Article Title: Reconstruction of Ancestral 16S rRNA Reveals Mutation Bias in the Evolution of ...

12 Pages·2013·0.4 MB·English

Reconstruction of Ancestral 16S rRNA Reveals Mutation Bias in
the Evolution of Optimal Growth Temperature in the
Thermotogae Phylum
Anna G. Green,1 Kristen S. Swithers,1 Jan F. Gogarten,2 and Johann Peter Gogarten*,1
1DepartmentofMolecularandCellBiology,UniversityofConnecticut
2DepartmentofBiology,McGillUniversity,Montreal,Quebec,Canada
*Correspondingauthor:E-mail:[email protected];[email protected].
Associateeditor:KoichiroTamura
Abstract
D
Optimalgrowthtemperatureisacomplextraitinvolvingmanycellularcomponents,anditsphysiologyisnotyetfully o
w
understood. Evolution of continuous characters, such as optimal growth temperature, is often modeled as a one- nlo
a
dimensional random walk, but such a model may be an oversimplification given the complex processes underlying d
e
tghroewetvholtuetmiopneoraftcuornetionfuaonucsiecnhtaorargcatenriss.mRsecfreonmtatrhteicgleusahnainveeuanseddcayntocseisnteraclosneqteunetnocfetrheecosntesmtrurcetgiioonnstoofinrfiebrotshomeoapltRimNAal, d from
allowinginferencesabouttheevolutionofoptimalgrowthtemperature.Here,weinvestigatetheoptimalgrowthtem- h
ttp
peratureofthebacterialphylumThermotogae.Ancestralsequencereconstructionusinganonhomogeneousmodelwas s
used to reconstruct the stem guanine and cytosine content of 16S rRNA sequences. We compare this sequence recon- ://a
c
a
structionmethodwithotherancestralcharacterreconstructionmethods,andshowthatsequencereconstructiongen- d
e
eratessmallerconfidenceintervalsanddifferentancestralvaluesthanotherreconstructionmethods.Unbiasedrandom m
ic
walk simulation indicates that the lower temperature members of the Thermotogales have been under directional .o
u
p
selection;however,whenasimulationisperformedthattakespossiblemutationsintoaccount,itisthehightemperature .c
o
lineages that are, in fact, under directional selection. We find that the evolution of Thermotogales optimal growth m
/m
temperaturesisbestfitbyabiasedrandomwalkmodel.Thesefindingssuggestthatitmaybeeasiertoevolvefromahigh
b
e
optimalgrowthtemperaturetoaloweronethanviceversa. /a
Key words: complex trait evolution, thermophiles, ancestral character reconstruction, nonhomogeneous models, rticle
Thermotogae,randomwalk. -a
b
s
tra
c
Introduction t/3
randomwalk,whichdonotcapturetheprocessesunderlying 0
/1
Theevolutionofcontinuousphenotypictraitsisoftenmod- phenotypictraitevolution. 1
/2
eled as a simple one-dimensional random walk. Under this Optimal growth temperature (OGT) is an example of a A46
3
model,thevalueofatraitmayincreaseordecreaseindefined, continuousphenotypictraitwithmanycomplexunderlying r/1
t2
distincttimeintervals,andthemagnitudeofchangeiscon- factors,anditsevolutionwassuggestedtoproceedaccording i62
c5
stantwitheachstep(Felsenstein1985;Gingerich1993).With to a random walk (Dahle et al. 2011). To date, research has 9
l5
increasingnumberoftimesteps,therateofnetchangeinthe focusedonmanyoftheadaptationsofbiologicalmacromol- e b
y
trait decreases predictably (Gingerich 1993). Modeling trait ecules (protein, RNA, and DNA) to extreme temperatures, gu
e
evolutionasarandomwalkprovidesausefulnullhypothesis butunderstandingofhowtheseadaptationsinteracttopro- s
t o
againstwhichtotestadataset,todeterminewhetheratrait duceaphenotypeandhowsuchphenotypesevolveislimited. n
1
is evolving according to a deterministic process, such as di- In thermophilic organisms, virtually every molecule must 3
A
rectional selection. If the rate of change does not decrease adapt in particular ways to remain stable and functional at p
overthesumofthetimeintervals,thisisgroundstorejectthe high temperatures. RNA is likely to undergo 30 to 50 bond ril 2
0
nullhypothesisthatevolutionofthisparticulartraitproceeds hydrolysisathightemperatures(GrosjeanandOshima2007). 19
viaarandomwalk.However,eventhoughtheevolutionofa RibosomalandtransferRNAsofthermophileshaveacharac-
particulartraitappearstoberandom,thisdoesnotmeanthat teristichighguanine–cytosine(GC)content,whichincreases
theprocessesunderlyingtheevolutionofthatparticulartrait the stability of secondary structures (Grosjean and Oshima
aretrulyrandom(Gingerich1993).Althoughtraitevolution 2007).RibosomalRNAinparticulardisplaysacorrelationbe-
may appear to proceed by a random walk, there may be tweenthestemGCcontentandtheOGToftheorganismin
underlying factors that influence and even bias evolution. whichitisfound(GaltierandLobry1997).DNAisproneto
Given the wealth of molecular data now available, it may loseitshelicalstructureandundergodepurinationanddepyr-
betimetomoveawayfromsimplemodelsliketheunbiased imidationathightemperatures(GrosjeanandOshima2007).
(cid:2)TheAuthor2013.PublishedbyOxfordUniversityPressonbehalfoftheSocietyforMolecularBiologyandEvolution.Allrightsreserved.Forpermissions,please
e-mail:[email protected]
Mol.Biol.Evol.30(11):2463–2474 doi:10.1093/molbev/mst145 AdvanceAccesspublicationAugust21,2013 2463
MBE
Greenetal. . doi:10.1093/molbev/mst145
Thermophiles have adapted to these thermodynamic chal- from extant sequences of different composition,
lenges by using small ligandbindingandcovalentmodifica- nonhomogeneous substitution models are superior for an-
tion of nucleic acids, generation of compact tertiary cestral state reconstruction (Boussau et al. 2008). To date,
structures, and efficient DNA repair (Grosjean and Oshima none of these studies have compared inferences about an-
2007).Theproteinsofthermophilesareknowntoundergoa cestraltemperatureobtainedfromsequencereconstruction
variety of adaptations, including certain amino acid biases with methods traditionally used to reconstruct ancestral
that increase stability (Suhre and Claverie 2003; Zeldovich, character states in evolutionary studies, including Bayesian
Berezovsky,andShakhnovich2007),andanincreaseinbind- Markovmodels,Parsimony,andmaximumlikelihood.
ing affinity for certain metabolites (Massant 2007). In addi- Although OGT evolution within the Thermotogales may
tion,certainmetabolicintermediatesmaybeunstableathigh conformtoapatternsimilartoarandomwalk(Dahleetal.
temperatures, leading thermophilic cells to compensate 2011), this does not mean that OGT evolution is truly
throughincreasedproductionorsequestrationofthemetab- random or unbiased. A priori, an unbiased random walk
olite in question (Massant 2007). Given the complex and appears unlikely, because proteins, DNA, and RNA in ther- D
o
numerous adaptations required to optimize a cell for mophiles are so meticulously adapted to have highly stable wn
growth at high temperatures, it is reasonable to assume structures that there are simply more mutations available loa
d
thattheevolutionofthistraitisinfluencedbymanyunder- that would disrupt these stable structures, and fewer that ed
lyingfactors. would continue to increase stability. Thus, we hypothesize fro
m
AninterestingmodelcladeforthestudyofOGTevolution that more mutations are available which can lower the h
is the Thermotogae phylum. The Thermotogales, currently OGT of an organism than increase it, and this bias should ttp
s
theonly recognizedorderwithin the phylum Thermotogae, be observable in the evolutionary history of a clade of ther- ://a
are an order of anaerobic bacteria whose ribosomal genes mophilicorganisms. ca
d
indicate a close relation to the Aquificae (Zhaxybayeva Here,westudytheevolutionofOGTintheThermotogales em
et al. 2009). However, in bacterial ribosomal phylogenies utilizing30representativesfromtheorder.Weusedaknown ic.o
rooted with an archaeal outgroup, the root is frequently correlate of OGT, the GC content of the stem regions of up
.c
based on the branch leading to the Aquificae, turning the ribosomalRNA(GaltierandLobry1997),andreconstructed o
m
Thermotogae into the second deepest branching lineage the stem GC content of the 16S rRNA molecule in the /m
b
(Reysenbach et al. 2005). In contrast to ribosomal proteins Thermotogalesateverynodeinthetree.Thisallowedusto e
/a
and RNAs, the majority of genes in the genomes of the trace the evolution of thermophily within the clade, and to rtic
Thermotogales appear to have been acquired by horizontal show that this trait is not evolving according to a random le
-a
genetransferfromClostridia(Zhaxybayevaetal.2009).OGTs walk. We compare the inference of the ancestral state for b
s
within the Thermotogales range from 37 to 80(cid:2)C. Previous OGTfromGCcontentof16SrRNAwithtraditionalancestral tra
c
work, based on five representatives of the order, suggested state reconstruction methods. We find that no traditional t/3
0
thattheancestortotheThermotogalesgrewatahigherOGT methodagrees withthe values obtainedby ASR, suggesting /1
1
than the extant members of the clade (Zhaxybayeva et al. that sequence reconstruction may be able to reconstruct /2
4
2009).ArecentstudybyDahleetal.(2011)suggeststhatOGT evolutionarypatternsmoreaccurately. 63
in the Thermotogales evolved according to an unbiased /12
randomwalk.Thisstudywasbasedonthepairwisecompar- Results 62
5
9
isons of quantitative phenotypes of extant organisms and 5
Inference of OGT of Ancestral Sequences b
produced wide confidence intervals (CIs) for predicted y
OGT and stem GC content in the Thermotogae show a g
values that did not reject the random walk hypothesis. We ue
strong correlation (r=0.845, P<0.001; fig. 1b). This correla- s
proposethatconsideringsequencedatawillallowforbetter t o
tionremainsdetectablewhenthephylogeneticsignalissub- n
resolutionofevolutionaryhistoriesandprocesses. 1
tractedfromtheanalysisusingFelsenstein’s(1985)methodof 3
Traditional techniques of ancestral character estimation A
phylogenetically independent contrasts (PIC) (r=0.377, p
rely on a single numeric values of the trait in the extant P<0.05;fig.1a). ril 2
organisms, and use different methods of averaging these 0
traitstoestimatetheancestralvalue.Becausetheyrelyonly We calculated a correlation line between the stem GC 19
content and OGT in the Thermotogae, T ¼(cid:3)53:53
onsinglenumericvaluesandmethodsofaveraging,theymay opt
+169:45(cid:4)C , where T is the OGT and C is the stem
not provide realistic reflections on evolutionary processes. GC opt GC
GCcontent.WeusedthislinetopredicttheOGTofthelast
Ancestral sequence reconstruction (ASR) to infer ancestral
common ancestor of the Thermotogae, with an estimated
characters has recently emerged as an alternative to these
OGTof76(cid:5)3.2(cid:2)C.Seetable1forcompletelistofancestral
traditionaltechniquesofancestralcharacterestimation,and
OGTs.
has proved particularly useful for examining historic OGTs
(Galtier et al. 1999; Boussau et al. 2008; Zhaxybayeva et al.
Random Walk Simulation
2009; Groussin and Gouy 2011; Hobbs et al. 2012). Instead
ofsimplyaveragingextantvalues,ancestralstatereconstruc- WesimulatedevolutionofstemGCcontentfromtheances-
tiontakesintoaccountalloftheaminoacidsornucleotides tral node of the tree, using a random walk simulation with
in a sequence, and can simulate their evolution using 10,000replicates,andusedthesimulatedvaluestoestablisha
complex models. When reconstructing ancestral sequences 95% CI of values produced by a random walk for each tip.
2464
MBE
EvolutionofOGTintheThermotogaePhylum . doi:10.1093/molbev/mst145
haveonstemGCcontent.Theresultsareshownintable3.
This table indicatesthat there are a greater numberof pos-
sible mutations that decrease the stem GC content
(n=2,279) than there are that increase (n=1,120) or do
notaffectit(n=1,278).
Levene’s Test for Homogeneity of Variance was used to
comparethevariancesofthesetwosetsofpossiblechanges,
positive and negative. We found that there is a significant
differenceinthevariances,withthevarianceofthenegative
groupbeinggreater(P=<0.001).
Biased Random Walk Simulation
D
To attempt to explain the tendency of the GC content to o
w
decrease,thedistributionofphenotypiceffectsproducedby nlo
a
pointmutationstotheT.maritimaMSB8ribosomewasused d
e
tloowsimsteumlatGeCevcoolunttieonntodfothneoTlohnergmeroftaollgoauletss.idTehethliene9a5g%esCwIfitohr d from
randomwalkevolution,unlikeintherandomwalksimulation h
ttp
that did not use the estimated substitution probabilities. s
However, many of the lineages with high stem GC content ://a
c
a
now fall outside the 95% CI, including T. maritima MSB8, d
e
m
Thermotogasp.RQ2,T.neapolitana,T.petrophila,T.neapo-
ic
litana,T.petrophila,T.naphthophila,andT.thermarum. .o
u
p
.c
o
Comparison of Ancestral Character Reconstruction m
/m
Methods b
e
Theresultsofeachancestralreconstructionmethod,exclud- /artic
ingBayesTraits,werecomparedwiththe ancestralstem GC le
contentvaluesobtainedfromsequencereconstruction.The -ab
s
G-testwasusedtotestthecongruenceoftheCIs.Thet-test tra
c
wasusedtotestfordifferenceinmeansizeofCIgenerated. t/3
0
The results are summarized in table 4. No reconstruction /1
1
method produced both values congruent with those esti- /2
4
mated by ASR and CIs of a similar size, as indicated by the 6
3
G-testandt-testresults. /12
FIG.1. (a)ThecorrelationbetweenOGTandstemGCcontentinthe CIsforreconstructedstemGCcontentwerecalculatedby 62
5
Thermotogae,whenindependentcontrasts(ICs)areused,withregres- 9
10,000 replicate sequence reconstructions in Bppancestor, 5
sionline.r=0.377,P=<0.05.Somepointshavenegativevaluesbecause b
sampling from the probability distribution for each site, y
independentcontraststakethedifferencebetweenexistingphenotypes. g
rather than using the most probable nucleotide (table 5). u
(b) The correlation between OGT and stem GC content in the es
Thermotogae,withregressionline.r=0.845,P=<0.001. The95%CIfortotalGCcontentwasused,becausecalcula- t o
n
tion of the secondary structures for 10,000 replicates was 1
3
computationally unfeasible. A comparison of CI generated A
p
Noneofthelineagesfellabovethe95%CI.Wefoundthatthe byeachmethodisprovidedintable4. ril 2
Using the constant-variance random walk model imple- 0
stem GC content in organisms Petrotoga mexicana, 1
mentedinBayesTraits,theancestralnodewasreconstructed 9
P. halophila, P. olearia, and P. sibirica was below the 95%
atastemGCcontentof0.727,whichfallsoutsideofthe95%
CI (see table 2). Taken alone, this could suggest direc-
CIgeneratedbytheGCcontentofthesequencereconstruc-
tional evolution to decrease stem GC content, or a
tion (0.751–0.777). The 95 credibility interval for the con-
higher probability of mutations that decrease the stem GC
stant-variance random walk model was 0.699–0.755, which
content.
overlapswiththe95%CIgeneratedbytheGCcontentofthe
sequencereconstruction,althoughthisisalargeCIrange.The
Simulation of Changes to the Thermotoga maritima
directionalrandomwalkforBayesTraitsestimatedanances-
ribosome
tralstemGCcontentof0.791,whichwashigherthantheCI
To determine whether it is more probable to accumulate produced by sequence reconstruction. The 95 credibility
mutations that increase or decrease GC content, we simu- interval for the directional random walk model was 0.752–
lated all possible point mutations to the T. maritima MSB8 0.829, which overlaps with the credibility interval produced
ribosome, and calculatedthe effectsthese mutationswould bysequencereconstruction,thoughtheCIisagainlarge.
2465
MBE
Greenetal. . doi:10.1093/molbev/mst145
Table 1. Predicted Ancestral OGTs for Each Node, with 95% CIs, Table 2. Actual rRNA Stem GC Content of Extant Thermotogales
Calculated from a Linear Regression between OGT and Stem GC Compared with Results of an Unbiased Random Walk Simulation.
Content of Extant Organisms.
Extant 2.5th 97.5th
Node Predicted OGT Lower Bound Upper Bound GC percentile percentile
1 75.93 72.70 79.17 Content
2 76.79 73.38 80.20 Fervidobacterium changbaicum 0.728 0.649 0.881
3 77.60 74.02 81.18 Fervidobacterium gondowanense 0.703 0.645 0.887
4 77.87 74.23 81.51 Fervidobacterium islandicum 0.733 0.655 0.875
5 78.12 74.43 81.81 Fervidobacterium nodosum 0.714 0.645 0.884
6 77.87 74.23 81.51 Kosmotoga olearia 0.732 0.686 0.844
7 77.87 74.23 81.51 Marinitoga camini 0.664 0.655 0.876
8 75.42 72.28 78.55 Marinitoga hydrogenitolerans 0.659 0.656 0.874
9 74.19 71.29 77.10 Marinitoga okinawensis 0.681 0.653 0.875 D
o
10 69.12 66.90 71.34 Marinitoga piezophila 0.688 0.636 0.893 wn
11 68.87 66.67 71.07 Mesotoga prima 0.664 0.654 0.875 loa
d
12 74.97 71.92 78.01 Petrotoga halophila 0.631 0.634 0.895 ed
13 71.49 69.02 73.96 Petrotoga mexicana 0.628 0.633 0.899 fro
14 68.96 66.75 71.16 Petrotoga mobilis 0.631 0.631 0.898 m h
15 69.23 67.01 71.46 Petrotoga olearia 0.637 0.642 0.887 ttp
s
16 70.01 67.72 72.31 Petrotoga sibirica 0.631 0.648 0.881 ://a
17 70.29 67.97 72.62 Thermosipho africanus 0.714 0.644 0.882 ca
d
18 70.11 67.80 72.41 Thermosipho atlanticus 0.731 0.660 0.868 em
19 71.80 69.28 74.31 Thermosipho geolei 0.720 0.670 0.857 ic.o
20 76.18 72.90 79.47 Thermosipho japonicus 0.729 0.668 0.862 up
21 71.16 68.73 73.59 Thermosipho melanesiensis 0.722 0.644 0.880 .co
m
22 66.10 63.96 68.23 Thermotoga elfii 0.719 0.656 0.875 /m
23 59.51 56.65 62.36 Thermotoga hypogea 0.745 0.673 0.856 be
24 59.20 56.29 62.11 Thermotoga lettingae 0.723 0.669 0.860 /artic
25 61.20 58.62 63.78 Thermotoga maritima MSB8 0.774 0.689 0.841 le
26 53.35 49.22 57.48 Thermotoga naphthophila 0.771 0.683 0.845 -ab
s
27 54.30 50.39 58.21 Thermotoga neapolitana 0.778 0.682 0.847 tra
c
28 53.26 49.12 57.41 Thermotoga petrophila 0.777 0.682 0.845 t/3
0
29 52.98 48.77 57.19 Thermotoga sp. RQ2 0.775 0.684 0.843 /1
1
Thermotoga subterranea 0.729 0.656 0.873 /2
4
Thermotoga thermarum 0.764 0.680 0.848 6
3
/1
Of the models of ancestral reconstruction tested, only NOTE.—Columns2and3showtheupperandlowerboundsofthe95%CIofthe 26
unbiasedrandomwalksimulation.Theitalicizedvaluesaretheextantvaluesthatfall 2
5
restrictedmaximumlikelihood(REML)performsadequately. outsidetheCIgeneratedbythesimulation. 9
5
PIC, generalized least squares with a Brownian (GLSB) and b
y
Grafen (GLSG) model fail the t-test for size of CIs, meaning gu
e
that the CIs produced by those methods are significantly s
Table 3. Effect of Simulated Point Mutations on Stem GC Content t o
larger than those produced by ASR. Squared change parsi- n
of Thermotoga maritima MSB8 16S rRNA. 1
mony(SCP)failstheG-testforoverlapofCIs,meaningthat 3
n % Mean Change Variance A
SCPfailstoproduceestimatessimilartothoseproducedby p
ASR.Theonlymethodthatdoesnotfailbothofthesetestsis Increase 1120 23.9 0.00118 7.19E(cid:3)009 ril 2
REML, indicating it produces similar estimates and CIs to Neutral 1278 27.3 0 0 019
Decrease 2279 48.7 (cid:3)0.0009938 8.14E(cid:3)009
thosemadebyASR.
Boththerandomanddirectionalmodelsimplementedin
BayesTraits produced ancestral values with CIs that over-
lapped the CI produced by ASR, though these CIs were
traditional methods of ancestral character state reconstruc-
quite large. The directional model produced an estimate of
tion.Theadvantageofthisapproachisthatitcalculatesthe
ancestral stem GC content much higher than ASR did, and
propertyattheancestralnodebasedonsequencereconstruc-
the random model produced a much lower estimate than
tion, a procedure that relies on well-established models of
ASR.
sequence evolution, and does not rely on averaging of
extant values of character states. In addition, this method
Discussion
does not reduce the value of the trait to a single numeric
InthisarticleweusedASRtoreconstructancestralcharacter estimate,andinsteadconsidersthenucleotidesequenceun-
states, and systematically compared this method with derlyingthatestimation.Theaccuracyofthemethodbased
2466
MBE
EvolutionofOGTintheThermotogaePhylum . doi:10.1093/molbev/mst145
Table 4. Comparison of Ancestral State Reconstruction Methods.
Method No. CI that Overlap Ga Pa Mean CI Sizeb Tc dfc Pc
REML 29 2.975 0.08456 0.0279 1.0243 28.584 0.3134
SCP 21 15.9239 6.59E-05 0.021 (cid:3)6.627 51.065 2.00E(cid:3)08
PIC 29 2.975 0.08456 0.9732 8.9922 28.001 9.51E(cid:3)10
GLSB 28 0.1642 0.6853 0.3347 8.1825 28.004 6.59E(cid:3)09
GLSG 28 0.1642 0.6853 1.1825 19.2598 28.002 2.20E(cid:3)16
NOTE.—Thefollowingmodelsweretested(seetextfordetails):Brownianmotionwithrestrictedmaximumlikelihood(REML),squaredchangeparsimony(SCP),phylogenetically
independentcontrasts(PIC),generalizedleastsquareswithaBrownianmotionmodel(GLSB),andGrafenmodel(GLSG).
aTheoverlapofCIsgeneratedbytheancestralstatereconstructionmethodswiththosegeneratedbyancestralsequencereconstructionwasexaminedusingaG-test.Columns2
and3showtheG-teststatisticandPvalue(alowPvalueindicatesthatthetestedCIsareunlikelytohavecomefromthesamedistribution).
bThemeansizeofCIgeneratedbyeachreconstructionmethod(meanCIsize)wascomparedwiththatgeneratedbyancestralsequencereconstruction(0.0246)usingat-test.
cThefinalcolumnsshowthet-teststatisticandPvalue(alowPvalueindicatesasignificantdifferenceinCIsize).
D
o
w
n
lo
a
doesnotrelyonpairwisedistancebetweentheextantspecies d
Table 5. Results of Evolutionary Simulation of rRNA Stem GC e
Content Using Distribution of Possible Point Mutations. to test for random walk evolution, as pairwise distances are d fro
not independent data points (Felsenstein 1985). Indeed, m
Organism ExCtoannttenGtC 2.50% 97.50% when our method of ancestral character estimation based http
on sequence reconstruction is compared with conventional s
Fervidobacterium changbaicum 0.728 0.659 0.741 methods, the conventional methods perform poorly, either ://ac
Fervidobacterium gondowanense 0.703 0.636 0.73 a
failing to reconstruct values within the CIs of our sequence d
Fervidobacterium islandicum 0.733 0.663 0.743 em
FKeorsvmidootboagcateorlieuamrianodosum 00..771342 00..66492 00..775371 bstarsuecdtemdeptohiondts,orrenpdroerdinugcinthgeveesrtyimlaartgeespCrIsacotincatllhyeuirserleecsos.n- ic.oup
Our analyses indicate that the ancestor to the .c
Marinitoga camini 0.664 0.634 0.726 o
ThermotogaegrewatarelativelyhighOGT,76(cid:5) 3(cid:2)C.The m
Marinitoga hydrogenitolerans 0.659 0.64 0.729 /m
majorityoflineagesinthiscladehaveundergoneadecreasein b
Marinitoga okinawensis 0.681 0.63 0.724 e
OGT over time (e.g., Mesotoga prima has undergone a de- /a
Marinitoga piezophila 0.688 0.621 0.718 creaseof39(cid:2)C),whereasafewhavemaintainedorincreased rtic
Mesotoga prima 0.664 0.642 0.732 le
theirOGT,butonlybyasmallamount(e.g.,thelineagelead- -a
Petrotoga halophila 0.631 0.596 0.703 b
Petrotoga mexicana 0.628 0.595 0.705 ingtoT.maritimaMSB8hasundergoneanincreaseof4(cid:2)Cin stra
Petrotoga mobilis 0.631 0.589 0.7 its OGT since its divergence from the ancestral node). ct/3
Althoughthisdiffersfromearlieranalysesthatsuggestedan 0
Petrotoga olearia 0.637 0.601 0.706 /1
Petrotoga sibirica 0.631 0.602 0.707 ancestralgrowthtemperaturemorethan80(cid:2)C(Zhaxybayeva 1/2
4
Thermosipho africanus 0.714 0.682 0.752 et al. 2009), our results are based on a data set with many 63
more taxa and a more sophisticated method of sequence /1
Thermosipho atlanticus 0.731 0.677 0.75 2
Thermosipho geolei 0.72 0.675 0.748 reconstruction (i.e., nonhomogeneous implementation of 625
Thermosipho japonicus 0.729 0.68 0.751 substitutionmodels). 95
Thermosipho melanesiensis 0.722 0.663 0.743 Initialevolutionarysimulationsseemedtoindicatethatthe by g
Thermotoga elfii 0.719 0.663 0.743 lower temperature members of the Thermotogae are not ue
Thermotoga hypogea 0.745 0.677 0.749 evolving according to an unbiased random walk, and are in st o
Thermotoga lettingae 0.723 0.669 0.746 fact under directional selection. We have shown that the n 1
Thermotoga maritima MSB8 0.774 0.698 0.759 extantvaluesofstemGCcontentinthemajorityoflineages 3 A
p
TThheerrmmoottooggaa nneaapphothliotapnhaila 00..777718 00..669964 00..775587 faalflsewwitohrignatnhisem95s,%PC. Imperoxidcauncead, Pb.yhthaelospimhiulal,atPio.nosl.eHaroiaw,eavnedr, ril 20
1
Thermotoga petrophila 0.777 0.699 0.759 P.sibiricafallbelowthe95%CIofthevaluesobtainedfrom 9
Thermotoga sp. RQ2 0.775 0.695 0.758 thesimulation.Theseresultswouldindicatethattheseline-
Thermotoga subterranea 0.729 0.666 0.745 ages have been undergoing directional evolution toward a
Thermotoga thermarum 0.764 0.689 0.755 lower stem GC content, and thus a lower OGT, if not for
themutationbiasthatfurthersimulationsrevealed.
NOTE.—ThistableshowstheextantGCcontentalongwithupperandlowerbounds
ofthe95%CIofsimulatedvalues.Extantvaluesthatareaboveorbelowthe95%CI Weprovideevidencethatatraitmayappeartobeunder
areitalicized. directional selection when in fact there is a mutation bias
affecting that trait. This was done by a simulation of point
on sequence reconstruction has been demonstrated in pre- mutationstotheT.maritimaMSB816SrRNA,whichdem-
vious studies, which were able to successfully reconstruct onstratedthatthenumberofpossiblemutationsthatwould
sequences from deep nodes in the tree of life, and use decreasethestemGCcontentisalmosttwiceaslargeasthe
those sequences to make evolutionary inferences (Boussau number of mutationsthatwouldincrease it.Thereisalsoa
etal.2008;GroussinandGouy2011).Inaddition,ourmethod significantly greater variance in the number of possible
2467
MBE
Greenetal. . doi:10.1093/molbev/mst145
decreases,indicatingthatitmaybepossibletomakegreater
changes over a shorter period of time. However, when this
distributionofpossiblemutationsisusedtosimulateevolu-
tion along the tree, the high stem GC content lineages are
foundtobeoutsidethe95%CIforevolutionaccordingtoa
random walk. The high stem GC content lineages, and not
the low ones, are the lineages under directional selection,
whichisonlydetectedwhentheproperdistributionofpos-
sible mutations is used to simulate evolution. This demon-
strates that a one-dimensional random walk does not
adequatelyreflectthepossiblemutationsthatunderlieevo-
lution ofOGT,andmay failtodetectevolutionary patterns
suchasdirectionalselection. D
o
ThelineagesthathaveundergoneadecreaseinstemGC w
n
content,andthereforeadecreaseinOGT,tendtobefound loa
d
onlongerbranches,evenwhenbranchlengthsarecalculated e
d
independentlyofstemGCcontent(fig.2).Onemayassume fro
m
that this sharp decrease in stem GC content, and therefore h
OGT,isduetodirectionalselection,mutationbias,oracom- ttp
s
binationofthetwo.However,thisfailstoprovideanexpla- ://a
c
nationforwhythoselineagestendtobefoundatagreater a
d
distance from the root, because the branch lengths are cal- em
culatedindependentlyofstemGCcontent,whichispresum- ic.o
ably the trait under selection (fig. 3). This may be due up
.c
to increased directional selection on certain organisms to o
m
adapt to lower OGT, so more mutations are allowed to /m
b
reachfixation,orduetoanincreasednumberofmutations e
/a
thatbecomenondetrimentaltotheorganismasitgainsthe rtic
ability to survive at lower temperatures. A previous study le
-a
providesevidenceforthesecondexplanation,demonstrating b
s
bybiophysicalsimulationthatthemaximumpermissiblemu- tra
c
tation rate, that is, the rate above which populations go t/3
0
extinct, in thermophiles is less than one-third of the maxi- /1
1
mum permissible rate in mesophiles (Zeldovich, Chen, and /2
4
Shakhnovich2007),insupportoftheearlierobservationthat 63
high temperature lineages are more slowly evolving (Woese /12
6
1987). 25
We acknowledge the possibility that one substitution FIG.2. ChangesinstemGCcontentasafunctionofthelengthofthe 95
branchonwhichtheyoccur(a),andpercentchangeinstemGCcon- b
model is not adequate to reflect the evolution of both the y
tentforabranchlengthofoneasafunctionofthebranchlengthon g
stemandloopregionsofthe16SrRNAmolecule.However, ue
which each change occurs (b). Under a random walk simulation, we s
attempts,tocreateaphylogenyusingonlythestemregionsof expect the rates of change to converge to 0 as the branch lengths t on
the alignment, resulted in poorly supported tree topologies increase. 1
3
thatwereincongruentwiththewell-supportedtreeproduced A
p
byusingtheentiresequences.Itispossiblethatthealignment ril 2
ofonlystemregionsdoesnotcontainadequateinformation mutations in the 16S rRNA that cause a decrease in stem 0
1
to recover the correct tree topology (see alignment 3, sup- GC content, and therefore a decrease in OGT. Presumably, 9
plementarydata,SupplementaryMaterialonline),orthatthe there are also more possible mutations that would cause a
substitution models available do not adequately reflect the decrease in the fitness of finely adaptive protein and DNA
evolution of the stem regions. Advancement of models of structures in thermophiles than there are those that would
site-dependentevolutionwillprovidebettertoolstoaddress increase it. Therefore, adaptation to lower temperatures is
thisquestion(Williamsetal.2011). likely easier for a thermophile than further adaptation to
even higher temperatures. This suggests that it is relatively
Conclusions
moredifficulttoadaptfromalowertohighertemperature,
WehaveshownthatmanylineagesintheThermotogaehave and relatively easy to adapt from a higher to lower
adapted to lower OGT over time, whereas few have main- temperature.
tained or increased their OGT by a small amount. Because Itisimportanttoaddresstheorderofeventsinthispro-
there is a higher GC content in thermophiles, our observa- posed process of adaptation to lower OGT. Organisms that
tions can be explained by the larger number of possible have undergone a steep decrease in stem GC content, and
2468
MBE
EvolutionofOGTintheThermotogaePhylum . doi:10.1093/molbev/mst145
History of the Thermotogales
taxa. These sequences were downloaded from the NCBI
0
8 Thermotoga gp. 1 GenBankdatabase.
0. Thermotoga gp. 2
Thermosipho The 16S rRNA sequences were initially aligned using
● ●●● Fervidobacterium muscle v. 3.8.31 (Edgar 2004) with the default settings. The
● Marinitoga
●
5 Petrotoga alignmentwasthenrefinedtoincludestructuralinformation
7 Other
0. usingRNASalsa(Stocsitsetal.2009),usingstringencysettings
nt
e s1,s2,ands3=0.9,andaconstraintstructurefileforT.mar-
nt
o itima MSB8, obtained from the Comparative RNA Website
C 0
GC 0.7 and Project (Cannone et al. 2002). See supplementary data,
m Supplementary Material online, for the full structural align-
Ste mentof16Ssequences.
5 Todeterminethebestsubstitutionmodeltouseforcon-
6
0. structing a tree from the alignment, we used the phangorn D
o
package for R (Schliep 2011). This software tests the substi- wn
tutionmodelsJC,F81,K80,HKY,SYM,andGTR,withapro- loa
0 d
0.6 0.0 0.5 1.0 1.5 2.0 2.5 3.0 pnoeirtthioenrtooffiinnvdartihaentmsoitdese,lgwaimthmtaherabteestcafittegtoortiehse, bdoattah.,Tahnids ed fro
m
Length of Branches from Root manoddGelaswcuaselG2T00R3)+toGcr+eaIt.eWatereuesferdomPhtyhMeaLligvn.3m.0e(nGtu(ifingd.o4n), http
s
FIG.3. EvolutionofthestemGCcontentintheThermotogaelineage using the parameters specified by model test, and allowing ://a
overtime,whichisrepresentedbydistancefromtheroot.Thedifferent c
estimationoftheratecategoriesandproportionofinvariant a
d
genera in the clade are labeled, “other” represents Kosmotoga and sitesfromthedataset. em
Mesotoga. From this graph we can see that the lineages with higher ic
stemGCcontentareonshorterbranchesandhaveonlyincreasedtheir .o
u
stem GC content slightly over time, whereas the lineages that have Ancestral Sequence Reconstruction p.c
o
lowerstemGCcontentareonmuchlongerbranchesandhaveexpe- We used the BppML program in the Bio++ package m
rienced dramatic decreases over time. Note that for this figure, the (Dutheil and Boussau 2008) to refine the branch lengths of /m
b
branchlengthswerecalculatedfromsubstitutionsthatdidnotaffect e
thetreeproducedinPhyMLandoptimizethemodelparam- /a
thestemGCcontent. eters for ASR. To refine this tree, we defined a rtic
le
nonhomogeneous substitution model using GTR + G + I, -a
b
withadifferentsetofparametersoneachofthefourmajor s
tra
therefore OGT, must have already been able tosurvive at a clades on the tree: the more thermophilic Thermotogales c
lower temperature. After moving into a lower temperature group (genera Thermotoga, Thermosipho, and t/30
/1
niche,where thetemperatureissurvivablebutnotoptimal, Fervidobacterium), the less thermophilic Thermotogales 1
/2
theirproteinsandrRNAwouldhavebeguntoadapttofunc- group, the bacterial outgroup, and the archaeal outgroup. 4
6
tion optimally atthe new temperature.This would be facil- The parameters of the model were optimized in BppML 3/1
2
itatedbythegreaternumberofpossiblemutationsavailable (Dutheil and Boussau 2008), and used to reconstruct the 6
2
to decrease the stem GC content. In addition, a greater ancestral sequences at each node of the tree in 59
5
number of mutations would be permissible at these lower Bppancestor(DutheilandBoussau2008). b
y
temperatures(Zeldovich,Chen,andShakhnovich2007). g
u
e
Materials and Methods Gap Inferences in Ancestral Sequences st o
n
Weinferredthepositionofgapsintheancestralsequences. 1
Sequence Alignment and Tree Reconstruction Thereconstructedancestralsequenceswerethelengthofthe 3 A
p
The 16S rRNA sequences of 30 members of the originalalignment,butcontainednogaps,becausegapsare ril 2
Thermotogales were downloaded from the GenBank data- nottreatedasacharacterinthesubstitutionmodelweused. 0
1
base at NCBI. See table 6 for accession numbers and OGT Tocalculatethepositionofgapsintheancestralsequences, 9
information obtained from characterization articles. wechangedallofthegapsinexistingsequencestoCs,andall
Phenotype data on OGTs were obtained from characteriza- ofthenucleotidestoAs.WethenusedtheF84substitution
tion articles (see table 6). Organisms that have not model in Bppancestor to determine ancestral sequences,
been characterized were not included in the study. For which represent the position of gaps as nucleotides, for all
the purpose of tree and ASR, our data set also included oftheextantnodes.Thesubstitutionmodelisappropriatefor
bacterialandarchaealoutgroups:Thermoanaerobacterpseu- ourapplicationbecausenoguanineorthymine,whichdonot
dethanolis (CP000924.1), The. tengcongensis (NR_074701.1), represent anything in our model, will be introduced to the
Carboxydothermus hydrogenoformans (NR_074395.1), sequences.Thismodelhasthreeparameters:theGCcontent
Hydrogenobaculumsp.Y04AAS1(NR_074960.1),Aquifexaeo- is theta, the G/(G + C) ratio is theta1, and the A/(A + T)
licus (AJ309733.1), Sulfurihydrogenibium sp. YO3AOP1 ratio is theta2 (PERL scripts are available in supplementary
(NR_074557.1), Pyrococcus furiosus (NR_074375.1), and data,SupplementaryMaterialonline).Moreimportantly,the
Thermococcus kodakerensis (NR_028216.1), for a total of 38 GC content (or in this case C content, i.e., the number of
2469
MBE
Greenetal. . doi:10.1093/molbev/mst145
Table 6. Thermotogae Species Used in This Study, 16S Sequence Accession Numbers, and OGT.
Organism Accession No. OGT Source
Thermotoga maritima MSB8 NR_102775.1 80 Huber et al. 1986)
Thermotoga sp. RQ2 AJ872273.1 80 Swithers et al. (2011)
Thermotoga neapolitana NR_074959.1 80 Jannasch et al. (1988)
Thermotoga petrophila CP000702.1 80 Takahata et al. (2001)
Thermotoga naphthophila NR_074952.1 80 Takahata et al. (2001)
Thermotoga lettingae NR_074951.1 65 Balk et al. (2002)
Thermotoga elfii NR_026201.1 66 Ravot et al. (1995)
Thermotoga subterranea NR_025969.1 70 Jeanthon et al. (1995)
Thermotoga hypogea NR_029205.1 70 Fardeau et al. (1997)
Thermotoga thermarum CP002351.1 70 Windberger et al. (1989)
D
Thermosipho atlanticus NR_029020.1 65 Urios et al. (2004) o
w
Thermosipho geolei NR_025389.1 70 Haridon et al. (2001) n
lo
Thermosipho japonicus NR_024726.1 72 Takai and Horikoshi (2000) ad
e
Thermosipho africanus NR_102773.1 75 Huber et al. (1989) d
Thermosipho melanesiensis CP000716.1 70 Antoine et al. (1997) fro
m
Fervidobacterium islandicum NR_044730.1 70 Nam et al. (2002) h
Fervidobacterium changbaicum NR_043248.1 77.5 Cai et al. (2007) ttps
Fervidobacterium nodosum NR_074093.1 70 Patel et al. (1985) ://a
c
Fervidobacterium gondowanense NR_036997.1 66.5 Andrews and Patel (1996) ad
e
Kosmotoga olearia NR_044583.1 65 Dipippo et al. (2009) m
ic
Mesotoga prima CP003532.1 37 Nesbø et al. (2012) .o
u
Marinitoga hydrogenitolerans NR_042320.1 60 Postec et al. (2005) p.c
Marinitoga piezophila NR_027541.1 65 Alain et al. (2002) om
Marinitoga okinawensis NR_041466.1 57.5 Nunoura et al. (2007) /m
b
Marinitoga camini NR_028907.1 55 Wery et al. (2001) e/a
Petrotoga mexicana NR_029058.1 55 Miranda-Tello et al. (2004) rtic
Petrotoga halophila NR_043201.1 60 Miranda-Tello et al. (2007) le-a
Petrotoga mobilis NR_074401.1 60 Lien et al. (1998) bs
Petrotoga olearia NR_028947.1 55 L’Haridon et al. (2002) trac
Petrotoga sibirica NR_025466.1 55 L’Haridon et al. (2002) t/3
0
/1
NOTE.—Incaseswhereanoptimalgrowthrangewasgiveninthecharacterizationarticle,themidpointwasusedastheOGT. 1/2
4
6
3
/1
2
gaps),willremainconstantthroughoutthetree,becausethe ancestral sequences were inferred, we determined the stem 62
5
extantsequencesallhaveasimilarnumberofgaps,andpre- GCcontentofallofthesequences,ancestralandextant,using 9
5
sumablythe16Ssequenceshavenotshortenedorlengthened in-house PERL scripts (available in supplementary data, b
y
significantly over time. We used Bppancestor to calculate SupplementaryMaterialonline). gu
e
these three parameters based on the input sequences, and st o
thenreconstructedancestralsequencesfromthoseparame- n
Confirming Correlation between OGT and Stem 1
ters.WethenusedthepositionofCsintheoutputancestral 3
GC Content A
sequences to infer the position of gaps in the actual recon- p
structed ancestral sequences using in-house PERL scripts We calculated a regression equation between the OGT and ril 2
0
stemGCcontentofextantspecies.Ascomparisonsbetween 1
(available in supplementary data, Supplementary Material 9
traitsofrelatedspeciesarepronetocorrelationduetoshared
online).Acomparisonofthismethodwithanotheravailable
ancestry,thatis,thetraitsofrelatedspeciesarenotindepen-
methodofgapreconstructionisprovidedinthesupplemen-
dentdatapoints,weusedindependentcontrasts(Felsenstein
tarydata,SupplementaryMaterialonline.
1985),asimplementedintheapepackageforR(Paradisetal.
2004) to confirm that these two traits are correlated in the
Determination of Stem GC Content of Ancestral
Thermotogae.
Sequences
The structure of the ancestral sequences was inferred using
Calculating Time Intervals and Step Size for Random
RNASalsa(Stocsitsetal.2009)withstringencysettingss1,s2,
Walk Simulation
s3=0.9andaconstraintstructurefileforT.maritimaMSB8.
A full alignment of the extant and reconstructed ribosomal AfterweinferredtheOGTateverynodeinthetree,wethen
sequences can be found in the supplementary data, calculatedthechangeinstemGCcontentalongeachbranch
Supplementary Material online. After the structures of the ofthetree.Inourtree,thebranchlengthsaredeterminedby
2470
MBE
EvolutionofOGTintheThermotogaePhylum . doi:10.1093/molbev/mst145
P. sibirica
P. olearia
26-29
P. mobilis
P. halophila
22 P. mexicana
M. camini
24 M. okinawensis
20 23 25 M. piezophila
M. hydrogenitolerans
Ms. prima
21
K. olearia
F. gondowanense
19
18 F. nodosum
1 F. changbaicum
17 D
F. islandicum o
12 w
Ts. melanesiensis n
lo
13 Ts. africanus ad
Ts. japonicus e
1415 Ts. geolei d fro
Ts. atlanticus m
2 h
T. thermarum ttp
8 T. hypogea s
9 T. subterranea ://a
10 ca
T. elfii d
11 e
3 T. lettingae 0.01 m
T. napthophila ic.o
T. petrophila up
4-7 T. neapolitana .co
m
T. sp RQ2 /m
T. maritima str. MSB8 b
e
/a
FIG.4. 16SrRNAtreeproducedusingtheGTR+G+ImodelinPhyML.Thetreeisrootedusing16SrRNAsfrombacterialandarchaealgenomesas rtic
theoutgroup(seeMaterialsandMethodssection).Theblackdotsindicatenodeswithgreaterthan75%bootstrapsupport.Thegraydotsindicate le
-a
nodeswithlessthan75%bootstrapsupport.Thedashedlineindicatesthepositionoftheoutgroup.Thenodesarelabeledcorrespondingtotable1. b
s
tra
c
thenumberofdifferencesbetweensequences,andthetraitof branchesforuseinthesimulation.Thesimulationresultedin t/30
interestisabiasinthesequence—thereforethetwoarenot 10,000 values for the possible phenotype at each node, ob- /11
/2
independent. To determine a time-variable independent of tainedaccordingtoarandomwalkmodeloftraitevolution. 4
6
thetraitbeingmeasured,wewrotePERLscripts(availablein We simulated all possible point mutations to the 3/1
2
supplementary data, Supplementary Material online) to de- T. maritima MSB8 16S rRNA sequence using an in-house 6
2
terminethenumberofchangesalongeachbranchthatdid PERLscripttomakeallthepossiblepointmutations(available 59
5
notaffectstemGCcontent.Thesesumswerethenscaledso in thesupplementarydata,Supplementary Materialonline). b
y
thatthelowestnumberofchangesthatoccurredonabranch The resulting sequences each had one point mutation. The g
u
e
(n=0)correspondedtoabranchlengthof0,andthelargest secondarystructureofthesesequenceswasdeterminedusing s
number within the Thermotogae group (n=100) corre- RNAsalsa with constraint values of 1, appropriate for the t on
1
sponded to a branch length of 1. These calculated branch highlysimilarsequences.In-housePERLscriptswereusedto 3
lengthswereallincreasedby10(cid:3)9,toavoiddividingbyzeroin calculatethenumberofmutationsthatincreased,decreased, Ap
caseswhenthebranchlengthwas0.Thesecalculatedbranch ordidnotaffectthestemGCcontent(fig.5). ril 2
0
lengthswereusedforthefollowingrandomwalkanalysis. A second random walk simulation was performed using 1
9
the probability of the possible point mutations in the
Testing the Data against a Random Walk Simulation T. maritima ribosome. Evolution was again simulated from
theroottothetips,samplingfromthedistributionofpossible
Ancestralreconstructiondatawereusedtotestthehypoth-
point mutations. The number of samples per branch was
esisthatevolutionofOGTintheThermotogaeproceedsbya
calculated by multiplying the branch length (which gives
randomwalk,usingstemGCcontentasaproxyforOGT.This
the substitutions per site) by the length of the entire
wasdoneusingthefunctionrTraitCont,availableintheape
sequence, 1,462. Evolution was simulated in this fashion
packageforR(Paradisetal.2004),whichsimulatesevolution
10,000times.
ofacontinuouscharacterfromtherootofagivenphylogeny
tothetips.WesimulatedtheevolutionofthestemGCcon-
tent10,000times,alongthetreewithbranchlengthscalcu- Ancestral Character State Reconstruction
lated as explained above. One outlier was excluded when Usingthephylogenygeneratedabove,wecomparedvarious
calculating the standard deviation of the change along the methods of ancestral character state reconstruction. To
2471
MBE
Greenetal. . doi:10.1093/molbev/mst145
point estimates, it was also possible to generate CIs for
these estimates. As a much more conservative test of the
congruence between methods, we also used the number of
nodes on a tree for which the CIs for the nodes had any
overlap. By definition, our expected frequency of overlap in
CIs should be >0.05, whereas point estimates should fall in
theCIs95%ofthetime.WealsotestedwhetherCIswereof
significantlydifferentsizesbetweenthereconstructionmeth-
odsbyusingt-tests.
In addition, we tested two models of trait evolution in
BayesTraits (Pagel et al. 2004); a constant-variance random
walkmodelandadirectionalrandomwalkimplementedina
Bayesian framework. These two models of trait evolution D
o
w
were compared using likelihood ratio test (Pagel 1999). n
lo
BayesTraitsallowsforthereconstructionofthemostrecent a
d
commonancestorofalltaxainthetree,andestimatesfrom ed
thetwomodelsofcharacterevolutionwerecomparedwith fro
m
the results estimatedfrom the GC content of the sequence h
datatoseewhetherpointfellwithinthe95%CIandwhether ttp
s
CIsandcredibilityintervalsoverlapped.Defaultsettingswere ://a
c
used in BayesTrait with the exception of the rate deviance a
d
e
parameter,whichwassetat0.15fortherandomwalkand0.1 m
ic
for the directional walk to get an acceptance ratio between .o
u
20%and40%assuggestedintheBayesTraitsmanual. p
.c
o
m
Supplementary Material /m
b
e
Supplementary data are available at Molecular Biology and /a
Evolutiononline(http://www.mbe.oxfordjournals.org/). rticle
-a
b
s
Acknowledgments tra
c
The authors thank Kenneth Noll, Pascal Lapierre, David t/30
Williams,TimothyHarlow,andOlgaZhaxybayevafordiscus- /11
/2
sions and the Biotechnology Bioservices Center of the 4
6
UniversityofConnecticut,Storrs,USA,fortechnicalsupport. 3/1
2
This work was supported by a US National Science 6
FIG. 5. Histogram of the number of possible mutations that cause a 2
decrease(a)oranincrease(b)inOGT.Thehistogramsareshownwith FoundationGrant(DEB0830024toJ.P.G.andDGE-1142336 595
equalyaxesandxaxesofthesamelengthtodemonstratethehigher to J.F.G.) and the Canadian Institutes of Health Research’s b
y
numberofmutationsthatdecreasethestemGCcontent. Strategic Training Initiative in Health Research’s Systems gu
e
BiologyTrainingProgram(toJ.F.G.). s
t o
n
comparedifferentreconstructionmethods,weimplemented References 13 A
atemrsauxsiimngumaBlirkoewlihnoiaondmreoctoionnstmruoctdieoln(Sfocrhlcuotnetrinetuoalu.s19c9h7a)r,aacs- AlainEAK,,PrMieaurrteDin,sBsiorrnienVTJL,.M20ir0o2s.hMniacrhinenitokogaMpiLe,zoBpohnilcah-sOp.smnoovl.o,vaskroayda- pril 20
shaped, thermo-piezophilic bacterium isolated under high hydro- 1
well as reconstructions based on least squares (i.e., PIC; 9
static pressure froma deep-sea hydrothermal vent. Int J Syst Evol
Felsenstein 1985), GLSB and GLSG model (Martins and
Microbiol.52(4):1331–1339.
Hansen1997;Cunninghametal.1998),andSCP(Maddison AndrewsKT,PatelBK.1996.Fervidobacteriumgondwanensesp.nov.,a
1991);allimplementedintheAPEpackageforR(Paradisetal. new thermophilic anaerobic bacterium isolated from nonvolcani-
2004;RDevelopmentCoreTeam2012).REMLresultsinun- cally heated geothermal waters of the Great Artesian Basin of
Australia.IntJSystBacteriol.46(1):265.
biased estimates of the variance of the Brownian motion
Antoine E, Cilia V, Meunier JR, Guezennec J, Lesongeur F, Barbier G.
process, whereas maximum likelihood gives a downward
1997.Thermosiphomelanesiensissp.nov.,anewthermophilican-
biassoweusedtherestrictedmaximumlikelihoodapproach aerobicbacteriumbelongingtotheorderThermotogales,isolated
fortheBrownianmotionmodel(Heyde1997). from deep-sea hydrothermal vents in the southwestern Pacific
Totestthecongruenceofourreconstructedestimatesof Ocean.IntJSystBacteriol.47(4):1118.
Balk M, Weijma J, Stams AJ. 2002. Thermotoga lettingae sp. nov., a
ancestral stem GC content with those estimated from the
novel thermophilic, methanol-degrading bacterium isolated from
sequence data, we used G-tests (Sokal and Rohlf 1995).
a thermophilic anaerobic reactor. Int J Syst Evol Microbiol.
Although reconstruction estimates were often given as 52(Pt4):1361.
2472