Using lung cancer as a model to identify such modifiers, we conducted a screen for genes that, when silenced, enhance EGFR dependence in EGFR-mutant lung cancer cell lines. We intentionally selected H1650 cells for the primary screen because they are insensitive to EGFR TKIs despite expressing a mutant EGFR with an exon 19 deletion (ex19del) that predicts for erlotinib sensitivity in patients. H1650 cells harbour no known EGFR TKI resistance mechanisms other than functional PTEN loss, which does not fully account for their insensitivity⁵. To identify small hairpin RNAs (shRNAs) that might restore dependence on mutant EGFR in H1650 cells, we introduced a pooled, shRNA library targeting >2,000 ‘cancer-relevant’ genes⁶ (Supplementary Table 1) and treated these cells with vehicle or erlotinib, a standard EGFR TKI used in EGFR-mutant lung cancer patients. Hairpins targeting 36 genes reproducibly conferred erlotinib sensitivity in H1650 cells across three replicates (threshold 1.5-fold depletion in erlotinib/vehicle, P < 0.1; Supplementary Table 2).

Among the primary screen hits, 18 of the targeted genes, including FAS, could be directly or indirectly linked to NF-κB signalling, which is known to have a role in survival signalling (Supplementary Table 2). Because recent data demonstrated that FAS and NF-κB signalling can promote tumour growth^7–9, we proposed that FAS–NF-κB may rescue EGFR-mutant tumour cells from EGFR inhibition. First, we confirmed that shRNAs targeting six of the highest scoring NF-κB pathway-associated genes identified in the primary pooled screen effectively silenced expression of their targets in H1650 cells (Supplementary Fig. 1a). Then we validated their growth inhibitory effects in erlotinib-treated H1650 cells using independent siRNAs (Fig. 1a, Supplementary Fig. 1b). Importantly, this reduction in cell viability was associated with increased caspase 3/7 activity (Fig. 1b), indicating that knockdown of these genes promoted erlotinib-induced apoptosis. To address directly the role of the NF-κB pathway in EGFR TKI sensitivity, we knocked down the major NF-κB subunit RELA (not represented in the pooled library) and found that RELA knockdown also induced erlotinib sensitivity in H1650 cells using short term viability and apoptosis readouts as well as clonogenic assays (Fig. 1a, b and Supplementary Fig. 2). Furthermore, this sensitizing effect was specific to EGFR inhibition because stable RELA knockdown did not alter sensitivity to cisplatin, paclitaxel or ultraviolet treatment (Supplementary Fig. 2) or other TKI (imatinib) (Supplementary Fig. 3). Because c-FLIP and RIPK have been implicated as intermediate signalling proteins linking FAS to NF-κB¹⁰, we asked if these genes also regulate erlotinib sensitivity. Indeed, silencing of c-FLIP or RIPK induced erlotinib sensitivity in H1650 cells (Fig. 1a, b). The effects of the entire panel of siRNAs targeting nine different NF-κB pathway-associated genes were observed regardless of PTEN status (Supplementary Fig. 4a, b).

Next we asked whether the NF-κB pathway genes that scored in H1650 cells show similar effects in other EGFR mutant lung cancer cell lines. Whereas H1650 cells express EGFR^ex19del, 11-18 human lung cancer cells express EGFR(L858R) yet are also relatively insensitive to erlotinib without a resistance mechanism validated in patients¹¹. Each of the siRNAs that scored in H1650 cells also scored in cell viability and survival assays in 11-18 cells (Supplementary Fig. 4c, d). Silencing of the same set of genes also enhanced the growth suppressive effects of erlotinib in HCC827 cells (expressing EGFR^ex19del) and in H3255 cells (expressing EGFR(L858R)), both of which are relatively more sensitive to EGFR TKIs (Supplementary Fig. 4e–h).

Because unknown alterations in human lung cancer cells could also influence erlotinib sensitivity, we used an isogenic system of EGFR-transduced human bronchial epithelial cells (HBEC) to test whether silencing of these genes cooperates with mutant EGFR to induce oncogene dependence. Consistent with prior data¹², human bronchial epithelial cells (HBEC)-EGFR(L858R) and HBEC-EGFR^ex19del cells were not sensitive to erlotinib (100 nM). However, silencing of each of the nine genes studied in the lung cancer cell lines also induced erlotinib sensitivity in HBEC-EGFR^ex19del (Fig. 1c) and HBEC-EGFR(L858R) cells (Supplementary Fig. 5). Induction of erlotinib sensitivity seemed equivalent across both the exon19del and L858R EGFR genotypes. The erlotinib-sensitizing effect of silencing these genes was specific to mutant EGFR because no potentiating effect was seen in wild-type HBEC-EGFR^WT cells (Fig. 1d).

Because FAS knockdown promoted erlotinib-induced apoptosis, we measured the activation state of three signalling pathways linked to cell survival (AKT, ERK, also known as MAPK1, and NF-κB) to determine which, if any, was associated with erlotinib-induced cell death. In erlotinib-sensitive HCC827 cells, erlotinib treatment alone led to reduced levels of phosphorylated pAKT, pERK and pRELA (a measure of NF-κB activity), regardless of FAS expression (Supplementary Fig. 6). But in erlotinib-resistant H1650 cells, these effects of erlotinib were only observed when FAS was silenced (Supplementary Fig. 6). In HBEC-EGFR(L858R) cells, erlotinib suppressed AKT and ERK activation regardless of FAS expression, but RELA phosphorylation was abolished only upon FAS silencing (Supplementary Fig. 6). Together these data implicate persistent NF-κB signalling in resistance to erlotinib-induced cell death.

NF-κB signalling was recently shown to be essential for KRAS-driven tumour growth¹³. Oncogenic KRAS and EGFR may drive tumour growth through a common signalling pathway. Because in our studies NF-κB knockdown alone was not lethal in EGFR-mutant lung cancer cell lines (HCC827, H3255), we compared the effect of silencing NF-κB in HBEC-KRAS12V cells versus HBEC-EGFR(L858R) cells directly. Knockdown of RELA alone impaired the growth of HBEC-KRAS12V but not HBEC-EGFR(L858R) cells (Supplementary Fig. 7), indicating that mutant EGFR does not function identically to mutant KRAS in this model.

To extend our findings to in vivo models, we used shRNAs targeting FAS, RELA or c-FLIP (distinct from the primary screen) to achieve stable knockdown (Supplementary Fig. 8) of the cognate protein in erlotinib-resistant H1650 cells. Silencing of each gene by stable shRNA enhanced erlotinib sensitivity, with a 1–2 log-fold decrease in IC₅₀ (half maximal inhibitory concentration; Fig. 2a) and induction of apoptosis as measured by PARP cleavage (Fig. 2b). Similar results were observed in 11-18 and HBEC-EGFR(L858R) cells (Supplementary Fig. 9). We found that erlotinib treatment induced tumour regression and apoptosis only upon knockdown of FAS or RELA (Fig. 2c, d and Supplementary Fig. 10a, b) in H1650 human xenograft tumours.

We next asked whether activation of NF-κB might be sufficient to confer erlotinib resistance in EGFR-mutant tumours. Canonical NF-κB signalling requires downregulation of the NF-κB inhibitor IκB¹⁴. Therefore we predicted that decreased IκB levels, leading to increased NF-κB signalling, might promote EGFR TKI resistance. First, we noted that IκB expression was lower in three erlotinib-resistant EGFR-mutant lung cancer cell lines that are resistant to erlotinib compared with three sensitive cell lines (Fig. 3a and Supplementary Fig. 11). Furthermore, we found that knockdown of IκB increased NF-κB phosphorylation (Fig. 3b) and conferred partial resistance to erlotinib in HCC827 cells in vitro (Fig. 3c) and in vivo (Fig. 3d–f). These effects were specific to EGFR inhibition because stable knockdown of IκB did not protect HCC827 cells from cisplatin, paclitaxel or ultraviolet treatment, as measured by clonogenic assays (Supplementary Fig. 12). c-FLIP has been implicated as an intermediate signalling molecule that activates NF-κB downstream of FAS¹⁰. Ectopic expression of c-FLIP induced persistent phosphorylation of RELA and resistance to erlotinib in HCC827 cells (Supplementary Fig. 13). Together these data suggest that NF-κB activation can rescue EGFR-mutant tumour cells from EGFR TKI.

Acquired resistance to erlotinib in patients can occur through second site, drug-resistant mutations in EGFR or through amplification of the MET kinase^15,16. These same resistance mechanisms can also evolve in vitro with prolonged growth in EGFR TKI^17,18. We therefore asked if NF-κB activation occurs naturally during the derivation of erlotinib-resistant HCC827 cells through prolonged and continuous EGFR TKI treatment. In parental HCC827 cells acute erlotinib treatment (100 nM, 24 h) inhibited EGFR, MET, AKT, ERK and RELA phosphorylation. Continuous erlotinib treatment resulted in the outgrowth of eight resistant HCC827 subclones (over >8 weeks) in which EGFR phosphorylation remained inhibited by erlotinib. Consistent with prior data¹⁸, we observed compensatory upregulation of phosphorylated MET in four erlotinib-resistant subclones. We also observed increased FAS expression in the absence of MET upregulation in four additional subclones. Increased RELA phosphorylation and decreased IκB expression were observed in three out of four sub-clones in which FAS was upregulated (Supplementary Fig. 14a). Knockdown of FAS in those subclones with FAS upregulation decreased pRELA and enhanced erlotinib sensitivity, but did not result in decreased pAKT or pERK (Supplementary Fig. 14b, c). Thus, FAS-mediated EGFR TKI resistance is distinct from MET-mediated EGFR TKI resistance, which occurs primarily through reactivation of pERK and pAKT in HCC827 cells¹⁸. Our results are consistent with recent data establishing both MEK-dependent¹⁹ and MEK-independent²⁰ mechanisms of resistance of BRAF-mutant melanomas to the BRAF inhibitor PLX4032.

NF-κB pathway inhibition can be achieved through inhibition of IκB kinase (IKKβ, also known as IKBKB), the primary kinase that promotes IκB instability leading to NF-κB activation¹⁴. We therefore asked if pharmacologic or genetic inhibition of IKKβ enhances erlotinib sensitivity in our erlotinib-resistant lung cancer models. Treatment of H1650 (Fig. 4a) or HBEC-EGFR(L858R) cells (Supplementary Fig. 15a, b) with the IKKβ inhibitor BMS-345541 (ref. ²¹) inhibited RELA phosphorylation and restored erlotinib sensitivity. Because BMS-345541 has suboptimal pharmacokinetic properties we used an shRNA to knock down IKKβ in H1650 cells. The resulting H1650 cells, or those transduced with a non-target control shRNA, were then grown as xenograft tumours and treated with either vehicle or erlotinib. IKKβ knockdown resulted in increased levels of IκB and decreased NF-κB phosphorylation, as expected. Erlotinib treatment induced tumour regression and apoptosis only upon knockdown of IKKβ in this system (Fig. 4b and Supplementary Fig. 15c, d).

To determine the clinical relevance of our findings, we examined the status of NF-κB activation in a cohort of 52 patients (part of a previously characterized cohort²²) with EGFR mutant lung cancer who were treated with erlotinib and did not have evidence of the T790M mutation as a potential cause of erlotinib resistance. Because reduced IκB levels were associated with erlotinib resistance in cell lines and tumour models, we asked if IκB expression correlated with EGFR TKI response in patients with EGFR-mutant lung cancers. Low IκB expression (‘high-NF-κB’ activation state) was predictive of worse progression-free survival (Fig. 4c and Supplementary Table 4a) and decreased overall survival (Supplementary Fig. 16a and Supplementary Table 4b). This finding was specific to erlotinib because IκB expression did not predict outcome in a previouslycharacterized cohort²³ of 43 EGFR-mutant lung cancer patients that were not treated with EGFR TKIs but instead with chemotherapy and surgery (Fig. 4d and Supplementary Fig. 16b). We also observed a significant positive correlation between FAS expression and RELA, c-FLIP and IκB levels in the erlotinib-treated patients (Supplementary Table 5). Together, our findings suggest that the extent to which EGFR-mutant tumour cells engage the NF-κB pathway may, in part, explain the non-uniform response to EGFR TKI treatment observed in EGFR-mutant lung cancer patients and provide rationale for testing an IKK inhibitor in combination with an EGFR TKI in EGFR-mutant lung cancer patients.

S1