Proteomic Profiling Identifies Dysregulated Pathways in Small Cell Lung Cancer and Novel Therapeutic Targets Including PARP1

Small cell lung cancer (SCLC) accounts for 13% of lung cancers in the United States (1). Compared with the more common non–small cell lung cancer (NSCLC), SCLC is characterized by more aggressive behavior with a faster doubling time, higher growth fraction, and more rapid development of metastasis. These differences in clinical behavior are also reflected in distinct responses to treatment. Compared with NSCLC, SCLC is more responsive to chemotherapy and radiation initially but relapses quickly with treatment-resistant disease. As a result, outcomes remain dismal, with a 5-year survival rate of less than 10% (1).

Despite low overall response rates to standard chemotherapy, subsets of patients with NSCLCs with EGF receptor (EGFR) mutations or EML4–ALK fusions are highly responsive to targeted therapies (2–6). In SCLCs, genomic aberrations have been identified, including Rb loss (7, 8), c-Kit overexpression (9, 10), telomerase activation (11), c-Myc amplification (12), and p53 mutation (13–15). However, attempts to target these clinically have had limited success to date. Improved characterization of differences in signaling pathways between SCLCs and NSCLCs could identify novel therapeutic targets for SCLCs.

Previous gene expression studies have shown marked differences in mRNA profiles between SCLCs and NSCLCs (16–19). In the current study, however, we have conducted an integrative analysis to systematically assess the activation of critical intracellular signaling pathways and potential therapeutic targets using reverse-phase protein arrays (RPPA) and other approaches. Unlike gene expression profiling, RPPA enables high-throughput, quantitative assessment of both total and post-translationally modified proteins. Because most drugs act on protein effectors, proteomic profiling may be better able to identify targets that could be directly modulated by emerging or currently available therapeutics.

Here, we assess the expression of 193 total and phospho-proteins in 34 SCLC and 74 NSCLC cell lines to identify proteins and pathways differentially regulated in SCLCs and NSCLCs. This study represents the most comprehensive protein profiling of SCLCs to date, both in terms of number of cell lines and number of pathway proteins assessed. Among the proteins overexpressed in SCLCs, PARP1 was selected for further investigation on the basis of its potential as a therapeutic target. We analyzed PARP1 mRNA and protein expression levels in patient tumors and tested the effect of a PARP inhibitor, alone and in combination with chemotherapy, in cell lines.

A panel of 34 SCLC and 74 NSCLC cell lines was profiled by RPPA to identify differences in key oncogenic proteins and pathways. Protein targets analyzed included several tyrosine kinases, downstream pathways such as the PI3K/Akt/mTOR, Ras/Raf/MEK, LKB1/AMPK, and JAK/STAT pathways, as well as proteins involved in apoptosis, DNA repair, and epithelial–mesenchymal transition. The SCLC panel included cell lines with RB1, PTEN, and TP53 deletions and/or mutations (Supplementary Table S1). The NSCLCs included several histologic subtypes, including adenocarcinoma and squamous lines, as well as EGFR- and KRAS-mutated lines (Supplementary Table S2). To control for the possible effect of culture conditions on protein expression, protein lysates were collected from each cell line under 3 media conditions: 10% serum for 24 hours, 0% serum for 24 hours, and serum stimulation for 30 minutes before cell lysis.

Unsupervised hierarchical clustering of the cell lines based on their overall expression of 193 total and phosphoproteins separated SCLCs from NSCLCs at the first major branch of the cluster dendrogram (Fig. 1A). Similarly, first principal component analysis separated SCLCs and NSCLCs on the basis of their distinct protein profiles (Fig. 1B). An ANOVA was conducted to identify the proteins most differentially expressed between SCLCs and NSCLCs. Mean expression levels of 55 proteins differed by ≥1.5-fold between SCLC and NSCLC lines, independent of media condition, at a false discovery rate of ≤1% (P < 0.038; Fig. 1C and D). Notably, a large cell neuroendocrine carcinoma (LCNEC) cell line (H1155) and a large cell lung cancer cell line (H1770) were grouped with SCLCs on the basis of their similar protein profiles.

Several proteins known to be dysregulated in SCLCs were also assessed. Consistent with previous studies, we found higher c-Kit expression (9.67-fold higher mean expression in SCLC vs. NSCLC), Bcl-2 (4.03-fold), and stathmin (3.18-fold) in SCLC (P < 0.0001 for all, P values for fold change calculated from the t statistic). Similarly, we observed relatively lower levels of total and phospho-Rb (−2.55 and −2.64-fold relative expression, respectively, P < 0.0001 for both) in SCLCs, as compared with NSCLC lines, and relatively higher E2F1 expression (2.06-fold higher in SCLC, P < 0.0001). Although the highest total c-Myc protein expression across all cell lines was in a c-Myc–amplified SCLC line, mean total c-Myc was higher in NSCLCs whereas phospho-c-Myc (T58; associated with c-Myc degradation) was 1.35-fold higher in SCLCs.

Presumably, as a result of Rb loss and subsequent release of E2F1 repression, expression of several E2F1 targets was significantly higher in SCLCs than in NSCLCs, including several not previously described, such as thymidylate synthase (1.45-fold, P < 0.0001), enhancer of zeste homolog 2 (EZH2) (1.50-fold, P < 0.0001), and several DNA repair and apoptosis proteins. Notably, mean levels of total PARP1 (a DNA repair protein and E2F1 co-activator) were 2.06-fold higher in SCLC cell lines than in NSCLC cell lines (with a corresponding P < 0.0001 by t test). RPPA results for total PARP1 protein were confirmed by Western blotting in a subset of SCLC and NSCLC cell lines (Supplementary Fig. S1). Other DNA repair proteins more highly expressed in SCLCs included Chk2 (1.51-fold higher), ATM (1.59-fold), DNA PKcs (1.69-fold), proliferating cell nuclear antigen (PCNA; 1.56-fold), and 53BP1 (1.99-fold; P < 0.0001 for all; Tables 1 and 2; Supplementary Table S3). In addition to Bcl-2, apoptotic markers higher in SCLCs than in NSCLCs included cleaved PARP (4.24-fold), BIM (2.57-fold), and BAX (1.64-fold; P < 0.0001 for all). Of note, although both cleaved and total PARP1 were higher in SCLCs, there was no significant difference in the ratio of cleaved to total PARP between SCLCs and NSCLCs (P > 0.3).

In contrast to SCLCs, NSCLCs had higher total/phospho-EGFR (both were 1.7-fold higher in NSCLCs), as well as higher levels of other receptor tyrosine kinases (RTK) that heterodimerize with EGFR, including p-Her2 (1.5-fold), total/phospho–c-Met (2.5- and 3.4-fold), and total/phospho-Axl (1.8- and 1.3-fold; P ≤ 0.02, computed from t statistic). Proteins in pathways downstream of EGFR/RTK signaling were also expressed at lower levels in SCLCs, including the PI3K/Akt/mTOR, Ras/Raf/MEK, and JAK/Src/STAT pathways (Fig. 1D).

In NSCLC lines, we observed elevated expression of PI3K/Akt/mTOR pathway proteins, including p-Akt (1.5-fold higher in NSCLCs), and its downstream targets, phospho-p70S6K (1.3-fold), phospho-S6 (S240/242; 2.7-fold), and phospho-S6 (S235/236; 3.17-fold; P ≤ 0.0005). In contrast, SCLCs showed greater expression of targets normally inhibited by Akt (e.g., GSK3, AMPK, p21, and apoptosis proteins) further suggesting decreased Akt pathway activity in SCLCs (Fig. 1D). Similarly, activation of the AMPK pathway, a negative regulator of mTOR, was seen more in SCLCs than in NSCLCs, with higher levels of p-AMPK (1.4-fold), LKB1 (1.4-fold), and pTSC2 (1.2-fold; P ≤ 0.004). Other proteins with higher expression in NSCLCs than in SCLCs included those in the Wnt/Hedgehog/Notch pathway (e.g., E-cadherin, β-catenin, Notch3) and epithelial tumor markers (e.g., fibronectin).

We then compared the mRNA levels for genes corresponding to the total proteins identified by our analysis as differentially expressed in SCLCs versus NSCLCs. As expected, hierarchical clustering separated SCLC cell lines from NSCLC cell lines on the basis of differential mRNA expression of these genes (Fig. 2A). Among the DNA repair proteins, PARP1 had the greatest differential mRNA expression between SCLCs and NSCLCs (P < 0.0001 by t test; Fig. 2B). Other potentially druggable targets identified by RPPA that were more highly expressed at the mRNA level in SCLC included EZH2, Bcl-2, PRKDC (DNA PKcs), and PCNA (Fig. 2B, Supplementary Table S4). Using publicly available data, we then analyzed PARP1 expression across a panel of 318 cell lines from 30 cancer types (Fig. 2C; ref. 20). Remarkably, SCLC cells showed the highest median PARP1 expression of any solid tumor cells. Moreover, among individual cell lines, an SCLC cell line had the highest PARP1 expression of all solid tumor lines, including breast and ovarian cells.

Finally, we compared mRNA levels in patient tumors. Despite a limited number of SCLC tumor profiles available, 20 of the genes tested were expressed at significantly different levels between SCLCs and NSCLCs (P < 0.05 by t test), 9 of which were significantly different at P < 0.001 (Supplementary Table S5). Consistent with the cell line data, PARP1 mRNA was significantly higher in SCLC tumors compared with NSCLC tumors (P = 0.005) or normal lung (P ≤ 0.001; 2-sided t test), as were EZH2, Bcl-2, PRKDC, and PCNA (Fig. 2D).

Among proteins with elevated expression in SCLCs, several are potential drug targets including PARP1, EZH2, Chk1/2, DNA PKcs, and PCNA. Among these, we further investigated PARP1 because it was expressed at the highest relative levels among the DNA repair proteins and because PARP1 plays an independent role as an E2F1 co-activator (21, 22), suggesting that its inhibition may have a dual role, with direct effects on DNA repair and on other E2F1-regulated DNA repair proteins. Clinical trials with PARP inhibitors in breast and ovarian cancer have shown promise, particularly in patients with underlying defects in DNA repair or with platinum-sensitive tumors (23, 24). Because most SCLC tumors are highly sensitive to platinum-based treatment, PARP inhibitors may, therefore, be active in SCLCs. We also tested the effect of siRNA targeting of EZH2, a second potential therapeutic target with drugs currently being developed for cancer treatment.

To confirm elevated PARP1 protein expression in tumors from patients with SCLCs, we used immunohistochemical (IHC) analysis to measure total PARP1 in tissue microarrays of SCLCs and other neuroendocrine tumors (LCNEC, atypical carcinoid, typical carcinoid) and compared them with adenocarcinoma and squamous NSCLC tumors. Staining was scored on the basis of the percentage of cells staining positive (0%–100%) times the staining intensity (0–3+), for a total possible score of 300. In neuroendocrine lung tumors, total PARP1 protein levels correlated with the degree of differentiation. The highest levels were seen in SCLC (n = 12, mean IHC score = 262) and LCNEC tumors (n = 20, mean IHC score = 237). Intermediate levels were seen in atypical carcinoid (n = 9, mean IHC score = 230) and typical carcinoid tumors (n = 55, mean IHC score = 197; Fig. 3A and B). In contrast, PARP1 expression was significantlylower in NSCLCs with squamous (n = 15, mean IHC score = 120) and adenocarcinoma histologies (n = 24, mean IHC score = 104). PARP1 IHC expression was significantly higher in SCLCs than in squamous carcinoma (P = 2.3 × 10⁻⁴) or adenocarcinoma (P = 3.2 × 10⁻⁶ by ANOVA) but was not different between SCLCs and other neuroendocrine lung tumors (P = 0.11–0.94). There was no correlation between total PARP1 expression and Ki67 expression in SCLC or LCNEC tumors (P = 0.50 and 0.82, respectively, by Spearman rank correlation), suggesting that increased PARP1 is not a surrogate marker of increased proliferation.

We then tested the effect of PARP inhibition with AZD2281 in vitro. To confirm the inhibition of PARP1 activity by AZD2281, we treated SCLC cell lines H69, H82, and H524, the neuroendocrine lung cancer cell line H1155, and the NSCLC cell line A549 with AZD2281 for 24 hours and then measured poly [ADP-ribose (PAR)] levels by ELISA. In all cell lines tested, AZD2281 significantly reduced PAR levels in a dose-dependent manner, indicating inhibition of PARP1 activity (Fig. 4A). Because A549 is resistant to AZD2281 (as described below), these results suggest that target inhibition alone is not sufficient for cell line sensitivity.

In 35 lung cancer cell lines treated with increasing concentrations of AZD2281, we observed the greatest drug sensitivity in SCLC cell lines, with IC₅₀ values <2 μmol/L for H82 and H69 and <5 μmol/L for H524, H526, and H2107 (Fig. 4B). H1155, a LCNEC cell line with a protein signature similar to SCLCs, was moderately sensitive to AZD2281 in the 5-day assay with an IC₅₀ value of 5.6 μmol/L. In contrast, the majority of NSCLC cell lines had IC₅₀ values >60 μmol/L. Interestingly, the one SCLC line that was relatively more resistant to AZD2281, H841, had an NSCLC-like protein expression pattern and clustered in the middle of the NSCLC lines in Fig. 1.

We further evaluated the effect of AZD2281 as well as an additional PARP inhibitor, AG014699, on in vitro growth in a subset of cell lines after 14 days of treatment. Consistent with the results described above, SCLC cell lines were highly sensitive to 14-day PARP inhibition by AZD2281 with IC₅₀ values of ≤2 μmol/L in all SCLC lines except H841 (Fig. 4C). Similar to H1155 in the 5-day study, another LCNEC cell line (H1299) showed intermediate sensitivity with an IC₅₀ of 3.7 μmol/L. SCLCs and LCNECs were also highly sensitive to 14 days of treatment with AG014699, a highly specific PARP1 inhibitor (Fig. 4C). Consistent with the AZD2281 data, SCLC cell lines were highly sensitive to AG014699 (IC₅₀ <0.5 μmol/L for H82, H69, and H524 and 2.2 μmol/L for H526 and H841) and the NSCLC cell line A549 was resistant (8.6 μmol/L). IC₅₀ values are listed in Supplementary Table S6.

Because BRCA1/2 mutations and PTEN loss are associated with greater sensitivity to PARP inhibition in breast and ovarian cancer, we also tested the sensitivity of a BRCA1-mutated breast cancer cell line (HCC1395) and a PTEN-mutant breast cancer line (MDA-MB-468) as positive controls. Although the breast cancer lines were sensitive to both PARP inhibitors, SCLC cell lines were as sensitive or more sensitive in comparison (Fig. 4C). Because drug inhibitors may inhibit more than one target and because our analysis indicated that EZH2 and CHK1 may also be useful targets in the treatment of SCLCs, we targeted the expression of these proteins by siRNA as well. For PARP1, knockdown was confirmed by Western blotting as shown in Fig. 4D. Three independent siRNAs directed against PARP1 inhibited the proliferation of H69 (SCLC) cells. In contrast, there was no change in cell proliferation when we treated A549 (NSCLC) cells (PARP inhibitor–resistant) with multiple siRNAs targeting PARP1. In H69, knockdown of EZH2 also decreased cell growth compared with controls (mock transfected or scrambled siRNA; Fig. 4D). However, we did not observe an effect with CHK1 siRNA (data not shown).

Because NSCLC tumors expressed a range of PARP1 levels (Fig. 3A) and were higher than normal lung (Fig. 2D), we investigated whether PARP1 protein levels correlated with sensitivity to PARP inhibition. IC₅₀ values for AZD2281 in SCLC and NSCLC cell lines (including high-grade neuroendocrine) were correlated with protein expression levels by Spearman correlation. Higher PARP1 levels correlated with significantly greater sensitivity to AZD2281 (lower IC₅₀s; r = −0.48, P = 0.006). Other proteins that correlated with AZD2281 sensitivity included E2F1 (r = −0.35, P = 0.046) and several E2F1 targets, including EZH2 (r = −0.65, P < 0.001), pChk1 (r = −0.59, P < 0.001), and ATM (r = −0.52, P < 0.001; Supplementary Fig. S2).

The sensitivity of SCLCs to PARP inhibition suggests that there may be defects of DNA repair, particularly for double-strand breaks. To evaluate this further, we assessed radiation-induced RAD51 foci formation in A549, H69, and H82 cells using immunofluorescence staining. Our results show that in a PARP-resistant NSCLC cell line (A549), the percentage of cells expressing RAD51 foci increases after radiation, peaking at 6 hours, suggesting induction of DNA damage. However, at 18 hours after radiation and beyond, the damage is repaired, as reflected by reduction in RAD51 foci formation to baseline levels (Fig. 5A). For SCLCs (H69 and H82), however, RAD51 foci levels remained elevated at 24 hours (and were higher than unirradiated control samples). These results suggest that SCLCs may have a deficiency in DNA repair which could contribute to their sensitivity to PARP inhibitors (25).

We then conducted RPPA on SCLC cell lines following treatment with AZD2281 or AG014699 to investigate protein modulation following PARP inhibition (Fig. 5B and C). These data show a time-dependent downregulation of RAD51 (P < 0.001) and other DNA repair proteins after PARP inhibition that was maximally apparent at 14 days. These observations support our hypothesis that inhibition of PARP in SCLCs may suppress E2F1-mediated expression of several DNA repair proteins (due to its role as an E2F1 co-activator) which, in turn, may contribute to further DNA repair deficiency and account for the sensitivity of SCLC to these drugs.

AZD2281 leads to double-strand DNA breaks, as do the chemotherapeutics cisplatin and etoposide, the standard frontline agents for SCLC treatment. We therefore evaluated the effects of these agents in combination. H82 cells were treated with 1 μmol/L AZD2281 for 7 days. Cisplatin and etoposide were added, and the cells were incubated for an additional 7 days and counted. Treatment with either chemotherapeutic or with AZD2281 alone reduced the cell count by approximately 60% compared with control cells (P < 0.05; Supplementary Fig. S3). The cell count after treatment with chemotherapy plus AZD2281 was approximately 80% lower than in control cells and was significantly lower than the cell count after treatment with AZD2281 alone (P < 0.05 for both). We observed a similar trend in H69 with the combination of AZD2281 and cisplatin/etoposide, although this did not reach statistical significance (Supplementary Fig. S3). Treatment of H69 cells with AZD2281 in combination with irinotecan, another chemotherapeutic commonly used in the treatment of SCLCs, also resulted in a greater decrease in tumor cell viability than either agent alone (P = 0.03 AZD2281/irinotecan vs. control; P = 0.007 AZD2281/irinotecan vs. irinotecan alone).

In this study, we used proteomic and gene expression profiling to identify pathways dysregulated in SCLCs. We discovered higher expression of several E2F1-regulated proteins (e.g., EZH2, DNA repair, and apoptosis proteins) and PARP1, an E2F1 co-activator (21, 22). Conversely, our study revealed that EGFR, its associated RTKs (e.g., Her2, c-Met, Axl), and downstream targets in the PI3K/Akt/mTOR pathway and the RAS/Raf/MEK pathway were expressed at lower levels in SCLCs. Our analysis also detected known abnormalities in SCLCs, such as c-Kit and Bcl-2 overexpression and Rb loss.

Loss of the RB1 gene is a hallmark of SCLCs. Consistent with this, in our study, total and phospho-Rb protein expression was reduced or absent in SCLCs. Given that Rb is known to inhibit the transcription factor E2F1, this loss of Rb activity is a likely explanation for the observed increases in protein expression of E2F1 and E2F1 targets in SCLCs. To our knowledge, this study represents that first time several of these E2F1 targets have been shown to be highly expressed in SCLCs. These findings have potentially important clinical implications because of their role in drug resistance or as druggable targets. For example, thymidylate synthase has been implicated in pemetrexed resistance in patients with NSCLCs (26–28). Its high expression may account for the inferior outcomes in patients with SCLCs treated with pemetrexed plus carboplatin, compared with etoposide with carboplatin in a phase III clinical trial (29). EZH2 and Chk1 are other E2F1 targets highly expressed in SCLC cells (30, 31) and are both being explored as therapeutic targets in other malignancies (32, 33). Our data suggest they merit further investigation in SCLCs as well.

Of the potential therapeutic targets identified, we have first investigated PARP1 for 2 reasons. First, several PARP inhibitors are in advanced stages of clinical development for other tumor types. In breast and ovarian cancer clinical trials, data suggest that PARP inhibitors have increased activity in platinum-sensitive tumors, making PARP an attractive candidate for SCLCs, which are highly platinum sensitive (24). Second, PARP1 acts as an E2F1 co-activator, hence PARP inhibition could act by either directly blocking the repair of double-strand DNA breaks or by inhibiting the expression of E2F1-regulated DNA repair proteins, which would further impair DNA repair and potentially enhance the efficacy of therapies inducing double-strand breaks. We confirmed high PARP1 expression at the mRNA level in SCLCs and at the protein level in a tissue microarray of neuroendocrine tumors. High PARP1 expression was also detected in patient tumors with neuroendocrine histologies (SCLCs, LCNECs), whereas moderately high expression was seen in differentiated neuroendocrine tumors.

Previous studies suggest that PARP inhibitors are synergistic with radiation therapy or DNA-damaging drugs, such as topoisomerase inhibitors (34, 35). Therefore, we tested the effect of PARP inhibition alone and in combination with cisplatin and etoposide or irinotecan. We found that AZD2281 and AG014699 had single-agent activity in SCLCs and LCNECs, but not in most non-neuroendocrine NSCLC cell lines, and that PARP1 levels correlated with PARP inhibitor sensitivity. Strikingly, SCLC cells were as sensitive or more sensitive than 2 breast cancer cell lines tested that had BRCA1 or PTEN mutations. When combined with chemotherapy, AZD2281 further reduced SCLC viability relative to treatment with either single agent alone.

Consistent with our hypothesis that PARP inhibitor sensitivity is mediated, in part, through its effect on E2F1, when we measured protein expression after PARP inhibitor treatment, we observed decreased expression of multiple E2F1 targets (RAD51, PCNA, and others). Although there was a trend toward decreased BRCA1 levels after PARP inhibition, this was less significant than the modulation of other DNA repair proteins, suggesting that the mechanism of PARP inhibition is not dependent on BRCA1 specifically but may lead to a BRCA-like phenotype by decreasing expression of multiple DNA repair proteins. The abnormal RAD51 foci formation also suggests that SCLCs may have some underlying defect in homologous recombination at baseline and warrants further investigation. Interestingly, Garnett and colleagues have recently reported sensitivity of Ewing sarcoma to PARP inhibition in vitro, suggesting that there may be multiple mechanisms through which PARP inhibitors may be effective (36).

Our analysis also suggests that EGFR, Her2, c-Met, and other cell surface RTKs are present at lower levels in SCLCs, accompanied by decreased activation of downstream signaling pathways (PI3K/Akt and Ras/Raf/Mek). Consistent with these observations, inhibitors of EGFR and mTOR have not shown significant clinical activity in SCLCs (37, 38). These data suggest that approaches targeting PI3K/Akt/mTOR and Ras/Raf/MEK pathways may be more effective for NSCLCs than for SCLCs, although we cannot rule out the possibility that these pathways may be activated in subsets of SCLCs.

This study allowed us to leverage the proteomic differences between SCLCs and NSCLCs to elicit the biology underlying the distinct clinical behavior of SCLCs. However, because we directly compared protein expression between SCLCs and NSCLCs, we may have missed important pathways or targets that are highly expressed in both cancer types. Future studies comparing SCLCs with normal lung or a larger panel of tumor types may unravel additional pathways active in SCLCs.

In conclusion, we applied proteomic profiling of lung cancer lines to identify important differences in signaling pathways that differentiate SCLCs from NSCLCs. We identified several possible therapeutic targets regulated by E2F1, including PARP1, suggesting this pathway may be critical for SCLCs. Preclinical testing confirmed the sensitivity of SCLCs to a PARP1 inhibitor, supporting it as a potential target in SCLCs. Clinical studies evaluating the combination of PARP1 inhibition with chemotherapy and other agents in SCLCs merit further investigation and are currently in development.

Thirty-four SCLC, 74 NSCLC, and 2 breast cancer cell lines were provided by Dr. Minna (The University of Texas Southwestern, Dallas, TX) or obtained from American Type Culture Collection (ATCC). Cells were grown in RPMI unless otherwise specified by ATCC (Supplementary Tables S1 and S2). DNA fingerprinting was used to confirm the identity of each cell line at the time of total protein lysate preparation, as described in Supplementary Information.

Protein lysate preparation from subconfluent cultures, RPPA printing, and data analysis were conducted as detailed in Supplementary Information.

NSCLC and SCLC cell line microarray results were previously published and archived at the Gene Expression Omnibus repository (GEO accession GSE4824; refs. 39–42). To compare the cell lines of different tumor types, we analyzed the gene expression data of 318 cell lines arrayed by GlaxoSmithKline on Affymetrix Human Genome U133 Plus 2 arrays (Affymetrix). We downloaded array data (.CEL files) from ArrayExpress (20, 43) and used quantile normalization and a robust multi-array average algorithm to process the raw data for all 950 unique samples (30 Gb) in a single run. Cell lines were rank-ordered by their PARP1 mRNA expression.

Gene expression profiles from the International Genomics Consortium were used to assess PARP1 in 30 solid tumors (44). Profiles of 2,156 tumors arrayed on the Human Genome U133 Plus 2 platform (Affymetrix; .CEL files) were downloaded from the GEO (GSE2109; ref. 39). Raw data of 2,156 samples (65 Gb) were processed in a single run using quantile normalization and a robust multi-array average algorithm.

A tissue microarray was constructed from patients with lung cancer and IHC analysis was conducted for PARP1. PARP1 activity was evaluated using a PAR assay, and cell growth for the PARP inhibitors AZD2281 and AGO14699 was tested in an MTS assay. For siRNA, cells were transfected with control siRNA or siRNAs targeting PARP, EZH2, and CHK1 and then plated for cell growth assays with viability measured at days 1 and 5. Details about these methods are described in Supplementary Information.

G.B. Mills has received commercial research grants from AstraZeneca, Celgene, CeMines, Exelixis, GlaxoSmithKline, LPATH, Roche, SDI, and Wyeth/Pfizer; has ownership interest (including patents) in Catena and PTV Ventures; and is a consultant/advisory board member for Asuragen, Aushon, Catena, Daiichi, Foundation Medicine, and the Susan G. Komen for the Cure. J.D. Minna has received a commercial research grant from Geron Pharmaceuticals and is a consultant/advisory board member for Amgen. J.V. Heymach has received a commercial research grant from AstraZeneca and serves on their advisory board. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L.A. Byers, B.S. Glisson, J.D. Minna, J.V. Heymach

Development of methodology: L.A. Byers, J. Fujimoto, U. Giri, M. Peyton, B. Duchemann, J.D. Minna, J.V. Heymach

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.A. Byers, J. Wang, J. Fujimoto, J. Yordy, U. Giri, M. Peyton, Y. H. Fan, F. Masrorpour, B. Duchemann, P. Tumula, V. Bhardwaj, N. Kalhor, I.I. Wistuba, L. Girard, J.D. Minna, J.V. Heymach

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.A. Byers, J. Wang, P. Saintigny, M. Peyton, L. Diao, L. Shen, W. Liu, J. Welsh, S. Weber, I.I. Wistuba, L. Girard, K.R. Coombes, J.N. Weinstein, J.D. Minna, J.V. Heymach

Writing, review, and/or revision of the manuscript: L.A. Byers, M.B. Nilsson, J. Fujimoto, P. Saintigny, J. Yordy, M. Peyton, B.S. Glisson, I.I. Wistuba, L. Girard, S.M. Lippman, G.B. Mills, K.R. Coombes, J.N. Weinstein, J.D. Minna, J.V. Heymach

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.B. Nilsson, J. Fujimoto, U. Giri, V. Bhardwaj, J.V. Heymach

Study supervision: L.A. Byers, J.D. Minna, J.V. Heymach

The authors thank Ana M. Gonzalez-Angulo, MD, for her advice and input and Emily Brantley, PhD, for editorial assistance.

This work was supported by The University of Texas Southwestern Medical Center and The University of Texas MD Anderson Cancer Center Lung SPORE grant 5 P50 CA070907; DoD PROSPECT grant W81XWH-07-1-0306; Uniting Against Lung Cancer Grant; AACR-AstraZeneca-Prevent Cancer Foundation Fellowship for Translational Lung Cancer Research; MD Anderson Cancer Center Physician Scientist Award; Barbara Rattay Advanced Fellowship Program, CCSG grant 5 P30 CA016672; Chapman Fund for Bioinformatics in Personalized Cancer Therapy, 1 U24 CA143883; and the E.L. Wiegand Foundation.