Evidence That Mutation Is Universally Biased towards AT in Bacteria

13 Pages·2010·1.14 MB·English

Evidence That Mutation Is Universally Biased towards AT
in Bacteria
Ruth Hershberg*, Dmitri A. Petrov
DepartmentofBiology,StanfordUniversity,Stanford,California,UnitedStatesofAmerica
Abstract
Mutation is the engine that drives evolution and adaptation forward in that it generates the variation on which natural
selection acts. Mutation is a random process that nevertheless occurs according to certain biases. Elucidating mutational
biasesandthewaytheyvaryacrossspeciesandwithingenomesiscrucialtounderstandingevolutionandadaptation.Here
we demonstrate that clonal pathogens that evolve under severely relaxed selection are uniquely suitable for studying
mutational biases in bacteria. We estimate mutational patterns using sequence datasets from five such clonal pathogens
belongingtofourdiversebacterialcladesthatspanmostoftherangeofgenomicnucleotidecontent.Wedemonstratethat
acrossdifferenttypesofsitesandinallfourcladesmutationisconsistentlybiasedtowardsAT.Thisistrueevenincladesthat
havehighgenomicGCcontent.InallstudiedcasesthemutationalbiastowardsATisprimarilyduetothehighrateofC/GtoT/
Atransitions.Theseresultssuggestthatbacterialmutationalbiasesarefarlessvariablethanpreviouslythought.Theyfurther
demonstratethatvariationinnucleotidecontentcannotstementirelyfromvariationinmutationalbiasesandthatnatural
selectionand/oranaturalselection-likeprocesssuchasbiasedgeneconversionstronglyaffectnucleotidecontent.
Citation: Hershberg R, Petrov DA (2010) Evidence That Mutation Is Universally Biased towards AT in Bacteria. PLoS Genet 6(9): e1001115. doi:10.1371/
journal.pgen.1001115
Editor:MichaelW.Nachman,UniversityofArizona,UnitedStatesofAmerica
ReceivedFebruary9,2010;AcceptedAugust9,2010;PublishedSeptember9,2010
Copyright:!2010Hershberg,Petrov.Thisisanopen-accessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense,whichpermits
unrestricteduse,distribution,andreproductioninanymedium,providedtheoriginalauthorandsourcearecredited.
Funding:RHwasfundedbyaStanfordGenomeTrainingProgram(SGTP)fellowshipworksupportedbytheNationalInstitutesofHealthgrantGM077368to
DAP.Thefundershadnoroleinstudydesign,datacollectionandanalysis,decisiontopublish,orpreparationofthemanuscript.
CompetingInterests:Theauthorshavedeclaredthatnocompetinginterestsexist.
*E-mail:[email protected]
Introduction bacterial mutational biases must to be extremely variable to be
able to generate the extreme variation observed in bacterial
Mutationgeneratesthevariabilityonwhichnaturalselectionacts. nucleotide content.
Mutationisnotanentirelystochasticprocess,asitactsaccordingto Asecondpossibilityisthatitisnotvariationinmutationalbiases
certaindeterministicbiases.Becauseofthis,biasesintheoutcomeof thatleadstovariationinnucleotidecontent,butrathervariationin
theevolutionaryprocessresultnotonlyfromselectionbutalsofrom the relative probabilities of fixation of A/T to G/C and G/C to
thebiasesofmutation.Inordertounderstandevolutionitistherefore A/T mutations [1,3,8]. When considering changes to nucleotide
necessarytoelucidatemutationalbiasesandthewaysinwhichthese content, differences in fixation probabilities can stem both from
biasesthemselveschangeinevolution. differences in the strength and direction of natural selection and
Nucleotide content variation is much more pronounced in differencesintheratesofbiasedgeneconversion(BGC)[1,9,10].
bacteriacomparedtomulti-cellulareukaryotes[1].GCcontentsin Natural selection affects the probability of fixation of an allele
bacteria vary from less than 25% to over 75% [1–3]. Related based on the alleles fitness advantage or disadvantage and the
bacteriaevenfromrelativelybroadphylogeneticgroupingstendto effectivepopulationsize(N)oftheorganisminquestion.Similarly
e
show similar genomic nucleotide content [3]. For example BGC is also dependent on N and on the advantage or
e
bacterial genomes from the order Bacillales tend to be GC-poor, disadvantage the allele has. Here however, this advantage or
from the order Enterobacteriales to have intermediate GC disadvantage is not in fitness but rather in the increased or
contents, and from the phylum Actinobacteria to be GC-rich decreased probability of the allele to be passed on to the next
(Figure1A).Inaddition,GCcontentvaluesmeasuredatdifferent generation through gene conversion [1,10]. This increase or
functional site categories (intergenic, synonymous, and non- decrease is determined by recombination rates and by the
synonymous) show highly correlated patterns of variation across conversion bias, which has been shown in many eukaryotes to
bacteria [4] (Figure 1B). These observations suggest that forces beinfavorofGCnucleotidescomparedtoATnucleotides[1,9].
determining GC content in bacteria operate both genome-wide A recent study that showed that in Escherichia coli regions of low
and consistently over long periods of time. One possibility isthat recombination tend to be more AT rich demonstrates that BGC
the main force driving nucleotide content variation in bacteria is mayaffectnucleotidecontentinbacteriainasimilarmanner[11].
mutation. This possibility has often been assumed true (for Inordertogaininsightintomutationalbiasesitisnecessaryto
example see [2,4–7]). Under this assumption, clades that are GC investigate the results of mutation in isolation from those of
rich are clades in which mutation has been consistently biased selectionandBGC.Wheneffectivepopulationsizesaresmall,the
towards GC while clades that are AT rich are clades in which efficacy of both natural selection and BGC is severely reduced
mutation has been consistently biased towards AT. If true, relativetostochasticprocessesandthereforesequenceevolutionis
PLoSGenetics | www.plosgenetics.org 1 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
differences between them can be viewed as polymorphisms. In
Author Summary
addition, the lifestyle and clonality of such pathogens is likely to
Naturalselectionsortsthroughthevariabilitygeneratedby leadtosmallN,furtherreducingefficacyofnaturalselection[25].
e
mutation and biases evolution toward fitter outcomes. ThepatternsofSNPsamongcloselyrelatedstrainsofsuchclonal
However,becausemutationisitselfnotentirelyrandomit pathogensshouldthusreflectdirectlythepredominantmutational
can also bias the direction of evolution independently of biases.
selection. For instance, it is often assumed that the Here we estimate mutational biases by analyzing SNPs
extreme variation observed in nucleotide content among extracted from large sequence datasets of five lineages of clonal
bacteria(from,20%to,80%GC)ispredominantlydriven
pathogens (including MTBC) from four broad clades of bacteria
by extreme differences in mutational biases towards or
thatspanvirtuallythewholebacterialphylogenyandtherangeof
awayfromGC.Here,weshowthatbacteriallineagesthat
bacterial nucleotide contents (Figure 1A): Bacillales (AT rich),
recently developed clonal, pathogenic lifestyles evolve
Enterobacteriales (intermediate GC), Actinobacteria (high GC),
under weak selection and that polymorphisms in these
and Burkholderiales (high GC). We find that in all lineages
bacteriacanbeusedasafairproxyformutationalspectra.
mutationisbiasedtowardsAT,andthatG/CtoA/Ttransitions
We analyze large sequence datasets from five clonal
are always predominant. Previous studies indicate that mutation
pathogens in four diverse bacterial clades spanning most
oftherangeofgenomicnucleotidecontent.Wefindthat, may be universally AT-biased in eukaryotes [21,26–31]. Our
surprisingly, mutationisAT-biased ineverycaseto avery results together with additional studies that have focused on
similar degree and in each case it is dominated by Enterobacteriales (E. coli, Shigella, and Salmonella typhimurium)
transitions from C/G to T/A. This demonstrates that [32,33]demonstratethat mutationmaybeuniversally AT-biased
mutational biases are far les variable than previously in bacteria as well. These findings contradict the long-held view
assumedandthatvariationinbacterialnucleotidecontent that mutational biases are the main contributors to variation in
is not due entirely to mutational biases. Rather natural bacterial nucleotide content and are therefore highly variable
selection or a selection like process such as biased gene among bacteria. Rather they suggest that nucleotide content in
conversion strongly affectnucleotide content inbacteria. bacteria is strongly affected by variation in the relative rates of
fixation of AT to GC and GC to AT mutations and that
mutational biasesare farlessvariable thanpreviously thought.
affected strongly by mutational biases. Mutation-accumulation
experiments artificially reduce N of evolving laboratory cultures Results
e
[12]andcanthusbeusedtoassessmutationalbiasesinculturable
bacteria. Similarly, reporter constructs have also been used to Sequence data from lineages of clonal pathogens are
estimatemutationalbiases[13,14].However,withoutknowingthe extremely suited to the study of bacterial mutational
relativeamountoftimebacteriaspendindifferentgrowthphases biases
(logarithmic vs. stationary) and given that mutational rates and We focused on five lineages of clonal pathogens from four
patternsvarybetweengrowthphases[15,16]itcouldbedifficultto diversebacterialclades(Table1).Thefivelineagesweinvestigated
estimatethetruemutationalbiasesoperatinginnatureusingsuch areuniqueintheirsuitabilityforthistypeofanalysisbecausethey
experimental approaches. An additional approach is to examine provide us with sufficient amounts of available sequence data for
nucleotidesubstitutionsatsitesthatareexpectednottobesubject sufficiently closely related strains in which we can demonstrate a
toselectionduetoproteinfunctionality,suchaspseudogenes[17], genome-wide relaxation in the efficacy of natural selection. The
or fourfold degenerate sites [18]. This approach is also chosen strains are indeed very closely related with each lineage
problematic because while pseudogenes and fourfold degenerate exhibiting less than 0.5 pairwise differences per gene (Table 1).
sites are expected to be under no, or low selection for protein However, because of the availability of multiple whole genome
functionality,theyshouldbesubjecttothesamelevelsofselection sequencesineachlineagethetotalnumberofSNPsissubstantial
on nucleotide content astherest ofthegenome. (Table1),rangingfrom165to1877.Inaddition,theselineagesare
A good way to estimate mutational biases is to analyze the thought to be clonal [34]. Thus, horizontal gene transfer should
patternsofsinglenucleotidepolymorphisms(SNPs)withinspecies. occur only rarely, if at all and should not strongly influence our
Population genetic studies have shown that natural selection and ability to infer the ancestral and derived states of mutations.
otherselection-likeprocessesarelessefficientinaffectingpatterns Finally,theinferenceofSNPsfromsuchcloselyrelatedsequences
of nucleotide polymorphisms among very closely related strains is almost trivial, as alignment programs do much better when
compared to nucleotide differences between distantly related sequences are highly similar. Therefore, we expect to have no
strains or species [19,20]. Thus SNPs should better reflect the biasesintroduced through misalignment of sequences.
mutational patterns compared to substitutions between species. We assessed whether the patterns of SNPs in these data are
TheanalysisofSNPshasbeenusedtoinvestigatethemutational indeedweaklyaffectedbynaturalselectionbyestimatingtheratio
biases of a number of AT-rich eukaryotic genomes, such as of non-synonymous and synonymous differences per non-synon-
Drosophila[21–23].However,usingthismethodologyinbacteria ymous and synonymous site (dN/dS) [35,36] across all alignable
has been problematic due to a severe blurriness in species proteinswithineachdataset(MaterialsandMethods).Ifselection
boundariesamongprokaryotes[24].Asaresultofsuchblurriness is strong, dN/dS should be much smaller than 1 as it would
quiteoftenstrainssharinga‘‘species’’name(suchasE.coli)arein efficiently remove most non-synonymous mutations [35,36]. For
factquitediverged.Whileitisdifficulttodefinespeciesinbacteria, example, comparisons of E. coli strains yields dN/dS values of
severely reduced selection has been observed among very closely approximately 0.05 [37]. In contrast, for MTBC where multiple
related bacterial strains of some strictly clonalpathogens, such as linesofevidencesuggestthatnaturalselectionisseverelyreduced
strains belonging to the Mycobacterium tuberculosis cluster (MTBC) [25],dN/dSgoesupto0.59.Inourdataset,dN/dSvaluesforthe
[25]. This can be explained by the fact that strains belonging to other four lineages range between 0.45 and 0.64 (Table 1,
such lineages of pathogens are extremely closely related and thus Materials and Methods). This suggests that selection is indeed
these lineages may be good proxies of ‘‘species’’ and nucleotide relaxedinagenome-widemannerinthesegenomesandthusthat
PLoSGenetics | www.plosgenetics.org 2 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
PLoSGenetics | www.plosgenetics.org 3 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Figure1.PhylogeneticandgenomicvariationinGCcontent.(A)Thephylogenyofthefourbroadcladesexaminedinthisstudy.Itwasbuilt
usingtheiTOLwebpage(http://itol.embl.de/).AverageGCcontentofthedifferentbroadcladesareindicatedonthemargins.Smallbluetriangles
representthefivelineagesofclonalpathogensusedintheanalysis.(B)GenomewideobservedGCcontentofsynonymousandnon-synonymous
sitescorrelatewiththeintergenicGCcontentacrossbacterialgenomes.
doi:10.1371/journal.pgen.1001115.g001
thepatternofSNPsshouldbereflectiveofthemutationalbiasesin content, the predominant mutation is G/C to A/T transition
these lineages. (Figure 2).
Relaxedselectionoverlongevolutionarytimescalescanleadto It is important to remember that relaxation of selection in the
extremegenomereductionandthelossofmanyrepairpathways, studied lineages is fairly recent and that nucleotide content is a
which could affect mutational patterns [1,38]. The pathogens we slowly evolving trait. Therefore, if driven by selection or BGC,
useinthisstudyhavesufferedonlyashort-termrelaxationinthe nucleotide content should not have had time to reach a new
efficiencyofselectionandthereisnoindicationthatanyofthem mutational equilibrium. However, if nucleotide content is driven
have lost repair functions. To further substantiate that these predominantly by mutation, and selection and BGC do not
pathogensarenotlikelytohavesufferedlossofrepairfunctionswe stronglyaffectnucleotidecontentthegenomicnucleotidecontents
examined whether any of the repair genes annotated in close shouldalreadybeatthemutationalequilibrium.Wetestwhether
relatives of the examined pathogens that are not evolving under nucleotidecontent isatequilibrium bycomparingthenumber of
inefficient selection, have been lost in the pathogens. We found GCRATandATRGCchangesobservedineachdataset.Under
thatinB.anthracis,B.mallei,S.typhiandY.pestistherehasbeenno equilibrium these numbers will be equal. The results of such
lossofrepairgenesandthatinallcasestherepairgenesarehighly comparisons (Table 2) clearly show that the lineages with
similar to those found in the outgroups (Materials and Methods, intermediate(Salmonellatyphi&Yersiniapestis)andhigh(Burkholderia
TableS1).InthecaseofMTBC,sincethecloselyrelatedoutgroup mallei & MTBC) nucleotide contents are currently far from
strainM.canettiiisnotfullysequencedwecouldonlycomparethe equilibriumandthatGCRATchangesaremuchmorefrequent.
genespresentinfullysequencedstrainsofMTBCtothosepresent These results are statistically significant for all but the intergenic
inamoredistantlyrelatedoutgroup,M.marinum.Allbutoneofthe datasetofB.mallei,inwhichasmallnumberofSNPsleadstovery
repairgenesfoundinM.marinumarealsofoundinMTBC(Table low statistical power (Table 2).
S1). The M. marinum gene that is not found in MTBC is an un- Theaboveresultsindicatethatundercontinuedrelaxedselection
namedgeneofunclearfunction.Theseresultstogetherwithalack the lineages with intermediate or high GC content will evolve to
of previous evidence for loss of repair functions in these well becomemoreATrich.Itiseasytocalculatetheexpectedequilibrium
studiedpathogensmakesitunlikelythatthesepathogenshavelost GCcontentbasedonthemutationalrateswefound(GC ,Materials
eq
repair functions. It is even less likely that all of them suffered andMethods)[1,8,27].Suchacalculationshowsthatforalllineages
similar losses of repair functions. GC is lower than 50% (Figure 3, Table 3). In other words,
eq
mutational biases by themselves should lead to AT-rich genomes.
Mutation is AT-biased independently of the current Furthermore, if nucleotide content is primarily determined by the
nucleotide content of bacterial clades mutational biases of the genome, GC should approximate the
eq
We polarized the SNPs (Materials and Methods) and observed GC content genomewide, However, for the four clonal
classified changes into six possible types of mutations (G/C to pathogen lineages with either intermediate or high GC contents,
A/T, G/C to T/A, G/C to C/G, A/T to G/C, A/T to C/G, GC isalwayssignificantlylowerthanthecurrentgenomewideGC
eq
andA/TtoT/A).Therelativerateofeach ofthesixmutation at all types of sites (Figure 3, Table 3). Finally, no significant
types was calculated after normalizing for the current GC correlation was observed between GC and current GC across all
eq
contentatthestudiedpositions(MaterialsandMethods).Forall sitecategoriesinalllineages(r=0.09,Spearmancorrelation,n=14,
five lineages, irrespective of current genomewide nucleotide P#1. Note that the 14 data points examined are not entirely
Table1. Clonalpathogen lineages analyzedin study.
Maxpairwise
Clonalpathogen SourceofSNPs Outgroup variabilitya dN/dS #SNPs
Non-
Intergenic synonymous Synonymous
Bacillusanthracis Alignmentsof18fullyand B.thuringiensis 0.3 0.58 322 239 112
partiallysequencedstrains
Salmonellatyphi SNPsprovidedbyHoltetal.[52] None 0.4 0.45 260 961 656
basedonthesequencingof19 (phylogenetictree)
strains
Yersiniapestis Alignmentsof7fullysequenced Y.pseudotuberculosis 0.2 0.64 118 345 162
strains
Burkholderiamallei Alignmentsof11fullyand B.thailandensis 0.1 0.47 44 70 51
partiallysequencedstrains
MTBC Alignmentof89genessequenced M.canettii 0.5 0.59 NA 226 136
in107strains
aTheaveragepairwisediversitypergeneofthetwomostdivergedstrainswithinthelineage.
doi:10.1371/journal.pgen.1001115.t001
PLoSGenetics | www.plosgenetics.org 4 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Figure2.Relativeratesofthesixnucleotidepairmutations.ThemostcommonmutationisalwaysG/CtoA/Ttransitions.Theratesare
normalizedfortheunequalnucleotidecontentofthefivedifferentlineages(MaterialsandMethods).(A)non-synonymousSNPs.(B)synonymous
SNPs.
doi:10.1371/journal.pgen.1001115.g002
independent.ThecalculatedP-valueshouldthereforebetakenwitha spanalargeportionofbacterialphylogeny(Figure1A)mutationis
grainofsalt,andisonlyprovidedtodemonstratethatGC doesnot consistentlybiasedtowardsAT.Furthermore,theseresultssuggest
eq
appeartobecorrelatedtocurrentGC). that in bacteria that are GC rich or have intermediate GC
In order to show that these results are not an artifact of contents, it is an elevated fixation of GC-enriching mutations
sequencing errors, we recalculated GC after removing all cases ratherthanachangeinmutationbiasthatdrivestheelevatedGC
eq
in which the derived allele appears only in a single genome content.
(singletons). While this reduces the number of SNPs per dataset Differencesbetweengenomesthataremoredistantlyrelatedto
andincreasestheerroroftheestimate,GC valuesremainlower each other should reflect the effects of natural selection and/or
eq
than the current GC for all datasets from lineages with BGC better. The GC calculated based on such differences
eq
intermediate or high GC contents and the results remain should be more similar to the observed GC content than that
statistically significant in all the datasets except for the B. mallei calculated based on SNPs from more closely related strains. To
intergenicsites(TableS2).Theseresultstogethershowthatinfive examine these predictions we analyzed two additional datasets in
lineages of clonal pathogens, belonging to four broad clades that which sequences are still closely related enough to create reliable
PLoSGenetics | www.plosgenetics.org 5 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Table2. Nucleotidecontentsofclonal pathogenswith intermediate andhigh GCcontentsare far from equilibrium.
Clonalpathogen CurrentGCcontent Sites #GCRATa #ATRGCa
MTBC High Synonymous 103(84,123) 18(10,26)
MTBC High Non-synonymous 127(105,151) 58(43,73)
B.mallei High Synonymous 45(33,59) 2(0,5)
B.mallei High Non-synonymous 57(42,73) 7(2,12)
B.mallei High Intergenic 27(17,38) 12(6,19)
Y.pestis Intermediate Synonymous 116(95,138) 37(26,48)
Y.pestis Intermediate Non-synonymous 230(201,259) 79(63,97)
Y.pestis Intermediate Intergenic 69(52,85) 36(25,47)
B.anthracis Low Synonymous 44(32,57) 64(49,79)
B.anthracis Low Non-synonymous 136(114,158) 75(58,93)
B.anthracis Low Intergenic 141(119,166) 151(128,175)
S.typhi Intermediate Synonymous 570(524,619) 79(62,97)
S.typhi Intermediate Non-synonymous 707(654,760) 220(192,248)
S.typhi Intermediate Intergenic 189(164,215) 65(50,81)
a95%Confidenceintervalsappearinparenthesis.Numbersappearinboldifthereisastatisticallysignificant(p,0.05)differencebetweennumberofGCRATand
numberofATRGCchanges.
doi:10.1371/journal.pgen.1001115.t002
multiplesequencealignmentsforarelativelylargenumberoftheir thatintheseorganismsnucleotidecontentwillbestronglyaffectedby
genesandintergenicregions,yetshowhigherdivergencethanthe mutation.Thegenomesofobligatoryintracellularbacteriatendtobe
fivelineagesinTable1(Table4).Thefirstdatasetwascreatedby ATrich[38],indicatingthatmutationmaybebiasedtowardsATin
aligning sequences from Y. pestis and Yersinia pseudotuberculosis and these organisms. However, it is unknown whether the mutational
examiningthedifferencesbetweenthesetwolineages.Thesecond biases of obligate intracellular bacteria reflect those of their clade
dataset was created by aligning sequences from 10 strains of membersthatarenotlivingthislifestyle,sincetheseorganismslack
Burkholderia pseudomallei. These two datasets were selected because many of the repair genes which are present in their other clade
theycanbenaturallypairedtotwoofthefivedatasetsinTable1 membersthatarenotevolvingundersuchrelaxedselection[38].Itis
(B. mallei and Y. pestis). dN/dS values for these comparisons are unclear how the absence of repair pathways affects the mutational
lowerthanintheotherfivedatasets(x2test,P,0.00001,Table4), biases of these bacteria. Lind and Anderson [33] have carried out
suggestingthattheeffectsofselectionareindeedmoreevidentin mutation accumulation experiments in Salmonella typhimurium strains
thesedata.Aspredicted,theGCeqvaluesderivedfromdifferences lackingthemajorDNArepairsystemsinvolvedinrepairingcommon
between Y. pestis and Y. pseudotuberculosis and from B. pseudomallei spontaneous mutations caused by deaminated and oxidized DNA
comparisons are more similar to the current GC content at all bases. They found that in such strains of S. typhimurium mutation is
types of sites compared to GCeq calculated from the paired stronglyAT-biasedbutthatunlikewhatwefindforourpathogens,
datasets of clonal pathogens (Figure 4). These results are whentheserepairpathwaysareabsenttransversionsbecomemuch
statistically significant for the non-synonymous and synonymous morefrequentthantransitions[33].Thequestionhoweverremains
sitecomparisons(Figure4,Table4).Forintergeniccomparisonsa ofwhetherlossofrepairpathwayschangestheATbiasofmutationin
small number of identified B. mallei SNPs, and Y. pestis/Y. obligateintracellularbacteria.
pseudotuberculosis differences makes it difficult to demonstrate It is also unclear whether the nucleotide contents of obligate
statistical significance. Furthermore, the GCeq values calculated intracellular bacteria are currently at mutational equilibrium.
based on the more diverged datasets are significantly correlated Even though these organisms have been subject to prolonged,
with the current genomewide GC content in Y.pestis and B.
severe relaxations ofselection itispossiblethat theyhavenot yet
pseudomallei (r=0.99, Spearman correlation, n=6, P#0.03). It is
reached nucleotide content equilibrium. It is also theoretically
important to note that the GC values derived from differences
eq possible that the strength of selection in favor of GC nucleotides
betweenY.pestisandY.pseudotuberculosisandfromB.pseudomalleiare
increasesasgenomesbecomemoreATrich(afromofsynergistic
stilllowerthancurrentGCcontentforallcomparisons(Figure4).
epistasis[26,39,40]).Ifthisisthecaseitispossiblethatnucleotide
This indicates that even a relatively slight reduction in selection
content in obligate intracellular bacteria may be at a new
canleadtoacertainreductioninGCcontent.Asimilarresultwas
selection-mutation equilibrium.
recently found for E. coli and Shigella, where a relatively slight
The questions of whether the mutational biases of obligate
reductionintheefficiencyofselection[37]wasshowntoleadtoan
intracellular bacteria resemble those of their other clade members
excess of GCRAT changesinShigella compared to E. coli[32].
andofwhethertheirnucleotidecontentisatmutationalequilibrium
can be addressed by comparing our estimates of GC to the GC
eq
Nucleotide content of obligate intracellular bacteria
content of obligatory intracellular bacteria from the same clades.
matches those predicted based on estimates of Among the four analyzed clades, two (Enterobacteriales, and
mutational biases Actinobacteria)includesequencedobligatoryintracellularbacteria.
Obligate intracellular bacteria are known to be evolving under Intriguingly, the GC contents of the obligate intracellular bacteria
extremely prolonged relaxed selection [38]. It is therefore possible withinthesetwocladesaresimilartothevaluesofGC whichwe
eq
PLoSGenetics | www.plosgenetics.org 6 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Figure3.EquilibriumGCcontent(GC )andtheobservedGCcontentinthefivestudiedclonalpathogenlineages.GC iscalculated
eq r eq
from the estimated rates AT to GC and GC to AT mutations (rATRGC and rGCRAT) as GCeq~r ATz?rGC . GCeq values are
AT?GC GC?AT
significantly lower than the observed GC content at all site categories (intergenic, synonymous and non-synonymous) and for all
four lineages of clonal pathogens with either intermediate or high GC contents. MTBC intergenic SNPs were not available for
analysis.Errorbarsdepict95%confidenceintervalsforGC .NocorrelationisobservedbetweenGC andcurrentGCfortheclonal
eq eq
pathogen lineages (P#1).
doi:10.1371/journal.pgen.1001115.g003
calculatedusingclonalpathogenSNPdatafromtheircorresponding suggest that the GC content of obligate intracellular bacteria
clades: GC values calculated based on SNPs from the Enter- correspondstowhatwewouldexpecttofindintheirnon-obligate
eq
obacteriales S. typhi, and Y. pestis were ,22%, and ,25% intracellularclademembersatequilibriumifnucleotidecontenthad
respectively. While other members of the order Enterobacteriales been determined by mutation alone. This anecdotal evidence
haveacurrentGCcontentof43–57%,obligateintracellularbacteria suggeststhateventhoughobligateintracellularbacteriadotendto
withinthisorderhaveGCcontentsof23–30%.Withinthephylum lose many repair functions the extent of their mutational AT bias
ActinobacteriaallgenomeshaveGCcontentofover50%exceptfor resemblesthatoftheirotherclademembers.Additionally,itappears
theonlysequencedobligateintracellularbacteria,Tropherymawhipplei that nucleotide contents of these bacteria are close to mutational
that has a GC content of 46%. This is very close to the GC equilibrium. The observation that obligate intracellular bacteria
eq
calculated for the Actinobacteria MTBC (,42%). These results across additional clades tend to have high AT contents relative to
PLoSGenetics | www.plosgenetics.org 7 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Table3. Comparisonsofcurrent GCcontent and GC forfive clonal pathogenlineages.
eq
Current GCeq CurrentGC GCeq Current GCeq
Clonalpathogen GCintergenic intergenica non-synonymous non-synonymousa GCsynonymous synonymousa
Bacillusanthracis 32.5 34 39.5 26.5 23.4 30.8
(29.2,38.9) (21.3,32.2) (23.8,40)
Salmonellatyphi 46.1 22.8 50.1 23.8 62.3 18.6
(17.8.27.8) (21.2,26.6) (15.7,22.8)
Yersiniapestis 42.5 27.8 48.1 24.1 51.7 25.4
(20.2,36.3) (19.8,29) (18.9,32.7)
Burkholderiamallei 67.8 48.3 61.1 16.2 90.5 29.9
(29,63.3) (5.3,26.6) (0,54)
MTBC 62.8 Missingdata 60.8 41.5 80.2 41.6
(33.8,48.5) (29.1,52.3)
a95%Confidenceintervalsappearinparenthesis.BoldfontindicatesGCeqvaluesaresignificantlydifferentfromcorrespondingcurrentGCcontents.
doi:10.1371/journal.pgen.1001115.t003
their non-obligate intracellular clade members, and to have GC are being purged by selection. Yet it is still possible that our
contents lower than 50% (Table S3, Figure S1) may therefore estimatesofmutationalbiasesaresomewhataffectedbytheresidual
supportthegeneralityofAT-biasedmutationinbacteriabeyondthe effectsofnaturalselectiononstronglydeleteriousmutations.Ifsuch
fourphylogeneticallydiversebroadcladesexaminedinthisstudy. residualeffectsofselectionaffectourresultstheyarelikelytomake
ourestimatesoftheextentoftheATbiassomewhatconservative.
Discussion This is demonstrated by our finding that within a given clade a
stronger bias towards AT is observed in clonal pathogens where
Differencesinthepatternsofmutationhavebeenassumedtobe
selection is severely relaxed compared to closely related lineages
the most likely explanation for the vast variation observed in
subjecttomoreefficientselection.Thisbeingsaid,ourobservation
nucleotidecontentacrossbacteria(forexamplesee[2,4–7]).Given
thatwithinabroadcladeobligateintracellularbacteriaevolvetoa
that bacterial nucleotide content varies between .75% GC and
GCcontent that matchesverycloselytheonepredictedbasedon
.75%AT,thisassumptionimpliesthatpointmutationbiasesare
ourestimatesofmutationsuggeststhatourestimatesoftheextentof
extremely variable among bacteria. Our results demonstrate that
ATbiasarequitereliable.
mutationalbiasesareinfactverysimilaracrossbacteria.Mutation
Inanadditionalstudy,whichispublishedback-to-backwithour
appears to be dominated by C/G to T/A transitions and is AT-
biased in every studied instance, even in bacteria with high studyinthisissueofPLoSGenetics,Hildebrandetal.investigated
genomic GC contents. At the same time, it is important to note mutationalpatterns byexaminingchangesinfourfold degenerate
that mutational biases need not be entirely constant in different codons in a large number of bacterial lineages with divergence
bacteria.Infact,ourresultsdemonstratethatintheActinobacteria under 10% at these sites [41] and argue for the AT bias of
mutationappearstobelessstronglybiasedtowardsATthaninthe mutationinGCrichbacteria.TheargumentofHildebrandetal.is
other clades examined. complicatedbytwofactors.First,naturalselectioninsuchlineages
Due to a severe relaxation in natural selection, the recent canremainstrongasexemplifiedbyE.coli[37]andB.pseudomallei
evolutionoftheclonalpathogensinvestigatedinthisstudyshouldbe (studiedhere)andmuchstrongerthaninthefiveclonalpathogen
predominantlydrivenbymutation.However,itisimportanttonote strains studied here. Second, theinference ofmutational patterns
thatnaturalselectionisnotentirelyabsentinthesepathogensand from fourfold degenerate sites alone is complicated by a strong
that mutations of severedeleterious effectwill still be removedby correlationoftheGCcontentofselectivelyfavoredcodonsatthe
selection.InMTBCwehaveevidencethatselectionisrelaxedtothe leveloftranslationandtheGCcontentofgenomes[42],makingit
pointthatitdoesnotdistinguishbetweenmutationsthatalteramino possible that some of the detected effects are related to natural
acidsthatareconservedacrossallMycobacteriaandarepresumed selectionattheleveloftranslation.Bothofthesefactorsimplythat
to be under strong constraint and those that are variable in themutationalpatternsinferredbyHildebrandetal.aremuchless
Mycobacteria and are likely to be under much weaker constraint precisethanthosepresentedhere.Nevertheless,thetotalityofthe
[25],indicatingthatonlyextremelystronglydeleteriousmutations evidence in Hildebrand et al. is consistent with the generality of
Table4. Summaryof resultsfortwo more distantlyrelated lineages.
Maxpairwise CurrentGC GCeq CurrentGCnon- GCeqnon- CurrentGC GCeq
Dataset variabilitya dN/dS intergenic intergenicb synnonymous synonymousb synonymous synonymousb
Y.pestis/ 4.9 0.1 42.5 33 48.1 36.3 51.7 47
pseudotuberculosis (11,59.7) (29,44.3) (42.8,51.2)
B.pseudomallei 3.3 0.13 65.3 49.9 61.2 47 90 58.9
(41.1,58) (42.8,50.7) (55.5,61.7)
aTheaveragepairwisediversitypergeneofthetwomostdivergedstrainswithinthelineage.
b95%Confidenceintervalsappearinparenthesis.BoldfontindicatesGCeqvaluesaresignificantlydifferentfromcorrespondingcurrentGCcontents.
doi:10.1371/journal.pgen.1001115.t004
PLoSGenetics | www.plosgenetics.org 8 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
Figure 4. Comparison between GC calculated using data from clonal pathogens and GC calculated using data from more
eq eq
divergedlineages.GC estimatedusingmoredivergedlineagesisalwaysmoresimilartoandsignificantlycorrelatedwithcurrentGCcontent
eq
values(P=0.03).
doi:10.1371/journal.pgen.1001115.g004
AT-biasedmutationacrossbacteria,especiallygiventheirfocuson muchmorefrequentthantransitions[33].Incontrast,inallofthe
a much largernumber of bacterial lineagesthan presented here. fivepathogensweexaminedmutationwasstronglybiasedtowards
A prolonged severe relaxation of selection in obligate intracel- transitions, rather than transversions. We cannot show conclu-
lular bacteria has led to massive loss of repair genes in these sively that no repair functions have been lost in any of these five
bacteria [38]. In the clonal pathogens studied here relaxation of pathogens.However,webelievethatthelackofevidenceforany
selection occurred much more recently and there is no evidence suchloss,theconsistencyofthemutationalbiasesobservedacross
forlossofrepairfunctionsinthesegenomes.Infact,weshowthat allfivedatasetsexamined,andthebiastowardstransitionsrather
noneofthesepathogenshavelostanyoftherepairgenesencoded than transversions, make it reasonable to assume that our results
bytheircloselyrelatedoutgroupsthatareevolvingunderefficient are unaffected by asignificant lossinrepair functions.
selection. Additionally, a previous study that investigated the Weshowthatwithinbroadcladesobligateintracellularbacteria
patternofmutationinstrainsofS.typhimuriumwithdeficientrepair evolvetoaGCcontentthatmatchesverycloselytheGCcontent
functions has shown that in such strains transversions become predictedatequilibriumbasedonourestimatesofmutationrates
PLoSGenetics | www.plosgenetics.org 9 September2010 | Volume 6 | Issue 9 | e1001115
MutationIsBiasedtowardsATacrossBacteria
of clonal pathogens belonging to the same clade. This together It is possible to use the mutational rates we calculated to
with the long-standing observation that the vast majority of estimate the strength with which selection, and/or BGC were
obligate intracellular bacteria tend to have extremely AT rich acting on nucleotide content in the examined clonal pathogens
genomes [38] supports the generality of AT-biased mutation in prior to their recent relaxations of selection. Such calculations
bacteria. Itishoweverimportanttonotetherecentidentification (Table S4) show that such selection is always weak (s#1/N),
e
ofanobligateintracellularbacterium,CandidatusHodgkiniacicadicola which would be expected considering GC content is always
thathasaGC-richgenome(58.4%GC)[43].Itispossiblethatthis intermediate. This demonstrates that the selective or BGC
bacterium constitutes an exception to the universality of AT- advantage of GC over AT nucleotides need not be high in order
biasedmutationandthatmutationinthisorganismisGCbiased. toexplainthevastvariationinGCcontentsobservedinbacteria.
However, it is also possible that this is not the case and that this However, it is important to note that such calculations make a
bacterium is GC-rich for other reasons (such as due to natural number of assumptions that may not be reasonable. One such
selection or BGC). Further studies will be needed to determine assumption is that selection acts uniformly across sites and that
whether Candidatus Hodgkinia cicadicola indeed has exceptional there is no synergistic epistasis (i.e. that the intensity of selection
mutational biases. does not change with changes in nucleotide content). An
By demonstrating that variation in nucleotide content in additional assumption is that there are no competing selective
bacteriaisnotgenerallydrivenbydifferencesinmutationalbiases forces.Thissecondassumptionisclearlyincorrectwhenitcomes
we demonstrate that natural selection or another selection-like tonon-synonymoussites.Evenifselectiononnucleotidecontentin
process such as BGC must play a dominant role in nucleotide these sites were strong enough to drive them to use only GC
contentvariation,particularlyindrivingintermediatetohighGC nucleotides it is highly likely that selection for protein function
contentinmanybacteriallineages.Atthispointitisunclearhow would not allow this to happen. It is therefore unclear whether
much of this variation is driven by selection and how much by selection on nucleotide content isin factalways weak.
BGC. In this study we demonstrate the great utility of clonal
If natural selection plays a strong role in determining GC pathogens for the study of mutational biases in bacteria. We
content it suggests that in many bacteria there are no truly investigated five such clonal pathogens from four very diverse
neutrally evolving sites. The nature of such selection remains clades. We showed that in every studied case and across all site
obscure.BecauseGCcontentcorrelatesstronglyacrosscodingand categories point mutation is consistently biased towards AT and
noncoding sites genome-wide, natural selection acting on GC that the most frequent mutations are always G/CRA/T
content probably relates to genome-wide functions such as transitions, thus demonstrating that the biases of point mutation
replication or DNA maintenance and is less likely to be related aremuchlessvariablethanwaspreviouslyassumed.Byidentifying
to gene expression. Previous studies attempted to associate GC additional bacteria evolving under strongly relaxed selection and
content with environmental factors such as growth temperature conductingdeepsequencingofsuchbacteriaitshouldbepossible
[44,45], exposure to UV [8], oxygen requirements [46], and the to address additional questions regarding mutational patterns of
ability to fix nitrogen [47]. While these studies were thought by bacteria, including the variability in the rates of insertions and
sometobeinconclusive[3,44,48–51]theyprovidethebestcurrent deletions across bacteria and mutation clustering along bacterial
explanations for the possible involvement of selection in chromosomes.
determining nucleotide content. However, considering that Recent studies have shown that mutation appears to be
bacteria belonging to such broad clades as Actinobacteria have universally AT-biased in eukaryotes [21,26–31]. Our results
similargenomicnucleotidecontentseventhoughtheyareexposed demonstrating that this may also be the case in prokaryotes
to different environments it becomes tempting to speculate that thereforeshowthatmutationmayinfactbeAT-biasedinallliving
environmental variables may not be the only underlying organisms(althoughitisimportanttonotethatwedonotyethave
determinants for the natural selection acting on nucleotide good estimates of the mutational biases of Archea). Not only is
content. It is possible that selection on nucleotide content is also mutation AT-biased in all instances studied, but the specific
driven by more intra-organismal factors that can affect entire pattern of mutation is always consistent. The most common
clades irrespective of environment. Examples of such factors can mutationsarealwaysG/CtoA/Ttransitions.Theseresultsmake
betheabilityofthereplicationmachinerytoworkbetteronGCor ittemptingtospeculatethatthepredominantmutationsaresimply
AT rich sequences, DNA packaging, defense against phages or the result of the lability of cytosine to deamination, and that this
creatingbarriersforhorizontalgenetransfer.Morestudiesneedto pattern shows through despite possible variability in DNA
be carried out to probe the possible involvement of selection in replication andrepair mechanisms[27].
determining bacterial nucleotide content.
In order for an increase in BGC to explain GC richness, Concluding remarks
recombinationshouldbepervasiveenoughinGCrichbacteriato Inthisstudyweuseddatafromfivestrictlyclonalpathogensto
drive GC contents that are elevated substantially above those analyze the variation in point mutation biases in bacteria. These
observedeveninsexuallyreproducingeukaryotes.Itisverylikely pathogens are uniquely suitable for such analyses as they can be
that per generation advantage given to GC nucleotides through shown to be evolving under selection that is severely inefficient
geneconversion(whichisdeterminedbyrecombinationratesand relative to stochastic processes. Unlike obligate intracellular
by the conversion bias [1,10]) is significantly higher in sexually bacteria that have been evolving under inefficient selection for
reproducing eukaryotes than in prokaryotes for which recombi- long evolutionary times and have lost much of their repair
nation is assumed to be less frequent [1]. However, it is possible pathways these clonal pathogens have experienced only a short-
that BGC may still affect some bacteria more strongly than termrelaxationinselectionefficiencyandarelikelytohaveintact
eukaryotes, if N is increased in these bacteria by a higher factor repair mechanisms. Their mutational biases should therefore
e
than the factor by which the advantage given to GC nucleotides reflect those of their other clade members that are not subject to
through gene conversion is decreased. This is an intriguing inefficientselection.Wedemonstratedthateventhoughthesefive
possibility, opening up new avenues of research into recombina- pathogens belong to four very diverse clades with very different
tion ratesandvariability inN among bacteria. nucleotidecontentsmutationinallofthemisbiasedtowardsAT,
e
PLoSGenetics | www.plosgenetics.org 10 September2010 | Volume 6 | Issue 9 | e1001115