Small Interfering RNA Screens Reveal Enhanced Cisplatin Cytotoxicity in Tumor Cells Having both BRCA Network and TP53 Disruptions

RNA interference technology allows the systematic genetic analysisof the molecular alterations in cancer cells and how these alterationsaffect response to therapies. Here we used small interferingRNA (siRNA) screens to identify genes that enhance the cytotoxicity(enhancers) of established anticancer chemotherapeutics. Hitsidentified in drug enhancer screens of cisplatin, gemcitabine,and paclitaxel were largely unique to the drug being testedand could be linked to the drug's mechanism of action. Hitsidentified by screening of a genome-scale siRNA library forcisplatin enhancers in TP53-deficient HeLa cells were significantlyenriched for genes with annotated functions in DNA damage repairas well as poorly characterized genes likely having novel functionsin this process. We followed up on a subset of the hits fromthe cisplatin enhancer screen and validated a number of enhancerswhose products interact with BRCA1 and/or BRCA2. TP53^+/–matched-pair cell lines were used to determine if knockdownof BRCA1, BRCA2, or validated hits that associate with BRCA1and BRCA2 selectively enhances cisplatin cytotoxicity in TP53-deficientcells. Silencing of BRCA1, BRCA2, or BRCA1/2-associated genesenhanced cisplatin cytotoxicity $~$ 4- to 7-fold more in TP53-deficientcells than in matched TP53 wild-type cells. Thus, tumor cellshaving disruptions in BRCA1/2 network genes and TP53 togetherare more sensitive to cisplatin than cells with either disruptionalone.

INTRODUCTION

Top

Introduction

Gene silencing by RNA interference is a powerful genetic toolto identify genes involved in specific biological processesin model organisms and human cells (3, 14, 18, 38, 42, 56).It is now possible to perform unbiased genetic screens usingcultured human cells. Recently, small-scale siRNA (small interferingRNA) and shRNA (small hairpin RNA) screens in human cells havesuccessfully identified genes that modulate cell growth, apoptosis,chemoresistance, and chemosensitivity (4, 44, 52). While establishingproof-of-concept, most of these early screens tested gene setstoo limited in size to provide a comprehensive view of the genesinvolved in the cellular phenotypes being tested. Several groupshave employed nearly genome-wide screens of shRNAs expressedfrom vectors to identify genes that modify certain phenotypes(6, 40, 49, 59). To date, these shRNA screens have been limitedto phenotypes amenable to positive selection, but other approacheslike bar code screening have shown promise (6). Systematic well-by-welltesting of genomic-scale siRNA or shRNA libraries for modulationof nonselectable phenotypes in human cells has not been widelyattempted.

Cancer cells undergo numerous genetic changes that drive cellulartransformation from normal cell progenitors. Anticancer drugtarget discovery is now frequently directed toward understandingand exploiting genetic alterations that exist in tumor cells.Knowledge of genetic alterations in tumors may lead to betteruse of conventional therapeutics or development of new therapeuticsthat offer better therapeutic windows. Loss-of-function geneticscreens can identify genes whose loss of function enhances cytotoxicityof chemotherapeutics in cells with defined genetic lesions.

The TP53 transcription factor is a central mediator of the cellularresponse to DNA damage and a variety of other stresses and isone of the most frequently mutated genes in human cancers (19,57). TP53 status can also effect the chemosensitivity of tumorcells (15). The identification of genes that, when silenced,selectively enhance the chemosensitivity of TP53 mutant cancercells but not TP53 wild-type cells would make attractive drugtargets. Drugs developed to these genes have the potential toselectively increase the toxicity of the chemotherapeutic inthe cancer cell.

Genetic screens in model organisms have led to identificationof drug enhancers whose loss increases sensitivity to the drug(43, 54, 56, 61). We hypothesized that identification of enhancersfor commonly used cancer therapeutics would lead to new strategiesfor therapeutic application of these drugs and/or identificationof new targets for the development of combination chemotherapies.Here we used small- and genome-scale siRNA screens to identifygenes that enhance the sensitivity of human tumor cells to thecancer chemotherapeutic cisplatin. The screen hits includedknown genes that function in DNA repair, cell cycle checkpoints,and survival signaling, as well as genes with no annotated functions.Selected hits from these screens were validated and then testedon matched TP53-positive and TP53-deficient cells to identifyenhancers selective for tumor cells that have lost TP53 function.We show that silencing of BRCA1, BRCA2, and certain genes whoseproducts interact with BRCA1/2, selectively enhances cisplatincytotoxicity in TP53-deficient cells.

MATERIALS AND METHODS

Top

Materials and Methods

Cell lines and chemicals. HeLa, A549, and TOV21G cells were obtained from the AmericanType Culture Collection, Rockville, MD (catalogue numbers CCL-2,CCL-185, and CRL-11730, respectively). Isogenic pairs of TP53-negativeand -positive A549 and TOV21G cells were generated by transfectionwith a vector expressing an shRNA targeting TP53 (7) or an emptyvector, respectively. Stably transfected lines were selectedfor drug resistance and phenotyped for the functional statusof the TP53 DNA damage checkpoints(s). Cells designated "TP53positive" retained wild-type TP53 function (i.e., showed G₁cell cycle arrest after treatment with doxorubicin). Cells designated"TP53 negative" lacked the TP53 G₁ checkpoint following doxorubicintreatment and showed decreased expression of TP53 target genes(see Fig. S1 in the supplemental material). Cell lines are maintainedin exponential growth in Dulbecco's modified Eagle's mediumsupplemented with 10% fetal bovine serum (Invitrogen, Carlsbad,CA). Paclitaxel and cisplatin were obtained from EMD Biosciences,Inc. (San Diego, CA).

siRNA design and siRNA library composition. siRNA sequences were designed with an algorithm developed toincrease efficiency of the siRNAs for silencing while minimizingtheir off-target effects (33). siRNAs were ordered from Sigma-Proligo(The Woodlands, TX). The siRNA library targets $~$ 20,000 uniquehuman genes, with three siRNAs per gene. The library representsgenes comprising the druggable genome (30), membrane proteins,enzymes, pathways of therapeutic interest, and RefSeq (releases6 to 8; http://ncbi.nih.gov/RefSeq/). The sequences of siRNAduplexes used for follow up experiments are listed in TableS1 in the supplemental material.

siRNA transfection. HeLa cells were transfected with siRNAs using Oligofectamine(Invitrogen, Carlsbad, CA). For screening, three siRNAs targetingthe same gene were pooled at equal molarity (final concentrationof each siRNA, 17 nM; total siRNA concentration, 50 nM). TOV21Gcells and A549 cells were transfected with Oligofectamine andSilentFect (Bio-Rad, Hercules, CA), respectively. Semiautomatedtransfections on HeLa cells were carried out as follows. Oneday prior to transfection, cells were seeded in 384-well tissueculture plates (Matrix Technologies, Hudson, NH) at 600 cells/wellin 50 µl/well of complete medium using a WellMate (MatrixTechnologies, Hudson, NH). On the day of transfection, 18 µlof OptiMEM (Invitrogen, Carlsbad, CA) was added to siRNA poolsthat were preplated in a 384-well plate with 2 µl of 10µM siRNA pool per well. In a separate tube, 1.75 ml ofOligofectamine was mixed with OptiMEM to a final volume of 35ml, the mix was poured into a reservoir, and 20 µl ofthe mixture was then added to each well of the siRNA plate usinga Biomek FX laboratory automation workstation (Beckman Coulter,Inc., Fullerton, CA). Reagent/siRNA complexes were allowed toform for 15 to 20 min at room temperature. A Biomek FX was thenused to transfer 5 µl of this siRNA/reagent mixture toeach cell plate well. The plates were incubated for 4 h at 37°Cin a 5% CO₂ incubator. An aliquot of 10 µl/well of Dulbecco'smodified Eagle's medium-10% fetal bovine serum with or withoutdrug was added to each well to achieve the desired final concentrationof each agent. The plates were incubated at 37°C and 5%CO₂ for another 68 h before cell viability was determined. Fullyautomated transfections were performed in the same way exceptthat a robotic arm performed all plate transfers.

Cell viability assays. A cell viability assay was determined using AlamarBlue reagent(BioSource International, Camarillo, CA). Medium from 384-wellplates was removed and replaced with 25 µl/well completemedium containing 10% (vol/vol) AlamarBlue and 1/100 volumeof 1 M HEPES buffer, pH 7.4. The plates were then incubatedfor 1 to 4 h at 37°C before fluorescence was measured (544-nmexcitation, 590-nm emission) with an Envison multilabel platereader (PerkinElmer, Wellesley, MA). The fluorescence signalwas corrected for background (no cells). Cell growth was expressedas percent viability relative to the median value of wells transfectedwith an siRNA to luciferase. EC₅₀s (50% effective concentrations)were calculated using GraphPad Prism software (San Diego, CA).

Cell cycle analysis. Transfections were performed in six-well plates (50,000 cells/well).At 24 h after transfection, the supernatant from each well wascombined with cells harvested from each well by trypsinization.Cells were collected by centrifugation at 1,200 rpm for 5 min,fixed with ice cold 70% ethanol for $~$ 30 min, washed with phosphate-bufferedsaline, and resuspended in 0.5 ml of phosphate-buffered salinecontaining propidium iodide (10 µg/ml) and RNase A (1mg/ml). After a final incubation at 37°C for 30 min, cellswere analyzed by flow cytometry using a FACSCalibur flow cytometer(Becton Dickinson). Data were analyzed using FlowJo software(Tree Star, Ashland, OR).

Quantitative PCR. mRNA silencing was quantified by real-time PCR using an ABIPRISM 7900HT sequence detection system and Assays-on-Demandgene expression products (Applied Biosystems, Foster City, CA).The mRNA value for each gene was normalized relative to humanGUSB (catalogue no. 431088E) mRNA levels in each RNA sample.The following Assays-on-Demand reagents were used in this study:CHEK1 (Hs00176236_m1), BRCA1 (Hs00173233_m1), BRCA2 (Hs00609060_m1),BARD1 (Hs00184427_m1), RAD51 (Hs00153418_m1), SHFM1 (Hs00428232_m1),BCL2L1 (Hs00236329_m1), DVL2 (Hs00182901_m1).

RESULTS

Top

Results

siRNA library screens identify enhancers of chemotherapeutic cytotoxicity. Initial efforts were focused on validating an siRNA screeningplatform for drug enhancer screens in human tumor cells. Todo this, we screened an siRNA library targeting 2,400 genesto identify genes that enhanced sensitivity of HeLa cells totreatment with cisplatin, gemcitabine (Gemzar), or paclitaxel,anticancer drugs that affect different phases of the cell cycle(Fig. 1A). Cisplatin cross-links DNA and disrupts the S phaseof the cell cycle. Gemcitabine is a cytidine analog whose incorporationinto DNA results in chain termination and affects cell growthnear the G₁/S transition. Paclitaxel disrupts mitosis (M phase)by preventing depolymerization of microtubules. The experimentaldesign of our enhancer screens is outlined in Fig. 1B. HeLacells were plated and then transfected using a semiautomatedsystem 24 h later. Drugs were added (no drug, an $~$ EC₁₀ dose,or an $~$ EC₅₀ dose) 4 h posttransfection, and cell viability wasmeasured at 72 h posttransfection. The addition of drug at 4h posttransfection was selected because this allowed adequatetime for transfection to occur without possible interferencefrom drug effects and because it streamlined semiautomated andfully automated screening work flows. As positive controls forour viability assays, we included siRNAs targeting the polo-likekinase, which inhibited cell growth $≥$ 90% during a 72-h growthassay.

View larger version (9K):
[in this window]
[in a new window]

FIG. 1. Experimental design for identification of enhancers/suppressors of DNA damage. (A) Schematic view of the cell cycle and sites of action of selected anticancer drugs. Dotted horizontal lines indicate approximate phases of the cell cycle affected by the anticancer drugs gemcitabine, cisplatin, and paclitaxel. (B) Schematic view of the design and timing of siRNA screens.

The results of representative enhancer screens performed inHeLa cells with each drug are shown in Fig. 2A, B, and C. Thepercent viability of cells transfected with the siRNA pool inthe absence of drug is compared to the percent viability ofcells transfected with the siRNA pool in the presence of drug.The red diagonal lines on each plot represent a twofold enhancementor suppression of the growth inhibition by each drug. Hits ineach of the screens were defined as mean log₂ (% viability plusdrug/% viability no drug) – 2 standard deviations (SD).The overlap in hits between replicate screens for each drugis discussed further below. A few of the siRNA pools that reproduciblyand uniquely enhanced sensitivity to each of the drugs are selectedfor further discussion, and these genes are highlighted on eachof the plots in Fig. 2.

View larger version (31K):
[in this window]
[in a new window]

FIG. 2. Identification of genes that enhance cisplatin, gemcitabine, and paclitaxel cytotoxicity. HeLa cells grown in 384-well plates were transfected with individual siRNA pools targeting $~$ 2,400 human genes (3 siRNAs/gene). Four hours posttransfection, cells were treated with (y axis) or without (x axis) the following drugs: (A) cisplatin, 100 ng/ml; (B) gemcitabine, 7 nM; (C) paclitaxel, 1.2 nM. Cell growth was then measured at 72 h posttransfection using an Alamar blue assay and is expressed as the percentage of control viability (i.e., viability relative to wells transfected with an siRNA to luciferase). Diagonal red lines indicate twofold enhancement or suppression by drug treatment. Selected genes, which are labeled black, blue, or red, preferentially enhance cell killing by gemcitabine, paclitaxel, or cisplatin, respectively. (D) One-dimensional clustering of the overlap in consensus hits from cisplatin (Cis.), gemcitabine (Gem.), and paclitaxel (Tax.) siRNA screens. Consensus hits were defined as mean log₂ (% viability plus drug/% viability with no drug) – 2 standard deviations in at least 2 of 3 screens for cisplatin and gemcitabine. For paclitaxel, consensus hits were defined as being as 1.7 standard deviations from the mean in two of two screens. Red color indicates the pool scored as a hit, and the gray color indicates that the pools were represented on the hit list.

Noteworthy among the genes that enhanced growth inhibition bycisplatin were numerous genes involved in DNA repair or checkpointactivation. Some of these genes have scored as cisplatin enhancersin model organism screens. For instance, disruption of REV1(REV1L) (not highlighted; see also Table S2 in the supplementalmaterial) enhanced cisplatin cytotoxicity in Saccharomyces cerevisiae(61). Silencing of checkpoint genes, ATR and CHEK1, also enhancedsensitivity to cisplatin in our experiments. ATR and CHEK1 kinaseshave well-characterized roles in activating cell cycle checkpointsin response to DNA damage (1, 5). In addition, there may bedirect functional interactions between ATR/CHEK1 and the BRCApathway (62). However, the most prominent genes scoring as cisplatinenhancers are members of the BRCA pathway (BRCA1, BARD1, BRCA2,and RAD51, flagged in Fig. 2A).

The gemcitabine screen yielded a largely different set of hitsthan the cisplatin screen. Silencing of components of the BRCApathway did not enhance sensitivity to the replication stresscaused by gemcitabine. This likely reflects the use of differentDNA repair pathways to different types of DNA damage. One stronggemcitabine enhancer was the ribonucleotide reductase subunitM1 (RRM1). The activity of RRM1 is specifically targeted bya cellular metabolite of gemcitabine (53). Enhanced sensitivityto drugs following disruption of their target is seen in S.cerevisiae drug enhancer screens (43). Flagged in Fig. 2B aretwo pools targeting RRM1. Both pools enhanced gemcitabine cytotoxicity,but one pool was more effective at inhibiting cell growth thanthe second pool in the absence of drug. This is likely do todifferences in target mRNA knockdown, but we have not directlytested this hypothesis. A few gemcitabine enhancers were alsocisplatin enhancers, most notably CHEK1 (Fig. 2B).

The third drug screened was paclitaxel, which would not be expectedto initiate a direct DNA damage signal unlike gemcitabine orcisplatin. The mitotic kinase genes STK6 (AURKA, Aurora A) andBUB1 enhanced growth inhibition by paclitaxel when silenced(Fig. 2C). Our demonstration that STK6 silencing enhances sensitivityto paclitaxel is consistent with a report showing that overexpressionof STK6 increases resistance to paclitaxel (2) and confirmsa more recent report demonstrating that siRNAs targeting STK6sensitize pancreatic tumor cell lines to taxanes (29) Thus,silencing mitotic checkpoint genes enhances sensitivity to agentsthat act in mitosis, whereas silencing DNA damage checkpointgenes enhances sensitivity to DNA damaging agents that act inG₁ or S.

We did not find a substantial common set of enhancers for thedrugs we screened, as observed in S. cerevisiae drug enhancerscreens (50). Examples of genes that enhanced responses to multipledrugs in yeast include those that maintain integrity of thecell wall and ion channels and pumps that may regulate druguptake or export (50). A pairwise comparison of the hits definedby using a mean log₂ (% viability plus drug/% viability no drug)– x SD method revealed generally more significant overlap(P value <1 x 10^–6) in the hits from replicate screensperformed with the same drug than in hits from screens performedwith different drugs (Table 1). When consensus hits from eachdrug screen were compared then the hits were almost exclusivelydrug specific. Shown in Fig. 2D is the overlap in hits thatscored in two of three repeats of the cisplatin or gemcitabinescreens or 2 of 2 paclitaxel screens The consensus hit listfor each drug is provided in Table S2 in the supplemental material.There was one gene (CHEK1) in common between the cisplatin andgemcitabine lists, and one gene (CARS) in common between thegemcitabine and paclitaxel lists. Our failure to find a commonset of hits from all drugs in our screens may reflect the limitedsize of the siRNA library tested or the particular drugs tested.Taken together, however, our findings show that the siRNA screensyielded hits that were largely specific for each drug and includedgenes linked to the drug's mechanism of action.

View this table:
[in this window]
[in a new window]

TABLE 1. Hits from replicate screens of the same drug show more significant overlap than hits from screens comparing different drugs

Genome-scale siRNA screens identify BRCA1/2 pathway genes as enhancers of cisplatin cytotoxicity. To extend the identification of genes that enhance cisplatincytotoxicity to the complete human genome, we constructed agenome-scale siRNA library and developed a fully automated transfectionsystem to perform genome-scale screens. Our siRNA library comprisedsiRNA pools targeting $~$ 20,000 genes (3 siRNAs/pool). We screenedthis library for cisplatin enhancers using the same experimentalprotocols described above for the data in Fig. 2 but using afully automated platform. Hits from the primary screen wereselected based on a mean log₂ (% viability of drug treated/%viability of no drug) – 2 SD selection criteria (Fig.3A) and run in a confirmation screen in triplicate under thesame conditions as the primary screen (Fig. 3B). For follow-upexperiments, we selected the top-scoring hits from the confirmationscreen [defined by mean log₂ (% viability of drug treated/%viability of no drug) – 2 SD] (Fig. 3B). As depicted inFig. 3C, we started with over 22,000 siRNA pools and followedup on 146 siRNA pools in the assays described below.

View larger version (34K):
[in this window]
[in a new window]

FIG. 3. A genome-scale siRNA screen identifies enhancers of cisplatin cytotoxicity in HeLa cells. HeLa cells grown in 384-well plates were transfected with individual siRNA pools targeting $~$ 20,000 human genes (3 siRNAs/gene). Four hours posttransfection, cells were treated with (y axis) or without (x axis) cisplatin at 100 ng/ml. Cell growth was then measured at 72 h posttransfection using an Alamar blue assay and is expressed as the percentage of control viability (viability relative to wells transfected with an siRNA to luciferase). The log₂ (% viability plus drug/% viability with no drug) is plotted for each pool in the genome (A) or confirmation (B) screen. Pools that were 2 standard deviations (std dev) from the mean log₂ (% viability plus drug/% viability with no drug) were defined as hits and are indicated in blue. Nonscoring pools are red. (C) A "screening funnel" depicting the number of hits followed up on at each stage of the follow up process.

To confirm that siRNA pools selected as hits from the cisplatinscreen were enhancers of growth inhibition by cisplatin in HeLacells, the single siRNAs comprising the pools used in the screenwere resynthesized and the siRNA pools regenerated. We reasonedthat if silencing of the gene enhanced sensitivity to cisplatin,then the dose titration curve for cisplatin would shift to theleft (i.e., EC₅₀ shift). Therefore, we tested whether hits fromthe siRNA screens enhanced growth inhibition by cisplatin insemiautomated EC₅₀ shift assays. Cells were transfected witha constant concentration of siRNA (50 nM) and then subjectedto increasing concentrations of cisplatin (nine doses). Cellgrowth was measured in a 72-h assay, and EC₅₀s for growth inhibitionby cisplatin were calculated. The selected hits were processedin two sets. The 74 top-scoring siRNA pools in the confirmationscreen comprised one set (Fig. 4A), and the next best 72 siRNApools comprised the second set (Fig. 4B). Of the 74 top-enhancinghits selected from the cisplatin screen, 38 decreased the EC₅₀of cisplatin $≥$ 2-fold (average EC₅₀ of three replicate experiments)relative to control siRNA (Fig. 4A; see also Table S3 in thesupplemental material). Thus, $~$ 50% of the hits in this gene setwere confirmed as cisplatin enhancers in the EC₅₀ shift assays.

View larger version (23K):
[in this window]
[in a new window]

FIG. 4. Verification of siRNA pools that enhance cisplatin cytotoxicity in HeLa cells. EC₅₀ shift assay to confirm cisplatin enhancers. siRNA pools from the screen for which results are shown in Fig. 3B were further tested in the cisplatin EC₅₀ shift assay in HeLa cells. These assays tested siRNA pools for enhancement of cisplatin (cis) cytotoxicity as in Fig. 2, except drug was tested at nine cisplatin doses. HeLa cells were transfected with siRNA pools and then treated with increasing concentrations of cisplatin. Cell growth was measured at 72 h posttransfection using an Alamar blue assay and is expressed as the percentage of control viability. The top 74 hits were followed up on as one set (A), and the next 72 top scoring hits were followed up on as a second set (B). Red curves denote confirmed enhancers (i.e., siRNA pools that decreased the cisplatin EC₅₀ by $≥$ 2-fold); gray lines denote siRNA pools that decrease cisplatin cytotoxicity <2-fold; blue lines are a luciferase control siRNA. Panel A is representative of 3 repeat experiments, and panel B is representative of 2 repeat experiments. (C) Deconvolution of siRNA pools targeting BRCA pathway genes. EC₅₀ shift assays were performed as above except that we tested both the pool and the single siRNAs comprising the pool. The bar graph represents the relative decrease in the EC₅₀ of cisplatin in HeLa cells transfected with each siRNA or pool.

We tested an additional set of 72 hits in the EC₅₀ shift assay.In this set, 15 genes decreased the EC₅₀ of cisplatin by $≥$ 2-fold(average EC₅₀s of two replicate experiments) relative to controlsiRNA for a confirmation rate of $~$ 21% (Fig. 4B; see also TableS3 in the supplemental material). Therefore, weaker scoringgenes from the screen were more likely to be false-positivehits, as defined by the arbitrary cutoff criteria of a twofoldEC₅₀ shift. Thus, we confirmed that 53 pools decreased the cisplatinEC₅₀ by at least twofold. At this stage in our follow-up approach,these hits are considered confirmed, meaning the pools reproducedthe phenotype observed in the screen, but the hit is not yetvalidated, meaning that the phenotype is observed with morethan one pool or single siRNA targeting the gene. Table S3 inthe supplemental material contains the list of confirmed hitsas well as hits that were subsequently validated with multiplesiRNAs.

The list of confirmed cisplatin enhancers contained multiplemembers of the BRCA pathway as well as multiple members of theRAD6/RAD18 DNA repair pathway, suggesting an enrichment of genesthat mediate DNA repair. To examine the list of confirmed cisplatinenhancers using an unbiased analysis, we tested whether theconfirmed that the hit list from the genome screen was enrichedfor annotation terms in the Gene Ontology Biological Processhierarchy. As expected, the enhancer gene list was significantlyenriched for genes involved in DNA repair and recombination(E values between 10^–3 and 10^–5) (see Table S4 inthe supplemental material). Thus, genome-scale siRNA screenscan be used to reveal pathways that are important in regulatingthe response to a specific chemotherapeutic. In addition togenes implicated in DNA repair, the confirmed list of cisplatinenhancers contains genes involved in regulating cell cycle checkpointsand cell survival signaling and also many genes with no annotatedfunction. Indeed, 3 of the top 10 confirmed hits are genes (RFWD3,MGC33214, and C5orf5) that have no defined cellular function.

Silencing of BRCA1/2 pathway genes selectively enhances cisplatin cytotoxicity in TP53-deficient cells. We elected to follow up on the BRCA pathway genes (BRCA1, BARD1,BRCA2, RAD51, and SHFM1) in greater detail. First, because siRNAshave off-target activity (33, 35) and the experiments aboveused a pool of siRNAs to target the mRNAs, we tested the singlesiRNAs targeting each gene to determine if the phenotype ofenhancing cisplatin cytotoxicity was observed with multiplesiRNAs in the pool. For BRCA1, BARD1, BRCA2, RAD51, and SHFM1,we found that at least two of the siRNAs in the pool decreasedthe EC₅₀ of cisplatin, validating these genes as enhancers ofcisplatin cytotoxicity. In contrast, siRNAs targeting EPHA1and DVL2, genes that did not score as hits in our screens, didnot enhance cisplatin cytotoxicity (Fig. 4C). We also validatedthese genes with a second approach. We used siRNAs with chemicalmodifications on both the antisense and sense strands. Thesemodifications significantly reduce off-target silencing withonly slight reductions in on-target silencing (34). BRCA1, BRCA2,RAD51, and SHFM1 were also validated using this method withat least two chemically modified siRNAs enhancing cisplatincytotoxicity at least twofold or greater (see Table S3 in thesupplemental material). BARD1 did not have two chemically modifiedsiRNAs that enhanced cisplatin cytotoxicity at a 2-fold cutbut did have two siRNAs at 1.6-fold or better. The reduced rateof validation by the chemically modified siRNAs relative totheir unmodified counterpart is likely due to their slightlyreduced on-target silencing efficiency. Pool deconvolution andchemically modified siRNAs were also used to validate a secondDNA repair pathway that had multiple members score as hits.Multiple members of the RAD6/RAD18 pathway, which mediates translesionDNA repair (51), were validated using pool deconvolution andchemically modified siRNAs (see Table S3 in the supplementalmaterial). Thus, it is unlikely that the observed phenotypeof silencing BRCA1, BARD1, BRCA2, RAD51, and SHFM1 or membersof the RAD6/RAD18 pathway is due to siRNA off-target activity.

Having ruled out that enhancement of cisplatin cytotoxicityby members of the BRCA1/2 pathway was due to siRNA off-targetactivity, we next determined whether BRCA1, BARD1, BRCA2, RAD51,and SHFM1 would function equivalently in isogenic cell pairswith or without TP53 function. The experiments above utilizedHeLa cells, which are TP53-deficient (26, 63). We constructedisogenic cell pairs by transfecting cells having wild-type TP53(TOV21G and A549) with an empty plasmid vector or a vector directingexpression of an shRNA targeting TP53 (7). Silencing of TP53mRNA was confirmed by PCR, and functional loss of TP53 was confirmedby exposing cells to the DNA-damaging agent doxorubicin andmonitoring cell cycle progression. The levels of TP53 mRNA werereduced greater than 90% in cells containing the shRNA targetingTP53 than empty vector control cells (see Fig. S1 in the supplementalmaterial). The majority of cells transduced with the empty vectorarrest in G₁ following treatment with doxorubicin. However,the cells transduced with the shRNA targeting TP53 no longerhad an intact G₁ cell cycle checkpoint and arrested almost exclusivelyin G₂ (see Fig. S1 in the supplementa; material). In addition,gene expression profiling revealed down-regulation of TP53 targetgenes such as CDKN1A (p21) and TP53INP1 (36) in both the TOV21GTP53 shRNA line ( $~$ 3.4-fold and $~$ 4.7-fold, respectively) (see Fig.S1 in the supplemental material) and the A549 TP53 shRNA line( $~$ 2.6-fold and $~$ 3.4-fold, respectively) (see Fig. S1 in the supplementalmaterial).

siRNA pools targeting BRCA1, BARD1, BRCA2, RAD51, and SHFM1were tested in nine-point cisplatin titration experiments onTOV21G control (TP53⁺ TOV21G) and TOV21G TP53sh (TP53^–TOV21G) cells using the same method as described for Fig. 4A.We also tested CHEK1 because it is a well-characterized regulatorof DNA damage checkpoints whose disruption can selectively sensitizeTP53-deficient cells to DNA damage (41, 58). In addition, wetested BCL2L1 (BCLx), a regulator of apoptosis, and EPHA1 andDVL2, which did not enhance growth inhibition by cisplatin inHeLa cells. As shown in Fig. 5A, silencing of BRCA1, BARD1,BRCA2, RAD51, SHFM1, and CHEK1 enhanced growth inhibition bycisplatin approximately four- to sevenfold more in the TP53^–TOV21G cells than in TP53⁺ TOV21G cells. Silencing of BCL2L1enhanced growth inhibition equivalently in both TP53⁺ TOV21Gand TP53^– TOV21G cells ( $~$ 3.5-fold and 5-fold, respectively),whereas silencing EPHA1 or DVL2 did not enhance growth inhibitionby cisplatin relative to the control siRNA. The selectivityof EC₅₀ enhancement could not be attributed to differentialsilencing, since we measured equivalent mRNA knockdown acrossthe eight different genes we tested in TP53⁺ TOV21G and TP53^–TOV21G cells (Fig. 5B).

View larger version (29K):
[in this window]
[in a new window]

FIG. 5. Silencing of BRCA1 selectively enhances apoptosis in TP53^– cells in response to DNA damage. (A) Identification of BRCA pathway genes as selective cisplatin enhancers in TP53^– cells. TP53^– TOV21G or TP53⁺ TOV21G cells were transfected with siRNAs prior to treatment with increasing amounts of cisplatin (9 doses). Cell growth was measured at 72 h posttransfection. Shown is the relative decrease in cisplatin EC₅₀ caused by transfection with each siRNA pool in TP53^– TOV21G or TP53⁺ TOV21G cells. These results are representative of three experiments. (B) siRNAs exhibit similar silencing efficiency in TP53^– TOV21G and TP53⁺ TOV21G cells. TP53^– TOV21G and TP53⁺ TOV21G cells were transfected with the siRNAs used for panel A. RNA was isolated at 24 h posttransfection, and levels of target mRNAs were determined by quantitative PCR. Shown are the mRNA levels for cells transfected with the siRNA pool relative to cells transfected with a control siRNA targeting luciferase (Luc). (C) Isogenic TP53⁺ (top row) or TP53^– (bottom row) TOV21G or A549 cells were transfected with an siRNA to luciferase (left column) or BRCA1 (right column) prior to treatment with 1 µg/ml or 2 µg/ml cisplatin for TOV21G and A549 cells, respectively. At 72 h posttransfection, cells were fixed, stained with propidium iodide, and analyzed for cell cycle distribution by flow cytometry. The labeled gates indicate the percentage of sub-G₁ (dead) cells, and the arrows indicate the relative positions of the 2N and 4N cell populations.

For a subset of the genes (BRCA1, RAD51, and CHEK1) we alsotested each of the single siRNAs comprising the pool in theEC₅₀ assay on the TOV21G TP53^+/– matched pair. For eachpool, at least two of three siRNAs selectively enhanced cisplatincytotoxicity in the TP53-deficient cell line (see Fig. S2 inthe supplemental material). We also observed that each of theindividual siRNAs mediated equivalent levels of silencing ineach of the matched-pair cell lines (see Fig. S2 in the supplementalmaterial). We conclude, therefore, that enhancement of cisplatincytotoxicity in TP53^– cells by siRNAs targeting the BRCAnetwork is not likely due to siRNA off-target effects or differencesin siRNA silencing efficiency.

The AlamarBlue assays we have used did not distinguish betweenreduced cell numbers caused by cell cycle arrest or caused bycell death (cytotoxicity). To better understand the mechanismof cisplatin enhancement, we used flow cytometry to measureincreases in cell death following BRCA1 siRNA transfection andcisplatin treatment (i.e., increases in sub-G₁ amounts of DNA).As shown in Fig. 5C, silencing of BRCA1 in the presence of cisplatinincreased the amount of cell death (sub-G₁ cells) in TP53^–TOV21G cells much more than in TP53⁺ TOV21G cells. BRCA1 silencingwas equivalent in the two lines (71 to 75% silencing of BRCA1mRNA in each lines). We obtained similar results using isogenicTP53⁺ A549 and TP53^– A549 cells (Fig. 5C) (70 to 73% silencingin each line) and with siRNAs targeting BRCA2, BARD1, and RAD51(data not shown). In contrast, silencing of BCL2L1 caused equivalentincreases in the amounts of sub-G₁ DNA in both TP53⁺ and TP53^–cells following cisplatin treatment (data not shown). In additionto the TOV21G TP53 shRNA cell line and the A549 TP53 shRNA celllines used above, we also tested cell lines made TP53-deficientby overexpression of the human papillomavirus E6 oncoprotein,which targets p53 for degradation (46). These matched-pair celllines gave similar results to the TP53 shRNA matched-pair celllines (data not shown). Thus, targeting multiple componentsof the BRCA1/2 pathway selectively enhanced cytotoxicity ofcisplatin in TP53^– cells compared to TP53⁺ cells.

DISCUSSION

Top

Discussion

Although the utility of genome-scale RNA interference screensusing model organisms is well established, relatively few genome-scalescreens have been performed using human cells. Our data demonstrateseveral advantages to performing genome-scale screens in humancells. One major advantage is that genome-scale screens permitthe unbiased identification of pathways or networks involvedin a phenotype. For instance, we identified two signaling networksthat enhance sensitivity of TP53-deficient cells to cisplatin(BRCA1/2 and RAD6/RAD18 networks). Small-scale screens usingsubgenome libraries will by definition yield more biased and/orincomplete results.

Screens of genome-scale libraries can also assign function touncharacterized genes. We identified many genes with no annotatedfunction in our screens, several of which probably functionin DNA damage repair pathways (see Table S3 in the supplementalmaterial). RFWD3 contains two WD domain G-beta repeats and aC3HC4 type (RING) zinc finger, a domain associated with ubiquitinligase activity. RFWD3 is a weak homologue (reciprocal BLASTmatch score, E = 6e–05) of the yeast protein BRE1, anE3 ubiquitin ligase with Rad6p (22, 32, 60). The cisplatin enhanceractivity of RFWD3 suggests, therefore, that this protein mayfunction in the human RAD6/18 pathway. Another gene of unknownfunction on the confirmed hit list is C5ORF5, a putative GTPaseactivating protein for the Rho family (InterPro domains, RhoGTPase activation protein). The closest yeast homolog for C5ORF5is YDR379W (RGA2, reciprocal best BLAST hit, E = 1e–08).RGA2 binds to PSY2 (23, 24), a novel yeast gene conferring sensitivityto platinum agents (PSY, "platinum sensitivity") (61). In otherexperiments, we found that silencing of C5ORF5 increased theintensity of DNA damage foci containing phosphorylated H2AFX,suggesting a role for this protein in DNA damage repair (notshown). The nonannotated confirmed enhancers from our screensrequire validation, but we expect the confirmed hit list willbe a rich source of genes that have important but uncharacterizedfunctions in the cellular response to cisplatin.

Another advantage of using human cells for genome-scale screensis that they reveal pathways that may not be operative in modelorganisms. Simon et al. (54) observed in yeast drug enhancerscreens that the mechanism of action of a drug and the typeof DNA damage it induces determined which specific genetic defectsenhanced cytotoxicity of the drug being tested. We observedsimilar results in our subgenomic screens, where each of thethree drugs we tested gave a unique profile of drug enhancers,and at least some of the enhancers could be linked to the drugsmechanism of action. Similar to Simon et al. (54), we also showthat genetic disruption of multiple members in the same DNArepair pathway resulted in a similar sensitivity profile tothe chemotherapeutic agents tested. Another point of similaritybetween our observations and those of Simon et al. (54) is thatdisruption RAD6/RAD18 pathway enhances sensitivity to cisplatin.Members of this pathway that scored as hits in the genome screenincluded UBE2A (RAD6A), REV3L, REV1L, MAD2L2, and POLH (seeTable S3 in the supplemental material). Thus, there are manysimilarities in our findings and those from yeast drug enhancerscreens. However, a particularly striking difference in ourstudies and those of Simon et al. (54) was the enhancement ofcisplatin cytotoxicity by multiple members of the BRCA1/2 signalingnetwork, a network that does not exist in yeast.

BRCA1, BARD1, and BRCA2 are members of a multiprotein signalingnetwork which recruits the RAD51 recombinase to nuclear focifollowing DNA damage (12). SHFM1 interacts with BRCA2 and regulatesincorporation of RAD51 into DNA damage foci (27). One of thephenotypes of BRCA1, BRCA2, or RAD51 mutant cells is their hypersensitivityto DNA cross-linking agents such as cisplatin (11). Cisplatinmediates DNA damage by forming both intra- and interstrand cross-links.Our results suggest that genes in the BRCA1/2 molecular complexhave largely nonredundant functions critical to the cellularresponse to the cisplatin-induced DNA damage.

BRCA1 has direct and indirect interactions with many differentproteins to provide a variety of functions in cell cycle progression,DNA repair, cell cycle checkpoint activation, transcriptionalregulation, and others (48). Even in responding to DNA damage,BRCA1 differentially interacts with cellular proteins to functionin DNA repair and regulating DNA damage checkpoints (25, 48).One of the pathways that BRCA1 interacts with is the Fanconianemia (FA) pathway (a genetic cancer susceptibility syndrome)(9, 11). BRCA2 (FANCD1) deficiency defines one of the complementationgroups for FA (31). It is striking that none of the FA genesscored as enhancers in our cisplatin screens. This may indicatethat the siRNAs targeting the FA genes were not effective atsilencing the target mRNAs. However, we found that FANCD2 andFANCF transcripts were reduced by $≥$ 75% following transfectionof siRNA pools targeting these genes (data not shown). Thislevel of silencing is similar to what we measured with BCRA1and BRCA2 siRNA pools that markedly enhance cisplatin activity(Fig. 5B). However, the FA genes may require higher thresholdsof silencing to enhance cisplatin cytotoxicity. Alternatively,the FA pathway may not be as rate limiting for the responseto DNA damage induced by cisplatin in the cell lines used inthis study, or the stability of the FA proteins is such thatprotein levels were not adequately reduced in the time frameof the assays used in this study. Finally, it has been reportedthat FANC mutant cells have mild defects in homology-directedDNA repair compared to severe defects observed in BRCA1 or BRCA2mutant cells (47). The viability assay used here may not besensitive enough to detect changes that result from a mild defectiverepair phenotype. Thus, while BRCA1 and BRCA2 interact withmany proteins, only a subset of those interactions enhancedcisplatin cytotoxicity in our assays.

Yet another advantage of using human cells for genome-scalescreens is that it permits the direct testing for targets ofpotential therapeutic benefit. Hartwell et al. (28) describedstrategies for identifying drugs or drug targets that selectivelykill cells having a molecular context like one found in tumors.These approaches resemble synthetic lethal screening, a genetictechnique used in S. cerevisiae (13), which identifies mutationsindividually tolerated, but which cause lethality in combination.This approach was proposed as a method for identification ofnew anticancer targets and/or drugs (20, 21, 37).

In this study, we identified genes that selectively enhancesensitivity of TP53-deficient cells to cisplatin. While similarconceptually, our results differ in important ways from syntheticlethality in yeast. For one thing, simultaneous disruptionsof TP53 and BRCA1 are not lethal in short-term assays in theabsence of DNA damage. Another difference is that our enhancersconfer quantitative rather than qualitative differences in viabilityto TP53^– cells in the presence of cisplatin. These differencesmay result from our use of short-term assays rather than long-termgrowth used to demonstrate synthetic lethality. We do not currentlyknow how combinations of these gene disruptions and/or cisplatinaffect cell growth over a longer term. The experimental approachused in this report is not effective in longer term assays becauseof the transient silencing achieved using siRNAs.

Our results demonstrate synergistic or synthetic interactionsresulting from loss of TP53 and BRCA1/2 network genes in thepresence of DNA damage. Previous reports showed that that lossof BRCA1 function significantly enhances sensitivity of TP53^–/–mouse embryo fibroblasts to DNA-damaging agents. These studies,however, did not evaluate the effects of BRCA1 disruption onTP53⁺ and TP53^– cells or the effects of disruption ofother pathway members identified here (BRCA2, RAD51, BARD1,and SHFM1) (17, 64). Here we confirm and extend these previousfindings by demonstrating the contribution of TP53 deficiencyto the hypersensitivity of BRCA1-deficient cells to DNA-damagingagents. We also demonstrate that disruption of multiple componentsof the BRCA1/2 network besides BRCA1 confers this hypersensitivity.This may be important therapeutically, since many more tumorsexhibit signs of "BRCA-ness" than actually have mutations inBRCA1 or BRCA2 (55). Our results extend to DNA-damaging agentsother than cisplatin, since silencing of multiple members ofthe BRCA network also selectively enhance sensitivity of TP53-deficientcells to camptothecin and doxorubicin (data not shown). In addition,recent reports showed that poly(ADP-ribose)polymerase inhibitorsare selectively cytotoxic for BRCA1- or BRCA2-deficient cells(8, 16). Taken together, these findings suggest that mutationsin a subset of BRCA network genes may expose an ever wideningvariety of cancers to therapeutic sensitivity to DNA damagingagents. Our results also suggest that cells with defects inthe BRCA pathway may be more responsive to treatment with cisplatinthan BRCA1/2-proficient cells. Indeed, clinical results supportthat patients with defects in BRCA1 or BRCA2 have better responsesto platinum-based therapies than patients with BRCA1/2 intact(10, 39, 45). The ability to perform genome-scale siRNA screensin cells of different genetic backgrounds in the presence ofdifferent combinations of approved chemotherapeutic drugs mayprovide a rational way to guide clinical trials.

ACKNOWLEDGMENTS

We thank members of the following departments at Rosetta Inpharmatics:the Gene Expression Lab for automation assistance; the InformaticsDepartment for siRNA design and database development; RosettaBiosoftware for development of data analysis software; and theCancer Biology Department for invaluable assistance and support.

Rosetta Inpharmatics, LLC, is a wholly owned subsidiary of Merck& Co., Inc.

FOOTNOTES

* Corresponding author. Mailing address: Rosetta Inpharmatics, LLC, 401 Terry Ave N., Seattle, WA 98109. Phone: (206) 802-6451. Fax: (206) 802-6388. E-mail: steven_bartz{at}merck.com.

Published ahead of print on 25 September 2006.

Supplemental material for this article may be found at http://mcb.asm.org/.

These authors made equal contributions to the work.

REFERENCES

Top