Proteasome Inhibitors Block DNA Repair and Radiosensitize Non-Small Cell Lung Cancer

Despite optimal radiation therapy (RT), chemotherapy and/or surgery, a majority of patients with locally advanced non-small cell lung cancer (NSCLC) fail treatment. To identify novel gene targets for improved tumor control, we performed whole genome RNAi screens to identify knockdowns that most reproducibly increase NSCLC cytotoxicity. These screens identified several proteasome subunits among top hits, including the topmost hit PSMA1, a component of the core 20 S proteasome. Radiation and proteasome inhibition showed synergistic effects. Proteasome inhibition resulted in an 80–90% decrease in homologous recombination (HR), a 50% decrease in expression of NF-κB-inducible HR genes BRCA1 and FANCD2, and a reduction of BRCA1, FANCD2 and RAD51 ionizing radiation-induced foci. IκBα RNAi knockdown rescued NSCLC radioresistance. Irradiation of mice with NCI-H460 xenografts after inducible PSMA1 shRNA knockdown markedly increased murine survival compared to either treatment alone. Proteasome inhibition is a promising strategy for NSCLC radiosensitization via inhibition of NF-κB-mediated expression of Fanconi Anemia/HR DNA repair genes.


Introduction
Radiation therapy (RT) is a critical modality in the treatment of lung cancer. It is highly effective when disease is localized, and curative doses can be safely administered. For example, Stage I non-small cell lung carcinomas (NSCLC) can be treated with sufficiently high doses of ionizing radiation (IR) to yield 3-year local control rates around 90% [1]. This is largely due to the high biologically effective doses (BED) that can be administered to isolated lung tumors using stereotactic body radiotherapy. Unfortunately, only 15% of patients present with such tumors [2]. More advanced tumors, which are significantly more common, cannot be treated to higher BED, due to infiltration of, or proximity to, radiosensitive structures including the lungs, spinal cord, esophagus and heart [3]. This limitation accounts for the higher local failure rates of around 30% [4]. Systemic agents are therefore administered to enhance the radiation response. Such agents include cytotoxic chemotherapeutics which themselves have significant dose-limiting toxicities and limited effects on tumor control. Only a minority of tumors exhibit genetic features, such as EGFR activating mutations [5] or EML4-ALK translocations [6], that allow for targeted therapies; additional options are needed.
Several genome-wide studies have been completed in NSCLC, including DNA sequencing [7], copy number analysis [8] and gene expression profiling [9]. These studies provided an unbiased identification of the most frequent and important genetic and epigenetic alterations among the 20-25,000 genes of the human genome [10]. They revealed intriguing associations among genes and clinical covariates, but were unable to directly demonstrate cause-and-effect relationships between alterations and therapeutic outcomes. In contrast, RNA interference (RNAi) can directly demonstrate the effects of reduced gene expression on cell physiology and survival [11].
Pooled short hairpin RNA (shRNA) screens, in which cancer cells are exposed to several thousand different shRNA sequences, averaging one gene knockdown per cell, have several compelling features. First, the effect of stable shRNA knockdown over several cell doublings can be explored, compared to transient transfection of small interfering RNA (siRNA) sequences with short-lived effects. Accordingly, shRNA expression mimics drug treatments that are typically given over several weeks rather than days. Also, cells harboring different shRNA sequences effectively compete with each other within the pool as they proliferate, giving rise to hits that yield more pronounced effects on proliferation [12].
Because radiotherapy is the mainstay of NSCLC treatment, gene silencings that result in synergistic cytotoxicity when combined with ionizing radiation (IR) are desirable. Many cancer treatments are additive in nature, and are combined because they result in differential side effect profiles [13]. Synergistic treatments that result in greater effects when administered concurrently, compared to the additive effects of each treatment given individually, may serve particularly well as radiosensitizers, since RT delivery may be constrained to a limited volume, and side effects may be less pronounced outside the irradiated volume.
Demonstrating the mechanism of a radiosensitizing shRNA may also reveal biomarkers for patient selection and treatment assessment. DNA double strand breaks (DSBs) are among the effects of IR that best correlate with its cytotoxicity. A single unrepaired DSB is sufficient to result in reproductive cell death via G2 arrest or mitotic catastrophe [14]. DSBs are predominately repaired through two pathways, homologous recombination (HR) and non-homologous end-joining (NHEJ) [15]. Gene silencings or small molecule inhibitors of either pathway may promote radiosensitization.
Here we examine proteasome inhibition as a strategy for NSCLC radiosensitization via inhibition of DNA DSB repair. Proteasome inhibition has been explored in multiple clinical trials enrolling NSCLC patients, with variable results. For example, a Phase II study of 114 patients treated with bortezomib plus gemcitabine and carboplatin as first-line treatment of advanced NSCLC, showed a response rate of 23% and a disease control rate (responses+stable disease) of 68%, thus warranting further studies [16]. Bortezomib showed no activity as monotherapy in a Phase II study of 14 patients, none of whom showed objective responses, and three (21%) had stable disease lasting 3.4-11.5 months [17]. Based on our data, we propose that proteasome inhibitors may be useful as radiosensitizers, given their repressive effects on NF-kBmediated expression of genes required for the HR pathway.

Multiple Proteasome Genes are Top Hits in NSCLC whole Genome RNAi Screens
A whole genome shRNA screen was conducted in two NSCLC cell lines, A549 and NCI-H460. These lines share common genetic features, including mutations in KRAS and STK11 (aka LKB1). These mutations are found in tumors that are particularly aggressive and resistant to therapy, and for which targeted therapies are unavailable [18,19]. The cell lines are wild-type in TP53 and therefore less likely, compared to mutated lines, to exhibit genomic instability over the course of a screen spanning multiple weeks [20]. Also, their genetic similarity, in terms of key oncogenic and tumor suppressor genes, increases the probability that screen hits in one line will be reproduced in the other (Table  S1).
The Hannon-Elledge library contains 74,705 distinct shRNA sequences and targets nearly 18,000 genes [21]. After transduction with this library and puromycin selection for stable integrants, cells were passaged, and the relative representation of each sequence within a pool was determined before and after twelve population doublings. ShRNA sequences directed against genes essential for cellular proliferation were selectively lost. 1,667 genes were the targets of at least one shRNA sequence whose abundance decreased by at least two-fold during passage in both cell lines (Table S2). Multiple proteasome subunits numbered among the hits in both cell lines, including the top hit PSMA1, a subunit of the core 20 S proteasome [22] (Fig. 1 and Table S3).

Proteasome Inhibitors Sensitize NSCLC Cells to Radiation
The doxycycline-inducible shRNA knockdown of PSMA1 in A549 and NCI-H460 resulted in loss of protein expression of both PSMA1 and PSMB5, another subunit of the core 20 S proteasome and the catalytic site of chymotrypsin-like (CTL) activity of the proteasome [22] (Fig. 2A). This result confirms the essential role of PSMA1 in 20 S proteasome assembly. Treatment with the small molecule proteasome inhibitor bortezomib or PSMA1 shRNA knockdown caused a loss of CTL activity (Fig. 2B).
Bortezomib is an active agent in vitro in NSCLC lines, including A549 [23] and NCI-H460 [24], and it has previously demonstrated radiosensitizing properties in preclinical studies [25]. The ability of proteasome inhibitors to augment the effects of fractionated radiation, a clinical method of delivering daily radiation in small doses to minimize long-term toxicity [26], was therefore explored. Bortezomib and fractionated radiation yielded synergistic effects (Fig. 2C). Synergy was also observed with singledose radiation (Fig. S1), although the result with fractionated radiation is more clinically relevant. This synergistic effect on cell death was mediated at least in part by apoptosis; 2 Gy IR alone did not induce apoptosis, while the combination of IR and bortezomib induced significantly increased apoptosis compared to either treatment alone (Fig. 2D).

Proteasome Inhibition Impairs Radiation-induced DNA Double Strand Break Repair by Decreasing Homologous Recombination
Given the synergy between proteasome inhibition and radiation, we next determined whether proteasome inhibitors affect radiation-induced DNA double strand breaks (DSBs) [26] ( Figure 3). In neutral comet assays, NSCLC cells repaired the vast majority of DNA DSBs generated by high-dose irradiation within one hour. In contrast, bortezomib treatment prior to radiation significantly delayed the repair of DNA DSBs at least up to 8 hours after irradiation (Fig. 3A-B). This persistence of DSBs may result in increased mitotic catastrophe [27], which is believed to be the predominant cause of cell death in irradiated solid malignancies.
To observe the effects of bortezomib or RNAi knockdown of PSMA1 on DNA repair, a GFP reporter construct for HR [28] was introduced into A549 and NCI-H460 cells. Compared to untreated cells, those treated with either bortezomib or PSMA1 siRNA showed a statistically significant decrease in HR-mediated repair of I-SceI-induced DNA DSBs (Fig. 3C). Similar results were observed for NHEJ (Fig. S2). These results demonstrate a direct effect of proteasome inhibition on the mechanisms of DNA DSB repair, consistent with previous studies [29][30][31].
To understand the mechanism of HR inhibition, we explored the effect of bortezomib on nuclear focus formation of key proteins in the Fanconi Anemia (FA)/HR pathway. IR-induced RAD51 nuclear focus formation, which is a biomarker for HR [32], was substantially decreased by bortezomib or PSMA1 RNAi (Fig. 4A,   Fig. S3-S4). FANCD2 protein expression and IR-induced FANCD2 focus formation were similarly decreased by bortezomib or PSMA1 RNAi (Fig. 4B, Fig. S3-S4). Finally, IR-induced BRCA1 focus formation was significantly decreased by bortezomib or PSMA1 RNAi (Fig. 4C, Fig. S3-S4). Taken together, these data suggest that proteasome inhibition may impact HR-mediated repair of IR-induced DNA DSBs by reducing availability or recruitment of key FA/HR proteins to damage sites.

Proteasome Inhibition Blocks NF-kB Induced Expression of Fanconi Anemia/Homologous Recombination Genes
We next explored the relationship between proteasome inhibition and the FA/HR pathway via NF-kB signaling. Dalton and colleagues [33] previously showed that bortezomib decreases FA/BRCA gene expression in multiple myeloma cells, and that NF-kB transcriptionally upregulates the FA/BRCA pathway. For instance, NF-kB binds directly to the promoter of FANCD2, based on the electrophoretic mobility shift assay (EMSA) (Fig. 5A). One mechanism by which bortezomib downregulates the NF-kB pathway is through inhibition of proteasome-mediated degradation of IkBa [34]. IkBa, which sequesters NF-kB in the cytoplasm, and phosphorylation of IkBa by the inhibitor of NF-kB kinases (IKKs) releases NF-kB, allowing its entry to the nucleus [35]. Bortezomib also downregulates the NF-kB pathway by inducing nuclear translocation of IkBa, resulting in suppression of NF-kB p65/50-dependent transcription [36].
We therefore sought to determine whether the IR-induced transcription of FA/HR genes FANCD2 and BRCA1 is diminished by proteasome inhibition. Transcription of FANCD2 and BRCA1 increased within 30 minutes of irradiation, and was substantially blocked by either bortezomib or PSMA1 RNAi ( Fig. 5B-C). This short time scale could account for the short-term effect on DSB repair observed in the comet assay. To establish causality, we performed a rescue experiment in NSCLC cells pretreated with IkBa siRNA. Cells were then subjected to proteasome inhibition and irradiation. IkBa knockdown rescued radioresistance and cell survival of bortezomib-treated cells (Fig. 5D). These data establish a role for proteasome inhibitors as radiosensitizers, via blockade of NF-kB induced expression of FA/HR genes (Fig. 6).

Combination of Proteasome Inhibition and Radiation Enhances Tumor Control
We next sought to determine whether proteasome inhibition may be a useful strategy in vivo. Poor penetration of bortezomib into tumors may limit its efficacy in solid malignancies [37]. We therefore tested whether inducible PSMA1 RNAi knockdown enhances control of xenografts treated with fractionated radiotherapy (RT).
To administer fractionated RT, we used a small animal radiation research platform (SARRP) that can deliver CT-guided conformal RT to tumors in mice [38,39]. This platform ensures complete treatment of the tumor while minimizing irradiation of uninvolved tissues. It also facilitated volumetric assessments of tumor growth and partially correlated (r = 0.61) with caliper measurements (Fig. 7A-B).
Mice were implanted with NCI-H460 cells harboring a doxycycline-inducible PSMA1 shRNA construct. One week after initiation of PSMA1 knockdown, tumors were irradiated to a dose of 20 Gy in five fractions. After RT treatments, PSMA1 knockdown was discontinued, and mice were followed for up to 100 days after tumor formation (Fig. 7C). Untreated mice showed rapid development of tumors. Mice treated with either RT or the inducible PSMA1 knockdown in tumor cells also showed progressive tumor growth and poor survival. Mice treated concurrently with both treatments however showed minimal to no xenograft growth and 100% survival (Fig. 7D-E). IR-induced FANCD2 focus formation was also reduced in tumor specimens showing inducible PSMA1 knockdown (Fig. 8A). The reduction in these foci may serve in the future as a biomarker for reduction of IR-induced activation of FA/HR pathway following proteasome inhibition. Significantly increased IR-induced c-H2AX foci in tumor specimens persisted at 1, 6, and 24 hours with PSMA1 knockdown compared to control, indicating delayed DNA DSB repair. (Fig. 8B-C).
The dramatic results observed with PSMA1 knockdown are in contrast with those observed previously with bortezomib in vivo, including genetically engineered mouse models (GEMMs) of lung adenocarcinoma [40]. Given the results with PSMA1 knockdown, the ability of bortezomib to radiosensitize NCI-H460 xenografts was examined to determine whether there may be greater tumoricidal effect with this combination therapy. There was no significant difference in survival of mice treated with radiotherapy with or without bortezomib ( Figure S5). To determine whether this was due to poor tumor drug penetration, chymotrypsin-like proteasome activity was examined in tumors treated with or without bortezomib compared to tumors with or without PSMA1 knockdown. There was substantially more proteasome inhibition following PSMA1 knockdown compared to that following bortezomib ( Figure S6). This suggests that the limited efficacy observed with bortezomib in vivo may be related to poor tumor drug penetration, which would explain the discrepancies in results in vitro and in vivo.

Discussion
The purpose of this study was to establish a rationale for proteasome inhibition as a strategy for NSCLC radiosensitization. We reasoned that since DNA double strand breaks (DSBs) correlate with the cytotoxicity of ionizing radiation (IR) [26], agents that delay the repair of these breaks may act as radiosensitizers. Accordingly, we observed decreases in homologous recombination (HR) following proteasome inhibition, significant delays in DNA DSB repair both in vitro and in vivo, and reduced levels of IR-induced expression and damage site localization of DNA repair proteins. Proteasome inhibitors function, at least in part, by blocking the NF-kB pathway by interfering with degradation of IkBa, which inhibits downstream NF-kB activity [34,35]. Bortezomib decreased the IR-induced expression following proteasome inhibition of Fanconi Anemia (FA)/BRCA genes, the majority of which have putative NF-kB binding sites [33]. Importantly, we rescued bortezomib-induced radiosensitivity by silencing IkBa, supporting the proposed mechanism. Proteasome inhibitors therefore appear to radiosensitize NSCLC by interfering with DNA DSB repair by reducing the expression of key NF-kB-inducible HR genes.
An alternate explanation for the effects of proteasome inhibition on radiosensitivity is through the induction of apoptosis. Multiple studies have demonstrated an increase in apoptosis following bortezomib treatment [41]. Apoptosis has not been shown as the predominant mode of cell death in irradiated NSCLC. For example, high doses such as 20 Gy irradiation of A549 or NCI-H460 cells yield rates of apoptosis of only 5-35% [42]. NCI-H460 cells treated with both 2 Gy IR and bortezomib showed increased apoptosis compared to either treatment alone. Of note, the . Proteasome inhibition impairs radiation-induced DNA double strand break repair by decreasing homologous recombination. (A) After bortezomib624 hours or PSMA1 knockdown672 hours, A549 or NCI-H460 cells were irradiated then fixed after 6 hours. RAD51 foci were detected by immunofluorescence. Cells with $5 foci were scored as positive (n .100). All results are mean 6 SEM. P values were calculated using a two-tailed Student's t test. Photos show representative images for NCI-H460 treated with bortezomib, bar = 10 mm. (B) As in (A), but for FANCD2, including Western blot and immunofluorescence. (C) As in (A), but for BRCA1. doi:10.1371/journal.pone.0073710.g004 persistence of DNA DSBs following irradiation of cells exposed to proteasome inhibitors may also itself contribute to apoptosis.
We observed reduced growth of NSCLC xenografts following proteasome inhibition combined with RT. GEMMs of lung cancer may be used to confirm the effect in primary lung tumors treated in situ. Exploring this in multiple GEMMs may have the additional advantage of identifying genotypes that particularly respond to proteasome inhibition. For instance, tumors arising from mice harboring inducible KRAS and TP53 mutations responded to bortezomib, unlike tumors from mice with mutant KRAS and wild-type TP53 [40]. We observed marked differences in survival with inducible proteasome knockdown in NSCLC xenografts expressing wild-type TP53, which is in contrast with these results. We propose that the combination with radiation may unlock the potential of proteasome inhibition to a wider range of tumor genotypes.
The adoption of proteasome inhibitors in the clinic will require improvement in tumor delivery. Bortezomib yielded relatively little proteasome inhibition in our xenograft studies compared to doxycycline-induced PSMA1 shRNA knockdown. Strategies in- cluding liposomal encapsulation, which has been successfully employed for doxorubicin [43], or use of second-generation proteasome inhibitors such as carfilzomib, marizomib or MLN9708 [44] may improve solid tumor penetration and proteasome inhibition.
The results of proteasome inhibition in NSCLC patients treated without radiotherapy have been mixed [16,17]. Proteasome inhibition was examined with concurrent chemoradiotherapy in one Phase I trial of twelve patients [45]. In this dose escalation study, patients with pathologically documented Stage IIIA-B disease received weekly carboplatin and paclitaxel, together with bortezomib 0.3-0.7 mg/m 2 twice weekly, during radiotherapy to a dose of 61.2 Gy in 34 daily fractions, followed by surgical resection. There were no unanticipated acute toxicities during chemoradiotherapy; Grade 2-3 myelosuppression was common, as expected with carboplatin and paclitaxel. Regrettably however, three of nine patients who underwent surgical resection died postoperatively; two died two to three days postoperatively and a third died 21 days postoperatively. It was concluded that delayed toxicity was severe and unpredictable. However, all three patients had undergone a right pneumonectomy after high-dose neoadju-vant chemoradiotherapy. Right pneumonectomy following neoadjuvant therapy has been associated with significantly increased mortality risk, regardless of addition of novel systemic agents. For example, there has been a reported 18% treatment-related mortality rate after right pneumonectomy, compared to a 4% rate after left pneumonectomy, after neoadjuvant radiotherapy to a median dose of 54 Gy with concurrent chemotherapy [46].
What was notable about the Phase I study [45] was the high rate of pathologic complete response (pCR) observed in patients treated with neoadjuvant chemoradiotherapy including bortezomib. Specimens from five of the nine patients (56%) who underwent surgical resection showed a pCR and an additional two showed 99% necrosis. This compares favorably to the 17.7% pCR rate observed in INT-0139 following neoadjuvant chemoradiotherapy among the 164 patients who underwent resection [47]. This data suggests that, with care, proteasome inhibition may be worth exploring further combined with radical chemoradiotherapy, perhaps in nonsurgical candidates or for left-sided or other tumors that will not require right pneumonectomy for resection.
Finally, proteasome inhibitors may disproportionally benefit patients whose tumors rely on HR-mediated repair of radiationinduced DNA DSBs. For example, high level of RAD51 protein expression has been associated with decreased survival of NSCLC patients [48]. Also, poorly differentiated NSCLC tumors have increased expression of HR genes [49]. In an analysis of published microarray data from 442 patients [9], high levels of RAD51 and BRCA1 were each associated with decreased overall survival ( Figure S7). These features may serve as biomarkers predictive of response to proteasome inhibitors as radiosensitizers. Also, reductions in radiation-induced DNA DSB repair protein foci, including FANCD2 and BRCA1, following treatment with proteasome inhibitors, may serve as biomarkers for response in treated tumors.

Reagents
Bortezomib was purchased from Selleck Chemicals (Houston, TX). 800 mM stocks in DMSO were stored at 220uC, and prior to use were diluted and stored at 4uC for up to one week. All treatments were at a final concentration of 30 nM for A549 or 50 nM for NCI-H460, unless stated otherwise. The following antibodies were used at the listed dilutions for Western immuno-

Whole Genome Pooled shRNA Screen
The screen was performed essentially as previously described for DLD-1 cells [21]. The Hannon-Elledge whole genome pooled shRNA library consists of six viral pools each containing approximately 13,000 different MSCV-PM retroviral shRNA particles targeting human genes. For each pool, three replicates of at least 1.3610 7 cells of A549 or NCI-H460 cells were incubated with an equivalent number of retroviral colony-forming units in media containing 8 mg/ml polybrene (Sigma-Aldrich, St. Louis, MO), for a 1000-fold representation of each shRNA sequence at a multiplicity-of-infection (MOI) of 1. After selection for stable integrants using 1 mg/ml puromycin, cells were passaged for a total of twelve population doublings (PD), at all times maintaining a minimum of 1.3 6 10 7 cells per replicate. Genomic DNA was extracted from cells harvested both before and after the twelve population doublings. Half-hairpin shRNA-containing sequences were amplified by PCR, purified by agarose gel electrophoresis  and labeled with Cy5 (PD 0) or Cy3 (PD 12). Competitive hybridization to custom Agilent microarrays was performed.
For scoring, the mean log 2 (Cy3/Cy5) ratio was determined for each shRNA sequence using triplicate data. Those shRNAs for which the standard deviation of log 2 (Cy3/Cy5) ratios was greater than the absolute mean log 2 (Cy3/Cy5) ratio were excluded from further analyses to reduce false positive hits. For each gene, one point was assigned for each different shRNA sequence targeting that gene for which the mean log 2 (Cy3/Cy5) ratio was less than or equal to 21, representing a decrease in representation of the shRNA sequence of at least 2-fold during passaging. One half point was assigned for each shRNA sequence for which the mean log 2 (Cy3/Cy5) ratio was less than or equal to 20.5. A negative point was assigned for each shRNA sequence for which the mean log 2 (Cy3/Cy5) ratio was greater than or equal to +1, and a negative half point was assigned for each shRNA sequence for which the mean log 2 (Cy3/Cy5) ratio was greater than or equal to +0.5, to penalize discordant results among shRNA sequences targeting a particular gene. Finally, genes were ranked based on descending total score, with higher rankings given to genes with equivalent total scores but fewer different shRNA sequences targeting the gene.

PSMA1 shRNA Inducible Cell Lines
The TRIPZ plasmid containing PSMA1 shRNA sequence V2THS_170669 was purchased from Thermo Scientific (Rockford, IL). The non-silencing shRNA control (RHS4743, Thermo Scientific) was also purchased. Lentiviral particles were generated using the Trans-Lentiviral Packaging Mix (Thermo Scientific) in HEK293T cells per manufacturer's protocol. Viral titers were assessed by serial dilution using U2OS cells. A549 and NCI-H460 cells were infected at a MOI of 1.5. Transfected cells were selected with puromycin, and individual subclones were tested for doxycycline-inducible PSMA1 shRNA knockdown via Western blot.
Proteasome Activity Assay 96-well white, clear bottom polystyrene tissue culture plates (#3903, Corning, Tewksbury, MA) were seeded with 1000 NSCLC cells in 200 ml media. After 24 hours, media was replaced with media containing bortezomib, or PSMA1 shRNA was induced with 1 mg/ml doxycycline. Cells were incubated for up to 3 days prior to chymotrypsin-like proteasome activity assays using the Proteasome-Glo cell-based luminescent assay (Promega, Madison, WI) per manufacturer's protocol. All results were normalized to cells treated with DMSO vehicle control for bortezomib or non-silencing control for RNAi knockdown.

Clonogenic Cell Survival Assay
NSCLC cells were seeded into 6-well tissue culture-treated dishes (#3516, Corning, Tewksbury, MA) at 10% confluence. Bortezomib was initiated 24 hours after seeding and continued during IR initiated 24 hours after bortezomib. IR was given using a Gammacell 40 Exactor Cs-137 irradiator (Best Theratronics, Ottawa, Ontario, Canada) in 1 or 2 Gy daily fractions over three days. Six hours after the final fraction, cells were trypsinized, counted, and seeded into 10 cm dishes in triplicate to generate isolated colonies. Once colonies became visible by eye, plates were stained with 1% crystal violet (Sigma-Aldrich) and colonies of at least 50 cells were counted by eye.

Apoptosis Assay
NCI-H460 cells were seeded into 6-well tissue culture treated dishes (Corning) at 10% confluence. Bortezomib was initiated 24 hours after seeding and continued during IR. 2 to 4 Gy IR was administered 24 hours later using a Gammacell 40 Exactor Cs-137 irradiator (Best Theratronics). Cells were prepped and stained 24 hours later with the FITC Annexin V Apoptosis Detection Kit II (#556570, BD Biosciences, San Jose, CA) as per manufacturer's protocol. Quantification was performed by flow cytometry (BD LSRFortessa, BD Biosciences, San Jose, CA).

Immunofluorescence
Cells were fixed with 4% formaldehyde (Sigma-Aldrich) followed by ice-cold methanol 6 hours post IR. Slides were incubated on a rocker at room temperature for 2 hours in blocking buffer containing 10% normal goat serum and 0.3% Triton X-100, then incubated in primary antibody overnight followed by secondary antibody for one hour in antibody buffer containing PBS, 1% bovine serum albumin and 0.3% Triton X-100). ProLong Gold antifade reagent with DAPI (Life Technologies, Grand Island, NY) was used for mounting. An Axio Imager Z1 fluorescence microscope with AxioCam MRc and MRm CCD cameras (Zeiss, Thornwood, NY) was used for visualization. DNA damage-inducible foci were counted manually in 100 cells per sample, with a minimum of 3 replicates.

Comet Assay
Cells were seeded into 6-well plates at 20% confluence. Bortezomib was initiated 24 hours after seeding and refreshed 1 hour prior to IR the following day. 40 Gy IR was given using a Gammacell 40 Exactor with cells on ice, and cells were collected 1, 4 or 8 hours after initiation of IR. The OxiSelect Comet Assay Kit (Cell Biolabs, San Diego, CA) was used per manufacturer's protocol, and comets were visualized using an Axio Imager Z1 fluorescence microscope with an AxioCam MRm CCD camera (Zeiss, Thornwood, NY).

Mouse Xenograft Experiments
NCI-H460 cells were trypsinized and suspended at 5610 6 cells/ ml in 0.9% saline. 1610 6 cells were injected subcutaneously into the right flanks of NCr nude mice. Tumor formation was assessed by caliper, with tumor volume determined by 0.56longest diameter6shortest diameter 2 . Once tumors reached at least 200 mm 3 , PSMA1 shRNA knockdown was induced by including 0.5 mg/ml doxycycline in drinking water. In a separate cohort, 0.6 mg/kg bortezomib given intravenously by tail vein was started. One week later, IR was initiated to give five fractions of 4 Gy up to three times per week using a small animal radiation research platform (SARRP) as described below. In the bortezomib cohort, treatment was administered 1 hour prior to RT. Doxycycline or bortezomib was discontinued after the final fraction of RT. Tumors were assessed until they reached 2 cm diameter or showed necrosis, or mice showed evidence of morbidity, or 100 days after initiation of treatment, whichever came first. Mice were euthanized with carbon dioxide. For immunofluorescence, tumor samples were collected immediately and embedded in O.C.T. (VWR, 4583) in Tissue-tek cryomold (VWR, 4557) plastic cassettes. Cassettes with sample and O.C.T. were then placed above liquid nitrogen until frozen and stored at 280uC. For immunohistochemistry, tumor samples were collected and fixed in 10% formalin acetate (SF99-4, Fischer Scientific) for 24 hours then stored in 70% ethanol prior to paraffin embedding and sectioning. c-H2AX (05-636, Millipore, Billerica, MA) staining at a dilution of 1:20,000 was performed by the Dana-Farber/Harvard Cancer Center Research Pathology Core. An Axio Imager Z1 fluorescence microscope with an AxioCam MRc camera (Zeiss) was used for visualization. c-H2AX foci were counted manually with a minimum of 400 cells per sample. For proteasome activity assays, tumor samples were collected immediately following euthanasia and snap frozen and stored at 280uC. PBS-EDTA buffer was then added and samples sonicated for 60 seconds. Proteasomal activity of the lysate was then assessed using the Proteasome-Glo assay (Promega) per manufacturer's protocol.

Small Animal Radiation Research Platform (SARRP)
Mice were anesthetized via isoflurane inhalation for the duration of each treatment. For each treatment, tumors were visualized using cone beam computed tomography (CT) using 65 kVp and 0.6 mA photons. Tumors were treated using a 1.2 cm circular collimator selected to give 0.25-0.5 cm margins around gross tumor, using 220 kVp and 13 mA photons given with a lateral en face field prescribed to a depth of 5 mm. The SARRP was calibrated and maintained as previously described [38,39].

Statistics
All quantitative data were analyzed and graphs formed using Prism 6 (Graphpad Software, La Jolla, CA). All differences between treatment groups were determined via two-tailed parametric Student's t test, with a p value ,0.05 viewed as significant.

Declaration of Ethical Approval
All animal experiments were performed in accordance with a protocol (11-024) approved by the Dana-Farber Cancer Institute Animal Care and Use Committee (DFCI ACUC) and in adherence to the NIH Guide for the Care and Use of Laboratory Animals. Figure S1 Clonogenic survival assay of A549 (left) and NCI-H460 (right) following single fraction IR and bortezomib. Marked bars show the percent kill of bortezomibtreated samples compared to DMSO vehicle control at each IR dose. All results are mean 6 SEM and normalized to DMSO vehicle control. (EPS) Figure S2 Proteasome inhibition reduces non-homologous end joining in NSCLC cells. GFP reporter assay for nonhomologous end joining (NHEJ) after proteasome inhibition for 24 hours via bortezomib (right) or PSMA1 siRNA knockdown (left) in A549 (top) and NCI-H460 (bottom). All results are mean 6 SD and normalized to DMSO vehicle control (Veh) or scrambled vector control. P values were calculated using a two-tailed Student's t test. (TIF) Figure S3 Bortezomib reduces RAD51, FANCD2 and BRCA1 IR induced foci in A549. These photos show representative images for the quantified data in Figure 4 of the main text; refer to that figure legend for additional details. Bar = 10 mm. (TIF) Figure S4 PSMA1 shRNA reduces RAD51, FANCD2 and BRCA1 IR induced foci in A549 and NCI-H460. These photos show representative images for the quantified data in Figure 4 of the main text; refer to that figure legend for additional details. Bar = 10 mm. (TIFF) Figure S5 Bortezomib does not improve control of irradiated NSCLC xenografts. 10 6 NCI-H460 cells were injected into the flanks of 6-8 week-old NCr nude mice. Once tumors reached 3 mm diameter (day 0), bortezomib was injected into the tail vein at 0.6 mg/kg twice per week. One week later, RT was initiated to give a total of five 4 Gy fractions every other day using a small animal radiation research platform (SARRP) with 0.6 mg/kg bortezomib given 1 hour prior to RT. Mice were subsequently followed until tumors reached 2 cm diameter, animals became moribund or for 100 days. (TIF) Figure S6 Chymotrypsin-like proteasome activity in NSCLC xenografts following bortezomib vs. inducible PSMA1 shRNA knockdown. Tumors were collected and frozen 1 hour after the last of three doses of bortezomib given over one week, or proteasome inhibition via PSMA1 shRNA knockdown for one week. Control mice were treated with DMSO (Veh) or non-silencing shRNA tumors. All results are mean 6 SEM. (EPS) Figure S7 High BRCA1 and RAD51 gene expression are each associated with decreased overall survival. Kaplan-Meier analysis was performed using a published dataset of 442 lung adenocarcinomas [9], comparing survival among patients with greater than median expression versus less than median expression of each gene. The log-rank test was performed to assess for statistically significant differences in overall survival. (TIF)  Whole genome shRNA screen hits in A549 and NCI-H460. Complete list of whole genome shRNA screen hits in A549 and NCI-H460. Each of the 1,667 genes was the target of at least one shRNA sequence whose representation decreased by at least two-fold during passage in both cell lines. Listed are the numbers of shRNA sequences in the screen targeting each gene, and the score for each cell line. To derive the score, one point was given for each shRNA sequence whose representation decreased by at least two-fold during passage, and one half point was given for each sequence whose representation decreased by 2 0.5 to 1 -fold. One point was subtracted for each shRNA sequence whose representation increased by at least two-fold during passage, and one half point was given for each sequence whose representation increased by 2 0.5 to 1 -fold, to penalize discordant effects. (XLSX)

Supporting Information
Table S3 List of 26 S proteasome genes highlighting whole genome shRNA screen hits. Each of the highlighted genes was the target of at least one shRNA sequence whose representation decreased by at least two-fold during passage in both cell lines, with the top hit PSMA1 shaded red, strong hits (.1 shRNA sequence per gene in both cell lines) shaded dark orange and minor hits (1 shRNA sequence per gene in both cell lines) shaded light orange. Listed are the numbers of shRNA sequences in the screen targeting each gene, and the score for each cell line. To derive the score, one point was given for each shRNA sequence whose representation decreased by at least two-fold during passage, and one half point was given for each sequence whose representation decreased by 2 0.5 to 1 -fold. One point was subtracted for each shRNA sequence whose representation increased by at least two-fold during passage, and one half point was given for each sequence whose representation increased by 2 0.5 to 1 -fold, to penalize discordant effects. (XLSX)