Received 14October 2002 Accepted 12December 2002 Publishedonline 19March 2003

Site-specificselfish astools for the control andgenetic engineering of naturalpopulations Austin Burt Departmentof Biological Sciences and Centrefor Population Biology, Imperial College, Silwood Park,Ascot, BerkshireSL5 7PY, UK ([email protected] ) Site-specificselfish genes exploit hostfunctions to copy themselvesinto a definedtarget DNAsequence, andinclude homing endonucleasegenes, group IIintronsand some LINE-like transposable elements.If suchgenes can be engineered to target newhost sequences, then they canbe usedto manipulate natural populations,even if thenumber of individuals releasedis asmall fraction ofthe entire population. For example, ageneticload sufficientto eradicate apopulation canbe imposed in fewerthan 20 generations, if thetarget isan essential host , the knockout is recessive and the selfish gene has anappropriate promoter. There will beselection for resistance,but several strategies are available for reducingthe likeli- hoodof it evolving. Thesegenes may also beused to genetically engineernatural populations,by means ofpopulation-wide gene knockouts, gene replacements and genetic transformations. By targeting sex- linked loci justprior tomeiosis one may skewthe population sexratio, andby changing thepromoter onemay limit thespread of the gene to neighbouring populations.The proposedconstructs are evol- utionarily stable in theface of the mutations most likely toarise during their spread,and strategies are also available for reversing themanipulations. Keywords: population eradication; population geneticengineering; homing endonucleasegenes; vector-bornediseases

1. INTRODUCTION notbe useful. Whichever classof site-specific selfish gene is used,all thevarious proposals presupposethe ability to Somespecies— arelatively small number—cause substan- engineersuch genes to recognize a newtarget sequence. tial harm tothe human condition;most prominent are Work ondesigning enzymesto recognizea specifiedDNA thosethat causedisease, transmit diseaseor reduceagri- sequenceis ongoing, motivated in part by their potential cultural output.Many such have long beentargets usein functionalgenomics and gene therapy ofpopulation control,with varying degreesof success, but (Chandrasegaran &Smith 1999; Segal et al. 1999; Bibi- somespecies are still beyondcontrol by currentmethods, kova et al. 2001; Buchholz &Stewart2001; Chevalier et andnew approaches are required.Genetic methods of al. 2002; Santoro& Schultz2002; Seligman et al. 2002; engineering or eradicating natural populations have been Takahashi &Fujiwara 2002). Engineering group IIintrons muchdiscussed (Knipling 1979; Curtis1985; Hastings totarget newsequences is particularly simple, asrecog- 1994), mostrecently in thecontext of using transposable nition largely dependsupon RNA– DNA basepairing (Guo elementsor bacterial symbiontsto drive novel genesof et al. 2000). Despitethis activity, theuses of engineered interestinto a population (Ribeiro &Kidwell 1994; selfishgenes for manipulating natural populations appear Beerntsen et al. 2000; Braig &Yan2002). However,there notto be recognized, and an exploration ofthe possi- are inherentdifficulties with theseproposals, in particular bilities seemswarranted onthree grounds: to motivate relating tothe stability ofthe proposed constructs, and more rapid developmentof the technology; to warn of goodreasons to think they may notwork (Turelli & containmentissues that ought tobe addressed during Hoffmann1999; Braig &Yan2002; Spielman et al. 2002; development;and to stimulate discussionson the desir- and§ 6, below).In this paper Iexplore aseriesof alterna- ability oferadicating or genetically modifying particular tive geneticapproaches basedon the use of site-specific species. selfishgenes— genes that exploit hostfunctions to copy themselvesinto a particular target sequence.These alter- 2. THE BASICCONSTRUCT native approaches appear tohave anumberof desirable features,including evolutionary stability andreversibility. HEGsare selfishor parasitic genesthat canspread Naturally occurring examples ofsite-specific selfish through populations owing totheir biased‘ super-Mendel- genesinclude homing endonucleasegenes (HEGs), group ian’inheritance (Chevalier &Stoddard2001; Goddard et IIintronsand some site-specific LINE-like transposable al. 2001). They encodean that recognizesand elements(Chevalier &Stoddard2001; Belfort et al. 2002; cleavesa 20–30 bp sequencefound on chromosomes not Eickbush2002). Outof the three types, HEGs have the containing acopy ofthe HEG. The HEGitself is inserted simplest mechanism ofaction (describedfurther in §2), in themiddle ofits ownrecognition sequence,and so while theother twospread via anRNA intermediate and chromosomescarrying theHEG are protectedfrom being reversetranscription. For simplicity ofexposition, HEGs cut.The brokenHEG 2 chromosomewill typically be will beused as exemplars throughout thepaper, though repaired by thecell’ s recombinational repair system,which this is notmeant toimply that theother twotypes will usesthe intact HEG 1 homologue asa template. After

Proc.R. Soc.Lond. B (2003) 270, 921–928 921 Ó 2003 TheRoyal Society DOI10.1098/ rspb.2002.2319 922 A. Burt Site-speciŽc selŽsh genes and population control

recognition site 1.0 s

s 0.8 e n t i f

essential gene r 0.6 o

y c

n 0.4 e u q e r

f 0.2

essential HEG gene 0 10 20 30 40 50 time (generations)

Figure 2. Frequency of theHEG (solid curve) and population mean fitness (dashed curve) assuming e = 0.9 and an initial release frequency of 1%. Theseresults, and all meiosis- or germline- no self-splicing others in thepaper, are for an idealized population, from specific promoter or intein whichall real populations will deviate in some way.They should, therefore, betaken as rough indications, not precise Figure 1. Aconstruct for biological control: aHEG predictions. engineered to recognize asequence in an essential gene for whichthe knockout phenotype isrecessive. Note thatthe weassume that thepopulation is large andmates ran- HEGis inserted into themiddle of its own recognition sequence. domly, that theknockout is arecessivelethal andthat TRDoccursequally in males andfemales, then the equi- librium frequencyof the HEG ( qˆ)canbe shown to be repair, both chromosomeswill containa copy ofthe HEG, qˆ = e, where e isthe probability that theHEG 2 allele in a anda heterozygote will have beenconverted into a homo- heterozygote isconverted to aHEG 1 allele; e = 0 for Men- zygote. Thus,the biased inheritance arisesfrom acombi- delian inheritance.The load imposedupon the population nationof asequence-specificendonuclease inserted in the (i.e.the fraction ofthereproductive effort that is rendered middle ofits ownrecognition sequenceand the cell ’s own unproductive)is thenequal tothe frequency of homozy- recombinational repair pathway. gotes, L = qˆ2 ,andthe mean fitness of the population is 1 The simplicity ofthis mechanism suggeststhat it may minusthis, or wˆ = 1 2 e2.For example, HEGsof yeasts beopen to human artifice. The proposedconstruct is illus- canshow extreme TRD, with e < 0.99 (Jacquier &Dujon trated in figure 1, andthe essential features are asfollows. 1985; Wenzlau et al. 1989). If oneconsiders, conserva- tively, anengineered HEG with aTRDof e = 0.9 (this (i) AHEGis engineeredto recognize andcut a sameassumption will bemade in all numerical examples sequencein themiddle ofan essential gene, and the in this paper), thenthe equilibrium meanfitness of the HEGis insertedinto themiddle ofits ownrecog- population will be wˆ = 0.19. That is,four-fifths of zygotes nition sequence,simultaneously disrupting thegene producedwill die,and only one-fifthwill survive torepro- andprotecting thechromosome from being cut. duce.Moreover, this load will arise relatively quickly. If Naturally occurring HEGsdo notusually disruptthe theHEG is introducedinto 1% ofthe population, thenit

functionof the host gene because they are associated will take only t1 ,9 0 = 12 generationsfor theload toreach with self-splicing group Iintronsor inteins 90% ofits equilibrium value. If onecan manage torelease (Chevalier &Stoddard2001), butthe engineered aninitial frequencyof only 0.01%, thenit will take 19 elementwould not have these. generations.The progression toequilibrium isshown in (ii) The target geneis chosensuch that theknockout figure 2. mutation has little phenotypic effectin thehetero- Theseresults are fairly robustto changes in thefitness zygousstate, but is severely deleteriouswhen homo- scheme.For example, if thehomozygote has someresidual zygous(i.e. the knockout is recessive). viability (i.e.the knockout is sub-lethal),then this can (iii) Finally, theHEG is underthe control of ameiosis- actually increasethe genetic load. Load is highest justat specificpromoter, sothat heterozygouszygotes thepoint at whichthe HEG cango tofixation. The results developnormally, buttransmit theHEG toa dispro- are also robustto a certain level ofheterozygote impair- portionate fraction oftheir gametes. ment(i.e. the knockout is incompletely recessive).For example, if thehomozygote islethal andthe heterozygote This last conditionmay berelaxed, dependingupon has afitnesslevel 90% ofthe wild-type, then wˆ = 0.192 whenthe target geneis expressed: if thetarget geneis (insteadof 0.19) and t1 ,9 0 = 14 generations(instead of 12 expressedonly in larvae, or in somatic tissues,then the generations;calculations notshown). promoter canbe adult-specific, or germline-specific. If sucha constructis introducedat lowfrequency into (a) Evolutionary stability apopulation, theninitially it will appear mostly in thehet- Akey featureof this constructis that it is evolutionarily erozygousstate, and so it will showtransmission-ratio dis- stable,in thesense that themutant forms most likely to tortion (TRD) butfew harmful effects.It will therefore arise asit spreadsthrough apopulation will beselected increasein frequency,until it reachesan equilibrium fre- against andlost. For example, amutantHEG that loses quencyat which theharmful effectsbalance theTRD. If theability torecognize or cutthe target DNA will have

Proc.R. Soc.Lond. B (2003) Site-speciŽc selŽsh genes and population control A. Burt 923

1.0 achievable in mostoutcrossed . Even in species that are predominantly haploid, with only abrief diploid s s

e 0.8 phase,one could target ageneencoding a protein needed n t i

f for theentry intomeiosis, and then have homing occur

r 0.6 o

during meiosis.Alternatively, somepredominantly hap- y

c loid taxa (including malarial Plasmodium) have an n

e 0.4

u extendedpost-meiotic syncytial phase,and so in these q e

r speciesone might also target aprotein neededduring mei- f 0.2 osis(e.g. a synaptonemal complex protein) andhave hom- ing occurduring thesyncytial phase.Highly inbredand 0 20 40 60 80 100 wholly asexual specieswill beless amenable tocontrol by time (generations) engineeredselfish genes. Figure 3. Recalling aHEG.Aresistant allele is introduced at1% in generation 40 (dotted curve). All other parameters 3. INCREASINGTHE LOAD are asin figure 2. Having describedthe logic ofthe construct, I nowcon- siderhow to increase the load imposedupon a population. reducedTRD, withoutreducing the harm doneto the The extentto which this is necessarywill dependupon host,and so will belost from thepopulation. With a theefficacy ofthe construct in showingTRD. If onecan mutation rate of1%, there will belittle effecton the engineera highly effectiveHEG with e = 0.999, thentar- ˆ dynamics: w = 0.22 (insteadof 0.19) and t1 ,90 = 12 gener- geting asingle recessivelethal genewill give an equilib- ations(as before). Even if themutation rate is ashigh as rium meanfitness of wˆ = 0.002 (i.e.99.8% ofreproductive ˆ 10%, theload will still besubstantial ( w = 0.5, t1 ,90 = 13 outputis wasted,and only 0.2% is viable), with generations).Similarly, mutant HEGsthat are active in t1 ,90 = 11, probably sufficientto eradicate many popu- somatic aswell asgermline tissues,or that showless lations.However, such high levels ofTRD may notbe sequencespecificity, will bemore harmful tothe host achievable. If,as assumed here, one can achieve aTRD withoutincreasing theTRD, andso will also beselected of only e = 0.9, thenonly four-fifthsof the population will out.Simulations showthat suchmutations will have even die,and for somepests this may nothave asubstantial lesseffect on the invasion dynamicsthan mutationsto effecton their population dynamics,as it may merely relax non-functionalHEGs (not shown). density-dependentpressures on survival andrepro- This evolutionary stability ofthe proposed construct duction.It is thereforeworthwhile investigating howone gives it amarked advantage over thealternative strategy might increasethe load further.Two possibilities will be ofusing a non-Mendeliangenetic element to drive atoxic considered:targeting alternative loci andtargeting mul- geneinto a population. In thelatter case,knockout tiple loci. mutationsdestroying thefunction of the toxic geneare sureto arise, andthese mutations will spreadat the (a) Loadas a function ofthe gene targeted expenseof the toxic gene.Unless one can achieve quite In themodel analysed in § 2, thefrequency of theHEG high releasefrequencies, such an approach is probably increasesdue to TRD, anddecreases due to the homo- limited tomodifying apopulation, rather than eradicating zygouslethality. For many speciesthat onemight wantto it (discussedfurther in § 6). control,killing males is worsethan uselessbecause it reducesthe frequency of the HEG butwill dolittle to (b) Reversibility reducepopulation growth ratesor equilibrium density, Afurther attractive featureof theproposed construct is whichwill largely bedetermined by female productivity. that it is fully reversible. If onetargets agenethat, when Oneway toavoid the ‘wastage’ ofkilling males wouldbe knockedout, is strongly deleterious,then there will be totarget agenethat, whenknocked out, kills only females. strong selectionin favour ofresistant alleles —sequences For theTRD assumedhere, targeting sucha genegives a that are functional,but are notrecognized and cut by the three-foldreduction in mean fitnesscompared with tar- HEG.Onecould engineer resistant alleles by, for geting arecessivelethal gene(table 1). Knockoutscausing example, usingthe degenerate property ofthe genetic femalesto be sterile will beequally effective.Indeed, codeto create a DNA sequencethat codedfor thesame knockoutscausing male sterility will have thesame effect, sequencebut differed in nucleotidesequence if they have noeffect on the male ’sfertilization success from thetarget (e.g.by changing many third-position (e.g.the only effectis tomake sperm that are defective sites).Releasing suchresistant alleles couldbe used to after karyogamy). effectively ‘recall’ aHEG,asthe resistant allele would Targeting other classesof loci canbe even more effec- spreadthrough thepopulation, driving theHEG extinct tive, though finding suitable candidateloci may bemore (figure 3). difficult.If theknockout causes both sexesto be sterile, thenmean fitness will bethe square of what it is under (c) Wide applicability lethality, becauseboth parentshave tobe fertile in order The logic ofthe approach requiresthat theknockouts for azygote tobe formed. For theTRD assumedhere, belargely recessive,and that homing occurseither at mei- this gives afive-foldreduction in meanfitness compared osis,or in sucha way that thefitness of heterozygous with targeting alethal gene(table 1). If onetargets a zygotes is notimpaired butthey producea predominance ‘grandchildless ’ mutation (i.e.homozygous femalespro- of HEG1 meiotic products.These criteria shouldbe duceviable butsterile sonsand daughters; Ashburner

Proc.R. Soc.Lond. B (2003) 924 A. Burt Site-speciŽc selŽsh genes and population control

Table1. Equilibrium mean fitness asa function of theclass of gene targeted.

knockout phenotype general e = 0.9 t1,90

lethal 1 2 e2 0.19 12 femalelethality; unisexual sterility (1 2 4e2)/(1 1 3e2) 0.055 11 bisexual sterility (1 2 e2)2 0.0361 11 maternal effect bisexual sterility ( ‘grandchildless ’) [(1 2 4e2)/(1 1 3e2)]2 0.0031 12 conditional lethal a 0.16 a Assumesfive (summer) generations of no selection on thegene, followed byone (winter) generation of selection. Meanfitness in theoverwintering generation will be w¯ˆ = 1.6 ´ 102 5.

1989), thenmean fitnesscan be 60-fold lower than for a with someredundancy (as naturally occurring HEGs lethal gene(table 1). Finally, in somespecies one might have) torecognize all sequencevariants detected,one beable totarget loci that are essentialonly in somegener- couldthereby ensurethat theinitial frequencyof resistant ations,creating aconditional lethal. This cangive amod- alleles wasless than 1in 10 3 –104 .Indeed,one might be estincrease in meanload compared with anunconditional able togo further,and engineer a HEGthat couldrecog- lethal (table 1); more importantly, with densitydepen- nizeall sequencevariants actually detected,plus, say, all dence,the effect on the population may besubstantially possiblesingle-nucleotide variants ofthe observed greater (Knipling 1979). sequences.Insertion or deletionmutations in thetarget sitemay beparticularly difficult for theHEG torecognize, (b) Multiple loci andso one will wantto choose regions that are well con- Anothermeans of increasing theload is totarget mul- servedfor length acrossspecies, or for whichstructural tiple loci simultaneously.In thesimplest casewhere the information suggeststhat any length variant is likely tobe TRD andphenotypic effectsat onelocus are independent non-functional. ofgenotype at theother locus,the equilibrium meanfit- This isnot to say that recognition-siteredundancy nesswill bethe product of the fitnesses of the two loci shouldbe maximized (or,put another way,that sequence separately. For example, if oneengineers HEGs to target specificity shouldbe minimized). If theendonuclease n differentloci essentialfor female fertility and e = 0.9 at cleavesnon-homologous sites, then it will reducefitness n each,then mean fitnesswill be0.055 .If onetargets five evenwhen heterozygous, slowing or preventing its spread. ˆ 2 7 loci, thenmean fitness will be w = 5 ´ 10 , enough to Also,for safety,one will wantto be able torelease a resist- drive any population extinct.Simulations showthat it ant allele. Finally, it may also besafer if theendonuclease makeslittle differencewhether recombination betweenthe doesnot recognize thehomologous sequencein closely loci is0 or1/2, and,even if it is0, it makeslittle difference related non-targetspecies, so asto reduce the risk ofhori- whetherthe HEGs are introducedin coupling orrepulsion zontal transfer.Ideally, onewants to target asitethat (notshown). This isbecause the gene conversion events showslittle variation within species,but considerable act analogously torecombination tobreak upcorrelations divergencebetween species (at least at thenucleotide betweenloci, andso the alleles endup in linkage equilib- level). rium. In Drosophilamelanogaster ,at least,there are thought ‘Combination therapy ’—theuse of multiple drugssim- tobe some 3000 essentialgenes, and more than 100 ultaneously —has recentlybeen acknowledged to slowthe required for fertility (Ashburner1989, p.435; Miklos & evolutionof resistance in human pathogens(White et al. Rubin1996; Ashburner et al. 1999). Mostprospective tar- 1999; Palumbi 2001), andthe same approach canbe used get speciesare likely tohave anabundance of suitable tar- get loci. with engineeredHEGs, virtually withoutlimit. First, one couldrelease, say, 10 differentHEGs, attacking 10 differ- entsites along thelength ofasingle gene.The more HEGs 4. DEALINGWITH NATURALRESISTANCE that are released,the lower the likelihood that resistant The simulation shownin figure 3demonstratesthat if sequenceswill exist for all ofthem, and, even if they do afunctionalhost gene exists that is resistantto the HEG, exist,the longer it will take for amultiply resistant thenit will increaserapidly in frequencyand drive the sequenceto be stitched together by recombination. The HEGextinct.Care must therefore be taken to minimize geneticload imposedin themeantime canbe substantial, thelikelihood that suchsequences exist, or that they can andmay well beenough to drive thepopulation extinct. arise beforethe population iseradicated. The first step Onecould also engineertwo HEGs with adjacentrecog- wouldbe to use mutagenesis experiments andstructural nition sites,such that if either onewas able tocut, then studiesto choose target genes,and sites within genes,that both wouldbe transmitted. Since alleles resistantto only seemunlikely tobe able tochange (at theamino acid oneof the HEGs would have little selectiveadvantage level) withoutseriously compromising thefunction. One (arising only from non-functionalmutant HEGs), the wouldalso wantto choose sites that showlittle sequence evolutionof the doubly resistantallele wouldbe substan- variation in thetarget population. Onecould plausibly tially retarded.The only limit onthe number of adjacent sequence 103 –104 alleles, andif oneengineered a HEG HEGsone could use would be in thelength ofsequence

Proc.R. Soc.Lond. B (2003) Site-speciŽc selŽsh genes and population control A. Burt 925 that canbe copied from onechromosome to another dur- 1999; Koufopanou et al. 2002). However,demonstration ing recombinational repair. that horizontal transmissionoccurs regularly onan evol- The secondform of ‘combination therapy ’ would be to utionary time-scaleof millions ofyears doesnot mean it target multiple loci simultaneously.Even if resistance is asubstantial risk during a10-year population control couldevolve at eachlocus separately, in acombinedattack programme. Transposable elementshave beendemon- thegenetic load canbe sufficient to drive thepopulation stratedto transfer readily between Drosophila species extinct.As notedin § 3b, thereis unlikely tobe a shortage (Jordan et al. 1999), butthe likelihood ofaHEGtransfer- ofsuitable target loci. Supposeone is targeting n recessive ring is probably substantially lower,as they have noextra- female sterility genesand that at every onethere is a resist- chromosomal part ofthe life cycleand no time whenthe ant allele at afrequencyof 10 2 6 .If releasefrequencies are protein is boundto the gene. For horizontal transmission 1% andall HEGshave e = 0.9, thenthere is afive-gener- tooccur, one needs the DNA containing theHEG some- ation window(from generations11 to15 inclusive)in howto get intoa germline nucleusof another species,and which themean fitness of the population isabout 0.08 n. besufficiently intact that it canbe transcribed and then With 10 HEGs,mean fitness will beabout 10 21 1 , enough usedas a template for repair. This is unlikely tooccur in todrive any population extinct. many prospective target species,and, indeed, may bethe Twoother classesof resistant genotypes are also poss- main reasonwhy HEGsappear tobe absent from animals ible, though whethereither oneis likely toarise in any real with segregated germlines. Assuggested in § 4, one way population isunclear. First, amutation might arise that toreduce further theprobability ofhorizontal transferis compensatesfor theknockout of the target genebut is toengineer the HEG notto recognize the homologous otherwiseneutral. Were such a mutation toarise, it too sequencein related species.Indirect evidence from yeasts wouldincrease rapidly in frequency,and population mean suggeststhis canbe an effectiveway tolimit transfer fitnesswould return to normal. Asimple duplication of (Koufopanou et al. 2002). In addition,one could target a thetarget locusis unlikely tobe sufficient, as the HEG region with lowoverall nucleotidesimilarity betweenspec- will readily transferover tothe new locus. If suchresist- ies(to reduce the likelihood ofhomologous recom- ancedoes turn out to be a problem, thenit will beone bination), andhave theHEG regulated by aspecies- more reasonto attack multiple genessimultaneously. specificpromoter. Finally, amutation might arise that somehowreduces or eliminates thehoming activity, butis otherwise neutral. 6. POPULATIONGENETIC ENGINEERING Several pointshere seem relevant. Ihave focusedthus far onthe problem ofimposing a (i) Homing dependsupon very basic cellular processes geneticload soas to control or eradicate apopulation. (transcription, translation, nucleartransport, recom- Moresubtle approaches may oftenbe desirable. In parti- binational repair), andso it is notclear howsuch a cular, onemay notwant to eradicate apopulation, but mutation might arise. rather totransform it genetically suchthat it is lessnoxi- (ii) Manyprospective target speciesdo not appear to ous.For example, onemight useengineered HEGs to per- have HEGsnaturally, andso are unlikely tohave form apopulation-wideknockout of, say, a genenecessary evolved general defencesagainst them. for mosquitoesto transmit malaria. If theknockout is not (iii) Naturally occurring HEGsfall intothree or fourdis- tooharmful, thenthe HEG will spreadto fixation, and tinctprotein families (Chevalier &Stoddard2001) this may have little or noeffect on population numbers. andartificial HEGscan be different again (e.g. Indeed,if thehomozygous knockouthas fitnessgreater fusionsof a sequence-specificzinc finger protein than 1/(e 1 1), thenthe HEG is expectedto spreadto fix- with anon-specificendonuclease ; Bibikova ation evenif its expressionis notlimited tothe germline et al. 2001), soif resistanceevolves against oneof (calculations notshown). Many (perhaps most)genes them,there may notbe cross-resistance to others. couldbe knocked out in this way.Such an approach also (iv) Site-specificselfish genes are also available that use hasthe advantage that resistantgenotypes will beless reversetranscription rather than recombinational strongly selected. repair topropagate, including group IIintronsand Rather than knocking outa gene,one might instead someLINE-like transposable elements,and cross- wantto change somespecific aspect of it (e.g.its resistanceto these is highly unlikely. promoter). If thegene was essential, so that therewas strong selectionagainst theknockout, then one could per- In conclusion,the unthinking useof HEGs for popu- form apopulation-widegene replacement by engineering lation controland eradication may lead totheevolution of aHEGthat attackedthe unwanted sequence but not the resistance,as for any other methodof pest management. desiredone, and then simultaneously releasing individuals However,numerous strategies exist for minimizing this carrying theHEG andthe resistant allele intothe popu- likelihood. At thevery least,it isnot obvious a priori that lation. Asshown in figure 4, theresistant allele comesto resistancewill inevitably evolve. predominate relatively quickly after introduction,with only arelatively small andtemporary reductionin mean fitness. 5. PREVENTINGHORIZONTAL TRANSMISSION Finally, onemight also wantto drive anovel geneinto Horizontal transmissionbetween species has beendem- apopulation —toperform apopulation-widetransform- onstratedfor HEGsin plant mitochondria andin yeast ation.To achieve this,the novel genecould be linked to mitochondria andnuclei, and probably occursin all taxa aHEGtargeting aneutral region ofthe , and the in whichHEGs exist (Cho et al. 1998; Goddard& Burt wholeconstruct inserted into the recognition site.As the

Proc.R. Soc.Lond. B (2003) 926 A. Burt Site-speciŽc selŽsh genes and population control

1.0 latter separate from thehost chromosome. Introducing the novel genelinked toa resistancelocus should be even s s

e 0.8 safer.Finally, transformations usinga HEGshouldbe n t i

f fully reversible, by releasing aHEGengineeredto target

r 0.6 o

thenovel gene. y c n

e 0.4 u

q 7. OTHERUSES e r

f 0.2 Twoother potential usesfor engineeredHEGs can briefly bementioned. First, it has long beenrecognized 0 10 20 30 40 50 that if aYchromosomewere to show TRD andspread to time (generations) fixation in apopulation, thenthe sex ratio wouldbecome Figure 4. Population-wide gene replacement. TheHEG and male biased,and if theTRD wasextreme, then the popu- theresistant allele are introduced simultaneously ata 1% lation couldbe driven extinct for wantof females frequency. All other parameters are asin figure 3. (Hickey &Craig 1966; Hamilton 1967). In Aedes and Culex mosquitoes,there are Ychromosomesthat some- howcause the X chromosometo break during thefirst HEGspreadthrough thepopulation, it wouldbring the meiotic division,and thus show TRD (Newton et al. 1976; novel genewith it. Other strategies are also possible.For Sweeny& Barr 1978). Though themolecular mechanism example, aHEGmight beengineered to target an essen- isnot yet known(the breaks may representfailed attempts tial gene,and the novel genelinked toa resistantallele, at crossing-over),these observations suggest the following possibly usingan inversion. This wouldgreatly expandthe strategy. Insertonto a Ychromosomeone or more endo- sizeof the gene one could introduce, even allowing mul- nucleasegenes that recognize andcut sequences specific tiple genesto be introduced simultaneously. It wouldalso tothe X chromosome,and have them underthe control reducethe rate at whichnon-functional mutant genes ofpre-meiotic-specific promoters. Then, during sperma- arise, if DNA replication associatedwith cell division has togenesis,the X chromosomewould be cut, and, as there alowererror rate than that associatedwith DNArepair wouldnot be an appropriate template for repair, theY andgene conversion. chromosomewould show TRD. It wouldspread through Regardlessof how exactly thenovel geneis driven into thepopulation, andif theTRD wassufficiently extreme, thepopulation, akey limitation ofthis approach is that, thepopulation couldbe eradicated. Such an approach if it isharmful, theconstruct will notbe evolutionarily wouldnot rely onrecombinational repair or homing. In stable.Mutant constructs in whichthe novel geneis somespecies the sex chromosomes are inactivated prior deletedor otherwisedefective will beselected, because tomeiosis, which could complicate thedesign of thecon- they causeless harm tothe host but still have thetrans- struct,but this isnot so in all species,including many missionadvantage. However,if thenovel geneis nottoo dipterans(McKee & Handel1993). harmful, andthe mutation rate isnot too high, thenit Second,all themanipulations discussedthus far are canpersist for aconsiderablelength oftime beforegoing ‘inoculative ’,in that therelease of relatively fewengine- extinct.Suppose, for example, theresistant allele in figure eredindividuals will drive thepopulation manipulation. 4has linked toit anovel genethat is fully dominant, Oftenthis will bean advantage, butnot always: if,for reducesfitness by 10% andmutates to a non-functional example, onewanted to eradicate only onepopulation and form with afrequencyof 10 2 6.This novel genewould leave othersin therest of the species range undisturbed, have afrequencygreater than 50% within 15 generations theninoculative methodsmay notbe appropriate. ‘Inun- andgreater than 95% within 40 generations,and would dative’ strategies suchas the release of sterile males remain above 95% for about 4000 generations(not (Knipling 1979) are inherently self-limiting, andso more shown).Full dominanceis critical here,for thenhetero- appropriate for suchpopulation-specific targeting. Engi- zygousmutants have noselective advantage. Introducing neeredHEGs could be used in aninundative strategy if multiple copiesof thegene, linked to one or more resistant they causeddominant female lethality or sterility. Knock- alleles, shouldalso help toreduceselection in favour ofthe outscausing dominant female-specific effects are rare, but non-functionalmutants and prolong thetransformation. if theHEG wasengineered to be constitutively active in Someauthors have suggestedusing transposable all tissues,then even if azygote startedheterozygous, the elementsor cytoplasmic incompatibility agentsas vectors organism wouldbe converted to ahomozygote. Thus,one todrive novel genesthrough apopulation (Ribeiro & couldstill target arecessivefemale-specific locus. Females Kidwell 1994; Turelli &Hoffmann1999), butthe use of inheriting theHEG wouldbe dead or sterile,and males HEGsis likely tohave anumberof advantages. Con- wouldpass on the HEG tothe next generation. As long structsusing transposable elementsare likely tobe less asthe HEG wasnot perfectly efficient( e , 1), it would stable,owing tothe high mutation rate during transpo- slowly disappear from thepopulation, butcould cause a sition,and give lessprecise control over genomic location substantial load beforedoing so.Thus, simply bychang- andcopy number.Constructs using cytoplasmic incom- ing thepromoter, thethreat posedby rare emigrants to patibility agentsare likely tospread more slowly and neighbouring populations canbe avoided. The useof such require larger introductionfrequencies and do not allow engineeredHEGs would be more efficientthan therelease thegene to be in thenucleus. Both alternatives are per- ofsterile males,allowing either fewerindividuals tobe haps more likely totransfer the novel geneto another released,or larger populations tobe targeted (Thomas et speciesthan isa HEGconstruct,because at notime is the al.2000; A.Burt,unpublished data).

Proc.R. Soc.Lond. B (2003) Site-speciŽc selŽsh genes and population control A. Burt 927

8. PROSPECTS Beerntsen, B.T., James, A.A.&Christensen, B.M.2000 Genetics of mosquito vector competence. Microbiol. Mol. The idea ofdriving aforeign geneinto a natural popu- Biol. Rev. 64, 115–137. lation usinga non-Mendeliangenetic element has been Belfort, M., Derbyshire, V.,Parker, M.M.,Cousineau, B.& muchdiscussed, particularly in thecontext of rendering Lambowitz, A.M.2002 Mobile : pathwaysand pro- mosquitoesunable to transmit malaria (e.g.Ribeiro & teins. In Mobile DNAII (ed. N.L.Craig, R.Craigie, M. Kidwell1994; Turelli &Hoffmann1999; Beerntsen et al. Gellert &A.M.Lambowitz), pp. 761 –783. Washington, 2000; Ito et al. 2002). Aswehave seen,engineered HEGs DC:ASMPress. may beuseful tools for suchpopulation geneticengineer- Bellaiche, Y., Mogila, V.&Perrimon, N.1999 I- SceI endonu- ing. However,this remains arelatively complex manipu- clease, anew tool for studying DNAdouble-strand break repair mechanismsin Drosophila. Genetics 152, 1037–1044. lation. The alternative approach, ofusing HEGs to Bibikova, M.,Carroll, D.,Segal, D.J., Trautman, J.K., Smith, perform apopulation-widegene knockout, is muchsim- J., Kim, Y.-G. &Chandrasegaran, S.2001 Stimulation of pler, whetherthe goal is tomodify thepopulation or to homologous recombination through targeted cleavageby eradicate it. The requisite constructconsists of only one chimeric nucleases. Mol.Cell. Biol. 21, 289–297. geneinstead of two, and it is evolutionarily stable in the Braig, H.R.&Yan, G.2002 Thespread of genetic constructs faceof the most obvious mutations that will arise asit in natural insect populations. In Genetically engineered organ- spreadsthrough thepopulation. Knockoutscan also be isms: assessing environmental and human health effects (ed. more easily reversed,by releasing alleles resistantto the D.K.Letourneau &B.E.Burrows), pp. 251 –314. Boca HEG.Theseadvantages suggestthat geneknockouts may Raton, FL: CRCPress. bea more suitable approach topopulation geneticengin- Buchholz, F.&Stewart, A.F.2001 Alteration of Cre recombi- nase site specificity bysubstrate-linked protein evolution. eering than geneintroductions, particularly in thefirst Nature Biotechnol. 19, 1047–1052. attempts. Chandrasegaran, S.&Smith, J.1999 Chimeric restriction Beforeany field trial canbegin, further progress is : whatis next? Biol. Chem. 380, 841–848. neededin thebasic molecular biology ofsite-specific self- Chevalier, B.S.&Stoddard, B.L.2001 Homing endonucle- ish genes.First, further developmentsare neededin the ases:structural and functional insight into thecatalysts of re-targeting ofthese selfish genes to novel hostsequences, intron/intein mobility. Nucleic Acids Res. 29, 3757–3774. though important progress hasbeen made, both by design Chevalier, B.S.,Kortemme, T., Chadsey,M. S., Baker, D., andby selection(Segal et al. 1999; Guo et al. 2000; Monnat Jr, R.J.&Stoddard, B.L.2002 Design, activity, Buchholz &Stewart2001; Wilson et al. 2001; Chevalier and structure of ahighlyspecific artificial . et al. 2002; Santoro& Schultz2002; Seligman et al. 2002; Mol. Cell. 10, 895–905. Takahashi &Fujiwara 2002). Second,it mustbe demon- Cho, Y., Qiu, Y.-L., Kuhlman, P.&Palmer, J.1998 Explosive invasion of plant mitochondria bya group Iintron. Proc. stratedthat theseelements can be made to work in the Natl Acad.Sci. USA 95, 14 244–14 249. prospectivetarget species:insects, for example, are natural Curtis, C.F.1985 Genetic control of insect pests:growth hostsonly for theLINE-like elements.Encouragingly, a industry or lead balloon? Biol. J.Linn.Soc. 26, 359–374. yeastHEG hasrecently been shown to beactive in stimul- Eickbush,T. H.2002 R2and related site-specific non-long ter- ating recombinational repair in D.melanogaster (Bellaiche minal repeat retrotransposons. In Mobile DNAII (ed. N. L. et al. 1999; Rong &Golic2000; Rong et al. 2002). Craig, R.Craigie, M.Gellert &A.M.Lambowitz), pp. 813 – In carrying outthis work,it shouldbe noted that the 835. Washington, DC:ASMPress. easeand rapidity with whichthese selfish genes can invade Goddard, M.R.&Burt, A.1999 Recurrent invasion and apopulation applies notjust to planned releases, but also extinction of aselfishgene. Proc.Natl Acad.Sci. USA 96, tounintentional releases of laboratory escapees.Proper 13 880–13 885. Goddard, M.R., Greig, D.&Burt, A.2001 Outcrossed sex attentionto containment issues is neededto preventnatu- allows aselfishgene to invade yeastpopulations. Proc. R. ral populations of D.melanogaster ,totake anobvious Soc. Lond B 268, 2537–2542. (DOI 10.1098/rspb.2001. example, from being infected(and possibly endangered) 1830.) by engineeredselfish genes that weremeant tostay in the Guo, H., Karberg, M., Long, M., Jones III, J.P., Sullenger, laboratory. Finally, wide-ranging discussionsare needed B.&Lambowitz, A.M.2000 Group IIintrons designed to onthe criteria for decidingwhether to eradicate or geneti- insert into therapeutically relevant DNAtarget sitesin cally engineeran entire species. Clearly, thetechnology human cells. Science 289, 452–457. describedhere is notto beused lightly. Giventhe suffering Hamilton, W.D.1967 Extraordinary sexratios. Science 156, causedby somespecies, neither is it obviously oneto be 477–488. ignored. Hastings, I.M.1994 SelfishDNA asamethod of pestcontrol. Phil. Trans.R. Soc. Lond. B 344, 313–324. Hickey, W.A.&Craig Jr, G.B.1966 Distortion of sexratio M.Ashburner, C.Godfray, B.Sinden and B.Trivers provided in populations of Aedes aegypti . Can.J. Genet. Cytol. 8, useful discussions and/or comments, and theBiosocial 260–278. Research Foundation and theWellcome Trust supplied finan- cial support. Ito, J., Ghosh, A., Moreira, L.A., Wimmer, E.A.&Jacobs- Lorena, M.2002 Transgenic anopheline mosquitoes impaired in transmission of amalaria parasite. Nature 417, REFERENCES 452–455. Jacquier, A.&Dujon, B.1985 An intron-encoded protein is Ashburner, M.1989 Drosophila: alaboratory handbook . Cold active in agene conversion process thatspreads an intron Spring Harbor, NY:Cold Spring Harbor Laboratory Press. into amitochondrial gene. Cell 41, 383–394. Ashburner, M.(and 25 others) 1999 An exploration of the Jordan, I., Matyunina, L.&McDonald, J.1999 Evidence for sequence of a2.9-Mb region of thegenome of Drosophila therecent horizontal transfer of long terminal repeat retro- melanogaster : the Adh region. Genetics 153, 179–219. transposon. Proc.Natl Acad.Sci. USA 96, 12 621–12 625.

Proc.R. Soc.Lond. B (2003) 928 A. Burt Site-speciŽc selŽsh genes and population control

Knipling, E.P.1979 The basic principles ofinsect population sup- Seligman, L.M., Chisholm, K.M., Chevalier, B.S.,Chadsey, pression and management .Washington, DC:USDepartment M.S., Edwards, S.T.,Savage,J. H.&Veillet, A.L.2002 of Agriculture. Mutations altering thecleavage specificity of ahoming Koufopanou, V.,Goddard, M.R.&Burt, A.2002 Adaptation endonuclease. Nucleic Acids Res. 30, 3870–3879. for horizontal transfer in ahoming endonuclease. Mol. Biol. Spielman, A., Beier, J.C.&Kiszewski, A.E.2002 Ecological Evol. 19, 239–246. and community considerations in engineering arthropods to McKee, B.D.&Handel, M.A.1993 Sexchromosomes, suppressvector-borne disease. In Genetically engineered recombination, and chromatin conformation. Chromosoma organisms:assessing environmental and human health effects (ed. 102, 71–80. D.K.Letourneau &B.E.Burrows), pp. 315 –329. Boca Miklos, G.L.G.&Rubin, G.M.1996 Therole of thegenome Raton, FL: CRCPress. project in determining gene function: insights from model Sweeny, T.L.&Barr, A.R.1978 Sexratio distortion caused organisms. Cell 86, 521–529. bymeiotic drive in amosquito, Culex pipiens . L. Genetics 88, Newton, M.E., Wood, R.J.&Southern, D.I.1976 Acyto- 427–446. genetic analysisof meiotic drive in themosquito, Aedes Takahashi,H. &Fujiwara, H.2002 Transplantation of target aegypti (L.). Genetica 46, 297–318. site specificity byswappingthe endonuclease domains of two Palumbi, S.2001 Humans asthe world ’sgreatest evolutionary LINEs. EMBO J. 21, 408–417. force. Science 293, 1786–1790. Thomas,D. D.,Donnelly, C.A., Wood, R.J.&Alphey, L.S. Ribeiro, J.M.C.&Kidwell, M.G.1994 Transposable 2000 Insect population control using adominant, repress- elements aspopulation drive mechanisms: specification of ible, lethal genetic system. Science 287, 2474–2476. critical parameter values. J.Med.Entomol. 31, 10–16. Turelli, M.&Hoffmann, A.A.1999 Microbe-induced cyto- Rong, Y.S.&Golic, K.G.2000 byhomolo- plasmicincompatibility asa mechanismfor introducing gous recombination in Drosophila. Science 288, 2013–2018. transgenes into arthropod populations. Insect Mol.Biol. 8, Rong, Y.S., Titen, S.W., Xie, H.B., Golic, M.M., Bastiani, 243–255. M., Bandyopadhyay, P., Olivera, B.M.,Brodsky, M., Wenzlau, J., Saldanha, R., Butow, R.&Perlman, P.1989 A Rubin, G.M.&Golic, K.G.2002 Targeted mutagenesis by latent intron-encoded maturase is also an endonuclease homologous recombination in D.melanogaster . Genes Dev. needed for intron mobility. Cell 56, 421–430. 16, 1568–1581. White, N.(and 16others) 1999 Averting amalaria disaster. Santoro, S.W.&Schultz, P.G.2002 Directed evolution of Lancet 353, 1965–1967. thesite specificity of Cre recombinase. Proc.Natl Acad.Sci. Wilson, D.S.,Keefe, A.D.&Szostak, J.W.2001 Theuse of USA 99, 4185–4190. mRNAdisplay to select high-affinityprotein-binding pep- Segal, D.J., Dreier, B., Beerli, R.R.&BarbasIII, C.F.1999 tides. Proc.Natl Acad. Sci. USA 98, 3750–3755. Toward controlling gene expression atwill: selection and design of domains recognizing eachof the5 9- GNN-39 DNAtarget sequences. Proc.Natl Acad.Sci. As this paper exceedsthe maximum lengthnormally permitted, the USA 96, 2758–2763. author has agreedto contributeto production costs.

Proc.R. Soc.Lond. B (2003)