Acquired Resistance of EGFR-Mutated Lung Cancer to Tyrosine Kinase Inhibitor Treatment Promotes PARP Inhibitor Sensitivity.

Lung cancers with oncogenic mutations in the epidermal growth factor receptor (EGFR) invariably acquire resistance to tyrosine kinase inhibitor (TKI) treatment. Vulnerabilities of EGFR TKI-resistant cancer cells that could be therapeutically exploited are incompletely understood. Here, we describe a poly (ADP-ribose) polymerase 1 (PARP-1) inhibitor-sensitive phenotype that is conferred by TKI treatment in vitro and in vivo and appears independent of any particular TKI resistance mechanism. We find that PARP-1 protects cells against cytotoxic reactive oxygen species (ROS) produced by nicotinamide adenine dinucleotide phosphate (NADPH) oxidase (NOX). Compared to TKI-naive cells, TKI-resistant cells exhibit signs of increased RAC1 activity. PARP-1 catalytic function is required for PARylation of RAC1 at evolutionarily conserved sites in TKI-resistant cells, which restricts NOX-mediated ROS production. Our data identify a role of PARP-1 in controlling ROS levels upon EGFR TKI treatment, with potentially broad implications for therapeutic targeting of the mechanisms that govern the survival of oncogene-driven cancer cells.

INTRODUCTION

In patients with non-small-cell lung cancer (NSCLC) harboring activating mutations in the epidermal growth factor receptor (EGFR), the mainstay of treatment has been administration of an EGFR-directed tyrosine kinase inhibitor (TKI), such as erlotinib, gefinitib, or osimertinib (Mok et al., 2009; Sequist et al., 2008; Soria et al., 2018). However, over time virtually all tumors acquire resistance to TKI through a variety of mechanisms (Jänne et al., 2015; Piotrowska et al., 2015; Sequist et al., 2011). As a result, a majority of patients develop disease progression within 1–2 years. In many cases, mechanisms of acquired resistance remain unknown or cannot be currently targeted (Sequist et al., 2011). Furthermore, more than one resistance mechanism may arise in the same patient (Niederst et al., 2015). Thus, heterogeneity of acquired TKI resistance is a major clinical problem. Common therapeutic vulnerabilities in EGFR mutant tumors with different TKI resistance (TKI-R) remain to be identified.

Pre-clinical studies have shown that EGFR mutant tumor cells that initially survive TKI treatment can persist and adapt over months to develop bona fide genetic mechanisms of TKI resistance (Hata et al., 2016; Sharma et al., 2010). This ‘‘persister’’ state likely harbors multiple vulnerabilities, which may or may not be relinquished once TKI resistance is acquired (Arasada et al., 2018; Sharma et al., 2010). An unanswered question is whether elimination of these persister cells will substantially delay the development of acquired TKI resistance.

Poly (ADP-ribose) polymerase (PARP) comprises a large family of proteins involved in numerous nuclear and cytoplasmic processes (Bai, 2015; Kraus, 2015). PARP-1 is the most abundant, chromatin-associated enzyme mediating post-translational polyADP-ribosylation (PARylation), which is involved in DNA repair, transcriptional control, genomic stability, cell death, and transformation (Andrabi et al., 2008; Chiu et al., 2011; Peralta-Leal et al., 2009). Since its discovery, most studies have focused on the role of PARP-1 in DNA damage detection and repair (D’Amours et al., 1999). For DNA repair, PARP-1 binds damaged DNA through its N-terminal zinc-finger motifs, thereby activating the C-terminal catalytic domain to hydrolyze NAD+ and produce poly ADP-ribose (PAR) chains (Murai et al., 2012). Over the past decade, however, the role of PARP-1 in gene regulation has received increasing attention (Kraus, 2008; Krishnakumar et al., 2008; Luo and Kraus, 2012). PARP-1 also has been reported to affect mitochondrial content and metabolism as well as reactive oxygen species (ROS) production through controlling the levels of NAD+ and key metabolic transcriptional regulators, including NRF2 (Schiewer and Knudsen, 2014).

Catalytic PARP inhibitors (PARPis) that are in clinical use trap PARP-1/2 on DNA single-strand breaks (SSBs) (Murai et al., 2012). The collision of these complexes with DNA replication forks is synthetically lethal with defects in homologous recombination repair (HRR), such as those conferred by BRCA1/2 mutations (Bryant et al., 2005; Farmer et al., 2005). Additional PARylation targets of PARP-1/2 under conditions of genotoxic stress have been reported, but it is unknown whether they can be therapeutically exploited (Jungmichel et al., 2013). There exists a great need to identify biomarkers of synthetic lethality, other than BRCA1/2 mutations, to guide the rational use of PARPis in cancer patients, including those with lung cancer. In addition, one central question is whether the roles of PARP-1 in processes unrelated to DNA repair impact the anti-cancer activity of PARPis (Lord and Ashworth, 2017).

Because mutation of EGFR can be associated with sensitivity to different DNA damaging agents, including PARPis (Liccardi et al., 2011; Pfäffle et al., 2013), we set out to investigate whether EGFR TKI exposure alters the response of EGFR mutant NSCLC cells to PARPis. We used a panel of established and patient-derived EGFR mutant cell lines that have proven clinically relevant models for studying EGFR TKI resistance mechanisms (Engelman et al., 2007; Hata et al., 2016; Ramirez et al., 2016). Unexpectedly, we find that TKI-R cells are markedly more sensitive to PARPis than their TKI sensitive (TKI-S) controls. We demonstrate that PARPi hypersensitivity is not linked to HRR but is due to a requirement for PARP-1 to restrict cytotoxic ROS production through PARylation of RAC1 and suppression of nicotinamide adenine dinucleotide phosphate (NADPH) oxidase (NOX) activity. Our data unravel a common vulnerability associated with TKI resistance and the TKI persister state that has potential for clinical translation into future treatments of EGFR mutant lung cancers.

RESULTS

Diverse TKI Resistance Mechanisms Share a Common Sensitivity to PARPis

We subjected a panel of 8 TKI-R EGFR mutant NSCLC cell lines and their TKI-S controls to treatment with the PARPi olaparib (Figure S1A). EGFR TKI resistance was associated with enhanced albeit variable sensitivity to olaparib in 7 of 8 TKI-R cell lines, including patient-derived tumor cell lines, except for H3255GR cells (Figures 1A, 1B, and S1B). Increased PARPi sensitivity was also seen with other clinically relevant inhibitors and confirmed in a colony formation assay (Figures 1C, S2A, and S2B). Furthermore, heterotopic tumor xenografts from established and patient-derived tumor cell lines with acquired TKI resistance demonstrated olaparib sensitivity in contrast to TKI-naive tumors (Figure 1D). In addition, in all TKI-R models except H3255GR cells, there was a significant increase in olaparib-induced γ-H2AX foci compared to TKI-S controls (Figures 1E and 1F). Taken together, PARPi sensitivity that is accompanied by DNA damage is a common phenotype of TKI-R tumors and appears independent of any particular TKI resistance mechanism.

Figure 1

PARP Inhibitor Sensitivity of EGFR Mutant TKI-Resistant (TKI-R) Cells In Vitro and In Vivo

(A and B) Fraction of PC9 cells (A) and patient-derived cell lines (B) after 5 days of treatment with olaparib at the concentrations shown. Gefitinib-resistant lines are marked as GR or TKI-R.

(C) Left, IC50 values for different PARPis in the PC9/GR7 pair. Right, clonogenic survival of different TKI-S/R pairs following PARPi treatment.

(D) Growth of heterotopic TKI-R and TKI-S EGFR mutant NSCLC xenografts as a result of PARPi or vehicle treatment.

(E) Left, representative immunofluorescence microscopy images showing DNA damage in patient-derived cell lines caused by olaparib (2.5 µM). Green, γ-H2AX; blue, 4′,6-diamidino-2-phenylindole (DAPI). Scale bars, 20 µm. Right, percentage of cells with ≥20 γ-H2AX foci after 24 h of olaparib.

(F) Percentage of cells with ≥20 γ-H2AX foci after 24 h of olaparib. Data points and error bars represent means with SE based on ≥3 biological repeats with at least 2 to 3 technical repeats in the cell survival experiments and 6 to 9 tumors per data point in the mouse experiments. In the immunofluorescence experiments, each repeat is based on 50 to 100 nuclei per data point.

*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.

See also Figures S1 and S2.

PARPi Sensitivity Can Be Dissociated from PARP Trapping and DNA Repair

Even though PARPis are generally thought to cause cytotoxicity through DNA replication-fork-associated double-strand breaks, we unexpectedly observed large numbers of γ-H2AX foci outside the S- and G2-phase of the cell cycle in cell lines and patient tumor tissue (Figures 2A and 2B). Furthermore, there was no difference in RAD51 foci formation between TKI-R and TKI-S cells (Figure 2C), raising the possibility that PARPi-induced DNA damage was not related to replication forks colliding with PARP molecules trapped on DNA. Indeed, TKI-R cells were sensitive to the PARPi ABT-888, which has relatively little trapping activity (Figure 2D), and when we depleted PARP-1 in TKI-R cells, we observed cytotoxicity that was comparable to olaparib treatment (Figure 2E). Lastly, knock down of PARP-1’s partner in base excision and SSB repair, XRCC1, did not abrogate the PARPi sensitivity of TKI-R cells (Figures 2F and S2C). In keeping with this observation, PARPi treatment did not yield any differential increase in DNA breaks in TKI-R compared to TKI-S cells by using the alkaline Comet assay (Figure 2G).

Figure 2

Dissociation of PARPi Effects from PARP Trapping and DNA Repair

(A) Representative images of DNA damage (green, 53BP1) in cells in different phases of the cell cycle (red, PCNA positive = S- and G2-phase). White arrows indicate PCNA-negative cells that are in G1 phase. Scale bars, 20 µm.

(B) Representative images of tumor explant tissue taken from a patient with EGFR mutant TKI-R disease. Top, 40× image with hematoxylin and eosin staining showing viable tumors cells. Bottom, 53BP1 and PCNA co-staining as in (A). Scale bars, 20 µm.

(C) Percentage of cells with at least 10 RAD51 foci following treatment with olaparib (2.5 µM) for 24 or 48 h.

(D) Fraction of cells after 5-day treatment with the PARPi ABT-888.

(E) Left, western blot depicting partial depletion of PARP-1 by using validated small interfering RNA (siRNA) in PC9 GR7 cells. Right, fraction of cells after treatment with 5 µM olaparib or after partial depletion of PARP-1.

(F) Left, western blot depicting knock down of XRCC1 with validated siRNA in PC9 GR7 cells. Right, fraction of cells after treatment with 5 µM olaparib with or without XRCC1 depletion.

(G) Normalized tail moment in a modified alkaline Comet assay, with or without FPG treatment, following 24-h treatment of cells with olaparib (5 µM). Data points and error bars represent means with SE based on ≥3 biological repeats with ≥3 technical repeats. In the immunofluorescence experiments, each repeat is based on 50 to 100 nuclei per data point.

*p ≤ 0.05, **p ≤ 0.01.

See also Figure S2.

We conclude that the PARPi sensitivity of TKI-R cells is at least partially independent of PARP trapping and SSB repair. To further study the type of DNA damage caused by PARPis, we used a modification of the alkaline Comet assay by using formamidopyrimidine-DNA glycosylase (FPG) to investigate the presence of oxidative damage. We observed a statistically significant increase in FPG-dependent tail moment when TKI-R but not TKI-S cells were treated with olaparib (Figure 2G).

TKI-Resistant Cells Treated with PARPis Display Characteristics of a Pro-oxidant State

Given our data in Figure 2 and because TKI treatment can induce ROS (Okon et al., 2015; Raha et al., 2014), we next investigated ROS levels in our tumor models. There was increased ROS in TKI-R PC9 GR7 cells treated either with PARPis or subjected to PARP-1 depletion (Figures 3A and S3A). A similar phenotype was seen in other TKI-R but not TKI-S cell lines, with the exception of H3255GR cells (Figures 3B and S3B). A similar observation was made in tumor tissues from patients with EGFR mutant cancers, although we do not have matched samples before and after acquisition of TKI resistance in the same patient (Figures 3C and S3C). Notably, even at baseline in the absence of PARPis, ROS were slightly increased in TKI-R compared to TKI-S cells (Figures 3A and S3B). Other signs of a pro-oxidant state in TKI-R cells included susceptibility to olaparib-induced 8-oxoguanine production, presence of EGFR cysteine sulfenylation, protein oxidation, as well as phosphorylation of apoptosis signal-regulating kinase 1 (ASK1) (Figures 3D–3F and S4A). Consistent with the known ROS-mediated activation of ASK1 (Noguchi et al., 2008), olaparib triggered a pronounced apoptotic response in TKI-R but not TKI-S cells (Figures 3G, S4B, and S4C).

Figure 3

PARP Inhibition Increases ROS and Oxidative Damage in TKI-R Cells

(A and B) Normalized 2’,7’-dichlorofluorescein (DCF) fluorescence as a measure of intracellular ROS after treatment of PC9 cells (A) and patient-derived cell lines (B) with olaparib (10 µM for 24 h) followed by incubation with 2′,7′-dichloro-dihydrofluorescein diacetate for 30 min and flow cytometry.

(C) Top panel, representative immunofluorescence microscopy images showing CellRox staining of patient tumor tissue explants after 24 h of olaparib treatment (10 µM). Scale bars, 20 µm. Bottom, average CellRox intensity per nuclear region of interest (ROI) in tumor explants as a function of olaparib treatment and acquired EGFR TKI resistance. Horizontal lines present median, and statistical comparison by Mann-Whitney test.

(D) Left, representative images showing olaparib-induced oxidative DNA damage (8-oxoguanine) in TKI-R versus TKI-S cells. Right, average 8-oxoguanine (8oxoG) staining intensity per nucleus in HCC827 cells. Scale bars, 20 µm.

(E) Sulfenylation assay on TKI-S and TKI-R cell lines as shown.

(F) Western blot showing phosphorylation of ASK after 24-h treatment with olaparib (5 µM).

(G) Fraction of apoptotic cells normalized to untreated TKI-S controls as measured by annexin V after 48 h of olaparib (10 µM).

(H) Fraction of TKI-S and TKI-R cells following treatments with olaparib (5 µM), glutathione (1 µM), or N-acetyl-L-cysteine (NAC; 400 nM) for 5 days as indicated. Data points and error bars represent means with SE based on ≥3 biological repeats with ≥3 technical repeats in the cell survival experiments. For the flow-cytometry-based experiments, each sample was run once. For the immunofluorescence experiments, the total number of nuclei counted is displayed.

*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.

See also Figures S3, S4, and S5.

Next, we asked whether ROS was the cause of cell death rather than an indirect result of olaparib-induced cytotoxicity. We used the anti-oxidants N-acetyl cysteine and glutathione, which had little to no toxicity under the experimental conditions chosen (Figure S5A). At the selected drug concentrations, these scavengers reduced ROS levels in TKI-R cells and were able to reverse their olaparib sensitivity but did not affect TKI-S cells (Figures 3H and S5B). Although the data do not exclude a contribution of ROS to cell kill at higher doses of olaparib in TKI-S cells, we conclude that PARPi treatment induces cytotoxic ROS in a more pronounced fashion in TKI-R than in TKI-S cells.

PARPi Sensitivity Arises in the TKI Drug Persister State Before TKI Resistance Is Acquired

The observation that TKI-R cell lines demonstrated PARPi sensitivity despite harboring distinct mechanisms of TKI resistance, such as T790M mutation, Met amplification, or epigenetic mechanisms (Figure S1A), suggested that PARPi sensitivity may have arisen before TKI resistance was acquired, i.e., in the early persister state of TKI treatment (Sharma et al., 2010). We, therefore, treated parental TKI-S cells with an EGFR TKI for 0 to 4 weeks and assayed ROS, apoptosis, and cell survival in response to PARPis at different time points. To increase the robustness of our findings, we used different EGFR TKIs (gefitinib and osimertinib) and PARPis (olaparib and niraparib). We observed significant increases in ROS and apoptosis levels upon PARPi exposure by 2 to 3 weeks into TKI treatment (Figures 4A, 4B, S5C, and S5D). At the same time, there was noticeably enhanced cellular sensitivity to PARPis by week 3, as evidence by a sustained shift in drug 50% inhibitory concentration (IC50) values for all treatment combinations (Figures 4C, S6A, and S6B). Further arguing for a TKI-inducible, PARPi sensitive cellular state, we observed that de novo TKI-resistant PC9 cells exogenously expressing the T790M resistance mutation can be made PARPi sensitive by mixing the cell population with TKI-S cells and treating with EGFR TKI for 2 weeks (Figure S7A).

Figure 4

PARPi Sensitivity Induced During the Initial Response to EGFR TKI

(A) DCF fluorescence measured in TKI-S PC9 cells after 2 weeks of 1-µM gefitinib treatment followed by reseeding and 48 h of 5-µM PARPi treatment with or without continued gefitinib.

(B) Left, representative flow cytometry diagrams of propidium iodide and annexin V double-stained cells after 3 weeks of 1-µM gefitinib treatment followed by reseeding and 48 h of 5-µM PARPi treatment with or without continued gefitinib. Right, relative fold induction of apoptosis compared to untreated controls.

(C) IC50 values for 5-day treatment with 2 different PARPis plotted against preceding treatment with different EGFR TKIs for up to 4 weeks in TKI-S PC9 cells.

(D) Relative tumor growth of heterotopic PC9 xenografts treated with osimertinib.

(E) Relative tumor growth of PC9 xenografts pre-treated with osimertinib for 3 weeks following randomization to one of three treatment groups as shown. Data points and error bars represent means with SE based on ≥3 biological repeats with ≥3 technical repeats and 7 tumors per data point in the mouse experiments.

*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

See also Figures S5, S6, and S7.

To confirm our findings in vivo, we used osimertinib and niraparib given that this combination appeared to be the most effective one in vitro (Figure 4C; Figure S6B) and because of the increasing importance of osimertinib in the clinic (Soria et al., 2018). PC9 xenografts were induced to regress by treating mice with osimertinib, with the rate of tumor regression starting to plateau after about 3 weeks of treatment (Figure 4D). At that point, animals were randomized to treatment with a PARPi, EGFR TKI, or a combination of the two (Figures 4E and S7B). Strikingly, the addition of TKI to PARPi treatment resulted in a highly statistically significant further tumor regression compared to continued TKI in mono-therapy (Figures 4E and S7C). Interestingly, PARPi treatment alone was associated with tumor regrowth within 2 weeks. This observation stands in contrast to our in vitro results where the efficacy of PARPi treatment was not influenced by the presence of EGFR TKI, at least not over the 2- to 5-day course of the respective assay (Figures S5C, S5D, S6A, and S6B).

In conclusion, EGFR TKI treatment induces sensitivity to PARPis within ~3 weeks in vitro and in vivo, which is consistent with the notion that PARPi sensitivity seen in TKI-R resistant models is generally not dictated by the type of acquired TKI resistance mechanism.

PARP-1 Regulates NOX-Dependent ROS through RAC1 PARylation

Turning next toward possible mechanisms of how PARPis may increase ROS, we considered that mitochondria and the NOX family are major sources of cellular ROS (Holmström and Finkel, 2014). Although a mitochondrial ROS scavenger was not able to reduce olaparib-induced ROS (Figure S8A), two different inhibitors of NOX reduced ROS (Figures 5A and S8B). Consistent with this observation, NOX activity in TKI-R cells was enhanced by olaparib treatment (Figure 5B). PARP-1 is a transcriptional regulator of a large number of metabolic regulators, including the redox regulator NRF2 (Wu et al., 2014). However, PARPi did not suppress NRF2 expression in TKI-R cells and produced DNA damage within only 5 h of treatment (Figures S8C and S8D). Therefore, the data suggested a non-transcriptional mechanism of ROS regulation by PARP-1.

Figure 5

RAC1 Is PARylated by PARP-1 and Controls ROS Levels

(A) DCF fluorescence after 24-h treatment of TKI-S and TKI-R cells with olaparib or NOX inhibitor VAS-2870 (2 µM) as indicated, normalized to untreated TKI-S cells.

(B) NOX activity in TKI-S and TKI-R cells after drug treatments as indicated.

(C) Normalized DCF fluorescence in TKI-S and TKI-R cells following treatment with olaparib (5 µM) or RAC1 inhibitor NSC23766 (RACi) (25 µM).

(D) Fraction of cells after 5-day treatment of cells with drugs as indicated.

(E) Lysates from TKI-S and TKI-R cells treated with olaparib for 0 to 5 h were subjected to immunoprecipitation (IP) in which RAC1 immunoprecipitates were probed for PAR and for RAC1 itself.

(F) Left, sequence alignment of RAC1 protein from different species using clustalW. Conserved amino acids are indicated as well as mutated sequences for mutational analysis. Right, western blot showing expression and PARylation of myc-tagged RAC1 PARylation mutants in PC9 GR7 cells. WT, wild-type RAC1 sequence; E156A, glutamine mutant that does not affect parylation and serves as positive control (ctrl).

(G) DCF fluorescence of PC9 GR7 cells transfected with myc-tagged RAC1 constructs as shown, with or without olaparib (PARPi) treatment of cells expressing wild-type (WT) RAC1.

(H) Fraction of cells surviving after transfection of PC9 cells with the respective PARylation mutants compared to PARPi treatment.

(I) Model of PARP-1 as gatekeeper in EGFR mutant NSCLC cells treated with TKI where it restricts the production of cytotoxic ROS through PARylation of RAC1. Whether RAC1 is PARylated by a cytoplasmic PARP-1 fraction or in the nucleus remains to be determined. Data points and error bars represent means with SE based on ≥3 biological repeats with ≥3 technical repeats.

*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

See also Figures S8, S9, and S10.

To discover protein targets for ROS regulation by PARP-1, we surveyed differentially expressed genes in TKI-R versus TKI-S EGFR mutant cells (Sharma et al., 2010). Interestingly, we observed signs of altered RAC1 pathway activity in TKI-R cells (Figure S9A) potentially linking PARP-1 function to RAC1, a small G-protein in the Rho family, which is an activator of NOX enzymes (Ueyama et al., 2006). Accordingly, the RAC1 protein was expressed at higher levels in TKI-R than in TKI-S cells (Figure S9B). Furthermore, pharmacological inhibition or depletion of RAC1 suppressed PARPi-induced ROS(Figures 5C, S10A, and S10B) and at least partially reversed the PARPi sensitivity of TKI-R cells (Figures 5D, S10C, and S10D).

We then elected to directly investigate whether RAC1 is modulated by PARP-1 catalytic function. In TKI-R cells, RAC1 was PARylated, which could be inhibited by PARPi (Figure 5E). Notably, RAC1 PARylation in TKI-S cells was more pronounced, consistent with lower ROS levels in these cells, but it was resistant to PARPi. We identified several potential PAR-binding consensus motifs in RAC1 that are evolutionary conserved (Figures 5F and S10D). We constructed a panel of loss-of-function RAC1 PARylation mutants, which were transfected into TKI-R cells, postulating dominant effects over the endogenous protein as described (Farrar et al., 2010). We confirmed that several of these sites, namely amino acids 91, 130–134, and 148, are important for PARP-1-dependent PARylation (Figure 5F). Transfection of the respective PARylation mutants, but not a control mutant (E156A), phenocopied the effects of PARPi with regard to ROS production and cytotoxicity (Figures 5G and 5H). Together, our data demonstrate a role of PARP-1 in regulating NOX complex activity and ROS production in TKI-R cells.

DISCUSSION

PARP-1 is a pleiotropic protein with functions that include regulation of DNA repair, transcription, chromosome organization, mitosis, and metabolism (D’Amours et al., 1999; Luo and Kraus, 2012; Schiewer and Knudsen, 2014). We describe a role of PARP-1 in restricting the production of cytotoxic ROS through controlling PARylation of RAC1 and NOX activity, specifically in EGFR mutant NSCLC cells treated with EGFR TKI. This is not observed in TKI-naive parental cells and appears distinct from PARP-1’s canonical function in DNA repair.

ROS are important for cell growth, stress adaption, and cell death in a concentration-dependent manner (Gorrini et al., 2013). Cancer cells often harbor high ROS levels to support proliferation and survival, and yet they may also be more sensitive to any further ROS increase and prone to cell death because of it (Gorrini et al., 2013; Trachootham et al., 2009). Excessive oncogenic ROS production by the NOX complex generates DNA damage with deleterious consequences (Weyemi et al., 2012). Owing to the pleiotropic nature of NOX-derived ROS, tumor cells with upregulated NOX may maintain activated growth pathways, evade cell death, and initiate angiogenesis and metastasis (Block and Gorin, 2012). Accordingly, TKI treatment of EGFR mutant tumor cells is associated with a ROS increase, which may aid in cell survival (Okon et al., 2015; Raha et al., 2014). We speculate that excessive NOX-derived ROS requires these cells to engage protective processes.

RAC1 belongs to the Rho family of small GTPases and plays important roles in G1 cell cycle progression, actin cytoskeleton organization, and downstream gene expression (Heasman and Ridley, 2008; Jaffe and Hall, 2005). RAC1 is spatially and temporally regulated by a wide variety of post-translational modifications, deregulation of which is found in different diseases, including cancer (Hobbs et al., 2015; Levental et al., 2010; Navarro-Lérida et al., 2012; Schwarz et al., 2012). RAC1 PARylation has not been reported previously. We observed that RAC1 is PARylated and that this modification is reduced specifically in TKI-R cells by PARP-1 inhibition within 5 h. Consistent with the reported binding of RAC1 to the NOX complex, which activates NOX enzymes and promotes superoxide production (Ueyama et al., 2006), we found RAC1- and NOX-dependent ROS production in TKI-R but not TKI-S cells. Of note, RAC1-induced superoxide production can induce a feedback loop where RAC1 protein turnover is accelerated by the increased degradation of RAC1 by the ubiquitin-mediated proteasome (Torrino et al., 2011). However, even though PARylation can serve as a signal for the ubiquitination and degradation of PARylated proteins (Huang et al., 2009; Wei and Yu, 2016), we have not observed reduced RAC1 expression following PARP-1 inhibition in our TKI-R lung cancer cells, suggesting diverse mechanisms of RAC1 regulation.

Our data make an important contribution to the view of PARP-1 as a redox regulator. For example, PARP-1 has been implicated in the regulation of mitochondrial function and oxidative metabolism (Bai et al., 2011). PARP-1 can also promote the transcription of the redox regulator NRF2 (Wu et al., 2014), although in our cell lines we did not find this. In a Parp-1 knockout mouse model, an increase in ROS levels was seen with age and a significant number of genes linked to oxidative stress and ROS production were dysregulated compared to wild-type mice (Deschênes et al., 2005). Inhibitors of PARP also protect against hydrogen-peroxide-mediated cell death (Cristóvão and Rueff, 1996). However, to date it has been unclear how to exploit these PARP-1 functions for therapeutic gain.

Our data support a concept of synthetic lethality for PARPis that specifically occurs upon TKI treatment in EGFR mutant lung cancer cells. In this model (Figure 5I), PARP-1 functions as a gatekeeper to maintain cell viability by limiting the cellular stress response to TKI exposure by PARylation of RAC1, which serves to restrict NOX-dependent ROS production. As a result, persister cells exhibit PARPi sensitivity within ~3 weeks of TKI treatment, which appears to be preserved even in clones that acquire TKI resistance at a later stage. We thus predict that a fraction of TKI-resistant tumors exhibits some degree of PARPi sensitivity in the clinic. A phase I clinical trial combining a PARPi with an EGFR TKI in TKI-resistant EGFR mutant NSCLC has opened at our institutions (ClinicalTrials.gov: NCT03891615). We suggest that our data have broad implications for a better understanding of metabolic vulnerabilities and acquired TKI resistance not only in EGFR mutant tumors but also potentially in other oncogene-driven cancers.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Henning Willers (hwillers@partners.org).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

All cell lines have been described previously and are listed in Figure S1A (Crystal et al., 2014; Engelman et al., 2006; Hata et al., 2016; Sharma et al., 2010; Soucheray et al., 2015; Turke et al., 2010). We used 5 TKI-S cell lines, i.e., PC9, HCC827, HCC4006, HCC3255, MGH 119–1, and 8 TKI-R derivatives with different mechanisms of TKI resistance, i.e., PC9 GR7 (drug tolerant/non-T790M), PC9 B (T790M), PC9 C (T790M), HCC827 GR6 (Met amplified), HCC4006 GR6 (EMT driven), HCC3255 GR (T790M), MGH 121–1 (T790M), and MGH 134–1 (T790M). Cell lines were cultured as described (Crystal et al., 2014; Hata et al., 2016; Pfäffle et al., 2013; Sharma et al., 2010; Turke et al., 2010). Established and patient-derived cell lines were cultured in RPMI-1640 (Sigma-Aldrich) except for MGH 119–1 which was cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Sigma-Aldrich). Cell culture media was supplemented with 10% and 15% Bovine Growth Serum (HyClone) for established and patient-derived cell lines, respectively, 20 mM HEPES (Sigma-Aldrich), 1% penicillin/streptomycin (Sigma-Aldrich), and 2 mM L-Glutamine (Sigma-Aldrich). TKI-R cell lines were maintained in gefitinib (2 mµ) for passaging, but gefitinib was removed when seeding for experiments to facilitate comparisons with TKI-S cells unless specified otherwise. Each cell line was generally kept in passage for no more than 2 months and a passage number window was defined in which the experiments were conducted.

Xenografts

All mouse studies were conducted through Institutional Animal Care and Use Committee–approved animal protocols in accordance with institutional guidelines. Cell line suspensions were prepared in 1:1 matrigel and approximately 5 × 106 cells were injected subcutaneously into the flanks of female athymic nude (Nu/Nu) mice (6–8 weeks old) (Charles River). Visible tumors developed in approximately 2–4 weeks. Tumors were measured with calipers and the tumor volume was calculated according to the formula Vol = 0.52 × L × W2.

Tumor tissues

Tumor tissue explants from NSCLC patients (female) were obtained under Institutional Review Board approved protocol and subjected to ex vivo treatment as described (Birkelbach et al., 2013; Willers et al., 2015).

METHOD DETAILS

Drug treatments

Gefitinib, olaparib, ABT-888, osimertinib (LC Laboratories), niraparib (MedKoo Biosciences), rucaparib (Selleck Chemicals), VAS-2870 (Sigma-Aldrich), NSC23766 (Selleck Chemicals), diphenyleneiodonium (DPI) (Sigma-Aldrich) were dissolved in dimethyl sulfoxide following manufacturers’ recommendations, and aliquots were stored at −20°C or −80°C. N-Acetyl-L-cysteine (NAC), glutathione, MitoTEMPO, and lucigenin (Sigma-Aldrich) were dissolved in ddH2O. In vitro drug concentrations and incubation times are provided in the Figure legends.

For measuring the effects of olaparib on TKI naive/resistant models in vivo (Figure 1), mice with heterotopic xenografts reaching ~350 mm3 size were randomized to drug versus control groups. Sample size (n = 6–9 per treatment group) was chosen to ensure satisfactory inter-animal reproducibility. For the persister experiment (Figure 4), tumors were grown similarly and mice with established tumors at ~1,000 mm3 were treated with osimertinib by oral gavage once daily, 5 days a week for 3 weeks until the tumor size reduced to ~350 mm3 before randomizing them into 3 different drug groups of 7 animals each. Treatments were given by oral gavage once daily, 5 days a week, for an additional 3 weeks. Depending on the experiment, mice which were treated by oral gavage with 50 mg/kg olaparib (LC Laboratories), 50 mg/kg niraparib (MedKoo Biosciences), or/and 5 mg/kg osimertinib (LC Laboratories). Tumor volumes were measured twice weekly for the entire span of the experiment.

NSCLC research specimens not needed for pathological diagnosis were placed into RPMI medium typically within 30 minutes of resection or biopsy and arrived in the laboratory 30–60 minutes later. Specimens were evenly divided into samples of < 5 mm size depending on the amount of tissue available. Samples were mock treated or incubated with olaparib (10 µM) for 24 hours at 37oC and 5% CO2, followed by flash freezing with OCT (Optimal Cutting Temperature) compound, as described (Birkelbach et al., 2013).

Cell survival assays

Cell proliferation/survival assays were performed as described (Crystal et al., 2014; Hata et al., 2016; Pfäffle et al., 2013). Briefly, exponentially growing cells were seeded into 96-well plates and treated with drug at different concentrations for 5 days, after which fractions of viable cells were determined with the CellTiter-Glo® Luminescent Cell Viability Assay (Promega) following the manufacturer’s protocol or with a nucleic acid stain (syto 60, Invitrogen). For clonogenic survival assays, cells were seeded at low density through a limited dilution series and drug was added after ~16 hours. After 2–4 weeks incubation, depending on cell line, viable colonies containing at least 50 cells were counted and the surviving fraction relative to the plating efficiency without drug was determined.

Immunofluorescence microscopy

Cells were fixed with 4% paraformaldehyde for 15 minutes and then permeabilized with PBS/0.5% Triton-X for 20 minutes at 4°C. Cells were blocked for one hour at room temperature with 5% goat serum/PBS. Primary antibodies γ-H2AX (Millipore) and RAD51 (Santa Cruz) were added for 2 hours at room temperature in 3% goat serum/PBS/0.1%Triton-X and subsequently incubated with fluorescent secondary antibody Alexa488 (Invitrogen) or Alexa555 (Invitrogen) for 1 hour at room temperature. For costaining experiments, cells were incubated with anti-53BP1 rabbit polyclonal antibody and anti-PCNA mouse monoclonal antibody (Abcam). Nuclei were counterstained with DAPI and subnuclear protein foci were scored by fluorescence microscopy (Olympus BX51). Primary antibody rabbit polyclonal anti-53BP1 (Abcam) and mouse monoclonal anti-PCNA (Abcam) were added for 24 hours at 4°C and secondary antibody Alexa Fluor488 goat anti-mouse (Invitrogen) and Alexa Fluor555 goat anti-rabbit (Invitrogen) were added for 1 hr at room temperature. Nuclei were stained with DAPI (1 µg/ml in ddH2O) for 2 minutes and subnuclear foci were scored by fluorescence microscopy (Olympus BX51). For measurement of oxidative DNA damage, fixed cells were stained with anti 8-oxoguanine mouse monoclonal antibody (Millipore). For cell-based ROS measurements, cryosections were incubated with CellROX reagent (Molecular Probes/Thermo Scientific) at a final concentration of 5 µM in PBS, incubated for 30 minutes at 37°C, followed by fixing in 3.7% formaldehyde for 15 minutes. Samples were air-dried and mounted with ProLong Diamond Antifade Mountant (Molecular Probes/Thermo Scientific) with DAPI, allowed to cure for 21 hours and imaged within 3 hours. For ROS measurements, random images were analyzed by NIS-Elements AR (Nikon). Nuclear regions of interest (ROI) were defined and average fluorescence staining intensity per ROI recorded while normalizing for unspecific extracellular staining.

Antibodies and western blot

SDS-PAGE and western blotting was carried out using standard procedures. Primary antibodies against EGFR (Santa Cruz), PARP-1 (Cell Signaling), XRCC1 (Cell Signaling), PAR (Trevigen), RAC1 (Millipore), NRF2 (Santa Cruz), Actin (Santa Cruz), Tubulin, Lamin A (Millipore) were used. Horseradish peroxidase (HRP)-conjugated rabbit anti-mouse and goat anti-rabbit secondary antibodies were purchased from Santa Cruz.

Immunoprecipitation (IP)

Cells were lysed in IP buffer (50mM Tris-HCl [pH 7.5], 150mM NaCl, 1% (v/v) NP-40, and protease inhibitor (Roche)) for 30 minutes on ice, centrifuged for 10 minutes at 13,000 x g and supernatant was removed for analysis. The lysate was then pre-cleared with 40 µL slurry of protein G beads (Santa Cruz) for 45 minutes at 4°C, centrifuged for 3 minutes at 2,000 x g and the beads were discarded and the supernatant removed for IP. A total of 500–1,000 mg of protein was incubated with 1–2 µg/ml of antibody overnight at 4°C and immunoprecipitated the following day with 40 mL slurry of protein G beads (Santa Cruz) for 2 hours at 4° C. Precipitates were washed three times for 5 minutes in IP buffer at 4°C before addition of SDS-gel-loading buffer and immunoprecipitated proteins were detected by western blot. In case of IP Rac and Probe Rac, the secondary antibody used was Mouse TrueBlot® ULTRA: Anti-Mouse Ig HRP (Rockland antibodies & assays) without hindrance by interfering immunoprecipitating immunoglobulin heavy and light chains.

RNA interference

siRNA knockdown was performed using X-tremeGENE siRNA transfection reagent (Roche) and validated siRNA sequences (working concentration 200 nM) against PARP-1 (Santa Cruz), RAC1 (Millipore Sigma), and XRCC1 (Ambion) for 48 hours. A scramble sequence (Ambion) was used as a negative control.

Fpg-modified alkaline Comet assay

Oxidative DNA damage was detected using E. coli Formamidopyrimidine-DNA Glycosylase (Fpg) in conjunction with the CometAssay® single cell gel electrophoresis kit (Trevigen). Tail moment was analyzed using TriTek CometScore.

Detection of DNA and protein oxidation

Oxidative DNA damage was detected by treating cells with 10 µM olaparib for 24 hours, or hydrogen peroxide as a control, followed by fixation with 25% Acetic Acid/75%MeOH for 20 minutes at −20° C and permeabilization with 0.5% Triton-X buffer for 10 minutes at room temperature. Cells were stained with mouse monoclonal anti-8-oxoguanine antibody (Millipore) in 3% goat serum (1:200). As a secondary antibody goat anti-mouse Alexa Fluor 488 (1:1,000) was used. Oxidized proteins were detected using the OxyBlot Protein Oxidation Detection Kit (Millipore) according to the manufacturer’s instructions.

S-Sulfenylation assay

Cells were lysed in lysis buffer (50 mM HEPES, 50 mM NaCl, 1 mM EDTA, 10% glycerol,1% Triton x-100) supplemented with 0.5 mM DCP-BIO1, 0.1 mM N-ethyl malemide, 0.1 mM Iodacetamide, 1 mM Na3VO4 and protease inhibitors. Samples were sonicated, kept on ice for 30 minutes, and centrifuged at 12,000 x g for 10 minutes at 4°C. The pellet was discarded and the supernatant removed to a new tube and rotated for 1 hour at room temperature to allow for labeling of sulfenic acids. After incubation, protein was precipitated by acetone to remove any excess probe and pelleted by centrifuging at 12,000 x g for 5 minutes. The pellet was washed in 70% acetone and suspended in lysis buffer. The amount of protein was determined by the BCA assay and 1 mg of total protein was added to a 50 µL slurry of streptavidin beads. Then, beads were rotated overnight at 4°C. After 24 hours, beads were centrifuged at 1,000 x g for 3 minutes, supernatant was discarded, and beads were washed in 1 µL of lysis buffer. This washing step was repeated three times and beads were eluted in 30 mL of 2 x LDS buffer (Life Technologies) supplemented with 2-ME.

Flow cytometry

To measure ROS in cells, 2′, 7′ dichlorodihydrofluorescein diacetate (H2DCF-DA) fluorescent probe (Invitrogen) was employed at a concentration of 10 µM and detected by flow cytometry (SORP 4 Laser BD LSRII (BD Biosciences)) using standard protocol (Ameziane-El-Hassani and Dupuy, 2013). For quantification of apoptosis, cells and media were collected, centrifuged, and resuspended in Annexin binding buffer with cell density adjusted to approximately 106/mL 48 hours after treatment with olaparib. Cells were stained with propidium iodide (Sigma-Aldrich) and Annexin V–Cy5 following the manufacturer’s protocol (BioVision), and then analyzed by flow cytometry.

NADPH assay

NOX Activity was measured using lucigenin as a substrate. Lucigenin is a reagent that interacts with superoxide and generates luminescence. This emitted luminescence can be quantitatively recorded and analyzed by a luminometer. Cell were treated with (5 µM) and without Olaparib for 5 hours and harvested in ice cold phosphate buffer of 50 mM KH2PO4, 1 mM EGTA, and 150 mM sucrose (pH 7.4) with protease inhibitors. Cell homogenates (~50 ml) were incubated with 100 µM NADPH (Sigma-Aldrich) in the absence or presence of ROS inhibitor DPI (1 µM) for 20 minutes at 37°C. Lucigenin (50 µM) was added to the reaction mixture and the incubation continued for 20 minutes. Luminescence was read by the MultiLabel reader, 2140 Envision (Perkin Elmer) as relative-light units (RLU) and activity was normalized to protein concentration.

Generation of RAC1 mutants

RAC-1 mutagenesis was performed by site-directed mutagenesis using the Stratagene Quick-Change system according to the manufacturer’s instructions. All DNAs were sequenced to verify that only the desired mutations were present.

QUANTIFICATION AND STATISTICAL ANALYSIS

Information on experimental replicates can be found in each Figure legend. Statistical comparisons were performed with the Student’s t test (two-sided) unless otherwise stated.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-γ-H2AX (Ser139) (mouse monoclonal)	Millipore	Cat# 05-636; RRID:AB_310795
Anti-53BP1 (rabbit polyclonal)	Abcam	Cat# ab21083; RRID:AB_722496
Anti-PCNA (rabbit polyclonal)	Abcam	Cat# ab2426; RRID:AB_303062
Anti-PCNA (PC10) (mouse monoclonal)	Abcam	Cat# ab29; RRID:AB_303394
Anti-8-oxoguanine (mouse monoclonal)	Millipore	Cat# MAB3560; RRID:AB_94925
Anti-Rad51 (Ab-1) (rabbit polyclonal)	Millipore	Cat# PC130; RRID:AB_2238184
Anti-EGFR (1F4) (mouse monoclonal)	Cell Signaling	Cat# 2239S; RRID:AB_331373
Anti-PARP-1 (F-2) (mouse monoclonal)	Santa Cruz	Cat# sc-8007; RRID:AB_628105
Anti-XRCC1 (rabbit polyclonal)	Cell Signaling	Cat# 2735S; RRID:AB_2218471
Anti-PAR (monoclonal)	Trevigen	Cat# 4335-MC-100; RRID:AB_2572318
Anti-RAC1 (23A8)	Millipore	Cat# 05-389; RRID:AB_309712
HRP-conjugated rabbit anti-mouse	Santa Cruz	Cat# sc-358914; RRID:AB_10915700
HRP-conjugated goat anti-rabbit	Santa Cruz	Cat# sc-2030; RRID:AB_631747
Alexa Fluor 488 goat anti-mouse	Invitrogen	Cat# A11029; RRID:AB_138404
Alexa Fluor 555 goat anti-rabbit	Invitrogen	Cat# A21428; RRID:AB_141784
Mouse TrueBlot ULTRA: Anti-Mouse Ig HRP	Rockland antibodies and assays	Cat# 18-8817-30; RRID:AB_2610849
Anti-NRF2 (C-20) (rabbit polyclonal)	Santa Cruz	Cat# sc-722; RRID:AB_2108502
Biological Samples
Tumor tissue explants from NSCLC patients	MGH	2012P002255
Chemicals, Peptides, and Recombinant Proteins
Veliparib (ABT-888)	LC Laboratories	V-4703
Rucaparib	Selleckchem	S1098
Olaparib	LC Laboratories	AZ-2281
Niraparib	Selleckchem	MK-4827
Gefitinib	LC Laboratories	G-4408
Osimertinib	LC Laboratories	AZD9291
Diphenyleneiodonium	Sigma Aldrich	D2926
VAS-2870	Sigma Aldrich	SML0273
NSC 23766	Selleckchem	S8031
CellROX™ Green Reagent	Invitrogen	C10444
H2DCF-DA	Invitrogen	C6827
Critical Commercial Assays
CometAssay® single cell gel electrophoresiskit	Trevigen	4250-050-K
OxyBlot Protein Oxidation Detection Kit	Millipore	S7150
Annexin V-Cy5 kit	BioVision	K103-100
X-tremeGENE siRNA Transfection Reagent	Roche	04476093001
QuickChange Mutagenesis Kit	Stratagene	200519
Experimental Models: Cell Lines
MGH 119-1, 121-1, 134-1	Crystal et al., 2014	N/A
PC9, GR7	Sharma et al., 2010	N/A
PC9 B, C	Hata et al., 2016	N/A
HCC827, GR6	Engelman et al., 2007	N/A
HCC4006, GR6	Soucheray et al., 2015	N/A
H3255, GR	Engelman et al., 2006	N/A
Experimental Models: Organisms/Strains
Female athymic nude (Nu/Nu) mice	Charles River	N/A
Oligonucleotides
siRNA against PARP-1, validated	Santa Cruz	sc-29437
siRNA against XRCC1, validated	Ambion	121593
siRNA against RAC1, validated	Sigma Aldrich	EHU075591
Software and Algorithms
TriTek CometScore	Tritek Corp	N/A
GraphPad Prism 6	GraphPad Software	N/A