Genome-Wide Sirna Screen for Modulators of Cell Death Induced by Proteasome Inhibitor Bortezomib

Published OnlineFirst May 11, 2010; DOI: 10.1158/0008-5472.CAN-09-4428 Published OnlineFirst on May 11, 2010 as 10.1158/0008-5472.CAN-09-4428

Integrated Systems and Technologies Cancer Research Genome-Wide siRNA Screen for Modulators of Cell Death Induced by Proteasome Inhibitor Bortezomib

Siquan Chen1, Jonathan L. Blank2, Theodore Peters1, Xiaozhen J. Liu2, David M. Rappoli1, Michael D. Pickard3, Saurabh Menon1, Jie Yu2, Denise L. Driscoll2, Trupti Lingaraj2, Anne L. Burkhardt2, Wei Chen2, Khristofer Garcia1, Darshan S. Sappal2, Jesse Gray1, Paul Hales1, Patrick J. Leroy2, John Ringeling1, Claudia Rabino2, James J. Spelman2, Jay P. Morgenstern1, and Eric S. Lightcap2

Abstract Multiple pathways have been proposed to explain how proteasome inhibition induces cell death, but mechanisms remain unclear. To approach this issue, we performed a genome-wide siRNA screen to evaluate the genetic determinants that confer sensitivity to bortezomib (Velcade (R); PS-341). This screen identified 100 genes whose knockdown affected lethality to bortezomib and to a structurally diverse set of other proteasome inhibitors. A comparison of three cell lines revealed that 39 of 100 genes were commonly linked to cell death. We causally linked bortezomib-induced cell death to the accumulation of ASF1B, Myc, ODC1, Noxa, BNIP3, Gadd45α, p-SMC1A, SREBF1, and p53. Our results suggest that proteasome inhibition promotes cell death primarily by dysregulating Myc and polyamines, interfering with protein translation, and disrupting essential DNA damage repair pathways, leading to programmed cell death. Cancer Res; 70(11); OF1–9. ©2010 AACR.

Introduction active oxygen species (ROS), stabilization of Myc and Noxa, and induction of endoplasmic reticulum (ER) stress, any of The primary targets of most cancer chemotherapies are which may trigger cell death (1–4). Despite the pleiotropic known. However, the downstream effect on the cell is gener- effects of proteasome inhibition, bortezomib is an effective ally poorly understood, particularly the sequence of events cancer chemotherapy which exploits key weaknesses that dif- directly involved in the commitment of the tumor cell to ferentiate tumor cells from normal cells. Determining the pri- die. This lack of knowledge is particularly profound for pro- mary mechanisms by which bortezomib induces tumor cell teasome inhibition because the majority of intracellular pro- death would increase our understanding of these key weak- teins can be degraded by the proteasome. As such, simply nesses and allow for better utilization of bortezomib in the showing that proteasome inhibition can result in the stabili- clinic. zation of a protein does not necessarily imply that it is mech- RNA interference (RNAi) has been widely adopted for func- anistically relevant and important for cell death. Prior studies tional genomic studies, including defining drug mechanism have suggested that bortezomib has an antitumor effect by (5). Genome-wide RNAi allows for different mechanisms to inhibition of antiapoptotic proteins (e.g., NFκBandBcl2), be ranked against each other, providing an evaluation of stabilization of tumor suppressors (e.g., p53), disruption their relative phenotypic dominance. Some limitations of of cell cycle, dysregulation of Fas or tumor necrosis factor– RNAi are that protein knockdown is transient, never results related apoptosis-inducing ligand pathways, increases in re- in complete elimination of the protein, and has various de- grees of off-target effects. Genetic redundancy may also mask the phenotype of targeted genes. Nevertheless, RNAi studies – Authors' Affiliations: 1Discovery Technologies, 2Discovery Oncology have provided powerful insights into drug mechanism (6 8). Biology, and 3Medical Biostatistics, Millennium Pharmaceuticals, Inc., Bortezomib induces death of most cell lines in culture Cambridge, Massachusetts within 72 hours, with an LC50 of ∼10 nmol/L. We performed Note: Supplementary data for this article are available at Cancer the genome-wide siRNA screen to identify genes whose Research Online (http://cancerres.aacrjournals.org/). knockdown affected the cytotoxicity of bortezomib using Current address for S. Chen: Cellular Screening Center, Institute for the colon cancer cell line HCT-116 because of its highly re- Genomics and Systems Biology, University of Chicago, GCIS, WSB03, 929 East 57th Street, Chicago, IL 60637. Current address for T. Peters: producible sensitivity to bortezomib and high transfectability Constellation Pharmaceuticals, 148 Sidney Street, Cambridge, MA when compared with other cell lines. The effect of the top 02139. Current address for D.M. Rappoli and J.P. Morgenstern: hits from the screen were also compared in A375 and HeLa Monsanto Biotechnology, 325 Vassar Street, Cambridge, MA 02139. cells, highly transfectable lines derived from melanoma and Corresponding Author: Eric S. Lightcap, Discovery Oncology Biology, Millennium Pharmaceuticals, Inc., 40 Landsdowne Street, Cambridge, cervical cancer, respectively. MA 02139. Phone: 617-551-3717; Fax: 617-551-8906; E-mail: eric.lightcap@ Herein, we show that bortezomib leads to cell death by in- mpi.com. terfering with ribosome function, essential DNA damage path- doi: 10.1158/0008-5472.CAN-09-4428 ways, and dysregulation of Myc. In particular, we examine the ©2010 American Association for Cancer Research. effect of bortezomib on protein translation initiation,

www.aacrjournals.org OF1

Chen et al.

especially eIF2α and mammalian target of rapamycin (SABiosciences) that is under the control of a minimal cyto- (mTOR), the effect of upregulation of Myc and ODC1 on megalovirus (CMV) promoter and tandem repeats of the polyamine homeostasis, and DNA damage signaling. In ad- E-box sequence using pCMV firefly luciferase for normaliza- dition, we show that a common set of 39 genes is respon- tion. Forty-eight hours after transfection, cells were treated sible for conferring sensitivity to proteasome inhibition in with 30 nmol/L bortezomib or vehicle for the indicated times multiple cell lines. Although these results do not account under normal or serum-starved conditions, after which lucif- for differences in bortezomib efficacy across tumor types erase activities were determined using Dual-Glo reagents seen in the clinic and cannot account for the effect of (Promega) as described in the Supplementary Data. the tumor microenvironment on bortezomib sensitivity, they provide a foundation for future studies exploring Polyamine high-performance liquid chromatography these issues. assay Polyamines were quantified by high-performance liquid Materials and Methods chromatography on an AccQTag column using the AccQTag method (Waters) with a modified acetonitrile gradient as Genome-wide screen and hit deconvolution specified in the Supplementary Data. HCT-116 cells were transfected with 15 nmol/L siRNA oligos (siGENOME SMARTpool, Dharmacon) using Dhar- Data treatment and statistics maFECT 2 (DH2) reagent (Dharmacon) in Opti-MEM All ATPlite fluorescence intensity values were log2 trans- (Invitrogen) in a BioCoat poly-D-lysine (PDL)–coated 384- formed (V). The Bliss Independence (BI) scores were calcu- well daughter plates (BD Biosciences). Forty-eight hours af- lated from the log2-transformed data as ter transfection, cells were treated with 0, 4, or 7 nmol/L V þ V − V − V ; bortezomib and incubated for a further 48 hours, after which xn g0 gn x0 viability was assessed using ATPlite reagent (Perkin-Elmer) where V is the mean of RNAi oligos treated with 4 nmol/L and luminescence was measured using a LEADseeker imaging xn (or 7 nmol/L) bortezomib, V is the mean of GL2 control system (GE Healthcare). g0 cells, V is the mean of GL2 cells treated with 4 nmol/L SMARTpool hits were deconvoluted in HCT-116, A375, and gn (or 7 nmol/L) bortezomib, and V is the mean of RNAi oligos HeLa cells again in sextuplicate, except that CELLCOAT x0 treated with 0 nmol/L bortezomib. 384-well PDL-coated black, clear-bottom plates (Greiner) The Rescue scores were calculated using log -transformed were used. For HCT-116 on the Greiner plates, DH2 lipid 2 data as min(V , V ) − V . was reduced from 0.075 to 0.057 μL/well. Bortezomib was gn x0 xn The SE was calculated for each of the equation variables. modestly more potent under these conditions, so concentra- These estimates were then squared, and the square root of tions of 2 nmol/L (LC ) and 5 nmol/L (LC ) were used. For 15 60 their sum was taken to give the SE of the score. Probabilities A375, changes included the use of DharmaFECT 4 (DH4) that gene knockdown was on target were calculated by using transfection reagent at 0.07 μL/well, 300 cells per well, and hypergeometric distribution function based on deconvoluted bortezomib treatments at 4 nmol/L (LC )and7nmol/L 5 oligo data as specified in the Supplementary Data. (LC55). For HeLa, changes included the use of DharmaFECT 1 (DH1) transfection reagent at 0.2 μL/well, 900 cells per well, and bortezomib treatments at 12 nmol/L (LC25)and Results 25 nmol/L (LC50). Bortezomib will be provided to qualified researchers once a standard Materials Transfer Agreement Design of screen has been executed. Gene knockdown in HCT-116 by 15 nmol/L Dharmacon SMARTpool siRNA was allowed to progress for 48 hours to Western blot and immunoprecipitation provide for adequate depletion of protein, followed by HCT-116 cells were treated with 30 nmol/L bortezomib 48 hours with two concentrations of bortezomib (LC15 and (LC90) or vehicle (0.1% DMSO) for the times indicated. Sam- LC60 at 48 h) or a vehicle control. These concentrations allow ples were prepared for SDS-PAGE and quantitative immuno- for the evaluation of either enhancement or suppression of blot analysis with antisera as specified in the Supplementary bortezomib-induced cell death by the SMARTpools, by which Data using tubulin as a normalization control (Supplementa- both activators and inhibitors of key pathways can be de- ry Fig. S2B). Myc was immunoprecipitated using anti-Myc tected. The genome-wide screen was performed with agarose beads (Novus Biologics) from HCT-116 cells that 21,062 SMARTpools in duplicate. Three thousand eight hun- had been treated for 8 hours with either 0.1% DMSO or dred and seventy-seven SMARTpool hits were selected from 30 nmol/L bortezomib. Associated Myc (Epitomics) and the primary screen and rescreened in sextuplicate (Supple- Max (Santa Cruz Biotechnology) were quantified using an mentary Fig. S1A). Odyssey Infrared Imager (LI-COR Biosciences). The scoring of the screen using BI assumptions (see Sup- plementary Results) enabled assignment of phenotypes to Myc reporter assay the siRNAs, namely, synthetic lethality, suppressor, or epista- Myc activation was measured by transfection of HCT-116 sis (Supplementary Fig. S1B and C). The magnitude of the BI cells with the Cignal Myc-responsive luciferase reporter score indicates the relative importance of the effect for the

OF2 Cancer Res; 70(11) June 1, 2010 Cancer Research

Proteasome Inhibition RNAi Screen

induction and execution of cell death, although this ranking tion, 14 of 14 SMARTpools targeting 20S core subunits were is confounded by the efficiency with which the oligos are able synthetic lethal (BI = −0.46 to −1.99 at 4 nmol/L bortezomib), to knock down their respective mRNAs and ultimately re- indicating that knockdown of 20S core subunits sensitized duce protein levels. It remains a formal possibility that differ- the cell to bortezomib (Fig. 1A). Strong scores were obtained ences in knockdown efficiency rather than target biology will even when the SMARTpool alone reduced viabilities to <5% be a primary determinant of interaction strength. Neverthe- (e.g., PSMA1,BI=−1.17 ± 0.08 with 3% viability). Because the less, because knockdown efficiency is likely to be random data for the 20S core subunits are highly consistent, this in- across pathways, it is anticipated that pathways can be dicates that useful drug interaction data can be obtained ranked by this approach provided that at least some proteins even at viabilities <5%. Because <1% of the SMARTpools in within each pathway are efficiently knocked down. the primary screen gave viabilities <5%, SMARTpools with low viabilities were not excluded from further analysis. Behavior of SMARTpools Stabilization of proteins by inhibition of the proteasome is At clinically relevant concentrations (i.e., low nanomolar; expected to be the proximal determinant of bortezomib- ref. 9), bortezomib selectively inhibits the proteasome, pri- induced cell death, both by inducing unfolded protein marily through the β5 subunit (PSMB5, chymotrypsin-like stress and by stabilizing proapoptotic proteins. Indeed, protease) but also the β1 subunit (PSMB6, caspase-like cycloheximide abrogates bortezomib-induced apoptosis (11, protease or PGPH; ref. 10). Therefore, the behavior of 12). Based on this result, knockdown of ribosome subunits SMARTpools targeting proteasome subunits is of particular should result in resistance to proteasome inhibition by re- interest. From the genome-wide screen, 7% of all SMART- ducing the rate of protein synthesis and the accumulation pools had viabilities of <15% relative to the GL2 negative of proteins. control. In contrast, 14 of 14 SMARTpools targeting 20S Of the SMARTpools targeting the 34 small ribosome sub- core subunits had viabilities of <15% (Fig. 1A; Supplemen- units (40S; ref. 13), 21 resulted in viabilities of <15% (Fig. 1B; tary Table S3A). Thus, the proteasome is critically required Supplementary Table S3A). Of the SMARTpools targeting the for viability. Because all SMARTpools targeting constitutive 47 large ribosome subunits (60S; ref. 13), 44 resulted in viabil- proteasome subunits had similar effects on viability, the ities of <15% (Fig. 1C; Supplementary Table S3A). SMART- concentration of 15 nmol/L SMARTpool seems adequate pools targeting 29 of the 40S ribosome subunits and 44 of for genome-wide screening in HCT-116. the 60S ribosome subunits gave positive BI scores (BI = Because proteasome subunit knockdown leaves few viable 0.29–2.83 at 7 nmol/L bortezomib). cells, evaluation of the interaction of proteasome RNAi with In contrast to the expectation that ribosome activity bortezomib may be compromised. Contrary to this expecta- would be required for cell death induced by bortezomib, of

Figure 1. SMARTpools against proteasome or ribosome subunits behave in a highly coherent manner. SMARTpools targeting key genes were evaluated for effects on viability and interactions with bortezomib (Bz). Genes are listed in alphabetical order according to their HUGO gene names. A, 20S core subunits of the proteasome. The three subunits with the most modest viability effect are the immunoproteasome subunits PSMB8, PSMB9, and PSMB10. RNAi targeting 20S subunits at 4 nmol/L bortezomib all scored as synthetic lethal. B, 40S ribosome subunits. The data for the SMARTpool targeting FAU was removed. RNAi targeting most 40S ribosome subunits scored as epistatic at 7 nmol/L bortezomib. C, 60S ribosome subunits. RNAi targeting most 60S ribosome subunits scored as epistatic. Supporting data are included in Supplementary Table S3A. Columns, mean (n = 6); bars, SE.

www.aacrjournals.org Cancer Res; 70(11) June 1, 2010 OF3

Table 1. HCT-116 colon cancer cell line

Classification Ribosome biogenesis/ BNIP3- HR Myc/ Ubiquitin Apoptosis Cytoskeletal ER/ Insulin TGFβ Sterol/ Unassigned translation induced polyamine Golgi lipid cell death Published OnlineFirstMay11,2010;DOI:10.1158/0008-5472.CAN-09-4428 cancerres.aacrjournals.org Suppressor EIF4E ATG4A IKIP SNIP1 TAX1BP1 BAX NAV3* FUT11* SORBS1 SREBF1* CSE1L BXDC2 BNIP3* TP53 ODC1 NUP54 DDX27 SETX* ASF1B MYC NUPL1* † † EIF4G1 GADD45A* PMAIP1 CDC2L1 PPP1R15B SMC1A* DCLRE1A* † Epistatic EIF2B4 RAD21 DDB1 TMSB10 EIF3G RPA3 SPSB3 FAU RAD51* FCF1 on September 25, 2021. © 2010American Association for Cancer EIF3A

Research. RPL13 EIF2S1 BXDC5 UTP20† Synthetic STK11 CLTB FANCE YY1 PSMA5 BCL2L1 RASAL2 SEC61G IDE HNF1B OSBPL6 RNF10 lethal HSPA14 NDUFA13 H2AFX SPI1 PSMB5 CFLAR CLASP1 RYR1 MAPK3 FOXH1 MGST2 SKIV2L HAGH BTG1 SRM E4F1 BAD PDLIM7 ATF6 ENPP1 TGFB1 SS18L1 VPS35 EIF5A UBA3 BAT3 ARHGEF1 ATP2A3 KCNQ2 † SNX8 TRRAP VPS28 FAS CDC42EP1* PTH2 NFE2L2 FASTK CENPB GSTZ1 LGALS1 VPS4A MTHFR FOXO4* TAF11* DAK*

NOTE: Pathway modules inferred from genetic interactions of genes confirmed by deconvolution in colon cancer cell line HCT-116. Genes had hypergeometric P < 0.05, except as indicated. Genes are ordered by increasing P values. All genes included in the table interacted with bortezomib with P < 0.10 in at least one cell line. Underlined genes had similar phenotypes in HCT-116, A375, and HeLa cell lines. Eight hits with low expression data in HCT-116 were CLASP1, EIF3A, ENPP1, FOXH1, HNF1B, MTHFR, SORBS1, acrResearch Cancer and SPI1 (Supplementary Table S3C). Abbreviations: HR, homologous recombination; TGFβ, transforming growth factor β. *P < 0.10. † P < 0.15. Published OnlineFirst May 11, 2010; DOI: 10.1158/0008-5472.CAN-09-4428

Proteasome Inhibition RNAi Screen

76 significant SMARTpools targeting ribosome subunits, only RPS8 and RPL5 knockdown gave significant rescue, suggesting an off-target effect with these two pools. The very consistent lack of significant rescue indicating epistasis suggests that inhibition of protein translation and proteasome inhibition by bortezomib result in related stresses to the cell (see Discussion).

Deconvolution of genome-wide hits Because of the pleiotropic effect of proteasome inhibition by bortezomib, the frequency of interactions due to off-target effects of the SMARTpools was anticipated to be substantial. To determine the likelihood that the phenotype was due to knockdown of the target, the individual oligos making up each SMARTpool were further evaluated (four oligos per gene at 4 nmol/L individual oligo). Eight hundred and sixty- eight genes were selected based on the screen results or prior literature. Because of the coherence of their data, proteasome and ribosome subunits were underrepresented in this deconvolution set. The oligos making up the selected SMARTpools were individually evaluated for interaction with bortezomib in sextuplicate (Supplementary Table S3B). The top 160 hits were rearrayed and evaluated again in sextuplicate. Finally, these top 160 hits were reassayed twice using an updated set of individual oligos (Supplementary Table S3C).

One hundred genes gave significant results in HCT-116 Figure 2. Regulation of candidate proteasome substrates whose based on a hypergeometric evaluation of the deconvoluted stabilization induces cell death across a time course of bortezomib oligo data (Table 1). The 100 genes were classified into path- treatment of HCT-116. Regulation of proteins targeted by the suppressor way modules using their gene annotation (Supplementary oligos in Table 1 was determined by Western blot on HCT-116 treated Table S3D). Based on the statistical approach used to vali- with vehicle (0.1% DMSO, top blot) or 30 nmol/L bortezomib (LC90 concentration, bottom blot). The control lanes 8 and 9 (24-h treatment) date hits, it is anticipated that 5 to 10 hits within Table 1 show that exposure of the film for the top and bottom blots was are false positives. equivalent. Phosphorylation of SMC1A and TP53 on their ATM sites was Genes whose siRNAs score as suppressor (Table 1) repre- also evaluated. Full-length SREBF1 is indicated by the arrowhead. Two sent those whose presence is important for inducing cell death bands that are consistent with activation of the two spliceforms of SREBF1 by MBTPS1 (S1P) and MBTPS2 (S2P) are indicated by the two bars. in response to proteasome inhibition; removal of these pro- Full-length blots are presented in Supplementary Fig. S8, along with teins protects against bortezomib-induced cell death. It is pos- theoretical molecular weights of the proteins. sible that these proteins are proteasome substrates that when stabilized contribute to cell death. Therefore, we investigated whether these proteins were stabilized by proteasome inhibi- Myc pathway gene Max was modestly affected by bortezo- tion. ASF1B, BNIP3, Gadd45α,Myc,ODC1,PMAIP1(Noxa), mib treatment (Fig. 4A). Immunoprecipitation of Myc copre- p-SMC1A, SREBF1, and p53 showed increases in protein le- cipitates Max but bortezomib treatment does not result in a vels following bortezomib treatment (Fig. 2). ATG4A, BAX, commensurate increase in associated Max (Fig. 4B). A Myc BXDC2, DDX27, EIF4E, EIF4G1, SETX, SNIP1, and SORBS1 reporter assay showed that the stabilization of Myc results did not show clear increases (Supplementary Fig. S2A). in increased transcriptional activity both in the presence Genes associated with pathway modules were explored in and in the absence of serum (Fig. 4C). ODC1, a target of more detail, focusing on ribosome biogenesis and transla- Myc, is increased (Fig. 2) and the product of ODC1, putres- tion, Myc/polyamine, and homologous recombination (HR). cine, showed a marked increase following bortezomib treat- Among the translation pathway genes, phosphorylation of ment (Fig. 4D). targets of mTORC1, namely 4EBP1 (pS65) and S6K (pT389), Evidence was collected suggesting that bortezomib as a both decreased following bortezomib treatment (Fig. 3A). single agent potentiates DNA damage signaling. HR pathway Both the pT37/pT46 and total 4EBP1 antibodies indicate that gene RAD51 showed increases following bortezomib treat- 4EBP1 shifts toward the hypophosphorylated form. In agreement. The phosphorylation of SMC1A (pS957), TP53 (pS15), ment with ribosome knockdown being epistatic with borte- H2AFX (pS139), and CHEK2 (pT68) on their ATM sites zomib, we have found that cycloheximide is epistatic with and the ubiquitination of FANCD2 were also increased by bortezomib in HCT-116 viability and does not rescue cells bortezomib, particularly at 18 and 24 hours, whereas MDC1 (Fig. 3B). The phosphorylation of Ser51 on EIF2S1 (eIF2α) de- and RAD21 levels were unaffected (Figs. 2 and 5). The phos- creased probably due to increases in PPP1R15A (Gadd34) le- phorylation of CHEK1 (pT317) on its ATR site was modestly vels (Fig. 3C). increased.

www.aacrjournals.org Cancer Res; 70(11) June 1, 2010 OF5

Chen et al.

more correlated than those for MLN4924 (r2 < 0.11), a Nedd8-activating enzyme inhibitor, or thapsigargin (r2 < 0.18), an inhibitor of the SERCA pumps and an inducer of ER stress.

Discussion

The 100 genes within Table 1 highlight the critical biology involved in the response of HCT-116 cells to bortezomib. We have chosen to focus on candidate proteasome substrates and pathway modules where new biology has been identified that had not previously been characterized as being critical for proteasome inhibition–induced cell death, specifically mTOR inhibition, polyamine dysregulation, and potentiation of DNA damage signaling. A comprehensive analysis of the role of all of the genes within Table 1 will await further

Figure 3. Bortezomib inhibits translation initiation signaling. Regulation of proteins was determined in the same manner as Fig. 2. A, mTOR substrates. Evaluation of the phosphorylation state of 4EBP1 is consistent with substantial mTOR inhibition induced by proteasome inhibition. Phosphorylation of S6K is also reduced after 4 h, although total protein levels are unaffected. B, the dose-response curve for the effect of cycloheximide (CHx) on viability (ATPlite) in HCT-116 at 48 h was determined in the presence of vehicle (0.1% DMSO;▴) or 7 nmol/L bortezomib (LC74;▾). Cycloheximide was epistatic with bortezomib and did not rescue cells from bortezomib-induced cell death. C, eIF2α status. Phosphorylation of eIF2α is reduced following 6 h with no changes in total protein. PPP1R15A (Gadd34) levels increased with decreasing eIF2α phosphorylation. Full-length blots are presented in Supplementary Fig. S8, along with theoretical molecular weights of the proteins.

Evaluation of hits in other cell lines and with other proteasome inhibitors More than half of the 100 genes interacting with bortezomib in HCT-116 gave similar phenotypes in the melanoma cell line A375 (Supplementary Table S1) or in the cervical cancer cell line HeLa (Supplementary Table S2). Notably, Figure 4. Bortezomib activates Myc pathway. Regulation of proteins was the knockdown of 39 genes gave similar phenotypes in all determined in the same manner as Fig. 2. A, Max levels were modestly three cell lines (underlined genes in Table 1). Therefore, al- regulated by bortezomib. B, immunoprecipitation of Myc from vehicle (0.1% DMSO) or 30 nmol/L bortezomib–treated HCT-116 cells for 8 h. though details of the induction of cell death probably vary Lysate samples were normalized relative to immunoprecipitation samples between cell lines, these results indicate the existence of for Myc content by LI-COR and then evaluated for relative Max content. common themes. These common themes include translation C, HCT-116 cells were transiently transfected with an E-box firefly initiation, Myc/polyamines, and key ubiquitin pathway genes. luciferase (solid line) or a pCMV firefly luciferase (dashed line). Renilla The individual oligos for each of the 100 genes interacting Cotransfection of luciferase showed that viability was unaffected over 8 h. Addition of bortezomib, either in the presence or in the with bortezomib behaved consistently with multiple classes absence of serum, showed a significant increase in Myc transcription of proteasome inhibitors in HCT-116 cells. RNAi interactions activity. Fold induction is relative to the vehicle control. D, HCT-116 cells with the peptidyl boronate MLN2238, the peptide ML912, were treated with vehicle (0.1% DMSO, dashed line) or 30 nmol/L and the epoxide epoxomicin were highly correlated (r2 = bortezomib (solid line) for the time course indicated. A significant increase in putrescine levels was seen within 4 h, whereas no significant 0.86–0.91) with interactions with bortezomib (Supplementary β changes were seen in spermidine until levels were decreased at 24 h. Fig. S3). The interactions with the -lactone salinosporamide Full-length blots are presented in Supplementary Fig. S8, along with 2 A were less correlated (r = 0.69–0.70), although still much theoretical molecular weights of the proteins.

OF6 Cancer Res; 70(11) June 1, 2010 Cancer Research

Proteasome Inhibition RNAi Screen

tive sites, and less bortezomib is needed to reduce proteasome function below a critical limit. Therefore, these data are consistent with bortezomib inducing cell death by binding to the 20S core directly.

Ribosome subunits and translation Broadly speaking, oligos affecting ribosome function scored as epistatic in the screen (Table 1; Fig. 1), contrary to the expectation that inhibition of the ribosome should prevent the buildup of cytotoxic proteins caused by proteasome inhibition. This result suggests that inhibition of ribosome function induces a stress similar to inhibition of proteasome function. Indeed, an epistatic interaction between protein translation and the ubiquitin-proteasome system has been shown previously in yeast; inhibition of either results in the toxic depletion of free ubiquitin pools (14). Three of the genes with suppressor oligos are involved in translation initiation, EIF4E, EIF4G1,andPPP1R15B (CReP, constitutive repressor of eIF2α phosphorylation), suggesting that the activity of these genes is important for induction of cell death by bortezomib. eIF4E binds the cap structure of Figure 5. Bortezomib potentiates DNA damage signaling. Regulation mRNA and, in a complex with eIF4A and eIF4G, recruits of proteins was determined in the same manner as Fig. 2. Markers of DNA the mRNA to the ribosome. mTOR promotes protein synthe- damage response, including RAD51 stabilization and phosphorylation sis in part by phosphorylating 4EBP1, reducing its ability to of H2AFX and CHEK2 at their ATM sites, are increased following compete with eIF4G binding to eIF4E. HCT-116 cells are re- bortezomib, whereas CHEK1 phosphorylation at its ATR site is upregulated more modestly. Monoubiquitination of FANCD2 (top band) sistant to mTOR inhibition due to downregulation of 4EBP1 is increased relative to the unmodified FANCD2 (bottom band) beginning protein levels, reducing its ability to inhibit eIF4E (15). Our at 12 h. MDC1, the three closely spaced bands above 247 kDa, is not data suggest that siRNA against eIF4E could reduce eIF4E le- clearly differentially regulated between vehicle and bortezomib treatment, vels sufficiently that 4EBP1 would be capable of inhibiting contrary to prior reports (31). Neither full-length nor separase-cleaved protein translation. This inhibition would prevent protein ac- fragments of RAD21 increase following bortezomib treatment. Full-length blots are presented in Supplementary Fig. S8, along with theoretical cumulation following proteasome inhibition, rescuing the molecular weights of the proteins. cells from bortezomib-induced apoptosis. Therefore, the EIF4E and EIF4G1 hits suggest that bortezomib induces studies. Our results do support some previously suggested mTOR inhibition but that, in HCT-116 cells, this inhibition mechanisms, including key roles for ER stress, p53, and is unable to inhibit protein synthesis. In fact, bortezomib Myc. Surprisingly, we do not see evidence for the critical in- treatment does shift 4EBP1 to the inhibitory hypophosphory- volvement of ROS, NFκB, activator protein-1, or cell cycle lated form (Fig. 3A). Phosphorylation of S6K is also inhibited, proteins in HCT-116. We have provided some initial RNAi- consistent with mTOR inhibition. These data agree with a re- independent data to substantiate suggested pathway involve- cent publication (16). Many tumor cells have dysregulated ment. Below, we suggest the context within which these protein translation inhibition (17). In contrast, normal cells interactions are occurring. Additional modules are explored should respond to this bortezomib-induced mTOR inhibition in Supplementary Results. Specifically, regulation of apopto- with a reduction in protein translation, suggesting a rationale tic proteins by bortezomib is presented in Supplementary for the therapeutic window of bortezomib. Fig. S4. In addition, we show that BNIP3 stabilization does The other suppressor hit, PPP1R15B, is the constitutive not seem to result in ROS or mitophagy as shown in Supple- form of the inducible PPP1R15A (Gadd34). Although we were mentary Figs. S4 to S7. Although not all phenotypes that bor- not able to evaluate PPP1R15B levels, PPP1R15A was in- tezomib might affect (growth arrest, apoptosis, necrosis, creased quite dramatically (Fig. 3C). Both are regulatory sub- autophagy, and senescence) will be equally captured by the units for protein phosphatase 1, targeting EIF2S1 (epistatic ATPlite viability assay used in this screen, these results rep- gene, also known as eIF2α) for dephosphorylation. Phosphor- resent a genome-wide evaluation of genetic sensitivity within ylation of EIF2S1 results in general translation inhibition, al- the inherent biases of that assay. though some stress response mRNAs are still translated (17). It is possible that the translation of these stress mRNAs is Proteasome subunits involved in the induction of apoptosis. Salubrinal, an inhibi- Knocking down the 20S core subunits of the proteasome tor of EIF2S1 dephosphorylation, promoted cell death follow- gave a synthetic lethal phenotype (Table 1; Fig. 1). These data ing bortezomib treatment (18). Thus, phosphorylated EIF2S1 indicate that the function of the 20S core subunits compen- is important for induction of cell death by bortezomib. sates for bortezomib-induced cell death. RNAi against 20S In many multiple myeloma cell lines, bortezomib induces proteasome subunits reduces the number of proteasome ac- a robust and long-lived phosphorylation of EIF2S1 (19),

www.aacrjournals.org Cancer Res; 70(11) June 1, 2010 OF7

Chen et al.

consistent with this tumor type being particularly sensitive following DNA damage by another agent. We suggest that to bortezomib. bortezomib also blocks DNA damage repair of double-stranded Under hypoxic conditions, phosphorylation of EIF2S1 pro- breaks occurring during normal replication. vides acute and general translation inhibition, but this is lost A recent phase 3 trial in multiple myeloma showed that following the induction of PPP1R15A. More chronic transla- the combination of bortezomib with melphalan-prednisone tion inhibition is provided by inhibition of the mTOR path- was superior to melphalan-prednisone alone (32). Melphalan way, possibly via upregulation of BNIP3 or DDIT4 (Redd1; resistance is dependent on HR (33, 34). The ability of borte- ref. 17). Recent work suggests that induction of DDIT4 is im- zomib to inhibit HR could help to explain the clinical supe- portant for bortezomib resistance in multiple myeloma (20). riority of this combination, in agreement with recent studies In this way, bortezomib mimics hypoxia and hypoxia resis- in multiple myeloma cells (35), although we did not see re- tance may predict sensitivity to bortezomib. ductions in FANCD2 protein levels. In conclusion, the genome-wide RNAi study of the deter- Myc minants of bortezomib sensitivity suggests multiple mechan- Myc stabilization and its induction of PMAIP1 (Noxa) have isms by which proteasome inhibition results in cell death. previously been proposed to be critical for cell death follow- Ribosome function is inhibited following proteasome inhibi- ing proteasome inhibition (21). Myc, ODC1, and PMAIP1 are tion, particularly translation initiation. Tumor cells have dys- substantially increased by proteasome inhibition (Fig. 2). regulated translation inhibition and so are not fully protected ODC1 (ornithine decarboxylase) is both a known substrate by this homeostatic response. Myc stabilization promotes of the proteasome and relatively strongly induced target of protein translation, inducing further stress in the tumor cell, Myc (22, 23). Surprisingly, the oligo targeting SRM (spermi- as well as inducing the accumulation of putrescine. BNIP3 is dine synthase) is synthetically lethal with bortezomib, sug- also stabilized, although the downstream effectors are un- gesting that the intermediate between ODC1 and SRM (i.e., clear. The repair of DNA damage, probably originating from putrescine) is increased by bortezomib and, at least in this normal S-phase replication, is inhibited. Stabilization of p53 setting, toxic. The oligo targeting EIF5A, which is modified promotes cell death. ER stress also seems to contribute to by spermidine (24), is also synthetically lethal with bortezo- cell death. Because multiple homeostatic responses are in- mib, suggesting that putrescine affects the modification of hibited, the tumor cell engages programmed cell death. elF5A as well. We have observed a significant increase in pu- Our initial data are in support of these mechanistic hypoth- trescine levels in bortezomib-treated cells (Fig. 4D), followed eses and further studies will be needed to confirm these in- by a significant drop in spermidine levels at 24 hours. Protea- sights. We anticipate that such studies will improve the some inhibition by MG-132 or lactacystin has previously selection of new targets, the design of pharmacodynamic bio- been shown to prevent modification of eIF5A by spermidine markers, and the application of combination chemotherapy (25). Excess putrescine accumulation has been shown to in- regimens containing bortezomib. hibit the modification of eIF5A (26), suggesting that bortezomib blocks eIF5A modification by increasing putrescine Disclosure of Potential Conflicts of Interest levels. Myc, which regulates 15% of the genome (27), controls All authors were employed by Millennium Pharmaceuticals at the time of multiple steps within protein translation, including ribosome their contribution to this work. In addition, E.S. Lightcap receives royalties from Velcade sales as part of the LeukoSite agreement around the purchase biogenesis and translation initiation, glycolysis, mitochondrial of ProScript, Inc. homeostasis, polyamine biosynthesis, and iron metabolism (23). Myc has also been shown to cooperate with eIF4E in Acknowledgments transformation (28). Therefore, stabilization of Myc may play a role in all of the modules identified in this study and may We thank Ben Amidon, Jeff Ecsedy, Mark Rolfe, and Petter Veiby for their thus be a central player within bortezomib-induced death of careful review of this document and Kenneth M. Gigstad and Christopher Blackburn for generating ML912. HCT-116 cells.

Homologous recombination Grant Support Many genes associated with HR have oligos that give sig- Millennium Pharmaceuticals, Inc. nificant phenotypes within the screen. This result suggests The costs of publication of this article were defrayed in part by the payment that the function of HR is important to cell death following of page charges. This article must therefore be hereby marked advertisement in bortezomib treatment. We have shown that bortezomib po- accordance with 18 U.S.C. Section 1734 solely to indicate this fact. tentiates DNA damage signaling as a single agent (Fig. 3). Pri- Received 12/07/2009; revised 02/18/2010; accepted 03/24/2010; published or studies (29–31) have shown that bortezomib can block HR OnlineFirst 05/11/2010.

References 1. Orlowski RZ, Kuhn DJ. Proteasome inhibitors in cancer therapy: 3. Nencioni A, Grünebach F, Patrone F, Ballestrero A, Brossart P. Pro- lessons from the first decade. Clin Cancer Res 2008;14:1649–57. teasome inhibitors: antitumor effects and beyond. Leukemia 2007; 2. McConkey DJ, Zhu K. Mechanisms of proteasome inhibitor action 21:30–6. and resistance in cancer. Drug Resist Updat 2008;11:164–79. 4. Fribley A, Wang CY. Proteasome inhibitor induces apoptosis through

OF8 Cancer Res; 70(11) June 1, 2010 Cancer Research

Proteasome Inhibition RNAi Screen

induction of endoplasmic reticulum stress. Cancer Biol Ther 2006;5: induce a terminal unfolded protein response in multiple myeloma 745–8. cells. Blood 2006;107:4907–16. 5. Iorns E, Lord CJ, Turner N, Ashworth A. Utilizing RNA interference to 20. Decaux O, Clement M, Magrangeas F, et al. Inhibition of mTORC1 enhance cancer drug discovery. Nat Rev Drug Discov 2007;6:556–68. activity by REDD1 induction in myeloma cells resistant to bortezomib 6. Whitehurst AW, Bodemann BO, Cardenas J, et al. Synthetic lethal cytotoxicity. Cancer Sci 2009;101:889–97. screen identification of chemosensitizer loci in cancer cells. Nature 21. Nikiforov MA, Riblett M, Tang WH, et al. Tumor cell-selective regu- 2007;446:815–9. lation of NOXA by c-MYC in response to proteasome inhibition. Proc 7. Fotheringham S, Epping MT, Stimson L, et al. Genome-wide loss-of- Natl Acad Sci U S A 2007;104:19488–93. function screen reveals an important role for the proteasome in 22. Adhikary S, Eilers M. Transcriptional regulation and transformation HDAC inhibitor-induced apoptosis. Cancer Cell 2009;15:57–66. by Myc proteins. Nat Rev Mol Cell Biol 2005;6:635–45. 8. Tsui M, Xie T, Orth JD, et al. An intermittent live cell imaging screen 23. Schmidt EV. The role of c-myc in regulation of translation initiation. for siRNA enhancers and suppressors of a kinesin-5 inhibitor. PLoS Oncogene 2004;23:3217–21. One 2009;4:e7339. 24. Hyvönen MT, Keinänen TA, Cerrada-Gimenez M, et al. Role of 9. Papandreou CN, Daliani DD, Nix D, et al. Phase I trial of the protea- hypusinated eukaryotic translation initiation factor 5A in polyamine some inhibitor bortezomib in patients with advanced solid tumors depletion-induced cytostasis. J Biol Chem 2007;282:34700–6. with observations in androgen-independent prostate cancer. J Clin 25. Jin BF, He K, Wang HX, et al. Proteomic analysis of ubiquitin- Oncol 2004;22:2108–21. proteasome effects: insight into the function of eukaryotic initia- 10. Crawford LJ, Walker B, Ovaa H, et al. Comparative selectivity and tion factor 5A. Oncogene 2003;22:4819–30. specificity of the proteasome inhibitors BzLLLCOCHO, PS-341, 26. Tome ME, Fiser SM, Payne CM, Gerner EW. Excess putrescine ac- and MG-132. Cancer Res 2006;66:6379–86. cumulation inhibits the formation of modified eukaryotic initiation 11. NawrockiST,CarewJS,DunnerK,Jr.,etal.Bortezomibinhibits factor 5A (eIF-5A) and induces apoptosis. Biochem J 1997;328: PKR-like endoplasmic reticulum (ER) kinase and induces apoptosis 847–54. via ER stress in human pancreatic cancer cells. Cancer Res 2005;65: 27. Dang CV, O'Donnell KA, Zeller KI, Nguyen T, Osthus RC, Li F. The 11510–9. c-Myc target gene network. Semin Cancer Biol 2006;16:253–64. 12. Ding WX, Ni HM, Yin XM. Absence of Bax switched MG132-induced 28. Ruggero D, Montanaro L, Ma L, et al. The translation factor eIF-4E apoptosis to non-apoptotic cell death that could be suppressed promotes tumor formation and cooperates with c-Myc in lymphoma- by transcriptional or translational inhibition. Apoptosis 2007;12: genesis. Nat Med 2004;10:484–6. 2233–44. 29. Jacquemont C, Taniguchi T. Proteasome function is required for 13. Uechi T, Tanaka T, Kenmochi N. A complete map of the human DNA damage response and Fanconi anemia pathway activation. ribosomal protein genes: assignment of 80 genes to the cytogenetic Cancer Res 2007;67:7395–405. map and implications for human disorders. Genomics 2001;72: 30. Murakawa Y, Sonoda E, Barber LJ, et al. Inhibitors of the proteasome 223–30. suppress homologous DNA recombination in mammalian cells. 14. Hanna J, Leggett DS, Finley D. Ubiquitin depletion as a key mediator Cancer Res 2007;67:8536–43. of toxicity by translational inhibitors. Mol Cell Biol 2003;23:9251–61. 31. Shi W, Ma Z, Willers H, et al. Disassembly of MDC1 foci is controlled 15. Dilling MB, Germain GS, Dudkin L, et al. 4E-binding proteins, the by ubiquitin-proteasome-dependent degradation. J Biol Chem 2008; suppressors of eukaryotic initiation factor 4E, are down-regulated 283:31608–16. in cells with acquired or intrinsic resistance to rapamycin. J Biol 32. San Miguel JF, Schlag R, Khuageva NK, et al. Bortezomib plus mel- Chem 2002;277:13907–17. phalan and prednisone for initial treatment of multiple myeloma. 16. Wu WK, Volta V, Cho CH, et al. Repression of protein translation and N Engl J Med 2008;359:906–17. mTOR signaling by proteasome inhibitor in colon cancer cells. Bio- 33. Chen Q, Van der Sluis PC, Boulware D, Hazlehurst LA, Dalton WS. chem Biophys Res Commun 2009;386:598–601. The FA/BRCA pathway is involved in melphalan-induced DNA inter- 17. Wouters BG, Koritzinsky M. Hypoxia signalling through mTOR and strand cross-link repair and accounts for melphalan resistance in the unfolded protein response in cancer. Nat Rev Cancer 2008;8: multiple myeloma cells. Blood 2005;106:698–705. 851–64. 34. Wang ZM, Chen ZP, Xu ZY, et al. In vitro evidence for homologous 18. Schewe DM, Aguirre-Ghiso JA. Inhibition of eIF2α dephosphorylation recombinational repair in resistance to melphalan. J Natl Cancer Inst maximizes bortezomib efficiency and eliminates quiescent multiple 2001;93:1473–8. myeloma cells surviving proteasome inhibitor therapy. Cancer Res 35. Yarde DN, Oliveira V, Mathews L, et al. Targeting the Fanconi anemia/ 2009;69:1545–52. BRCA pathway circumvents drug resistance in multiple myeloma. 19. Obeng EA, Carlson LM, Gutman DM, et al. Proteasome inhibitors Cancer Res 2009;69:9367–75.

www.aacrjournals.org Cancer Res; 70(11) June 1, 2010 OF9

Genome-Wide siRNA Screen for Modulators of Cell Death Induced by Proteasome Inhibitor Bortezomib

Siquan Chen, Jonathan L. Blank, Theodore Peters, et al.

Cancer Res Published OnlineFirst May 11, 2010.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-09-4428

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2010/05/10/0008-5472.CAN-09-4428.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/early/2010/05/11/0008-5472.CAN-09-4428. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.