9, Slug and Snail, however, overexpression of Sox9 did not rescue expression of these genes after Notch1 knockdown. Furthermore, modulation of Sox9 expression did not affect mRNA levels of Vimentin, CEACAM1, Twist, Zeb1 or Zeb2 (Data not shown). To establish the correlation between Sox9 and E-cadherin expression in vivo, we stained the human lung ADC tumors for E-cadherin. We found a significant inverse correlation between E-cadherin and Sox9 staining intensities (Fig. 6D and E). Together, these findings suggest that Sox9 may be a gatekeeper partly mediating Notch's role in EMT by inducing loss of epithelial features, blocking E-cadherin expression, and increasing cell motility and invasion.

DISCUSSION
Sox9 is a developmental gene required for lineage commitment in numerous tissue and organs, including the lung. It is well-accepted that genes regulating development also play critical roles in cancer development and progression, and thus it is not surprising that Sox9 expression is dysregulated in cancer. Sox9 is overexpressed in several tumor types and associated with poor survival [8,16,54]. Upregulation of Sox9 expression was reported to increase cell proliferation [9,11,16], and induce invasion in pancreatic, prostate, and urothelial cancer [12,55,56]. Sox9 cooperates with Slug to induce EMT Oncotarget 3644 www.impactjournals.com/oncotarget in mammary stem cells, and tumor progression in breast cancer [10]. These reports in total indicate the importance of examining the pathways inducing Sox9 dysregulation and to fully delineate the mechanisms through which Sox9 participates in lung carcinogenesis and tumor progression.
We demonstrate that Sox9 is overexpressed at the mRNA level across the majority of publicly available datasets and at the protein level in 50 primary lung ADC tumors, consistent with previous reports [8,16]. We show that Sox9 is overexpressed within the lung ADC subgroup that harbors KRAS mutations, and confirm that Sox9 is overexpressed within the more malignant portions of tumor from the LSL-K-rasG12D lung ADC mouse model [9]. The Notch pathway cooperates with Kras in development of murine pancreatic intraepithelial neoplasia (PanIN) [57] and lung ADC [21]. Furthermore, Sox9 overexpression induces acinar-to-ductal reprogramming and cooperates with Kras during PanIN formation [48].
In KRAS-mutant lung ADCs, Sox9 overexpression may be secondary event associated with tumor progression [9], rather than directly downstream of the Kras pathway. Conversely, in the murine model of Notch1-induced lung cancer, Sox9 is overexpressed as early as 7 days after induction of Notch1 overexpression in the alveolar epithelium, specifically confined to the alveolar hyperplastic regions, and thus is likely to be directly downstream of Notch. We propose that Sox9 participates as a primary downstream Notch effector early in Notch1-induced carcinogenesis, but as a secondary event during KRAS-mutant lung carcinogenesis, and both  Oncotarget 3645 www.impactjournals.com/oncotarget with poor survival, CSCs, EMT, and chemoresistance in lung cancer [23,24,58,59]; however, very little is known about the downstream genes mediating the effects of Notch in lung ADC [26,27]. We show that Sox9 is not only a direct target of Notch1 in lung ADC but also functionally mediates Notch1-induced cell motility, invasion, and mesenchymal-like characteristics. While we observed that Sox9 overexpression decreased E-cadherin, the mode by which this happens is unclear. A search of 5 kb upstream of the E-cadherin transcription start site revealed no Sox9 binding sites. Future studies are needed to investigate the mechanisms by which Sox9 regulates E-cadherin expression. Work is also needed to examine if Sox9 mediates Notch-induced chemo-and radio-resistance, as well as CSC self-renewal.
This study has important clinical implications since Notch signaling is aberrantly over-activated not only in 40% of lung ADCs but across numerous tumor types [17]. Gamma secretase inhibitors that inhibit Notch activation, and monoclonal antibodies are in clinical trials, but they are largely ineffective due to dose-limiting toxicities or development of resistance [60][61][62]. Clearly, novel methods to target the tumorigenic effects of Notch while minimizing toxicities are necessary. The results of this study could be an impetus for investigations for inhibiting Sox9 transcriptional activity or expression.

Cell culture and drug treatments
Human bronchial epithelial Beas2b cells were a kind gift of Dr. John Langenfeld, A549, H1299, H358, H1650, H322, H441, H1975, H838, H1838, H23, and H1755 cell lines were obtained from the American Type Culture Collection (ATCC), and Sk-Lu-1, H2030, and HCC-44 cell lines were a gift from Bristol Myers Squibb (Princeton, NJ). Cells were cultured in media as recommended by ATCC and maintained at 37° with 5% CO 2 and ambient O 2 . For EDTA-induced Notch activation, cells were exposed to 1 mM EDTA or culture media for 30 min, then placed back in culture media for 1 hour. For inhibition of Notch activation, cells were treated with 1 µM RO4929097 or DMSO for 48 hours. Human recombinant TGF-β (PeproTech) was used at a concentration of 10 ng/ ml for 48 hours. TGF-β inhibitor S431542 (Sigma) was used at a concentration of 10 μM for 48 hours.

Human patient samples
Formalin-fixed paraffin embedded tissue specimens were obtained through the Tissue Retrieval Service, Rutgers Cancer Institute of New Jersey, from patients who underwent surgery at Robert Wood Johnson University Hospital. This study was approved by the Institutional Review Board of Rutgers Robert Wood Johnson Medical School.

Genetically engineered mouse models (GEMM)
Animal experimentation was conducted with the approval of the Institutional Animal Care and Use Committee of the University of California, San Francisco, CA. Mice expressing the doxycycline-responsive reverse tetracycline transactivator (rtTA) driven by the rat Clara cell secretory protein (CCSP) promoter, mice with a tetracycline-responsive promoter element controlling expression of human MYC, and mice with a tetracyclineresponsive promoter element controlling expression of human N1ICD, have been previously described [22]. Transgenic mice were fed with a supplemental diet containing doxycycline (200 mg/kg). Tissue sections were obtained and RNA samples were extracted as previously reported [22].

Immunohistochemical analysis and scoring
Surgical specimens were collected by the Tissue Retrieval Services, and a tissue array comprising 50 lung ADC and 6 normal adjacent tissue (NAT) samples were prepared as 1 mm diameter cores in duplicate on two slides, totaling 4 cores per tissue, by the Histopathology and Imaging Shared Resource, at Rutgers Cancer Institute of New Jersey. Tumor/tissue quality was monitored under the guidance of a clinical pathologist. Immunohistochemical staining was done by permeabilizing with 0.1% Triton X-100 in PBS for 10 minutes, quenching endogenous peroxides with 3% hydrogen peroxide for 10 minutes, followed by blocking, as well as primary and secondary antibody incubation. Immunoreactivity was visualized with 3,3′-diaminobenzidine (DAB, Sigma). The primary antibodies used were: Sox9 (1:50, HPA001758, Sigma), E-cadherin (pre-diluted, 760-4440, Ventana), Notch1 (1:100, EP1238Y, Epitomics).
Samples for which at least 25% of the tumor cells stained positive were considered positive. Staining intensity was scored as 0, negative; 1, weak; 2, moderate; and 3, strong. Scoring was performed in a blinded manner by two clinical pathologists and discordant results were discussed until a consensus was reached.

Real-time quantitative RT-PCR (Q-PCR)
Total RNA was isolated using TRIzol (Invitrogen) and further purified using Qiashredder columns then by the RNeasy Mini kit (both from Qiagen Ltd., Germany) following standard protocols. RNA quantity and quality was assessed using a Nanodrop spectrophotometer. Reverse transcription was carried out using the Qiagen QuantiTect Reverse Transcription Kit in accordance with the manufacturer's instructions, followed by Q-PCR on a Stratagene Mx3005P machine using SYBR Green according to standard procedures. Primers were synthesized by IDT and results were analyzed using Microsoft Excel. Each assay was done at least in biological and PCR triplicate. Specificity of PCR amplification was assessed regularly using a melting-curve analysis. β-actin was used as the internal standard for normalization.

Chromatin immunoprecipitation assay
The chromatin immunoprecipitation (ChIP) assay was done according to the manufacturer's protocol (Upstate Biotechnology) in triplicate and on 2 separate occasions. In short, chromatin was cross-linked to DNA by adding formaldehyde to tissue culture dishes to a final concentration of 0.75%. Formaldehyde was neutralized and cells were scraped, washed, and resuspended in lysis buffer. Samples were sonicated to 200-500 bp fragments, diluted in ChIP dilution buffer, pre-cleared with salmon sperm DNA-protein A agarose slurry, centrifuged, and the supernatant was collected. Samples were incubated with immunoprecipitating Notch1 or control IgG antibody overnight at 4°C with rotation. Immune complexes were collected with salmon sperm DNA-protein A agarose slurry, then washed with low-salt immune complex buffer, high-salt immune complex buffer, LiCl immune complex buffer, and twice with TE buffer. Input DNA was removed for later normalization, then remaining samples were eluted, cross-links were reversed by incubation at 65°C overnight, and DNA was purified using the Qiagen Gel Extraction kit. Quantitative PCR was performed as outlined above, using the following primer to amplify the Sox9 Site 4 (Fig. 4A) promoter region: forward primer: 5′-GGACTGCTGTGCTG-TGATTG-3′, reverse primer: 5′-AGTTTCGAGCTCCGCTTTCG -3′. Samples were subsequently run on a 1.5% gel.

Site directed mutagenesis and luciferase assay
The pGL2 firefly luciferase reporter plasmid under the control of the Sox9 promoter was a kind gift of Dr. William L. Farrar. Mutations were induced using sitedirected-mutagenesis primers in conjunction with the Agilent QuikChange II Site Directed Mutagenesis kit, according to the manufacturer's protocol. Cells were co-transfected with one of these plasmids and either the N1ICD or empty vector in triplicate. After 48 hrs, cells were harvested for determination of luciferase activity by a luminometer. A Renilla luciferase plasmid was used for normalization. Mutagenesis primers are listed in Supplementary Methods.

Wound healing motility assay
Cells were plated such that they would be 95-100% confluent at the start of the assay. Cells were transfected 48 hours prior to creating a single scratch wound in triplicate wells. The wound closure experiment was performed in 2% fetal bovine serum to reduce potential differences in cell proliferation. Images were obtained at various timepoints after wounding.

Matrigel invasion assay
H1299 cells were transfected 48 hours prior to trypsinization and plated onto Matrigel invasion chambers (BD Biosciences) without serum. The chemoattractant in the lower chamber was 10% fetal bovine serum. The invasion assay was conducted for a minimum of 24 hours to reduce confounding due to potential differences in cell proliferation. Uninvaded cells were removed, and invaded cells were fixed with 4% PFA for 10 minutes, and stained with 0.05% crystal violet. Invaded cells were manually scored.