Epithelial-to-Mesenchymal Transition Rewires the Molecular Path to PI3K-Dependent Proliferation

Epithelial-to-mesenchymal transition (EMT) is a process that underlies patterning and morphogenesis at multiple stages during development (1). EMT is induced by numerous signaling pathways in different contexts, ultimately with all pathways converging at the transcriptional level to induce expression of several pleiotropic transcription factors, including the Twist, Snail, and Zeb families. These factors repress genes that maintain the epithelial phenotype and induce the mesenchymal gene expression program (1, 2). Reactivation of transcriptional programs associated with developmental EMT is seen in human tumors. Expression profiles of circulating tumor cells show EMT-associated changes, and recent work suggests that reversible plasticity along the EMT spectrum is required for metastasis (3–7). Furthermore, tumors expressing mesenchymal markers have been correlated with aggressive disease, metastasis, and poor prognosis (2). As tumors show both intertumoral and intratumoral heterogeneity across the EMT spectrum, defining the molecular dependencies of cells in the two states will be critical for developing efficacious treatment strategies.

Although a broader picture of EMT-associated changes in gene expression and signaling pathways is emerging (8–11), the functional understanding of these changes remains incomplete. For example, a correlation between EMT status and EGF receptor (EGFR) dependency has been shown in cell lines and patients, where a more epithelial phenotype is associated with heightened sensitivity to EGFR inhibition (12–14). A similar relationship between EMT status and addiction to oncogenic KRAS and mitogen-activated protein kinase (MAPK) signaling has also been reported (15, 16). Understanding the mechanism underlying these correlations is critical to uncovering potential opportunities for therapeutic intervention. EMT-associated changes in expression of various receptor tyrosine kinases (RTK) have been reported, though the dependence on specific RTKs for proliferation and survival in different states remains to be fully elucidated (9, 10, 17, 18).

Expression of the EGFR family member ERBB3 has been associated with the epithelial phenotype in cell lines, as well as sensitivity to EGFR inhibition (9, 10, 12, 19–21). ERBB3 heterodimerizes with additional EGFR family members after stimulation with various ligands, including neuregulins (NRG). ERBB3 contains multiple binding sites for p85, the regulatory subunit of phosphoinositide 3-kinase (PI3K). This allows for direct recruitment and activation of PI3K signaling by ERBB3 (22). Although changes in expression have been observed, the functional consequences of altered ERBB3 expression and the relationship to downstream signaling after EMT have not been fully described.

Here, we use an inducible cell line model of EMT to investigate how induction of the transition alters signaling networks and requirements for proliferation. As tumor cells in vivo have more limited access to growth factors and nutrients but continue to survive and proliferate, we investigated these EMT state-specific dependencies in a serum-independent context. We demonstrate that after EMT, serum independence is reduced and that this switch is mediated by the downregulation of ERBB3, disconnecting an autocrine circuit maintaining PI3K-dependent proliferation. Furthermore, we show that alternate mechanisms of PI3K activation in mesenchymal cells are necessary to maintain proliferation and present evidence that in vivo, this may be achieved through the upregulation and/or amplification of PIK3CA. These findings inform potential therapeutic strategies for patients harboring tumors with an epithelial or mesenchymal phenotype or to circumvent the emergence of EMT-associated drug resistance.

To study the relationship between EMT, signaling, and requirements for proliferation, we established a cell line model in which EMT could be induced. H358 cells, a KRAS-mutant (G12C) non–small cell lung cancer (NSCLC) line, were transduced with 4-hydroxytamoxifen (4OHT)–inducible fusions of the EMT-promoting transcription factors Twist or Snail, and the hormone-binding domain of the estrogen receptor (ER). Treatment of H358-TwistER and H358-SnailER cells with 4OHT over 2 weeks led to EMT-associated morphologic changes (not shown), a gradual reduction in the epithelial marker E-cadherin, and a corresponding gain in the mesenchymal marker vimentin (Fig. 1A). Analysis of global gene expression before and after induction of Twist revealed additional EMT-associated changes such as downregulation of the epithelial genes MUC1, CLDN4, CLDN7, and TJP2 and upregulation of mesenchymal transcription factors including SNAI1 and ZEB1/2 and the mesenchymal marker FN1 (Supplementary Table S1).

To investigate epithelial versus mesenchymal state-specific requirements for proliferation in the H358-TwistER and H358-SnailER cells, we monitored signaling through major proliferation and survival pathways following EMT induction. Interestingly, after 2 weeks of 4OHT treatment, basal AKT phosphorylation under serum-free conditions was selectively reduced, whereas extracellular signal-regulated kinase (ERK) was not affected (Fig. 1B). ERK activity is likely maintained by constitutively active mutant KRAS in these cells. Phosphorylation of PRAS40 and p70S6K, effectors downstream of AKT, was also reduced after 4OHT treatment (Fig. 1B). To understand the functional consequences of decreased AKT signaling, we measured proliferation with and without 2 weeks of 4OHT pretreatment (Fig. 1C and D). In both the H358-TwistER and H358-SnailER models, the basal rate of proliferation in full serum was moderately reduced after 4OHT treatment. In serum-starved conditions, however, proliferation of cells in the epithelial state was unaffected, whereas cells in the mesenchymal state showed vastly reduced proliferation upon serum withdrawal (Fig. 1C and D).

To determine whether the decrease in basal AKT activity was related to the loss of serum-independent proliferation after EMT, we assessed whether signaling through PI3K–AKT was required for maintaining proliferation before EMT. Cells were treated with increasing doses of two class 1 PI3K inhibitors, XL147 and BKM-120, and serum-independent proliferation was measured over 72 hours. Inhibition of PI3K in the epithelial state reduced proliferation in a dose-dependent manner (Fig. 2A and B). Treatment with two AKT inhibitors, GSK690693 and MK-2206, also led to reduced serum-independent proliferation, though to a lesser extent than the PI3K inhibitors. This is likely due to observed feedback activation of ERK and/or AKT signaling (Supplementary Fig. S1A and S1B).

Expression of activating mutants of p110α (H1047R or E545K) or wild-type (WT) p110α after EMT was sufficient to restore AKT phosphorylation in TwistER and SnailER cells in the absence of serum (Fig. 2C), and, accordingly, expression of activated p110α as well as WT-p110α was able to restore serum-independent proliferation (Fig. 2D, left). Proliferation in the presence of serum was similar to serum-free conditions for cells expressing mutant p110α; however, serum increased proliferation of WT-p110α–expressing cells to a similar level as the oncogenic mutants (Fig. 2D, right). Together, these data show that basal PI3K–AKT signaling in the epithelial state is important for maintaining serum-independent proliferation in this model, and PI3K–AKT signaling is reduced in the mesenchymal state.

To determine what regulates AKT activity, phospho-RTK (p-RTK) arrays were used to assess RTK activity before and after the induction of EMT in H358-TwistER cells. Mesenchymal cells showed reduced phosphorylation of EGFR, ERBB2, and most prominently ERBB3 (Fig. 3A). These data were validated by Western blotting using phospho-specific antibodies. Loss of ERBB3 activity after EMT correlated with a reduction of total ERBB3 protein, whereas EGFR phosphorylation was reduced without a significant change in total protein levels (Fig. 3B). These changes were also apparent at the mRNA level (Fig. 3C and Supplementary Table S1). Changes in EGFR phosphorylation may be attributed to a downregulation of EGF ligands, including TGFA and EREG, seen after EMT (Fig. 3C and Supplementary Table S1).

To understand whether this dramatic loss of ERBB3 is a generalizable phenomenon in cells with a mesenchymal phenotype, we measured ERBB3 levels in additional models of EMT and in a panel of NSCLC cell lines. Treatment of parental H358 cells, as well as the H441 and H2122 NSCLC lines, with TGF-β1, recapitulated the reduction in ERBB3 protein, despite a less robust downregulation of E-cadherin (Fig. 3D). Furthermore, the loss of ERBB3 expression after TGF-β1 treatment in these cell lines coincided with reduced phosphorylation of AKT and its downstream effectors PRAS40 and p70S6K (Fig. 3D). H358 and H441 cells treated with TGF-β1 showed reduced serum-independent proliferation (Supplementary Fig. S2A), though H441 cells did not tolerate TGF-β1 treatment well and did not proliferate well even in the presence of serum. TGF-β1 treatment of two mesenchymal cell lines, H647 and H2030, showed no effect on AKT or downstream signaling (Supplementary Fig. S2B).

In a small panel of NSCLC cell lines, we observed that ERBB3 expression is present in cells with higher E-cadherin (Fig. 3E). In addition, gene expression data from a large panel of 75 NSCLC cell lines show that those expressing E-cadherin above the median level (CDH1^hi) compared with those expressing E-cadherin below the median level (CDH1^lo) had significantly higher ERBB3 expression (Fig. 3F). These data support the observation that high ERBB3 levels are seen in the epithelial state and ERBB3 expression is downregulated after the transition to a more mesenchymal state in NSCLC cell lines.

Next, we determined whether ERBB3 was directly responsible for driving basal AKT signaling in the epithelial state. In serum-free conditions, knockdown of ERBB3 in H358-TwistER cells before EMT induction led to a reduction in phosphorylation of AKT and its downstream effectors, suggesting that basal AKT signaling is maintained through an autocrine loop (Fig. 4A). To identify the autocrine ligand and ERBB3 heterodimerization partners, we knocked down ERBB2/HER2, EGFR, TGF-α, epiregulin (EREG), or neuregulin-1 (NRG1) and measured AKT pathway phosphorylation. Only loss of ERBB2 or NRG1 reduced Akt phosphorylation to levels comparable with that seen with ERBB3 knockdown (Fig. 4B and Supplementary Fig. S3A–S3E). Given that ERBB3 potently activates PI3K through recruitment of p85, the PI3K regulatory subunit (22), we immunoprecipitated p85 or ERBB3 in the absence or presence of siRNA-targeting NRG1, ERBB2, TGFA, or EREG. p85 was basally associated with ERBB3, and knockdown of NRG1 or ERBB2 selectively reduced this interaction (Fig. 4C and Supplementary Fig. S3F). Taken together, these data suggest that in the epithelial state, PI3K–AKT activation by autocrine NRG1 stimulation of ERBB3/ERBB2 heterodimers maintains serum-independent proliferation.

Accordingly, siRNA-targeting ERBB3, ERBB2, or NRG1, but not TGFA or EREG, ablated serum-independent proliferation in H358-TwistER cells before EMT (Fig. 4D and Supplementary Fig. S4A). Interestingly, in full serum, knockdown of ERBB3 in H358-TwistER cells before EMT reduced proliferation to the same level seen in cells after EMT induction, whereas knockdown of ERBB2 or NRG1 showed modest to no effect on proliferation in the presence of serum (Supplementary Fig. S4B).

Knockdown of ERBB3 and NRG1 in additional epithelial NSCLC lines, H441 and H2122, recapitulated the reduction in basal pAKT signaling seen in the H358-TwistER cells before EMT (Supplementary Fig. S5A). Immunoprecipitation of the PI3K regulatory subunit, p85, in these cells after transfection with NRG1 siRNA also led to the decreased association of ERBB3 with PI3K (Supplementary Fig. S5B). ERBB3 knockdown in these cells reduced serum-independent proliferation, though the effect was more modest than what was seen in the H358-TwistER cells (Supplementary Fig. S5C).

If downregulation of ErbB3 after EMT disrupts the autocrine circuit responsible for maintaining proliferation, we hypothesized that restoring ERBB3 should rescue serum-independent growth in mesenchymal cells. Indeed, exogenous expression of ERBB3, but not EGFR, in H358-TwistER or H358-SnailER cells after EMT increased basal phosphorylation of AKT, PRAS40, and p70S6K and restored serum-independent proliferation (Fig. 5A and B and Supplementary Fig. S4C). EGFR or ERBB3 expression in cells before EMT had minimal effect on downstream signaling (Fig. 5A). Transfection of ERBB3 into 4OHT-treated H358-TwistER cells promoted growth in full serum beyond the level seen in the control, suggesting that abundance of ERBB3 ligand may be limiting (Fig. 5B). Importantly, the ability of ERBB3 to reactivate AKT activity and rescue serum-independent proliferation in H358-TwistER cells after EMT was dependent on the presence of NRG1 as well as ERBB2, as concurrent knockdown of NRG1 or ERBB2 ablated the effect of ERBB3 reexpression on AKT activity as well as proliferation (Fig. 5C and D). Transfection of ERBB3 into additional mesenchymal NSCLC cell lines also elevated AKT phosphorylation. In the mesenchymal cell line, H2030, NRG1 siRNA ablated the increase in phospho-AKT (pAKT) following ERBB3 expression as was seen in the H358-TwistER cells (Fig. 5E). These data support a mechanism in which autocrine NRG1 stimulates the activity of ERBB3–ERBB2 heterodimers that maintain basal PI3K–AKT signaling driving serum-independent proliferation and loss of ERBB3 expression after EMT disrupts this circuit.

Given that tumor cells in the mesenchymal state in vivo still maintain proliferation in less than full-serum conditions, we investigated whether alternate mechanisms of PI3K–AKT activation could substitute for ERBB3 expression and rescue growth. H358-TwistER cells before and after EMT were stimulated with a panel of growth factors, and serum-independent proliferation was measured over 72 hours. Although none of the factors significantly affected proliferation in the epithelial state, a number of these were able to restore proliferation in the mesenchymal state, including hepatocyte growth factor (HGF), insulin-like growth factor I (IGF-I), EGF, and EREG (Fig. 6A and B). Importantly, each of the growth factors that restored proliferation after EMT also induced AKT phosphorylation (Fig. 6C). Interestingly, IGF-I stimulation showed an induction of pAKT but no stimulation of phospho-ERK (pERK), supporting the hypothesis that PI3K–AKT signaling is sufficient to restore proliferation in these conditions (Fig. 6C). Knockdown of ERBB3 in H358-TwistER cells before EMT affected only basal pAKT levels, while the ability of EGF to stimulate AKT activity was not affected, suggesting that although under normal conditions ERBB3 drives proliferation, the ability of growth factors to re-activate the pathway is not dependent on the presence of ERBB3 (Supplementary Fig. S6A).

Rescue by EGF and IGF-I was dependent on stimulation of PI3K–AKT signaling, as the presence of the PI3K inhibitor XL147, and to a lesser extent the AKT inhibitor MK-2206, ablated the ability of these growth factors to restore proliferation (Fig. 6D and E). As ERBB3 is absent to directly activate PI3K–AKT signaling required for proliferation after EMT, we sought to identify other adaptor proteins that mediate RTK activation of PI3K–AKT. We found that the adaptor GAB1 was required for EGF-mediated AKT phosphorylation, as knockdown of GAB1 ablated the ability of EGF to activate the pathway (Supplementary Fig. S6B). Furthermore, knockdown of GAB1 in H358-TwistER cells before EMT minimally affected proliferation of cells growing in serum; however, after the transition, these cells showed increased dependence on GAB1 for serum-derived proliferative signals (Supplementary Fig. S6C).

As H358 cells also harbor an activating mutation in KRAS, we sought to identify a relationship between EMT and KRAS dependence in our model. Knockdown of KRAS before and after the induction of EMT showed similar inhibition of proliferation in the presence of serum, without dramatic induction of apoptosis (Supplementary Fig. S7A–S7C). Similarly, EMT did not alter response to MAP–ERK kinase (MEK) inhibition in the H358-TwistER cells (Supplementary Fig. S7D). This suggests that in this cell line oncogenic KRAS is necessary but not sufficient to drive proliferation, as the cells require additional activation of the PI3K pathway from ERBB3. These data together support the conclusion that in this model of EMT, downregulation of ERBB3 after the transition disrupts an NRG1-dependent autocrine circuit driving PI3K–Akt signaling necessary for proliferation, and that reactivation of Akt signaling through alternate mechanisms is required to restore proliferation post-EMT.

As previous studies have found correlations between EMT status and sensitivity to RTK inhibitors, we sought to test these relationships in our models. EMT induction by 4OHT in H358-TwistER cells or TGF-β1 treatment in the H441 cell line both reduced sensitivity to erlotinib treatment over a range of doses in full-serum conditions (Supplementary Fig. S8A and S8B). Furthermore, as our data suggest that PI3K signaling may be important for the proliferation of cells in both the epithelial and mesenchymal states, we sought to test whether EMT altered sensitivity to PI3K inhibition. Treatment with BKM-120 showed no differential sensitivity before or after EMT in the H358-TwistER and H441 cells (Supplementary Fig. S8A and S8B). To determine whether ERBB3 may be directly involved in the change in sensitivity to EGFR inhibition, we treated H358 and H441 cells with a range of erlotinib doses after transfection with ERBB3 siRNA. Knockdown of ERBB3 shifted the erlotinib dose–response curve, suggesting that loss of ERBB3 after EMT may be related to the observed reduction in erlotinib sensitivity (Supplementary Fig. S8C).

To broaden these observations, we examined how sensitivity to erlotinib and the EGFR/HER2 inhibitor lapatinib was related to EMT status in a larger panel of cell lines from the Cancer Cell Line Encyclopedia (CCLE; ref. 23). The EMT status of the cell lines was examined using an EMT score for each cell line based on the differences in expression between known markers of the epithelial (CDH1, CLDN4, CLDN7, TJP3, and MUC1) and mesenchymal (FN1, VIM, CDH2, ZEB1/2, TWIST1/2, and SNAI1/2) states with more mesenchymal tumors having a higher EMT score (see Methods). The cell lines that were resistant to erlotinib and lapatinib had significantly higher EMT scores (Supplementary Fig. S8D and S8E). This was true for cell lines from diverse tissues, as well as in a subset of only lung cancer cell lines (Supplementary Fig. S8D and S8E). Applying this analysis to lung cancer cell lines treated with the PI3K inhibitor GDC-0941, however, showed no significant relationship between EMT score and response to PI3K inhibition. These data support our in vitro findings that EMT alters sensitivity to EGFR inhibition, but that both epithelial and mesenchymal cells use PI3K for proliferation.

To further validate our in vitro findings, we analyzed data from primary lung tumor samples collected by The Cancer Genome Atlas (TCGA). Tumors were analyzed for their relative expression of the epithelial and mesenchymal markers described above, and an EMT score for each sample was calculated and used to stratify tumors by EMT state (Fig. 7A). Relative ERBB3 expression levels were compared along the EMT spectrum. As predicted by our in vitro findings, ERBB3 mRNA expression was highly coexpressed with epithelial genes (Fig. 7A and B) and was strongly anticorrelated with tumors in the mesenchymal state (Fig. 7C).

Given that in vitro re-activation of PI3K–AKT signaling is necessary to maintain serum-independent proliferation in the mesenchymal state, we hypothesized that tumors with a more mesenchymal phenotype harboring low ERBB3 expression may be enriched for alternate mechanisms of PI3K pathway activation. Strikingly, expression of PIK3CA itself, encoding the p110α catalytic subunit of PI3K, clustered with expression of known mesenchymal genes (Fig. 7A and B) and was significantly correlated with a more mesenchymal phenotype (Fig. 7D). In addition, increased expression of PIK3CA mRNA was linked to copy number gains in PIK3CA, suggesting a selection for tumor cells harboring this amplification in mesenchymal tumor cells (Fig. 7A).

At the protein level, reverse phase protein array data from these human tumors recapitulated the correlation between EMT status and ERBB3, as the levels of E-cadherin and ERBB3 protein were positively correlated (not shown; r = 0.2; P = 1.7 × 10⁻⁶). Interestingly, there was no significant correlation between E-cadherin and the phosphorylation status of AKT (not shown; pS473: r = 0.05, P = 0.3; pT308: r = 0.06, P = 0.2), suggesting that despite the loss of ERBB3, through the increased expression of PIK3CA or alternate mechanisms, AKT activation is maintained in vivo. Analysis of this human tumor data supports our in vitro model that PI3K–Akt signaling is important for maintaining proliferation, and that cells in either an epithelial or mesenchymal state depend on divergent molecular mechanisms to activate the pathway.

Given the genetic heterogeneity within and between tumors (2, 24–28), an understanding of the best molecular targets for cells across a spectrum of differentiation states may facilitate strategies to target multiple populations and reduce the emergence of drug resistance. Identifying molecular vulnerabilities across the EMT spectrum may be particularly critical in light of studies that suggest that mesenchymal-like cells are generally more resistant to conventional treatments and are associated with poor prognosis (29).

Our data present a mechanistic example of how rewiring of signaling pathways after EMT relates to changes in specific dependencies for proliferation. Loss of ERBB3 expression after EMT disrupts an autocrine circuit required to activate PI3K-dependent proliferation before the transition. Previous studies have observed a critical role for autocrine NRG1 stimulation of ERBB3–ERBB2 heterodimers as a driver in various tumor contexts (30, 31). High levels of autocrine NRG1 stimulation of ERBB2–ERBB3–PI3K signaling has been shown to underlie sensitivity to the HER2 kinase inhibitor lapatinib in a subset of cell lines, whereas inhibition of EGFR–ERBB3-driven PI3K activity has been related to the efficacy of the EGFR inhibitor gefitinib and others (19, 30). Interestingly, RTK-driven PI3K activity has been shown to be important for growth and survival even in the context of KRAS-mutant colorectal cancer cells, though no relationship to EMT was described (32). Our data support and expand on this finding, as the cell lines tested in this study harbor mutations in KRAS, but still remain dependent on ERBB3 for PI3K signaling before EMT. This may be informative for designing treatments for the large percentage NSCLCs harboring KRAS mutations. Our current findings suggest that although targeting specific RTK-driven pathways may initially prove effective, induction of EMT or selection for preexisting genetic clones with a more stable mesenchymal phenotype may provide an escape mechanism or a path to drug resistance after these treatments.

In vitro, we showed that reactivation of PI3K by mutation, overexpression, or growth factor–mediated RTK activation is sufficient to restore proliferation. This defines many possible routes through which cells may circumvent EGFR family or other inhibitors through EMT. Analysis of primary tumor data revealed amplification or upregulation of PIK3CA as one potential mechanism to compensate for loss of ERBB3 in more mesenchymal tumors in vivo, supporting one of the mechanisms described to maintain proliferation in vitro. Whether increased levels of PIK3CA are sufficient to maintain proliferation or require additional signals remains to be elucidated. Recent work that suggests the broad potential for growth factors to drive resistance to RTK inhibition indicates that there may be several additional mechanisms of EMT-associated changes in drug sensitivity (33). Analysis of the changes in expression of RTKs after EMT may provide a starting point to define putative resistance mechanisms, though an understanding of microenvironmental context will be critical. Defining both the expression of RTKs as well as the presence of corresponding ligands and connection to downstream pathways is required to define the most relevant mechanisms able to drive proliferation in a given context. This is exemplified by our observation that ERBB3 reexpression can restore serum-independent proliferation after EMT, whereas EGFR cannot. This suggests that in the context of ERBB3, there is sufficient autocrine NRG1 present to stimulate receptor activity and drive recruitment of PI3K, whereas with EGFR there is either insufficient ligand or a lack of requisite connections to downstream pathways.

As this cellular plasticity may generate a moving target with respect to RTK inhibition, defining the relevant connections downstream may reveal alternate dependencies. In our data, loss of ERBB3-derived PI3K signaling rendered cells dependent on the adaptor GAB1 to transduce signals from upstream RTKs to stimulate PI3K–AKT-driven proliferation. A switch to signaling through GAB1 has been previously implicated in resistance to EGFR inhibition by transmitting signals from an alternate RTK, MET, highlighting the ability of downstream adaptors to provide flexibility to signaling pathways after initial blockades (34). Both before and after EMT cells can use PI3K–AKT for proliferation, but the mechanisms upstream of AKT activation are altered. This suggests that targeting PI3K in combination with relevant RTKs may prevent the selection for EMT-associated rewiring that drives pathway activity through alternate signals. Specifically, recent data suggest that mesenchymal cells show enhanced expression of the RTK AXL, and were more sensitive to AXL inhibition (35). Signaling through AXL has also been associated with resistance to EGFR-targeted therapies (36, 37). Accordingly, in our system, AXL was upregulated at the mRNA level and showed increased phosphorylation after the induction of EMT, supporting its role as a target in mesenchymal cells (Supplementary Table S1 and Fig. 3A).

This study defines how a specific molecular change associated with EMT affects signaling and proliferation. Future studies are needed to expand these findings to define the relationship between additional RTKs, EMT, and cellular dependencies in the context of different tumors. Establishing a more comprehensive mechanistic understanding of these relationships will allow for treatment strategies that account for inherent cellular plasticity or subclonality driving drug resistance.

Antibodies were obtained as follows: pAKT(S473), pERK, ERK, p110α, parp, pERBB3(Y1289), pEGFR(Y1068), Cl-caspase-3, pPRAS40, PRAS40, p-p70S6K, and p70S6K from Cell Signaling Technology; E-cadherin, ERBB3, EGFR, HER2, and GAB1 antibodies from Santa Cruz Biotechnology; actin and KRAS antibodies from Sigma; vimentin antibody from BD Biosciences; and AKT, p85, and pTyr antibodies from Millipore. Growth factors were obtained as follows: HGF, NRG1β, epiregulin, TGF-β1 (R&D Systems), EGF, platelet-derived growth factor (PDGF; Invitrogen), and IGF-I (Sigma). XL147 was a gift from Exelixis, and BKM-120, GSK-690693, and MK-2206 were obtained from Selleck Chemicals.

The H358 cell line was obtained from the laboratory of M. McMahon at University of California San Francisco (UCSF; San Francisco, CA). H441 and SKLU1 cell lines were obtained from the UCSF Cell Culture Facility. H23 and H2122 cell lines were from J. Luo at the National Cancer Institute (NCI; Bethesda, MD). H647, H2030, SW1573, and LU99A cells were obtained from C. Benes at Massachusetts General Hospital (Boston, MA). No additional cell line authentication was performed. Cultures were maintained in RPMI-1640 (GIBCO), supplemented with 10% penicillin–streptomycin. For stable cell line generation, pWZL-Blast-TwistER (Addgene plasmid 18799), pWZL-Blast-SnailER (Addgene plasmid 18798), or pWZL-Blast-GFP (Addgene plasmid 12269) vectors were obtained from Addgene and have been previously described (38, 39). Retrovirus was produced by transfection of plasmids into Phoenix cells. H358 cells were transduced with retrovirus and selected in 5 μg/mL of blasticidin. For induction of EMT, H358-TwistER or H358-SnailER cells were treated with 100 nmol/L 4OHT (Sigma) for 14 days before plating for experiments.

siRNA-targeting ERBB3, NRG1, TGFA, EREG, and GAB1 were obtained from Dharmacon/Thermo. Negative control siRNA, ERBB2, and EGFR siRNAs were obtained from Qiagen. KRAS-targeted sequences were optimized in collaboration with Mirimus. siRNAs were transfected at a 20 nmol/L concentration, using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's protocol. A list of full oligo sequences is available in the Supplementary Methods. For expression, 0.5 μg of each expression vector per well in a 6-well plate (GFP, p110α, ERBB3, EGFR cDNAs, in a pcDNA3 backbone in frame with a C-terminal flag tag) was transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol.

Cells were lysed in 1% Triton lysis buffer [25 mmol/L Tris pH 7.5, 150 mmol/L NaCl, 1% Triton X-100, 1 mmol/L EDTA, 1 mmol/L EGTA, 20 mmol/L NaF, 1 mmol/L Na₂VO₄, and 1 mmol/L dithiothreitol (DTT)] supplemented with a protease inhibitor cocktail (Roche) and cleared by centrifugation (13,000 rpm, 10 minutes). Protein concentration was measured using a Bio-Rad modified Bradford Protein Assay (Bio-Rad). Lysates were boiled in 1× sample buffer (NuPAGE; Invitrogen). Equivalent protein quantities were run on SDS-PAGE (NuPAGE; Invitrogen), transferred to nitrocellulose (iBlot; Invitrogen), and probed with the indicated primary antibodies. Primary antibodies were detected with secondary antibodies conjugated with IRDye800 (Rockland) or Alexa Fluor 680 (Molecular Probes). Blots were visualized using a LI-COR Odyssey scanner.

Cells were lysed as described above for immunoblot analysis. Equal amounts of lysates were loaded onto Protein G Sepharose (GE Healthcare) or Dynabeads Protein G (Invitrogen) and incubated on a rotator overnight at 4°C. Beads were washed three times with lysis buffer and then boiled in 1× sample buffer for immunoblot analysis (NuPAGE; Invitrogen).

Growth curves were generated by quantifying the relative number of viable cells on consecutive days using the CellTiter 96 Aqueous Non-Radioactive Cell Proliferation Assay (MTS; Promega). Each condition was normalized to the corresponding day 1 time point. For siRNA experiments, cells were plated into 96 wells containing RNAi-Lipofectamine complexes on day 0. For expression of RTKs or p110α, cells were transfected as described in 6-well plates on day 0 and incubated for 6 hours. After 6 hours, transfected cells were plated into 96 wells. For growth curves in the presence of siRNA and expression of ErbB3, cells were transfected in 6-well plates on consecutive days with siRNA followed by transfection with ErbB3, incubated 6 hours, and then plated into 96 wells on day 0.

EMT score for each tumor or cell line was defined on the basis of analysis of published gene expression data. The EMT score was calculated by the sum of expression of well-known mesenchymal marker genes (FN1 + VIM + ZEB1 + ZEB2 + TWIST1 + TWIST2 + SNAI1 + SNAI2 + CDH2) minus the total expression of known epithelial genes (CLDN4 + CLDN7 + TJP3 + MUC1 + CDH1). Data were taken from the June 3, 2013, release of the TCGA Lung dataset and downloaded through the UCSC cancer genome browser (https://genome-cancer.ucsc.edu/). In addition, expression data and drug sensitivity for cancer cell lines were obtained through the CCLE project (23).

F. McCormick has received a commercial research grant from Daiichi-Sankyo Co., Ltd., and is a consultant/advisory board member of the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.B. Salt, F. McCormick

Development of methodology: M.B. Salt

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.B. Salt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.B. Salt, S. Bandyopadhyay, F. McCormick

Writing, review, and/or revision of the manuscript: M.B. Salt, S. Bandyopadhyay, F. McCormick

Study supervision: F. McCormick

The authors thank Daiichi-Sankyo Co., Ltd. for support; A. Balmain, B. Braun, and members of the McCormick laboratory for discussion and feedback; and T. Tokuyasu and the UCSF Cancer Center Bioinformatics Core and the Gladstone Institute Genomics Core for preparation of RNA and microarray data generation and analysis. In addition, the authors thank R.A. Weinberg and the Weinberg laboratory for making the pWZL-TwistER, pWZL-SnailER, and pWZL-GFP plasmids available; J. Luo and C. Benes for cell lines; and T. Yuan, E. Mercado, and M. Holderfield for critical review of the article.

This work was supported by Daiichi-Sankyo Co., Ltd. and NCI grant U01 CA16837.