Gene Expression Signature-Based Screening Identifies New Broadly Effective Influenza A Antivirals Laurence Josset1,2*, Julien Textoris3,4,5, Be´atrice Loriod3, Olivier Ferraris1, Vincent Moules1, Bruno Lina1,2, Catherine N’Guyen3, Jean-Jacques Diaz4, Manuel Rosa-Calatrava1* 1CentreNationaldelaRechercheScientifique(CNRS)FRE3011VirologieetPathologieHumaine,Universite´Lyon1,Lyon,France,2LaboratoiredeVirologieCentrede BiologieetdePathologieEst,HospicesCivilsdeLyon,Lyon,France,3InstitutNationaldelaSante´etdelaRechercheMe´dicale(INSERM)U928TechnologiesAvance´es pourleGe´nomeetlaClinique,Universite´delaMe´diterrane´e,Marseille,France,4CentreNationaldelaRechercheScientifique(CNRS)UMR5534,CentreLe´onBe´rard, CentredeGe´ne´tiqueMole´culaireetCellulaire,Universite´Lyon1,Lyon,France,5Serviced’anesthe´sieetdere´animationHoˆpitalNord,AssistancePublique-Hoˆpitauxde Marseille,Marseille,France Abstract Classical antiviral therapies target viral proteins and are consequently subject to resistance. To counteract this limitation, alternativestrategieshavebeendevelopedthattargetcellularfactors.Wehypothesizedthatsuchanapproachcouldalso beusefultoidentifybroad-spectrumantivirals.TheinfluenzaAviruswasusedasamodelforitsviraldiversityandbecause oftheneedtodeveloptherapiesagainstunpredictablevirusesasrecentlyunderlinedbytheH1N1pandemic.Weproposed toidentifyagene-expressionsignatureassociatedwithinfectionbydifferentinfluenzaAvirussubtypeswhichwouldallow theidentificationofpotentialantiviraldrugswithabroadanti-influenzaspectrumofactivity.Weanalyzedthecellulargene expressionresponsetoinfectionwithfivedifferenthumanandavianinfluenzaAvirusstrainsandidentified300genesas differentiallyexpressedbetweeninfectedandnon-infectedsamples.Themost20dysregulatedgeneswereusedtoscreen theconnectivitymap,adatabaseofdrug-associatedgeneexpressionprofiles.Candidateantiviralswerethenidentifiedby theirinversecorrelationtothequerysignature.Wehypothesizedthatsuchmoleculeswouldinduceanunfavorablecellular environment for influenza virus replication. Eight potential antivirals including ribavirin were identified and their effects were tested in vitro on five influenza A strains. Six of the molecules inhibited influenza viral growth. The new pandemic H1N1virus,whichwasnotusedtodefinethegeneexpressionsignatureofinfection,wasinhibitedbyfiveoutoftheeight identified molecules, demonstrating that this strategy could contribute to identifying new broad anti-influenza agents acting on cellular gene expression. The identified infection signature genes, the expression of which are modified upon infection,couldencodecellularproteinsinvolvedinthevirallifecycle.Thisisthefirststudyshowingthatgeneexpression- basedscreeningcanbeusedtoidentifyantivirals.Suchanapproachcouldacceleratedrugdiscoveryandbeextendedto otherpathogens. Citation:JossetL,TextorisJ,LoriodB,FerrarisO,MoulesV,etal.(2010)GeneExpressionSignature-BasedScreeningIdentifiesNewBroadlyEffectiveInfluenzaA Antivirals.PLoSONE5(10):e13169.doi:10.1371/journal.pone.0013169 Editor:Man-SeongPark,HallymUniversity,RepublicofKorea ReceivedJune2,2010;AcceptedSeptember9,2010;PublishedOctober4,2010 Copyright: (cid:2) 2010 Josset et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalauthorandsourcearecredited. Funding:ThisworkwassupportedbyCNRS(CentreNationaldelaRechercheScientifique),theHospicesCivilsdeLyon,theEuropeanVigilanceNetworkforthe ManagementofAntiviralDrugResistance(VIRGIL),andbyagrantofthe"ProgrammedeRechercheA(H1N1)"co-ordinatedbytheInstitutdeMicrobiologieet MaladiesInfectieuses(INSERM).Thefundershadnoroleinstudydesign,datacollectionandanalysis,decisiontopublish,orpreparationofthemanuscript. CompetingInterests:Theauthorshavedeclaredthatnocompetinginterestsexist. *E-mail:[email protected](LJ);[email protected](MRC) Introduction vitro and in vivo inhibition of two different cellular pathways without inducing resistance has been reported, and both are Antiviral drug development is currently based on two currently undergoing preclinical trials(recently reviewedin [4]). approaches:i)theconventionalapproachofinhibitingtheactivity Targeting cellular proteins may provide another crucial of a viral enzyme which often leads to the emergence of drug advantage: if a cellular pathway is critical to the viral cycle, resistant viruses due to viral genomic variability and ii) the more agents that target such a pathway should represent potential recent approach of targeting cellular factors that are required for broad-spectrum antivirals. The influenza virus represents a viralreplication.Indeed,codingforalimitednumberofproteins, constant threat to public health due to the emergence of new viruses hijack the cellular machinery and rely on many host viral strains and is therefore an ideal model on which to test this proteinsfortheirreplication.Themajorrecognizedadvantageof hypothesis. targeting a host factor is therefore to limit the development of Belonging to the orthomyxoviridae family, influenza viruses resistanceastheviruscannotreplaceamissingcellularprotein[1]. havegenomescomposedofsingle-strandedRNAandareclassified Suchanapproachhasbeenusedinantiretroviraltherapywiththe into three types: A, B and C according to their internal protein development of a CCR5 antagonist showing promise as an anti- sequences[5].TheinfluenzaAvirusesarefurthersubtypedbased HIV drug [2]. We have also demonstrated that this strategy is on the antigenicity of the two envelope glycoproteins hemagglu- efficient at inhibiting the replication of herpes viruses resistant to tinin(HA)[H1toH16]andneuraminidase(NA)[N1toN9].All conventional antivirals [3]. In influenza research, the effective in influenza A subtypes are endemic in aquatic birds but only two, PLoSONE | www.plosone.org 1 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion H1N1andH3N2,arepresentlycirculatingamonghumans.Since dose50%(TCID50)inMDCKcellsasdescribedinourprevious the influenza genome is segmented, two different viral strains study [15]. infectingthesamecellareabletoreassorttheirgenomicsegments. Forthemicroarrayanalysis,A549cellswereinfectedfor24hat Variability can also be due to the low fidelity of the viral RNA 37uCwithinfluenzavirusesatamultiplicityofinfection(moi)of1in polymerase, whichcausesyearlyepidemicsowingtoan antigenic DMEMsupplementedwith2mML-glutamine,100Uofpenicil- drift in glycoproteins. Novel pathogenic strains of the influenza lin/mL,100mgofstreptomycinsulfate/mLand0.5mgoftrypsin/ virus have also emerged with antigenically different HA and/or mL(infectionmedium).Thismoiwaschosentoensurethat100%of NA and have caused three pandemics in the 20th century: the the cells were infected 24h postinfection. The microarray Spanish influenza (H1N1) in1918, responsible forapproximately experimentswereperformedinfiveindependentreplicates. 50 million deaths; the Asian influenza (H2N2) in 1957 during For kinetics on A549 cells, confluent cells were infected with which about 2–4 million people died; and the Hong Kong influenzavirusesatamoiof0.1or2foronehourunderaminimal influenza (H3N2) in1968responsible for1–2million deaths [6]. volumeofinfectionmediumat37uC.Thecellswerethenoverlaid Considering this pandemic potential and with up to 500,000 with fresh infection medium and incubated at 37uC. Samples of annualdeathsworldwideduringusualwinteroutbreaks,influenza supernatants were collected at defined time points and stored at Avirusesrepresentamajorpublichealthconcern[7].Prevention 280uCuntilendpointtitrationassays(TCID50)inMDCKcells. reliesonvaccinationwhichhasseveralmajorlimitationsincluding the lag time for vaccine preparation and the low vaccination 2 RNA preparation and hybridization to the gene chip coverage rate. Once a patient becomes infected, the current Total RNA was extracted from cell pellets using an RNeasy etiologic treatment of flu relies on M2 channel blockers or NA Mini Kit (Qiagen,Valencia,CA) for the BSL2 viruses. For H5N1 inhibitors[8].However,theseexistingtherapiesareinappropriate infections, total RNA was extracted with Trizol LS (Invitrogen). foruseincasesofsevereinfectionandmaybelimitedduetothe mRNAs were labeled with 33P for the reverse transcription using riskofrapidemergenceofdrugresistantviruses.Thusthereisan the Superscript III RT (Invitrogen), (a33P)dCTP and an obvious need to complement existing therapies with new anti- oligodT25. Generated cDNAs were hybridized on home-made influenza drugs. Nylon microarrays (HuSG9k) containing 9216 spotted IMAGE To search for new antivirals, we hypothesized that common human cDNA clones, representing 8682 genes and 434 control viral effects on cell metabolism should occur after infection with clones [16]. Further details on the HuSG9k microarray are different avianandhuman influenzavirusesandthatthispattern available on the TAGC website (http://tagc.univ-mrs.fr/). All shouldleadtotheidentificationofdrugseffectiveonallinfluenza membranes used in this study belonged to the same batch. After A viruses potentially. We first sought to identify a common gene hybridizationandexposureonMicroImager,arrayswerescanned expression signature following the infection with different human ina FujiBAS 5000machine andhybridization signalsquantified and avian influenza A viruses. While several microarray analyses using the BZ Scan Software [17]. Primary data, in accordance havealreadycomparedthepandemic1918H1N1virus[9,10]or with the proposed MIAME standards, are accessible through some H5N1 strain [11,12] to other less pathogenic strains, our GEOSeriesaccessionnumberGSE22319(http://www.ncbi.nlm. study is the first to demonstrate that a global influenza-induced nih.gov/geo/query/acc.cgi?acc=GSE22319). gene-expression signature can be defined. This proof-of-concept studywasconductedonahome-madenylonarrayusingahuman 3 Data normalization and analysis pulmonary epithelial cell line infected by five influenza A virus Data files were loaded and analyzed with R (v2.9.2) and subtypes (H1N1, H3N2, H5N1, H5N2 and H7N1). Using this Bioconductor(v2.4.1)[18],usingtheNylonArraylibrarydeveloped signature, we determined if molecules disturbing this pattern of by the TAGC to support BZScan2 files (library available upon infection would have a broad-influenza antiviral effect. By request). Raw data were normalized by quantile normalization. consulting the Connectivity Map, a database of drug-associated Supervisedanalysis(supervisedmethodsaimatfindingasetofgenes gene expression profiles [13,14], we identified molecules that whose expression profiles best correlate with a known phenotype) induced gene expression changes after cell treatment that were betweengroupsInfectedandMocksampleswasconductedusingthe mainly opposite to those induced by infection. These molecules SignificanceAnalysisofMicroarrayalgorithm(SAM)[19],usingthe weretestedinvitrofortheireffectonthefivedifferentviruses.To siggenes library (v1.18.0) [20]. All statistical analyses involved confirm our methodology, we took the opportunity of using the corrections for multiple comparisons (Benjamini and Hochberg) newemergingpandemicH1N1virusasamodeltotesttheeffect [21]. Agglomerative hierarchical clustering was performed by the of these moleculeson a newunknownvirus. pairwise average-linkage method using the Pearson correlation distance(Cluster3.0,Eisen,StanfordUniversity). Materials and Methods 4 Quantitative real-time RT-PCR validation 1 Cell lines and viruses To validate the microarray results with real-time RT-PCR CellsofthehumanlungepithelialcelllineA549weregrownas assay,anothersetofA549cellswereinfectedwithinfluenzaviruses monolayers in Dulbecco’s modified Eagle’s medium (DMEM) at a moi of 1 and total cell RNA was extracted at 24 hpi with supplemented with 10% fetal bovine serum, 2 mM L-glutamine, TrizolLS(Invitrogen).FivehundredngoftotalRNAwerereverse 100U of penicillin/mL, and 100mg of streptomycin sulfate/mL transcribed using oligo(dT)18 andRevertAid M-MuLV (Fermen- at 37uC. tas) according to the manufacturer’s instructions. OnemL of Influenza viruses A/New Caledonia/20/99 (H1N1), A/Mos- cDNA was then amplified and analyzed in the 7500 Real Time cow/10/99(H3N2),A/Lyon/969/09(H1N1SOIV),A/Turkey/ PCRSystem(AppliedBiosystems)usingthePlatinum(R)SYBR(R) 582/2006 (H5N1), A/Finch/England/2051/94 (H5N2), and A/ Green qPCR SuperMix-UDG kit (Invitrogen) according to the Chicken/Italy/2076/99(H7N1)wereproducedinMDCKcellsin manufacturer’s instructions. Six genes were chosen according to EMEM supplemented with 2 mM L-glutamine, 100U of penicil- theirlevelofexpression(Foldchangeinlog .2or,22)andthe 2 lin/mL, 100mg of streptomycin sulfate/mL and 1mg of trypsin/ availability of primers for the quantitative PCR (Table S1). mL. Viruses were titrated to determine tissue culture infection Glyceraldehyde3-phosphatedehydrogenase(GAPDH)mRNAwas PLoSONE | www.plosone.org 2 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion used as an internal control. The reaction mix contained a total 8 Neuraminidase assay volume of 20mL and the thermal cycling consisted of UDG Standard fluorometric endpoint assays used to monitor NA incubationat50uCfor2 min,40cyclesof95uCfor15sand60uC activity was recently shown to be suitable to quantify influenza for33sforamplification.Alldatawerenormalizedtotheinternal virus in a high-throughput screening test [26]. Briefly, cell standard GAPDH mRNA. For each single-well amplification supernatants (25ml) were transferred to a black 96-well plate reaction, a threshold cycle (Ct) was observed in the exponential and 75ml of 29-(4-methylumbelliferyl)-alpha-N-acetylneuraminic phase of amplification. Relative changes in gene expression were acid(MUNANA,SigmaChemicalCo.)toafinalconcentrationof determined using the 2DDCt method as previously described [22] 50mMwereadded.Afterincubationoftheplateat37uCfor1hr, and reported as the n-fold difference relative to a control cDNA 150mlstopsolution(0.05Mglycine,pH 10.4)wasaddedtoeach (mock cells) prepared in parallel with the experimental cDNAs well and the fluorescence read on a FluoStar Opima (BMG (infectedcells).StatisticalsignificancewascalculatedusingWelch’s Labtech) with excitation and emission filters of 355nm and two sample t-test between mock and infected samples using R 460nm respectively. Relative fluorescence units (RFU) were software. corrected by subtracting specific blanks, ie medium with or without molecules. 5 In silico experiment: query the Connectivity Map with For the NA activity test on L3 viruses (H5N1), viruses were the infection signature inactivated as previously described [27]. Cell supernatants were To select potential antivirals, an unbiased in silico search for mixedwithfreshlypreparedTritonX-100toafinalconcentration molecules that reverse the infection signature identified in the of 1% (vol/vol) Triton X-100 and incubated for 1 h at room presentstudywasperformedusingthepubliclyavailableConnec- temperature. Theinactivated supernatants werethentransported tivity Map database (build 02) [13]. The Connectivity Map (also outoftheBSL3totheBSL2laboratoryandusedforNAassaysas known as CMAP) is a collection of genome-wide transcriptional described above. data from cultured human cells treated with different kinds of Potential interference of test molecules on the NA enzymatic molecules. The 20 most differentially expressed genes in the activity was tested by incubating the A/Moscow/10/99 (H3N2) infectionstate(FoldChangeinlog2.2or,22)wereselectedfrom viral stock diluted in DMEM (107.8 TCID50/mL final) with theinitial300genesetidentifiedbySAM.Thesewerethenmapped increasing concentrations of the test molecule (or DMEM for to the U133Aprobesets in order toquery theConnectivity Map control) for 0.5h at room temperature. Specific blanks were database. In total, 28 U133A probe sets mapped to the selected measured for each molecule. 25mL were used for the NA test as genes from this study. The connectivity scores and p-values were describedaboveandresultswereexpressedasaratioofcorrected obtainedusingtheCMAPalgorithm[13]. RFU of the sample to RFU of controls. Two independent experiments were performed induplicate. 6 Molecules 2-aminobenzenesulfonamide (Sigma), calcium folinate (Sigma), 9 Viral growth assays in the presence of the molecules harmol hydrochloride (MP Biomedical), merbromine (Sigma), For the viral growth assays in the presence of the molecules, midodrine (Sigma) and ribavirin (Valeant Pharmaceuticals) were A549 cells were seeded into 96-well plates at 0.156105 cells per dissolvedinsterilewatertoastockconcentrationof5g/L,5g/L, wellandculturedfor3daysto100%confluence.Cellswerethen 4 g/L, 3.4 g/L, 5 g/L and 10mM respectively. Rilmenidine washedwithDMEMandincubatedwithvariousconcentrationsof (Sigma) was dissolved in dimethylsulfoxide (DMSO) to a stock the different molecules diluted in infection medium (DMEM concentration of 13g/L and brinzolamide was in suspension at supplemented with 2 mM L-glutamine, 100U/mL penicillin, 10g/L inthecollyrium AZOPT. 100mg/mL streptomycin, 20mM HEPES and 0.5 mg/mL Sulfameter (Sigma), pyrvinium (Sigma), moxalactam (Sigma) trypsin). Six hours after treatment, cells were infected with andmethylbenzethoniumchloride(Sigma)weredissolvedinsterile influenza viruses at a moi of 2 or 0.2 by adding 25mL per well water to a stock concentration of 50g/L. Alvespimycin (Sigma) of virus diluted in infection medium. Infection was allowed to was dissolved in sterile water to a concentration of 0.03g/L. proceed for 65h at 37uC, 5% CO2 after which 25mL of Sulodictil (Sigma) and DL-Thiorphan (Sigma) were dissolved in supernatant were collected for the NA activity test. Results are DMSO toaconcentration of 50g/L. expressed as a ratio of corrected RFU of the sample to RFU of control(incubationwithinfectionmediumwithouttestmolecule). 7 Viability assays To check for cytotoxicity, viability assays were performed in parallel toeachviral growthassay. Cell viability was measured by the neutral red assay, an indicatorofcytotoxicityusedinculturesofdifferentcelllines[23] with the same sensitivity as the MTT assay [24,25]. The neutral 10 Test of infection efficiency after cell or virus redassayisbasedontheinitialprotocoldescribedbyBorenfreund pre-incubation with the molecules and Puerner (1984) and determines the accumulation of the A549cellswereseededinto96-wellplatesat0.156105cellsper neutral red dye in the lysosomes of viable, uninjured cells. Cells well and cultured for 3 days to 100% confluence. For the ’Cell were seeded into 96-well plates and treated with molecules or Preincubation’test,cellswerewashedwithDMEMandincubated solvent. 72h after treatment, cells were incubated for 3 h with with various concentrations of the different molecules diluted in neutral red dye (100 mg/ml) dissolved in serum free medium 200mLperwellofinfectionmediumfor14h.Aftertwowashings (DMEM).Cellswerethenwashedwithphosphatebufferedsaline withDMEM,cellswereinfectedwithinfluenzaA/Moscow/10/99 (PBS) and fixed in a formol/calcium mix (40%/10%) for 1min (H3N2)virusatamoiof7during15minandwashedtwicewith beforebeinglysedwithEtOH/AcCOOH,(50%/1%)followedby infection medium. Infection was allowed to proceed for 5 h at gentleshakingfor15minuntilcompletedissolutionwasachieved. 37uC. For the ‘Virus Preincubation’ assay, the molecules were Absorbance at 550nm was measured using a microplate diluted in infectionmedium andA/Moscow/10/99 (H3N2)viral spectrophotometer system (Microplate Reader 2001, BioWhit- stock (108.8 TCID /mL) was treated with increasing concentra- 50 taker) and results were presented as a ratioofcontrol values. tions of the molecules for 14h. Cells were then washed with PLoSONE | www.plosone.org 3 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion DMEM and incubated for 15min with the virus and molecule A/Moscow/10/99 (H3N2) and avian A/Turkey/582/2006 mixdiluted12times.Infectionwasallowedtoproceedfor5 hat (H5N1), A/Finch/England/2051/94 (H5N2), and A/Chicken/ 37uC. In both assays, the number of infected cells was estimated Italy/2076/99 (H7N1) influenza viral strains. These viruses are withaNAtest.CellswerewashedwithPBSandlysedbyshaking hereinreferredtoasH1N1,H3N2,H5N1,H5N2andH7N1.A549 for 1h with 25mL per well of Triton 1X. The cell lysis extracts cells express both sialic acid a2,6- and a2,3-galactose receptors were used for a neuraminidase test as described above. Results [30,31]andwereshowntobeinfectedbyhuman,avianandswine were expressedasaratioofcorrected RFUof sampletoRFUof influenza viruses [32,33]. Infections were performed at 37uC, a control (solvent treated infections). Statistical significance was temperature at which both human and avian influenza viruses calculatedincomparisontoresultsforcontrolcellsusingtwotailed efficiently infect cell cultures [34] and at a moi of 0.1. In these Welch ttest. conditions,therewasevidenceofproductiveviralreplicationofall virusesbutwithsomekineticandyielddifferencesbetweenviruses, 11 EC and CC calculations as determined by infectious titers (TCID ) of supernatants of 50 50 50 Viability and antiviral data were analyzed using the following influenzavirusinfectedA549cells(Figure1).TheH5N1virustiters three-parameter non linear logisticregression(l3) function [28] peakedhigherandearlier(24hpi)comparedtoothervirusestiters. AvianH7N1andH5N2virusesreplicatedwithcorrectefficiencies, similar to the human H3N2 virus. In contrast, the human H1N1 D y~Dz virusstrainreplicatedslower(titerspeakedat65hpi)andgrewto (cid:2)x(cid:3)B 1z lowertitersthanotherviruses(p-value,0.05at24hpi). E Todeterminethehostgene-responsetoinfection,totalcellular RNA was extracted at 24 hpi and submitted to reverse were y is the response, D is the upper limit (response when the transcriptioninthepresenceof33P.Eachconditionwasperformed dose x is ‘infinite’), E is denoted EC or CC and is the dose in 5 independent replicates. All labeled cDNAs provided a good 50 50 producingaresponsehalf-waybetweentheupperlimitandlower radioactiveintensityandwerehybridizedontohome-madenylon limit (0), and B is the relative slope around E. This model is the microarrays containing 8782 IMAGE cDNA clones. All hybrid- shortened form of the four parameter logistic function where the izations were of good quality according to signals within lower limit is fixed to 0. Results were obtained by fitting the l3 acceptable range, number of features present, and signals from functionusingthepackagedrc[29]intheRStatisticalLanguage control spots. (version2.7.1). Parameters of the l3 model were estimated and Supervised analysis of normalized gene expression data was fitted curves were plotted only if the data set contained one conducted using theSAMalgorithm. Thisalgorithm was usedto response ,D/2. identifygeneswhoseexpressionlevelsweresignificantlyalteredby influenzainfection.WesetthedeltathresholdintheSAManalysis Results to allow an acceptable false discovery rate (FDR) of 10%. We found that the expression levels for a total of 300 genes 1GlobaltranscriptionalsignatureofinfluenzaAinfection (representing 3.4% of the genes considered present on the chip) To characterize the global cellular gene-expression response to differed significantly between mock and infected samples (Table influenza A infection, human pulmonary epithelium A549 cells S2). Using the DAVID Bioinformatics Resources database, we wereinfected with human A/New Caledonia/20/99(H1N1) and annotatedthissignatureusingthegeneontology(GO)terms.This Figure1.ComparisonofviralreplicationkineticsbetweeninfluenzavirusesinA549cells.A.A549cellswereinfectedatamoiof0.1with influenza virus A/New Caledonia/20/99 (H1N1), A/Moscow/10/99 (H3N2), A/Turkey/582/2006 (H5N1), A/Finch/England/2051/94 (H5N2), and A/ Chicken/Italy/2076/99(H7N1)andsupernatantswerecollectedat24,41,48and65hpiforendpointtitrationassays.B.Eachbarrepresentsthemean ofviraltitersat24hpifortwoindependentexperimentsandtiterswerestatisticallyanalyzedbytheonetailedWelcht-test;*:titersgreaterthan H1N1titers(p-value,0.05),#:titerslowerthanH5N1titers(p-value,0.05). doi:10.1371/journal.pone.0013169.g001 PLoSONE | www.plosone.org 4 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion revealed an enrichment of genes related to various cellular A subset of six genes with absolute fold changes in log (FCs) 2 processes such as protein complex biogenesis, membrane and above 2 was selected to validate the microarray analysis by microtubule organization, DNA metabolic and catabolic process- quantitative RT-PCR (RT-qPCR) analysis: DNMT1, NTE and es,cellproliferationregulation,cellcycleandcelldeath(Figure2). CAPN1thatwerefounddownregulatedininfectedcellsandG1P2, Figure 2. Gene enrichment analysis of the 300 genes of the infection signature. Discriminatory genes were analyzed by DAVID for associations with particular Gene Ontology terms. The negative Log (P-values) of enriched terms (plotted in bar) refer to how significant an 10 associationaparticularontologytermhaswiththegenelist.The30mostsignificantbiologicalprocessaregroupedaccordingtotheirbiological meaning. doi:10.1371/journal.pone.0013169.g002 PLoSONE | www.plosone.org 5 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion OAS1andICAM1thatwereupregulated.The6geneswerechosen 3020byH7N1and1455byH5N1.Themaindifferencebetween atrandomamongthemost20dysregulatedgenesuponinfection. H1N1 and other viruses lay in the number of down-regulated This quantification was performed on new samples equivalent to genesduringinfection.WhereasH3N2,H5N1,H5N2andH7N1 those used for the microarray analysis. Figure 3 shows the influenza viruses induced a down-regulation of most of the genes confirmationbyRT-qPCRofthemicroarraydata.Foreachgene tested,asimilarnumberofgenesweredown-andup-regulatedby andeachstrain,microarrayFCsarepresentedasablackboxplot H1N1 (highlighted by the blue vertical line in the heatmap and RT-qPCR results are depicted as a gray histogram. Results Figure 4). As H1N1 viral titer was lower at 24 hpi than titers of from RT-qPCR were in good agreement with the cDNA other viruses (Figure 1B), the scope of gene-expression changes microarrayanalysesforfiveoutofsixgenestested.Indeed,except induced upon infection correlated, at least partially, to the viral for CAPN1 (p-value=0.1), significant difference between infected replication efficiencyof thevirus-cell system usedinthisstudy. andnoninfectedcellswasalsoobservedinquantitativeRT-PCR Interestingly, out of the 300 genes of the global infection analysis(p-value,0.05,Welcht-test),similartoDNAmicroarray signature,only16wereupregulatedinallinfectedcells.These16 analysis. This result was acceptable considering that samples genes were associated to three GO biological process, including analyzed by RT-qPCR were different from those used in the two related terms, ‘‘viral reproductive process’’ and ‘‘viral microarray analysis. reproduction’’, that annotate genes encoding proteins involved TovisuallycomparethechangesinmRNAabundanceforthe in the virus life cycle. Two genes were associated to these terms: 300 genes found to be influenced by influenza infection, ICAM1, which is the major receptor for human rhinovirus [35], hierarchicalclusteringanalysisinbothdimensionswasperformed. and IRF7, which activates the expression of Epstein-Barr Virus Results are depicted in the heatmap representation of Figure 4. Latent Membrane Protein 1 [36] (p-value=0.07 and 0.08, Dendrogramsindicatethecorrelationbetweensamplesandgenes. respectively). While IRF7 has not been directly involved in We verified that mock samples were sorted together vs infected influenza virus life cycle yet, ICAM1 was recently identified as a ones. The H1N1 samples co-clustered with the mock samples proviralfactorthatmaybeco-optedbyinfluenzavirus[37].The suggesting that infection with this strain induced few gene third associated biological process was the term ‘‘immune expression changes. We verified this result by conducting a response’’ (p-value=0.04) annotating 4 genes (ICAM1, OASL, virus-specificSAManalysisonthemockvsonevirussamples.For OAS1 and CFD). Therefore, the upregulated genes were mostly aFDRof10%,only36geneswerefoundtoberegulatedbyH1N1 associatedwiththeimmunologicalresponse.Besides,sevenofthe infectionincomparisonto2298genesbyH3N2,1510byH5N2, 16 genes were interferon stimulated genes (ISGs): IFITM1, Figure3.Validationofmicroarrayresultsbyreal-timequantitativeRT-PCRanalysis.Expressionof6genesfromDNAmicroarrayanalysis (blackcross)iscomparedwithreal-timequantitativePCRdata(greybar).Crossrepresentfoldchanges(mean6S.D.)givenbythemicroarrayanalysis = Log (normalizedamountoftargetgeneininfectedsamples)-Log (normalizedamountoftargetgeneinmocksamples)(right).Theformulausedto 2 2 determinetheamountoftargetgeneininfectedcellsbyRT-qPCR,normalizedtoGAPDHandrelativetomockis:22DDCtwhereCtisthethresholdcycle andDDCt = (Cttargetinfected–Ctgapdhinfected)–(Cttargetmock–Ctgapdhmock).BarrepresentedLog2expressionlevel(mean6S.D.)oftarget genesintwoindependentquantitativeRT-PCRanalysis(left). doi:10.1371/journal.pone.0013169.g003 PLoSONE | www.plosone.org 6 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion PLoSONE | www.plosone.org 7 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion Figure4.Hierarchicalclusteringandheatmapofthe300genesthatdiscriminatemockandinfectedsamples.Heatmaprepresenting theexpressionlevelsofthe300genesdifferentiallyexpressedbetweeninfectedandmockcells.Redsquaresindicatetheexpressionlevelsabovethe medianofthegeneabundanceacrossthesampleandgreensquarestheexpressionlevelsbelow.Themedianvalueswereclusteredhierarchicallyin bothdimensions.Dendogramsindicatethecorrelationbetweengroupsofsamples(horizontal)andgenes(vertical).Verticallinesintherightportion indicatethe2distinctgeneexpressionpatterns:upanddown-regulatedgenesduringinfection.Thebluelinestandsforthegenesdown-regulatedin samplesinfectedwithH1N1influenzavirus.Geneexpressiondataforthe300genesarereproducedintableS2. doi:10.1371/journal.pone.0013169.g004 ICAM1, IFIT3, OAS1, G1P2 (or ISG15), IRF7 and OASL. These bythemolecule,noneofthemoleculesavailableinthisdatabank results were in accordance with previous studies showing the wereabletocompletelyreversetheinfectionsignature. upregulation of immune response associated genes in samples infected in vitro and in vivo with various influenza viruses 3 Evaluation of the antiviral potency of the selected [9,11,12,37,38,39,40]. Gene expression levels in each group of drugs on H3N2 viral growth samplesaredepicted inFigureS1.AllISGsweremarkedlymore Weassessedtheeffectoftheeightselectedmoleculesoninfluenza up-regulated in H5N1 infected cells than in other samples. This replication in vitro. Cell viability, as assessed by the neutral red hyperstimulation has been described in other transcriptional assay, and viral growth, as quantified by a neuraminidase (NA) studies [11,40,41] reinforcing the validity of the experimental activitytest,wereconductedinparallel.BeforeusingtheNAactivity cell-virus systemdeveloped inthepresent study. test asanindirect measurementforviralimpairment,wechecked firstly that the different influenza viruses used in this study had 2 In silico drug screening of the Connectivity Map sufficientneuraminidaseactivitiestobequantifiedusingthismethod The Connectivity Map is a collection of genome-wide (FigureS2A).Foralltestedvirusesandforasignaltobackground transcriptional expressiondatafromculturedhumancellstreated ratio between 2 and 70, the fluorescence was proportional to the with bioactive small molecules [13]. The associated website amountofviruspresent(TCID /mL).Duringtheevaluationofthe 50 provides tools tofindmoleculesconnected tothequery signature drugpanel,allsignaltobackgroundratioswereincludedbetween2 i.e.anylistofgenesassociatedwithabiologicaltest.Thesimilarity and70.Secondly,wecontrolledthatthedifferentmoleculesdidnot ofthequerysignaturetoeachofthereferenceexpressionprofilesis inhibittheenzymaticactivityofNAtobesurethatadropinRFU assessed and quantified by a normalized score, from -1 for a would only reflect a drop of viral titer. While concentrations of molecule that reverses the signature to +1 for a molecule which merbrominabove50mMandharmolabove500mMinhibitedNA induces geneexpression changessimilar tothequery signature. activity,incubationoftheviruswithincreasingconcentrationsofthe Our strategy was to query the Connectivity Map with a list of molecules otherwise resulted in no inhibition (Figure S2B). genes differentially expressed in infected cells to find molecules Therefore, for these two molecules below these concentrations that induced the opposite gene expression changes. We hypoth- and for other molecules of the drug panel, viral growth can be esized that such molecules may influence host cell metabolism in assessedbyaneuraminidasetest. such away that effectiveviral replication wouldbealtered. Evaluation of the drug panel was first conducted on influenza A critical step in this screening was to define the query A/Moscow/10/99 (H3N2) virus. A549 cells were incubated with signature. As the number of upregulated genes was very low increasingconcentrationsofthemoleculefor6hbeforeinfection. (5.3%) in the list of 300 genes defined by the analysis, a lack of Thistimewaschosenbasedonthedurationoftreatmentindicated specificity resulting from a loss of information for up-regulated intheConnectivityMaptoobtainsimilarcellularresponsebefore genescouldbeintroducedindrugselectionifthesignaturewasnot infection [14]. Infection was allowed to proceed for 65h which corrected for this bias. By selecting genes with the most drastic representsmultiplecyclesofinfection,howeversimilarresultswere changes in level of expression (fold change in log . 2 or ,22), observed at 24and48hpi(Figure S3). 2 wewereabletodefineasignatureof20genesforinfluenzaAvirus Theviabilitydataoffiveindependentexperimentsaregivenin infection with similar amounts of those up and down regulated Figure S4. The 50% cytotoxic concentrations (CC ) were 50 (Figure 5A,Table 1). determined by regression analysis. The CC of calcium folinate, 50 Byqueryingtheconnectivitymapwiththisconcisesignature,we 2-aminobenzenesulfonamide and midodrine could not be deter- obtainedc-scoresfor6100instances,representingmorethan1000 mined since none of these molecules was cytotoxic at the highest moleculesinvariousconditions[14].Weselectedthoseassociated tested dose. with the most strongly anticorrelated signatures (negative enrich- The effect of each of the molecules on viral growth was tested ment)andwhichhadap-valuelessthan0.5%(Figure5Band5C). usingtheH3N2virusatamoiof0.2and2.Dose-responsecurves Applying this filtering step left us with eight candidate molecules: were fitted by regression analysis (Figure S5) and used to harmol, rilmenidine, brinzolamide, ribavirin, calcium folinate, determine the 50% effective concentration (EC ) of each 50 2-aminobenzenesulfonamide, merbromin and midodrine (from molecule if at least one response was inferior to 50%. Selective the most negatively correlated to the least negatively correlated indexes (SI) were calculated as CC /EC and used to classify 50 50 drug).Therelevanceofourselectionwassupportedbythefactthat selectedmoleculesasinactive(SI,2),weakinhibitors(2,SI,10), ribavirin,analreadyknowninfluenzavirusinhibitor,wasidentified moderate inhibitors (10,SI,50) and strong inhibitors (SI.50) withanegativeenrichmentof-0.83andapvalueof0.00157.Except (Figure5).Inagreementwithpreviousobservations[42],wenoted for the topical antiseptic merbromin, theother selected molecules that SI were dependent on the moi, since molecules are more havevarioustherapeuticindications(depictedinTableS3)butare effective at lower moi. In our conditions, at a moi of 0.2, two notreferencedasantivirals. molecules (calcium folinate and 2-aminobenzenesulfonamide) Graphs in Figure 5C report how the different genes of the were ineffective, two (harmol and merbromin) were weak infectionsignaturebehaveintheexpressionprofileoftheselected inhibitors, two (brinzolamide and midodrine) were moderate molecules.Althoughthegenesdown-regulatedduringinfectionare inhibitorsandone(ribavirin)wasastronginhibitor.Atamoiof2, generallyup-regulatedinresponsetothemoleculeandconversely whereas brinzolamide was reclassified as a weak inhibitor, the theup-regulatedgenesofthesignaturearegloballydown-regulated othermoleculesremainedinthesameclassdespitetheirSIbeing PLoSONE | www.plosone.org 8 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion PLoSONE | www.plosone.org 9 October2010 | Volume 5 | Issue 10 | e13169 InfluenzaSignatureInversion Figure5.Geneexpression-basedscreeningidentifieseightpotentialantiviralmolecules.A.Listofcellulargeneschosentoquerythe ConnectivityMap.Acircumscribedsignatureofinfectionwasderivedfromthe300genesdiscriminatingmockandinfectedsamplesbyselecting geneswithafoldchange.2or,22.Thefoldchangeisdefinedastheratioofthemeangeneexpressionininfectedsamplestothemeanforthe correspondingmockinfectioninlog .Thisselectionresultedinalistof20genes,12beingup-regulatedduringinfectionand8down-regulated(see 2 Table 1). These genes constituted the signature used to query the online database Connectivity Map. B. Drugs with significant enrichment to influenzavirusinfectionsintheConnectivityMap.Significancecut-offwassetatp-value,0.005.Thepermutationp-valueestimatesthelikelihood thattheresultswouldbeobservedbyrandomchance.Mean:thearithmeticmeanoftheconnectivityscoresforthepost-dosechanges(orinstances) bythegivenmolecule;n:thenumberofinstancesofagivenmoleculeintheCMAPdatabase;Enrichment:ameasureoftheenrichmentofthose instancesintheorderlistofallinstances,Positiveenrichmentscoresareofinterestifperturbagensinducingthebiologicalstaterepresentedbythe signatureusedtoproducetheresultaresought.Likewise,ifreversalorrepressionofthebiologicalstateencodedinthequerysignatureisrequired, perturbagenswithnegativeenrichmentscoresareofinterest.Thespecificityvalueisdefinedasthefrequencyatwhichtheenrichmentofasetof instancesequalsorexceedsthatofthesamesetofinstancesinqueriesexecutedon312published,experimentallyderivedsignaturesusingthe MolecularSignaturesDatabase.Lowervaluesareassociatedwithagreaterspecificity;thenonnullpercentagerepresentsameasureofthesupport fortheconnectionbetweenasetofinstancesandasignatureofinterestbasedonthebehavioroftheindividualinstancesinthatset.C.8molecules arenegativelyconnectedtoinfluenzavirusinfection(p-value,0.005).Agraphicalrepresentationofthelocationofthesignatureofinfectionis depictedforeachmolecule,takingtheinstancewiththemostnegativeconnectivityscoreofeachmolecule.Thex-axisrepresentsthegenesofthe expressionprofileofthemolecule,rankorderedaccordingtotheirdifferentialexpressionrelativetothecontrol.Thelocationofeachgeneofthe infectionsignatureisappreciatedalongthex-axis. doi:10.1371/journal.pone.0013169.g005 weaker. As an example, the CC for midodrine was superior to used to obtain 1550.7mM of rilmenidine- and the EC was 50 50 4250mMandEC wascomprisedbetween322mM(moiof0.2) comprised between 1.0% (moi of 0.2) and 1.8% (moi of 2). The 50 and 532mM (moi of 2). Concerning rilmenidine, which was EC ofrilmenidinewassignificantlydifferentfromthatofDMSO 50 dissolved in DMSO, it was not possible to conclude on an effect. at a moiof 2(p=2e24)but not at a moiof0.2 (p=0.85). DMSO has previously been shown to be cytotoxic and to inhibit However,evenifthismoleculeisconsideredineffectiveagainst influenzainfectionabove4%(vol/vol)[43]howeveritisstillused theH3N2influenzavirus,wedidobtainaveryhighconfirmation asamajorsolventformoleculesinhigh-throughputscreening.In rate (62.5%) in comparison with the hit rate of classical high- thisstudy,theCC forDMSOwas2.9%(v/v)-theconcentration throughput screening (3.2% in [43]). This clearly indicates that 50 Table1. Genesof thecircumscribed signature ofinfection. Gene Genesymbol I.M.A.G.E.CloneID FoldChange TranscriptionRegulation Etsvariantgene3 ETV3 1473929 7.24 Myocyteenhancerfactor2D MEF2D 4209 5.13 FBJmurineosteosarcomaviraloncogenehomologB FOSB 79022 22.69 DNA(cytosine-5-)-methyltransferase1 DNMT1 768241 22.06 AnkyrinrepeatandBTB(POZ)domaincontaining2 ABTB2 204456 2.28 Immuneresponse ISG15ubiquitin-likemodifier,G1P2 ISG15 742132 3.48 29,59-oligoadenylatesynthetase1,40/46kDa OAS1 666703 2.07 ComplementfactorD(adipsin),DF CFD 666128 2.99 Cellsurfaceprotein Patatin-likephospholipasedomaincontaining6,NTE PNPLA6 431990 22.01 Heparansulfateproteoglycan2 HSPG2 3339 22.23 Solutecarrierfamily2(facilitatedglucosetransporter) SLC2A2 207963 2.23 PTPRFinteractingprotein,bindingprotein1(liprinbeta1) PPFIBP1 263094 2.04 Intercellularadhesionmolecule1(CD54) ICAM1 3383 3.36 Lysophosphatidicacidreceptor1 LPAR1 505524 3.63 Enzyme Hyaluronoglucosaminidase4 HYAL4 668088 2.90 Asparagine-linkedglycosylation2homolog ALG2 150443 2.35 Calpain1,(mu/I)largesubunit CAPN1 70555 22.20 Proteinphosphatase1,PP1 PPP1R14D 140525 23.13 Nuclearprotein Karyopherin(importin)alpha6 KPNA6 16650 22.21 HypotheticalproteinHSPC111 NOP16 505242 22.08 FoldChange = log2(Inf/Mock). doi:10.1371/journal.pone.0013169.t001 PLoSONE | www.plosone.org 10 October2010 | Volume 5 | Issue 10 | e13169