Compatibility with Killer Explains the Rise of RNAi-Deficient Fungi

NA interference (RNAi), lost in a recent an-cestorof Saccharomycescerevisiae (fig.S1),canberestoredbyintroducingArgonaute(AGO1) and Dicer (DCR1) from a close relative,Saccharomycescastellii (1).Thereconstitutedpath-waysilences endogenoustransposons, whichex-plains retention of RNAi in some lineages ( 1).Butwhyisitlostinothers?Toaddressthisquestion,wedeterminedtheconsequencesofrestoringRNAito S. cerevisiae . Restoration of RNAi had little ifanyeffectongrowthmeasuredunder50differentconditions,andhigh-throughputRNAsequencing(RNA-Seq) results indicated minimal changes innonrepetitive mRNA accu mulation (figs. S2 to S6andtablesS1andS2).However, restoring RNAi toS. cerevisiae profoundly affected maintenance ofkiller, an endemic viral system cytoplasmaticallyinheritedas adouble-stranded RNA (dsRNA) vi-rus,L-A,anditssatellitedsRNA,M( 2).Mencodesa protein toxin that killsnearby cells while con-ferringimmunitytocellsmakingthetoxin,andL-Ais required to maintain M(3).IntheRNAi-competentS. cerevisiae strain, endog-enousMandL-AdsRNAswere processed into smallinterfering RNAs and thenlost in most cells (Fig. 1Aand fig. S7), thereby ren-dering cells susceptible tokilling by toxin from cellsthatretainedkiller(fig.S7).The loss of killer in theRNAi-competent strainillustrated a circumstancein which the viral-defensefunction of RNAi, knownto be beneficial in othercontexts(4),imparteda netselective disadvantage.This disadvantage, whenconsidered together withthe relatively benign ef-fects of losing RNAi inS. castellii (fig. S8), whichlacksactivetransposons( 1),suggested a model to ex-plain why yeasts that lostRNAi have nonethelesssucceeded in evolution:Descendents of cells thatlose RNAi can acquire and retain killer, whichunder some circumstances more than offsetsthedisadvantageoflosingRNAi.Thekiller sys-tem is observed in close relatives of S. cerevisiae(i.e., yeasts of the sensu stricto clade) ( 5, 6) (fig.S9), whereas RNAi is absent in all sequencedsensustrictospeciesyetpresentinacloseoutgroup,S. castellii. Thus, in our model, loss of RNAi in arecent sensu stricto ancestor enabled one of itsdescendents to acquire killer, which provided anetselectiveadvantageoveritsRNAi-containing,nonkiller, and toxin-sensitive neighbors, ultimatelygivingriseto S.cerevisiae andothersensustrictospecies.Totestwhetherourmodelcouldapplythrough-outfungi,weperformedwhole-genomesequencingtoidentifyRNAigenesinspecieswithkillervirusesand RNA analyses to search for dsRNA killerviruses in species with RNAi. RNAi was absent inallspeciesknowntopo ssessdsRNAkiller,where-as killer was absent in closely related species thatretained RNAi (Fig. 1B, fig. S9, and tables S3andS4).ThisincompatibilitybetweenRNAianddsRNA killer extended even to Ustilago,anevo-lutionarily distant basidiomycete (fig. S10).AnalysesofsyntenyandphylogenyofArgonauteand Dicer proteins (fig. S11) indicated that thediscontinuous presence of the RNAi pathway infungi is best explained by its loss in at least nineindependentlineagesratherthanbyitsacquisitionthrough lateral transfer (Fig. 1B). At least four ofthese nine lineages included isolates with dsRNAkillerviruses(Fig.1B),whichsupportsourmodelin which the ability of RNAi-deficient strains tohostkillerexplainstheirevolutionarysuccess.Theotherfivelineages,forwhichdsRNAkillerhasnotyet been reported in extant descendants, mighthave succeeded for other reasons. Alternatively,killer might have been acquired but then lost, justas some descendants of the sensu stricto radiation(including some S. cerevisiae strains) have lostkiller (Fig. 1B and fig. S9). All nine RNAi losseswere relatively recent, suggesting that individualsthat lost RNAi earlier left no living descendants.Thus,althoughcompatibilitywiththekillersystemcan explain the persistence of RNAi-deficient fun-gal lineages for many millions of years, lineagesthat lose this elegant transposon defense might bedoomed to extinction over the longer term.