The Rooting of the Universal Tree of Life Is Not Reliable

Abstract

Several composite universal trees connected by an ancestral gene duplication have been used to root the universal tree of
life. In all cases, this root turned out to be in the eubacterial branch. However, the validity of results obtained from comparative
sequence analysis has recently been questioned, in particular, in the case of ancient phylogenies. For example, it has been
shown that several eukaryotic groups are misplaced in ribosomal RNA or elongation factor trees because of unequal rates of
evolution and mutational saturation. Furthermore, the addition of new sequences to data sets has often turned apparently reasonable
phylogenies into confused ones. We have thus revisited all composite protein trees that have been used to root the universal
tree of life up to now (elongation factors, ATPases, tRNA synthetases, carbamoyl phosphate synthetases, signal recognition
particle proteins) with updated data sets. In general, the two prokaryotic domains were not monophyletic with several aberrant
groupings at different levels of the tree. Furthermore, the respective phylogenies contradicted each others, so that various
ad hoc scenarios (paralogy or lateral gene transfer) must be proposed in order to obtain the traditional Archaebacteria–Eukaryota
sisterhood. More importantly, all of the markers are heavily saturated with respect to amino acid substitutions. As phylogenies
inferred from saturated data sets are extremely sensitive to differences in evolutionary rates, present phylogenies used to
root the universal tree of life could be biased by the phenomenon of long branch attraction. Since the eubacterial branch
was always the longest one, the eubacterial rooting could be explained by an attraction between this branch and the long branch
of the outgroup. Finally, we suggested that an eukaryotic rooting could be a more fruitful working hypothesis, as it provides,
for example, a simple explanation to the high genetic similarity of Archaebacteria and Eubacteria inferred from complete genome
analysis.

Full text

The Rooting of the Universal Tree of Life Is Not Reliable
Herve´ Philippe,
1
Patrick Forterre
2
1
Phyloge´nie et Evolution Mole´culaires (UPRESA 8080 CNRS), Baˆtiment 444, Universite´ Paris-Sud, 91405 Orsay-Cedex,France
2
Institut de Ge´ne´tique et Microbiologie (UMR 8621 CNRS), Baˆtiment 409, Universite´ Paris-Sud, 91405 Orsay-Cedex, France
Abstract. Several composite universal trees connected
by an ancestral gene duplication have been used to root
the universal tree of life. In all cases, this root turned out
to be in the eubacterial branch. However, the validity of
results obtained from comparative sequence analysis has
recently been questioned, in particular, in the case of
ancient phylogenies. For example, it has been shown that
several eukaryotic groups are misplaced in ribosomal
RNA or elongation factor trees because of unequal rates
of evolution and mutational saturation. Furthermore, the
addition of new sequences to data sets has often turned
apparently reasonable phylogenies into confused ones.
We have thus revisited all composite protein trees that
have been used to root the universal tree of life up to now
(elongation factors, ATPases, tRNA synthetases, car-
bamoyl phosphate synthetases, signal recognition par-
ticle proteins) with updated data sets. In general, the two
prokaryotic domains were not monophyletic with several
aberrant groupings at different levels of the tree. Further-
more, the respective phylogenies contradicted each oth-
ers, so that various ad hoc scenarios (paralogy or lateral
gene transfer) must be proposed in order to obtain the
traditional Archaebacteria–Eukaryota sisterhood. More
importantly, all of the markers are heavily saturated with
respect to amino acid substitutions. As phylogenies in-
ferred from saturated data sets are extremely sensitive to
differences in evolutionary rates, present phylogenies
used to root the universal tree of life could be biased by
the phenomenon of long branch attraction. Since the eu-
bacterial branch was always the longest one, the eubac-
terial rooting could be explained by an attraction be-
tween this branch and the long branch of the outgroup.
Finally, we suggested that an eukaryotic rooting could be
a more fruitful working hypothesis, as it provides, for
example, a simple explanation to the high genetic simi-
larity of Archaebacteria and Eubacteria inferred from
complete genome analysis.
Key words: Root of the tree of life — ATPase —
Carbamoyl phosphate synthetase — Elongation factor —
tRNA synthetase — Signal recognition particle — Mu-
tational saturation — Long branch attraction
Introduction
According to ribosomal RNA (rRNA) sequence com-
parisons, all extant cellular organisms have been classi-
fied into one of three domains Eubacteria, Archaebacte-
ria, and Eukaryota (Woese 1987). The classification of
the living world in three groups is supported by numer-
ous molecular phenotypic traits specific for each domain
(for recent reviews, see Brown and Doolittle 1997; For-
terre 1997; Olsen and Woese 1997). Comparison of mo-
lecular biology and central metabolism between Eubac-
teria, Archaebacteria and Eukaryota is expected to help
in reconstituting the characteristics of the last common
ancestor to all extant cellular life, or cenancestor, here
called the Last Universal Cellular Ancestor (LUCA).
This would require polarizing characters found in one or
two domains to determine if they are primitive or derived
features (plesiomorphies or synapomorphies sensu Hen-
nig 1966). The rooting of the universal tree of life would
Correspondence to: H. Philippe, e-mail: herve.philippe@bc4.u-psud.fr
J Mol Evol (1999) 49:509–523
©Springer-Verlag New York Inc. 1999

facilitate this task since homologous traits only shared by
two domains, which are not sister-groups, should be an-
cestral. For example, considering that Archaebacteria
and Eubacteria have a similar type of genome organiza-
tion (chromosome size and number, operons, mode of
cell division), rooting the universal tree in the eubacterial
or the archaebacterial branch would suggest that these
traits were already present in LUCA. In contrast, if the
universal tree is rooted in the eukaryotic branch, these
characters could either have been present in LUCA or
have appeared in the branch common to the prokaryotes,
i.e. correspond to an evolved state.
At the end of the eighties, two research teams tenta-
tively rooted the universal tree of life in the eubacterial
branch (Gogarten et al. 1989; Iwabe et al. 1989). This
was inferred from the construction of universal trees for
two pairs of paralogous proteins, which originated by
gene duplication before LUCA. The proteins used were
the elongation factors (EF), namely EF-1␣(Tu) versus
EF-2(G), and the catalytic versus regulatory subunits of
eubacterial F-ATPases, and V or V-like-ATPases found
in Eukaryota and Archaebacteria. The root always turned
out to be located in the eubacterial branch. Later on,
Brown and Doolittle (1995) used the same strategy to
root a universal tree of lle-tRNA synthetases (lle-tRS)
versus paralogous Val- and Leu-tRS, Lawson et al.
(1996) a carbamoyl phosphate synthetase (CPS) tree us-
ing an internal gene duplication, Brown et al. (1997) a
Tyr-tRS tree versus paralogous Trp-tRS, and Gribaldo
and Cammarano (1998) a Signal Recognition Particle
(SRP) 54kD protein using paralogous SRP receptor SR-
␣. In all cases, the root again separated Eubacteria from
the two other domains. Moreover, a reanalysis of the
elongation factor data set with more sequences and a
refined alignment strengthened the eubacterial rooting
(Baldauf et al. 1996).
The eubacterial rooting supports the current view that
LUCA was a prokaryotic-like organism since characters
shared by Archaebacteria and Eubacteria are considered
primitive. Furthermore, it fits intuitively well with the
finding that several features of the cellular information
processing system are more similar between Eukaryota
and Archaebacteria than between Archaebacteria and
Eubacteria (Olsen and Woese 1997), and the common
assumption that these features are more “evolved” than
their eubacterial counterparts. Accordingly, this rooting
was rapidly accepted and advertised in the community of
evolutionary biologists and beyond, being now system-
atically used to draw universal trees in review papers,
and even textbooks. The eubacterial rooting was also
endorsed to support several evolutionary hypotheses,
such as the origin of life at high temperature (Stetter
1992). Last but not least, Woese and coworkers recruited
this rooting to support their new nomenclature for the
three domains of life (removing the suffix bacteria from
Archaebacteria) (Woese et al. 1990), since Archaebacte-
ria are the sister group of Eukaryota, not of Eubacteria,
when the universal tree is rooted in the eubacterial
branch.
However, the validity of sequence comparison to infer
ancient phylogenies has been questioned on various
grounds. With more and more sequences available, it
turned out that most protein phylogenies contradict each
others as well as the rRNA tree (reviewed in Brown and
Doolittle 1997; Doolittle and Brown 1994; Forterre
1997). In several cases, archaebacterial proteins were
found more closely related to eubacterial ones than to
eukaryotic ones, whilst in some cases eukaryotic proteins
appeared close to eubacterial ones. This situation led to
two major reactions. Some people suggested new sce-
narios of early cellular evolution based on their favorite
proteins, or else diverse scenarios of fusion between
primitive lineages to take into account contradictions be-
tween different phylogenies (Gupta and Golding 1993;
Martin and Muller 1998; Moreira and Lopez-Garcia
1998; Rivera and Lake 1992; Sogin 1991; Zillig 1987).
Other evolutionists argued that the proteins from the in-
formation processing system were intrinsically better
than others because they are less prone to inter-domain
transfer than metabolic proteins. Accordingly, since
many proteins of the archaebacterial transcription, trans-
lation or replication apparatus resemble their eukaryotic
homologues more than their eubacterial ones, they sug-
gested that their phylogenies (even unrooted) testify for
the eubacterial rooting of the tree of life (see for example
Brown and Doolittle 1997).
However, it is possible that contradictions observed
between universal phylogenies obtained with rRNA and
various proteins do not require specific ad hoc hypoth-
eses but simply reflect the weakness of the tree recon-
struction methods that have been used to infer these phy-
logenies (Forterre 1997; Philippe and Laurent 1998). In
particular, when the elongation factor and tRNA synthe-
tase data sets were analyzed for the slowly evolving po-
sitions which should have been a priori the most infor-
mative, we did not found a significant signal for any
rooting (Forterre 1997; Forterre et al. 1992). Up to now,
these criticisms have not been sufficiently taken into ac-
count. Nevertheless this situation could change now, fol-
lowing recent developments in the study of early eukary-
otic evolution which showed that some molecular
phylogenies might be highly misleading. Indeed, in the
case of eukaryotes, strikingly different trees can be ob-
tained depending on the molecule analyzed (either
rRNA, actin or tubulin). The order of emergence of the
various groups at the base of the eukaryotic tree mainly
depends on the rate of evolution of the protein used (the
more rapidly evolving taxon emerging first) because the
long branches of these groups are attracted by the long
branch of the outgroup that roots the tree (Philippe and
Adoutte 1998).
These considerations prompted us to revisit in detail
510

all the phylogenies that have been used up to now to root
the universal tree of life, using updated data sets that
especially include many novel archaebacterial and eu-
bacterial sequences obtained from complete genome se-
quencing efforts. Here we report our studies on CPS,
SRP, elongation factors, lle-tRS, Trp/Tyr-tRS and
ATPase genes. We demonstrate that the phylogenies are
highly confusing due to the combining effects of gene
duplication, gene loss, lateral gene transfer and tree re-
construction artefact. Moreover the six genes appear to
be highly mutationally saturated, suggesting that very
few ancient phylogenetic signal remains. Finally we sug-
gest that the eubacterial rooting is the result of a long
branch attraction artefact and we discuss the hypothesis
of a eukaryotic rooting.
Materials and Methods
All sequences homologous to the carbamoyl phosphate synthetase
(CPS), ATPase, lle- and Val-tRNA synthetase (tRS), Trp- and Tyr-tRS,
signal recognition protein (SRP) proteins, and elongation factors EF1␣
and EF2 available in data banks were identified by a BLAST search
using the sequence from Escherichia coli as query sequence. The pro-
grams blast2retp and retp2ali (Philippe Lopez, personal communica-
tion) allowed us to retrieve all the sequences automatically and to write
them into a MUST-compatible file. The alignment of these sequences
was carried out visually with the help of the ED program of the MUST
package version 1.0 (Philippe 1993). Some sequences were discarded
because they were either partial or redundant or because they contained
likely sequencing errors. The Pyrobaculum aerophilum sequences were
kindly provided by Drs. Sorel Fitz-Gibbon and Jeffrey Miller. Prelimi-
nary sequence data were obtained from the Institute for Genomic Re-
search website at http://www.tigr.org. The resulting alignments con-
tained 339, 122, 344, 104, 118, and 103 sequences for ATPase, CPS,
EF, lle/Val-tRS, SRP, and Trp/Tyr/tRS, respectively. Due to computer
time limitation, only 124 and 75 sequences were selected for ATPase
and EF, while keeping the greatest possible phylogenetic diversity.
Positions that could not be unambiguously aligned were excluded from
the analysis, yielding 201, 271, 158, 322, 184, and 83 usable positions
for ATPase, CPS, EF, lle/Val-tRS, SRP, and Trp/Tyr-tRS, respectively.
All alignments are available from HP upon request.
Phylogenetic trees were constructed with maximum-likelihood
(ML), maximum-parsimony (MP), and distance-based methods, with
the programs PROTML (Adachi and Hasegawa 1996) version 2.3,
PAUP (Swofford 1993) version 3.1, and NJ in the MUST package
(Philippe 1993) version 1.0, respectively. The distances were computed
with the substitution model of Kimura (1983). MP trees were obtained
by 10 random addition heuristic search replicates. Due to the high
number of species used, the search for the ML tree was limited to a
reduced sample of species and to local rearrangement method (option
R) starting from the MP and the neighbor-joining (NJ) trees. The model
of amino acid substitution used was JTT. Bootstrap proportions (BP)
were calculated by analysis of 1000 replicates for NJ analysis (Saitou
and Nel 1987). The results obtained by MP and ML methods are not
shown because they are very similar to those of the NJ method.
The saturation level of the phylogenetic markers was estimated
with the use of the method of Philippe et al. (1994). The inferred
number of substitutions between each couple of species was estimated
from the MP or the ML trees as the sum of the lengths of all the branch
on the pathway linking these two species, using the program
TREEPLOT (Philippe 1993). Using the program COMP_MAT, a plot
was drawn to estimate the saturation level by displaying all the pairs of
species with an abscissa value equal to the number of inferred substi-
tutions and an ordinate equal to the number of observed differences.
The mutational saturation was revealed by the presence of a plateau
within which the number of substitutions increased, whereas the num-
ber of differences remained constant.
Results and Discussion
Confusing Phylogenies
Updated phylogenies are shown in Figs. 1–5 for the two
CPS domains, Ile- and Val-tRS, Tyr- and Trp-tRS, the V-
and F-ATPases, and the SRP and its receptor. The num-
ber of sequences has considerably increased during the
last 2 years, thanks to the numerous genome projects that
have been completed or are in progress (http://www.
tigr.org/tdb/mdb/mdb.html). Furthermore, the availabil-
ity of complete eubacterial and archaebacterial genomes
allows us to identify putative gene loss, gene transfer,
and gene duplication more safely. None of the six up-
dated trees offers the classical Woese’s picture, e.g., the
monophyly of the three domains and the eubacterial root-
ing of each subtree using the other as an outgroup.
The more puzzling phylogeny was observed for the
Tyr/Trp-tRS tree (Fig. 1), since the monophyly of both
types of synthetase was not recovered, bacterial Trp- and
Tyr-tRS being grouped together. This peculiar phylog-
eny was initially obtained by Ribas de Pouplana et al.
(1996), using a limited data set that did not include ar-
chaebacterial sequences. These authors speculated that
the divergence between Trp- and Tyr-tRS might have
occurred only after the separation of prokaryotes and
eukaryotes. Later on, the monophyly of both types of
synthetase was nevertheless recovered by Brown et al.
(1997) using an expanded data set and a different align-
ment. They concluded that the phylogeny previously ob-
tained by Ribas de Pouplana and co-workers was due to
long branch attraction (LBA) and that inclusion of ar-
chaeal sequences had allowed to infer the correct phy-
logeny by breaking the longest branches. However, in
our present analysis, we get anew the topology first ob-
tained by Ribas de Pouplana and coworkers (1996). It
should be noted that the alignment of these tRNA syn-
thetases is very difficult to perform. This is confirmed by
the fact that a blast search using the Tyr-tRS of E. coli as
a query sequence detects the eubacterial Tyr-tRS only,
and neither the other Tyr-tRS nor the Trp-tRS. We were
able to align only 83 positions unambiguously, which is
significantly fewer than Brown et al. (1997) (147 or 184)
and Ribas de Pouplana et al. (1996) (between 190 and
230). This difference in the alignment together with the
use of various species sampling can explain the instabil-
ity of the inferred phylogenies. Since the two eubacterial
branches are much longer than all others in this phylog-
eny, we think that LBA is indeed the most likely hypoth-
esis to explain the nonmonophyly of each type of syn-
thetase. Examination of the local part of the Tyr and
511

Fig. 1. Phylogenetic tree based on comparison of Trp- and Tyr-tRNA synthetase sequences; 83 unambiguously aligned positions were used. The
tree was constructed with the NJ method employing the Kimura method of distance calculation. Bootstrap proportions are indicated when greater
than 50%. A scale bar corresponding to 10 substitutions per 100 positions is given at bottom.
512

Fig. 2. Phylogenetic tree based on comparison of ATPase sequences; 201 unambiguously aligned positions were used. For method, see the legend
to Fig. 1.
513

Fig. 3. Phylogenetic tree based on comparison of Ile- and Val-tRNA synthetase sequences; 322 unambiguously aligned positions were used. For
the method, see the legend to Fig. 1.
514

Fig. 4. Phylogenetic tree based on comparison of CPS sequences; 271 unambiguously aligned positions were used. For the method, see the legend
to Fig. 1.
515

Fig. 5. Phylogenetic tree based on comparison of SRP sequences; 184 unambiguously aligned positions were used. For the method, see the legend
to Fig. 1.
516

Trp-tRS tree revealed other oddities. In particular, Pyro-
coccus (Archaebacteria) sequences of Trp-tRS branch
inside the eukaryotes, whereas plant Tyr-tRS branch in-
side Archaebacteria. Our updated phylogeny also iden-
tifies two groups of eubacterial Tyr-tRS (1 and 2). They
have probably originated from a duplication in the eu-
bacterial domain since they are both present in most
bacterial kingdom, two species, B. subtilis and Clos-
tridium acetobutilicum, containing the two genes. If this
hypothesis is correct, one of the two genes should have
been repeatedly lost during eubacterial evolution. All mi-
tochondrial sequences are related to group 2, but they
emerged at the basis of this group instead of branching
with ␣-proteobacteria. Our conclusion is that Tyr- and
Trp-tRS are very bad phylogenetic markers (probably in
part because of the low number of residues which can be
aligned) and cannot be used to root the tree of life con-
fidently.
The ATPase tree (Fig. 2) resembles the Tyr and Trp-
tRS tree in the difficulty to recover the presumed mono-
phyly of the two proteins that were originally supposed
to be paralogous, in that case catalytic versus regulatory
subunits of V- and F-type ATPases. The ATPase tree
exhibits five major groups with extremely long basal
branches, such that the monophyly of each group is well
supported but the relationships between these groups are
very difficult to ascertain. To evaluate the robustness of
the ATPase tree, we computed the likelihood of three
quite different topologies, (((F
␣
,F
␤
), FliI),(V
A
,V
B
)),
(((F
␣
, FliI), F
␤
),(V
A
,V
B
)), and (((F
␤
, FliI), F
␣
),(V
A
,
V
B
)), applying Kishino/Hasegawa’s test on a limited set
of 90 species. The difference of log-likelihood (⌬L) with
respect to the ML tree (similar to that in Fig. 2; logL⳱
−21252.9) turned out to be only −20.5 (2.3 SE). By com-
parison, ⌬Lfor the tree constraining the monophyly of
crenotes with the V
B
gene was −23.0 (1.6 SE). Since this
constraint was quite reasonable, a difference of 20 in the
log likelihood cannot be rejected. As a result, the orthol-
ogy between F
␣
and V
B
, on one hand, and between F
␤
and V
A
, on the other, is far from being strongly sup-
ported.
One of the five groups includes only eubacterial se-
quences corresponding to flagellar ATPases, suggesting
a duplication and functional specialization in Eubacteria.
The two groups of F-ATPases contain only bacterial se-
quences with a single exception, the archaeon Methano-
sarcina barkeri, suggesting a transfer from Eubacteria to
Archaebacteria. However, in that case, one should imag-
ine an LBA artifact to explain why M. barkeri emerged
at the base of the two eubacterial subtrees. In contrast to
F-ATPases, the two groups of V-ATPases contain se-
quences from the three domains. This type of ATPase
was originally discovered in Archaebacteria and consid-
ered the bona fide archaebacterial enzyme (Gogarten et
al. 1989). Later on, the discovery of this type of ATPase
in two Eubacteria was interpreted as a lateral gene trans-
fer from Archaebacteria to Eubacteria (Gogarten et al.
1996). Now V-ATPase appears to be present in several
major branches of the eubacterial tree (Fig. 2). However,
their phylogeny is very confused: Archaebacteria and
Eubacteria turned out to be paraphyletic in both subtrees
with eukaryotic sequences and most archaebacterial ones
branching inside eubacterial sequences. Accordingly, be-
sides the previous hypothesis of several gene transfers
from Archaebacteria to Eubacteria, one can now argue as
well for several transfers from Eubacteria to Archaebac-
teria.
Considering the general shape of the tree with both F-
and V-ATPases, the long branches of the two eubacterial
F-ATPases subtrees suggest that these proteins might
have appeared by gene duplication and functional spe-
cialization in Eubacteria, as in the case of flagellar
ATPases. It is also possible that F- and V-ATPases were
already present in LUCA and that V-ATPase are ortho-
logues in the three domains (Forterre et al. 1992). In any
case, the ATPase data set appears unsuitable to root the
universal tree of life since, as for the Tyr- and Trp-tRS,
the evolutionary relationships between the various
classes of enzymes are obscure.
Similar problems are now obvious with the Ile- and
Val-tRS tree (Fig. 3). It has been shown previously that
the Val-tRS phylogeny cannot be used to root the uni-
versal tree since the eukaryotic enzymes turned out to be
of mitochondrial origin (Brown and Doolittle 1995;
Hashimoto et al. 1998). This is confirmed by our analy-
sis. However, the Ile-tRS tree was supposed to be safe for
this rooting. This is no more true with our new data set.
The updated tree revealed the existence of a very diverse
group of eubacterial Ile-tRS, including sequences from
many eubacterial lineages, which branch between the
archaebacterial and eukaryotic enzymes (Brown et al.
1998). To save the Woesian structure of the Ile-RS tree,
the existence of this group of sequence should be ex-
plained by the ancient transfer of an eukaryotic gene to
Eubacteria and its rapid evolution in this new context.
However, there is no objective reason to consider that
one of the two eubacterial groups corresponds to the
bona fide eubacterial gene. These two genes might be
ancient paralogues that have been lost selectively during
eubacterial evolution. Furthermore, the eukaryotic gene
might have been recruited from one of these eubacterial
species by the ratchet mechanism proposed by Doolittle
(1998) or from mitochondria (Rickettsia possessing this
“abnormal” gene). Many alternative scenarios can be
proposed with no obvious possibility to make a rational
choice. An interesting feature in Fig. 3 is that Archae-
bacteria are polyphyletic, the majority of them clustering
with Eubacteria and only Pyrobaculum with eukaryotes.
However, with MP and ML method, the Archaebacteria
are paraphyletic and the sister group of eukaryotes. This
switch between the eukaryotic and the eubacterial root-
ing depending on the tree reconstruction method could
517

be due to the limited resolving power of this gene. In any
case, the only safe conclusion is that the Ile-tRS phylog-
eny cannot be used anymore to root the tree of life with
confidence.
The universal tree inferred from the two CPS domains
(D1 and D2) appears a priori less confused since the
monophyly of each CPS domains is clearly recovered
(Fig. 4). However, the root is no more in the eubacterial
branches of the two subtrees, as it was the case in pre-
vious analysis that included only one archaebacterial se-
quence (Lawson et al. 1996). Now Archaebacteria turned
out to be polyphyletic: several euryotes (Archaeoglobus,
Methanococcus, and Methanobacterium) clustered with
Eubacteria, and two crenotes (Sulfolobus and Pyrobacu-
lum) and one euryote (Pyrococcus furiosus) clustered
with eukaryotes. Moreover, one eubacterial sequence
(Porphyromonas) emerged within eukaryotes and the
complete genome sequencing of Pyrococcus horikoshi
and Pyrococcus abyssi has revealed that this gene is
absent in both species. Finally, gene duplication of CPS
occurred in Eubacteria (two genes in Bacillus) and in
eukaryotes (two genes in Saccharomyces). As a result,
inferring the species phylogeny from the CPS gene phy-
logeny is a very difficult task because of the numerous
gene duplications, gene loss and horizontal gene transfer.
One can argue for an eubacterial rooting by assuming
that the euryote sequences have been acquired by hori-
zontal gene transfers or for a eukaryotic rooting by as-
suming that the crenote sequences have been acquired
from eukaryotes.
In conclusion, all four of these genes (tRS, ATPase,
and CPS) cannot be used confidently to root the tree of
life because of the difficulty to choose between different
evolutionary scenarios, knowing that gene duplication,
gene loss, and lateral gene transfer have been frequent
during prokaryotic evolution.
Classical Phylogenies
Of the six trees examined, the elongation factor (shown
by Lopez et al. 1999) and the signal recognition particle/
receptor trees are those that are more like the classical
Woese’s tree. However, they are again plagued with
various problems. For SRP (Fig. 5), the three domains
are monophyletic, with the root in the eubacterial branch,
except for ffh, where Archaebacteria are paraphyletic
with Archaeoglobus branching first. The situation is
more complex in the SR␣/Ftsy/FIhF subtree since it con-
tains two eubacterial groups. The Ftsy group, which cor-
responds to functional analogues of SR␣branches first in
this subtree, whereas the FIhF, which corresponds to
proteins involved in the biogenesis of the flagellum, is
clustered with Archaebacteria and eukaryotes. The two
bacterial groups likely originated from gene duplication
in the bacterial domain and the long branch of the FIhF
probably reflected rapid evolution due to functional
change. The relations of orthology cannot be safely in-
ferred, so that this part of the tree cannot be used to root
the universal tree (Brinkmann and Philippe 1999). In the
elongation factor tree (Lopez et al. 1999), eukaryotes and
Eubacteria are monophyletic in the two subtrees, but Ar-
chaebacteria are paraphyletic in both cases. The EF-2
(EF-G) tree exhibits Lake’s topology with crenotes
grouped with eukaryotes. However, the euryotes are
paraphyletic, the euryote Halobacterium salinarium
branching between crenotes and other euryotes. Further-
more, the EF-1␣tree exhibits a different topology, Ar-
chaebacteria being monophyletic, except the early emer-
gence of the Pyrococcus/Thermococcus group.
In all six trees, the examination of intradomain phy-
logenetic patterns shows a mixture of correct groupings
(e.g., Thermococcus with Pyrococcus, Thermus with
Deinococcus) and wrong ones (e.g., the grouping in the
Trp-tRNA synthetase tree of Rickettsia (an ␣-proteobac-
terium) with those of Synechocystis (a cyanobacterium)).
There are too many of these aberrant grouping to be
described in detail here, all the more so as that they are
often not well supported. Although some of them can be
explained by horizontal transfers, the remaining oddities
should be explained by tree reconstruction artifacts
(LBA, for example).
Saturation Analysis
The reanalysis of the six genes that previously supported
the sisterhood between Archaebacteria and Eukaryota by
using many new sequences led to a picture that was
much more complex than that first reported. The failure
to obtain the same phylogeny from different markers
could thus be explained by the following: (1) the species
tree was different from the gene tree, and (2) the tree
reconstruction method was inappropriate. The first hy-
pothesis was supported by the existence of several
clearly enigmatic sequences, which implied at least sev-
eral horizontal transfers but implied that it was almost-
impossible to infer the good species tree. This point has
been discussed in many papers (Doolittle 1998; Feng et
al. 1997; Jain et al. 1999; Lawrence and Ochman 1998)
and is not discussed in detail here. It should be noted,
however, that even if horizontal transfers are relatively
frequent and can severely disturb some gene phylogenies
(Figs. 1–4), they do not mix up the genomes since the
phylogeny based on gene content is similar to the phy-
logeny based on rRNA (Snel et al. 1999). The second
hypothesis was of prime importance because (i) the phy-
logenetic relationships within the domains were incor-
rectly inferred, and (ii) the model of sequence evolution
used was quite oversimplified (see Sullivan and Swof-
ford 1997), as discussed in the accompanying paper
(Lopez et al. 1999). One can conclude from all these
analyses that the relationships among Eubacteria, Eur-
yarchaeota, Crenarchaeota, and Eukaryota were still not
solved, despite the use of six different genes.
518

In fact, such a question deals with very ancient events,
at least 1 billion years ago and possibly more than 3
billion, and one should expect molecular phylogenetics
to encounter many problems, since, for example, the
complete mitochondrial genome was not able to recover
the rodent monophyly (Philippe 1997), an event that oc-
curred less than 100 million years ago. During more than
1 billion years, the evolutionary rate could have varied in
the different lineages, generating erroneous phylogenies
because of the LBA phenomenon. A more dramatic
problem could be that numerous multiple substitutions
occurred after the divergence of the three domains and
masked the old phylogenetic signal. To test this hypoth-
esis, the method of Philippe et al. (1994) was applied to
the six data sets to evaluate the mutational saturation.
The principle of this method is to compare, for each
pair of species, the number of observed differences and
the number of substitutions inferred by the parsimony
method, which is able to detect a fraction of the multiple
substitutions occurring at a same position and thus gives
an estimate of the real number of substitution. If the
phylogenetic marker is saturated, the number of inferred
substitutions will still increase, whereas the number of
observed differences stays nearly constant, which gener-
ates a plateau in the graphical plot of the pairwise com-
parison. For the six genes analyzed, such a plateau was
obviously present (Fig. 6). For example, the number of
observed differences reached a maximum of 190 for
ATPase (Fig. 6C), but the number of corresponding in-
ferred substitutions varied from 200 up to 700. The pla-
teau always contained all the pairwise comparisons of
the paralogous sequences (displayed as an open circle).
The saturation level within one orthologous gene was
variable. The most saturated marker was the ATPase
gene (Fig. 6C), because all the comparisons between a
eubacterial sequence and its putative orthologous archae-
bacterial or eukaryotic sequence were located within the
major plateau. The Ile-RS was almost as saturated, be-
cause the orthologous comparisons were mixed with the
paralogous ones (Fig. 6B). Elongation factors were also
highly saturated, as evidenced by a large plateau just
below the plateau of the paralogous comparisons (Fig.
6D). The CPS was less saturated, because the ortholo-
gous comparisons displayed only a small tendency to
form a plateau (Fig. 6A).
Since maximum parsimony is not the most efficient
method to recover a phylogenetic tree and to detect mul-
tiple substitutions (Hasegawa et al. 1991), the same
analysis was carried out with the use of the maximum
likelihood method to infer the number of substitutions. A
very similar pattern was observed (Fig. 7), and indeed
the level of saturation appeared to be higher than with the
parsimony estimation. This point was evidenced by the
fact that the plateau of the paralogous comparisons began
farther from the line with a slope equal to 1, which rep-
resented the theoretical case where no multiple substitu-
tions occur. It was not surprising that the probabilistic
approach allowed us to detect more multiple substitu-
tions than the parsimony one, especially along the long
unbroken branches. Moreover, it was likely that even the
number of substitutions inferred by maximum likelihood
was severely underestimated, at least for the very large
distances.
As a result, the six genes used to root the universal
tree of life were found to be highly saturated, probably
much more than shown in Figs. 6 and 7. This raised a
new interpretation of the inferred phylogeny. The reli-
ability of the eubacterial rooting has been supported by
the fact that the paralogous genes could have a constant
evolutionary rate (Feng et al. 1997; Iwabe et al. 1989),
thus avoiding the LBA artefact (Felsenstein 1978).
Fig. 6. Mutational saturation curves. A, CPS; B, Ile-tRS and Val-
tRS; C, ATPases; D, EF-1␣and EF-2; E, Trp-tRS and Tyr-tRS; F,
SRP. Yaxis: the observed number of differences between pairs of
species sequences. Xaxis: the inferred number of substitutions between
the same two sequences determined using the maximum-parsimony
method. Each dot thus defines the observed versus the inferred number
of substitutions for a given pair of sequences. It can be seen that in the
six cases, the curve levels off after a given point, indicating that while
the number of inferred mutations still increases (Xaxis), they are no
longer detected as observed differences (leveling along the Yaxis).
Pairs of paralogous genes are represented by open circles. In the case
of ATPase, the comparisons between Eubacteria and Archaebacteria/
Eukaryota are also indicated by open circles. The straight line repre-
sents the ideal case, for which at most one substitution occurred by
position.
519

Brown and Doolittle (1997) also argue that orthologous
proteins evolve at about the same rate in the three do-
mains according to the relative rate test. But an important
effect of mutational saturation is precisely an illusory
molecular clockwise behavior of the phylogenetic
marker, even if the evolutionary rate varies greatly be-
tween lineages (Philippe and Laurent 1998). To detect
differences in evolutionary rate, the most commonly
used method is indeed the relative rate test (Sarich and
Wilson 1973). It consists in using an outgroup (O) and
comparing the distance between it and two ingroup spe-
cies, A and B. If the distance d(O,A) is significantly
greater than d(O,B), then one can infer that species A has
evolved faster than species B. On the other hand, if
d(O,A) is equal to d(O,B), one can assume the constancy
of the evolutionary rate. But mutational saturation pro-
duces exactly the result that the distance values reach a
plateau (Figs. 6 and 7), irrespective of the real number of
substitutions. In our case, saturation would be enough to
make d(O,A) equal to d(O,B) even if species A does not
evolve at the same rate as species B. The high level of
saturation indicated that a relative rate test was inappro-
priate to detect any difference of evolutionary rates in the
six paralogous genes.
Because of the mutational saturation, it was thus
highly probable that the substitutions that occurred in the
deep branches of the tree were completely masked by the
innumerable substitutions that occurred later. The phy-
logenetic signal for ancient events could thus have been
completely lost, suggesting a priori that all the phylog-
enies used to root the tree of life were prone to tree
reconstruction artifact.
Eubacterial Rooting as the Result of Long
Branch Attraction
Let us assume that, for a given gene, Eubacteria evolved
faster than Eukaryota and Archaebacteria. When this
gene is rooted through the addition of a paralogous gene,
the phylogeny will contain two long branches (Eubacte-
ria and outgroup) and two short branches (Eukaryota and
Archaebacteria) as shown in Fig. 8. Such a topology is
very similar to the one that has been used by Felsenstein
(1978) to demonstrate the LBA artifact. It is likely that
the two long branches will be grouped together because
of this artifact, locating the root of the tree of life in the
eubacterial branch.
In Fig. 9, the unrooted topologies of the ATPase, Tyr-
tRS, SRP54, EF-1␣, Ile-tRS, and 16S rRNA were dis-
played. The branch lengths were equal to the average
distances from extant species to the trifurcating point
estimated on an ML tree. A salient feature in this figure
is that the branch lengths of the three domains were quite
different according to the gene studied. For example, for
the ATPase, the eubacterial branch was 6 times longer
than the archaebacterial and eukaryotic ones, and for the
rRNA, the eukaryotic branch was 1.4 times longer than
the eubacterial one and 1.9 times longer than the archae-
bacterial one. This meant that the evolutionary rate var-
Fig. 7. Mutational saturation curves as in figure 5, except that the
inferred number of substitutions between the same two sequences was
determined using the maximum-likelihood method. The numbers of
substitutions are represented as the frequency.
Fig. 8. The long branch attraction artifact and the rooting of the tree
of life. Left: A hypothetical unrooted tree linking the three domains,
for which the branch of Eubacteria (B) is much longer than those of
Archaebacteria (A) and Eukaryota (E). The outgroup (O) represents a
paralogous gene which obviously also has a very long branch. The
resulting topology is very similar to the model used by Felsenstein
(1978) to demonstrate the long branch attraction phenomenon. The root
of the tree of life thus could artifactually be located in the longest
branch.
520

ied greatly between genes and between domains and that
probably none of these six genes was a good molecular
clock. As discussed above, their apparently clock-like
behavior is spurious and is due to a high level of muta-
tional saturation. More interestingly, when these un-
rooted protein trees were rooted through the use of a
paralogous gene, the root always fell within the longest
branch, producing a result similar to a midpoint rooting
approach. This strongly suggested that this rooting was
artifactual, due to the LBA phenomenon. Further evi-
dence for this hypothesis is provided by the ML analysis
performed on a limited sample of species. The difference
in likelihood between the eubacterial rooting and the
eukaryotic rooting (i.e., the monophyly of prokaryotes)
was proportional to the length of the eubacterial branch:
about 15 SE for the ATPase, about 3 SE for the EF-1␣,
and only 1.3 SE for the Ile-RS. The scenario in Fig. 8
explaining the eubacterial rooting by the LBA artifact
was thus probable.
This hypothesis also suggested an explanation for the
following paradox. The six paralogous genes used in this
study were highly saturated. This explained why the in-
tradomain phylogeny, even their monophyly, was not
correctly recovered and was supported by low bootstrap
values (see Figs. 1–5). But it raised a serious question:
Why were the deeper nodes such as the monophyly of
the duplicated genes (in the majority of the cases) and the
relationships between domains recovered and generally
supported by high bootstrap values? The saturation
should have erased all the signal for the relationship
between the three domains, since the more ancient the
phylogenetic signal is, the more the saturation should
obscure it. If the tree reconstruction method inferred
some groupings statistically supported by high bootstrap
values despite the absence of phylogenetic signal, this
must reflect some inconsistency of the reconstruction
method, such as LBA. The best answer to the previous
paradox, i.e., resolution of deep nodes despite the high
level of mutational saturation, was therefore that the ro-
bust resolution of the rooting of the tree of life observed
for these six genes was due to the LBA artefact. How-
ever, the explanation is probably more complex, because
of the variation of invariant sites discussed in the com-
panion paper (Lopez et al. 1999).
Exploration of a Possible Eukaryotic Rooting
We have shown in this paper that the rooting of the tree
of life in the eubacterial branch has been based on un-
reliable phylogenies. To locate correctly the root of the
universal tree, we must take up the challenge of inferring
good phylogenies from highly saturated data. One way is
to use the ML method with a very adequate model of
sequence evolution (Sullivan and Swofford 1997). How-
ever, as discussed in the companion paper (Lopez et al.
1999) and in many others (Cao et al. 1998; Goldman et
al. 1998; Halpern and Bruno 1998; Lockhart et al. 1998;
Naylor and Brown 1998; Voelker and Edwards 1998;
Yang et al. 1998), it is clear that current models poorly fit
the data and thus we have no guarantee of finding the
true phylogeny. Another way is to study the slowly
evolving positions, which are much less saturated. We
have applied this approach to the two genes that are
apparently not affected by horizontal gene transfer or a
paralogy problem, the elongation factors (Lopez et al.
1999), and the SRP (Brinkmann and Philippe 1999) and
found that the eukaryotic rooting is the best-supported
hypothesis.
This eukaryotic rooting would best explain the pres-
ence of many more eubacterial-like genes than eukary-
otic-like ones in completely sequenced Archaebacterial
genomes (Koonin et al. 1997), which cannot be easily
explained in the frame of the current scenario. This will
also best explain the presence of many unique processes
in eukaryotes that involve the participation of structural
RNAs or ribozymes reminiscent of the RNA world (Jef-
fares et al. 1998; Poole et al. 1998).
The existence of many eukaryotic features in the ma-
jor archaebacterial cellular information processing sys-
tems (replication, transcription, and translation) is gen-
erally explained by the sisterhood of Archaebacteria and
Eukaryota, implying that they are evolved states. But it
can be also explained by an acceleration of the rate of
evolution of these systems in the eubacterial lineage (and
also by differential gene loss), which is equivalent to
interpret them as primitive. The acceleration phenom-
enon can now be documented in the case of microspo-
ridia, in which proteins involved in the translation ma-
chinery (rRNA, EF-1␣, EF-2, and Ile-RS) all evolved at
an accelerated evolutionary rate, leading to the artifactual
early emergence of these peculiar fungi in the eukaryotic
tree, because of the LBA artifact (Germot et al. 1997;
Hirt et al. 1999).
A major difference between the Eubacteria and the
Fig. 9. Six unrooted trees linking the three domains. The branch
lengths of each domain are represented as the average of the branch
lengths on an ML tree between the trifurcation point and the species in
the domain. When a paralogous gene is added to the analysis, the root
is always located in the longest branch, as indicated by the ellipse. The
long branch attraction artifact explains this phenomenon well (see
Fig. 8).
521

Archaebacteria in the transcription and translation sys-
tems is the smaller number of proteins in the Eubacteria.
The differences in these systems are too complex to be
simply explained by loss or gain of genes, but there is a
clear trend toward simplicity in Eubacteria that could be
interpreted by (1) the loss of these proteins in the Eu-
bacteria or (2) their gain in the Archaebacteria. The sec-
ond hypothesis fits well with the idea that Archaebacteria
are en route toward the complex system of eukaryotes,
but the selective constraints in favor of this hypothesis
are very unclear, since apparently the archaebacterial
system is not more efficient than the eubacterial one. In
contrast, the first hypothesis is reasonable since it is ad-
vantageous to perform the same work with fewer pro-
teins.
The loss of some proteins in the eubacterial transla-
tion apparatus could have produced an acceleration of
the evolutionary rate of remaining proteins. Indeed, as
first noted by Dickerson (1971), the physical contacts
between a protein and several partners (proteins or
nucleic acids) induce constraints for its evolution. If one
or several of these contacts disappear, the corresponding
constraint is thus removed and the sequence will evolve
faster. This could well explain why some archaebacterial
and eukaryotic proteins involved in identical protein–
protein interactions with ribosomes, such as elongation
factors, have conserved more similarities between them
(ancestral characters) than each one with the eubacterial
proteins (see Brown and Doolittle 1997).
A strong selective pressure that could have favored
the loss of many proteins in Eubacteria is the coupling of
transcription and translation. For that, the mRNA must
be sufficiently accessible to the ribosome and its acces-
sory proteins and must not be masked by the transcrip-
tional apparatus. As a result, simply because of sterical
hindrance, the coupling of transcription and translation in
Eubacteria could have been favored by the loss of many
proteins. The coupling of transcription and translation is
generally assumed in Archaebacteria because of the ab-
sence of nucleus, but without any experimental evidence.
If this coupling indeed exists, the similarity between Ar-
chaebacteria and Eukaryota proteins of the translation
apparatus would be all the more striking because their
functions would be quite different. If it were not, it
would be a support for the proposed acceleration of the
evolutionary rate of the transcriptional and translational
proteins in Eubacteria and thus for an artefactual rooting
in this branch.
Another aspect of eubacterial evolution that could
have reduced the number of proteins involved in various
cellular processes and led to a general simplification of
their molecular mechanisms is nonorthologous gene dis-
placement (Forterre and Philippe 1999). Comparative ge-
nomics have shown that this phenomenon has occurred
frequently during the divergence of the three domains
and this could explain divergent rates of evolution such
as those responsible for the long bacterial branches of
many universal trees.
Acknowledgments. We would like to thank Henner Brinkmann, Jac-
queline Laurent, Philippe Lopez, David Moreira, and Miklos Mu¨ ller for
helpful comments and readings of the manuscript. We would also like
to thank Drs. J. Miller and S. Fitz-Gibbon, as well as TIGR for un-
published sequences and La Foundation des Treilles for giving a start-
ing point to our collaboration. We acknowledge Karine Budin for her
help in the elaboration of the figures. Genome sequencing was accom-
plished with support from DOE, MGRI, NIAID, and NIDR.
References
Adachi J, Hasegawa M (1996) MOLPHY version 2.3: Programs for
molecular phylogenetics based on maximum likelihood. Comput
Sci Monogr 28:1–150
Baldauf SL, Palmer JD, Doolittle WF (1996) The root of the universal
tree and the origin of eukaryotes based on elongation factor phy-
logeny. Proc Natl Acad Sci USA 93:7749–7754
Brinkmann H, Philippe H (1999) Archaea sister-group of Bacteria?
Indications from tree reconstruction artifacts in ancient phylog-
enies. Mol Biol Evol 16:817–825
Brown JR, Doolittle WF (1995) Root of the universal tree of life based
on ancient aminoacyl-tRNA synthetase gene duplications. Proc
Natl Acad Sci USA 92:2441–2445
Brown JR, Doolittle WF (1997) Archaea and the prokaryote-to-
eukaryote transition. Microbiol Mol Biol Rev 61:456–502
Brown JR, Robb FT, Weiss R, Doolittle WF (1997) Evidence for the
early divergence of tryptophanyl- and tyrosyl-tRNA synthetases. J
Mol Evol 45:9–16
Brown JR, Zhang J, Hodgson JE (1998) A bacterial antibiotic resis-
tance gene with eukaryotic origins. Curr Biol 8:R365–R367
Cao Y, Waddell PJ, Okada N, Hasegawa M (1998) The complete
mitochondrial DNA sequence of the shark Mustelus manazo:
Evaluating rooting contradictions to living bony vertebrates. Mol
Biol Evol 15:1637–1646
Dickerson RE (1971) The structures of cytochrome c and the rates of
molecular evolution. J Mol Evol 1:26–45
Doolittle WF (1998) You are what you eat: A gene transfer ratchet
could account for bacterial genes in eukaryotic nuclear genomes.
Trends Genet 14:307–311
Doolittle WF, Brown JR (1994) Tempo, mode, the progenote, and the
universal root. Proc Natl Acad Sci USA 91:6721–6728
Felsenstein J (1978) Cases in which parsimony or compatibility meth-
ods will be positively misleading. Syst Zool 27:401–410
Feng DF, Cho G, Doolittle RF (1997) Determining divergence times
with a protein clock: Update and reevaluation. Proc Natl Acad Sci
USA 94:13028–13033
Forterre P (1997) Protein versus rRNA: rooting the universal tree of
life? ASM News 63:89–92
Forterre P, Philippe H (1999) Where is the root of the universal tree of
life? BioEssays (in press)
Forterre P, Benachenhou-Lahfa N, Confalonieri F, Duguet M, Elie C,
Labedan B (1992) The nature of the last universal ancestor and the
root of the tree of life, still open questions. BioSystems 28:15–32
Germot A, Philippe H, Le Guyader H (1997) Evidence for loss of
mitochondria in Microsporidia from a mitochondrial-type HSP70 in
Nosema locustae. Mol Biochem Parasitol 87:159–168
Gogarten JP, Hilario E, Olendzenski L (1996) Gene duplications and
horizontal gene transfer during early evolution. In: Roberts DM,
Sharp P, Alderson G, Collins M (eds) Evolution of microbial life.
Cambridge University Press, Cambridge, pp 109–126
Gogarten JP, Kibak H, Dittrich P, et al. (1989) Evolution of the vacu-
522

olar H
+
-ATPase: Implications for the origin of eukaryotes. Proc
Natl Acad Sci USA 86:6661–6665
Goldman N, Thorne JL, Jones DT (1998) Assessing the impact of
secondary structure and solvent accessibility on protein evolution.
Genetics 149:445–458
Gribaldo S, Cammarano P (1998) The root of the universal tree of life
inferred from anciently duplicated genes encoding components of
the protein-targeting machinery. J Mol Evol 47:508–516
Gupta RS, Golding GB (1993) Evolution of HSP70 gene and its im-
plications regarding relationships between archaebacteria, eubacte-
ria, and eukaryotes. J Mol Evol 37:573–582
Halpern AL, Bruno WJ (1998) Evolutionary distances for protein-
coding sequences: Modeling site-specific residue frequencies. Mol
Biol Evol 15:910–917
Hasegawa M, Kishino H, Saitou N (1991) On the maximum likelihood
method in molecular phylogenetics. J Mol Evol 32:443–445
Hashimoto T, Sanchez LB, Shirakura T, Muller M, Hasegawa M
(1998) Secondary absence of mitochondria in Giardia lamblia and
Trichomonas vaginalis revealed by valyl-tRNA synthetase phylog-
eny. Proc Natl Acad Sci USA 95:6860–6865
Hennig W (1966) Phylogenetic systematics. University of Illinois
Press, Urbana
Hirt RP, Logsdon JM Jr, Healy B, Dorey MW, Doolittle WF, Embley
TM (1999) Microsporidia are related to fungi: Evidence from the
largest subunit of RNA polymerase II and other proteins. Proc Natl
Acad Sci USA 96:580–585
Iwabe N, Kuma K, Hasegawa M, Osawa S, Miyata T (1989) Evolu-
tionary relationship of archaebacteria, eubacteria, and eukaryotes
inferred from phylogenetic trees of duplicated genes. Proc Natl
Acad Sci USA 86:9355–9359
Jain R, Rivera MC, Lake JA (1999) Horizontal gene transfer among
genomes: The complexity hypothesis. Proc Natl Acad Sci USA
96:3801–3806
Jeffares DC, Poole AM, Penny D (1998) Relics from the RNA world.
J Mol Evol 46:18–36
Kimura M (1983) The neutral theory of molecular evolution. Cam-
bridge University Press, Cambridge
Koonin EV, Mushegian AR, Galperin MY, Walker DR (1997) Com-
parison of archaeal and bacterial genomes: Computer analysis of
protein sequences predicts novel functions and suggests a chimeric
origin for the archaea. Mol Microbiol 25:619–637
Lawrence JG, Ochman H (1998) Molecular archaeology of the Esch-
erichia coli genome. Proc Natl Acad Sci USA 95:9413–9417
Lawson FS, Charlebois RL, Dillon JA (1996) Phylogenetic analysis of
carbamoylphosphate synthetase genes: Complex evolutionary his-
tory includes an internal duplication within a gene which can root
the tree of life. Mol Biol Evol 13:970–977
Lockhart PJ, Steel MA, Barbrook AC, Huson D, Charleston MA, Howe
CJ (1998) A covariotide model explains apparent phylogenetic
structure of oxygenic photosynthetic lineages. Mol Biol Evol 15:
1183–1188
Lopez P, Forterre P, Philippe H (1999) The root of the tree of life in the
light of the covarion model. J Mol Evol (in press)
Martin W, Muller M (1998) The hydrogen hypothesis for the first
eukaryote. Nature 392:37–41
Moreira D, Lopez-Garcia P (1998) Symbiosis between methanogenic
archaea and delta-proteobacteria as the origin of eukaryotes: The
synthrophic hypothesis. J Mol Evol 47:517–530
Naylor GJP, Brown WM (1998) Amphioxus mitochondrial DNA, chor-
date phylogeny, and the limits of inference based on comparisons of
sequences. Syst Biol 47:61–76
Olsen GJ, Woese CR (1997) Archaeal genomics: An overview. Cell
89:991–994
Philippe H (1993) MUST, a computer package of Management Utilities
for Sequences and Trees. Nucleic Acids Res 21:5264–5272
Philippe H (1997) Rodent monophyly: Pitfalls of molecular phylog-
enies. J Mol Evol 45:712–715
Philippe H, Adoutte A (1998) The molecular phylogeny of Eukaryota:
Solid facts and uncertainties. In: Coombs G, Vickerman K, Sleigh
M, Warren A (eds) Evolutionary relationships among Protozoa.
Chapman & Hall, London, pp 25–56
Philippe H, Laurent J (1998) How good are deep phylogenetic trees?
Curr Opin Genet Dev 8:616–623
Philippe H, So¨rhannus U, Baroin A, Perasso R, Gasse F, Adoutte A
(1994) Comparison of molecular and paleontological data in dia-
toms suggests a major gap in the fossil record. J Evol Biol 7:247–
265
Poole AM, Jeffares DC, Penny D (1998) The path from the RNA
world. J Mol Evol 46:1–17
Ribas de Pouplana L, Frugier M, Quinn CL, Schimmel P (1996) Evi-
dence that two present-day components needed for the genetic code
appeared after nucleated cells separated from eubacteria. Proc Natl
Acad Sci USA 93:166–170
Rivera MC, Lake JA (1992) Evidence that eukaryotes and eocyte pro-
karyotes are immediate relatives. Science 257:74–76
Saitou N, Nei M (1987) The neighbor-joining method: A new method
for reconstructing phylogenetic trees. Mol Biol Evol 4:406–425
Sarich VM, Wilson AC (1973) Generation time and genomic evolution
in primates. Science 179:1144–1147
Snel B, Bork P, Huynen MA (1999) Genome phylogeny based on gene
content. Nature Genet 21:108–110
Sogin ML (1991) Early evolution and the origin of eukaryotes. Curr
Opin Genet Dev 1:457–463
Stetter K (1992) Life at the upper temperature border. In: Traˆn Thanh
VanJ,Traˆn Thanh Van K, Mounolou J, Schneider J, McKay C
(eds) Frontiers of life. Editions Frontie`res, Gif-sur-Yvette, France
Sullivan J, Swofford DL (1997) Are guinea pigs rodents? The impor-
tance of adequate models in molecular phylogenetics. J Mammal
Evol 4:77–86
Swofford DL (1993) PAUP: Phylogenetic analysis using parsimony,
version 3.1.1. Illinois Natural History Survey, Champaign
Voelker G, Edwards SV (1998) Can weighting improve bushy trees?
Models of cytochrome b evolution and the molecular systematics of
pipits and wagtails (Aves: Motacillidae). Syst Biol 47:589–603
Woese CR (1987) Bacterial evolution. Microbiol Rev 51:221–271
Woese CR, Kandler O, Wheelis ML (1990) Towards a natural system
of organisms: Proposal for the domains Archaea, Bacteria, and
Eucarya. Proc Natl Acad Sci USA 87:4576–4579
Yang Z, Nielsen R, Hasegawa M (1998) Models of amino acid sub-
stitution and applications to mitochondrial protein evolution. Mol
Biol Evol 15:1600–1611
Zillig W (1987) Eukaryotic traits in Archaebacteria. Could the eukary-
otic cytoplasm have arisen from archaebacterial origin? Ann NY
Acad Sci 503:78–82
523