266
L. Bargelloni and G. Lecointre
such an artefact: an asymmetric part (bovichtids and nototheniids) and a
symmetric part, with crown groups (artedidraconids, channichthyids). In
molecular systematics, outgroups often have longer branches than
members of the ingroup. When one or several member(s) of the ingroup
has a higher rate of molecular evolution, the lineages implied are
successively and artefactually branched at the base of the tree, producing a
"Hennig's comb shape" of the branching pattern, obscuring possible
monophylies for some groups. Longer branches are not directly detectable
from the first glance at the tree because the taxa implied are basal. This is
the reason why some authors [20,21] consider the asymetric part of a
molecular phylogenetic tree as not reliable, even when the basal position
of a lineage is supported with high bootstrap proportions. Indeed, long
branch attraction artefact is a pitfall that bootstrapping cannot reveal
[22,23]. In the case of the notothenioids, the question is far from being
meaningless: the two most basal families, i.e. the bovichtids and the
nototheniids, all appear as paraphyletic. Are these paraphylies artefactual?
There are two means to refute the possible nonreliability of basal
paraphylies. First, the use of several genes to confirm the same basal
paraphylies is the most powerful tool to detect such an artefact, because
two genes different in nature (for instance one mitochondrial and one
nuclear) have poor chance to concomitantly accelerate their mutational
rate in the same lineages. The second mean is the intrinsic quality of each
data set. In absence of mutational saturation, in absence of detectable
acceleration as shown with a relative rate test, and in absence of excessive
amounts of homoplasy [high retention index (RI)], there is little reason to
suspect such a long-branch attraction artefact.
The long-branch attraction artefact as initially described by Felsenstein
[18] needs some random noise in the data to occur. For a few taxa (for
instance in four or five-species trees), some homoplasy will very easily
provoke the artefact, i.e. grouping the species with the closest mutational
rates together with high bootstrap proportions [22,23]. The artefact will be
less likely to happen when the number of taxa increases, because in
principle increasing the number of taxa breaks long branches. In other
words, when the homoplasy content of a data set is low and when the
number of species is high (usually more than 15), the artefact will be
unlikely.
These elements led us to conclude that the paraphylies of most of the
notothenioid families revealed by the recent molecular studies are not
artefactual. First of all, both the paraphyly of the Bovichtidae (with
Pseudaphritis as the sister group of the rest of the nonbovichtid
notothenioids) and the paraphyly of the Nototheniidae are supported both
by the nuclear 28S rDNA [10] and the mitochondrial 12S-16S rDNA
L. Bargelloni and G. Lecointre
such an artefact: an asymmetric part (bovichtids and nototheniids) and a
symmetric part, with crown groups (artedidraconids, channichthyids). In
molecular systematics, outgroups often have longer branches than
members of the ingroup. When one or several member(s) of the ingroup
has a higher rate of molecular evolution, the lineages implied are
successively and artefactually branched at the base of the tree, producing a
"Hennig's comb shape" of the branching pattern, obscuring possible
monophylies for some groups. Longer branches are not directly detectable
from the first glance at the tree because the taxa implied are basal. This is
the reason why some authors [20,21] consider the asymetric part of a
molecular phylogenetic tree as not reliable, even when the basal position
of a lineage is supported with high bootstrap proportions. Indeed, long
branch attraction artefact is a pitfall that bootstrapping cannot reveal
[22,23]. In the case of the notothenioids, the question is far from being
meaningless: the two most basal families, i.e. the bovichtids and the
nototheniids, all appear as paraphyletic. Are these paraphylies artefactual?
There are two means to refute the possible nonreliability of basal
paraphylies. First, the use of several genes to confirm the same basal
paraphylies is the most powerful tool to detect such an artefact, because
two genes different in nature (for instance one mitochondrial and one
nuclear) have poor chance to concomitantly accelerate their mutational
rate in the same lineages. The second mean is the intrinsic quality of each
data set. In absence of mutational saturation, in absence of detectable
acceleration as shown with a relative rate test, and in absence of excessive
amounts of homoplasy [high retention index (RI)], there is little reason to
suspect such a long-branch attraction artefact.
The long-branch attraction artefact as initially described by Felsenstein
[18] needs some random noise in the data to occur. For a few taxa (for
instance in four or five-species trees), some homoplasy will very easily
provoke the artefact, i.e. grouping the species with the closest mutational
rates together with high bootstrap proportions [22,23]. The artefact will be
less likely to happen when the number of taxa increases, because in
principle increasing the number of taxa breaks long branches. In other
words, when the homoplasy content of a data set is low and when the
number of species is high (usually more than 15), the artefact will be
unlikely.
These elements led us to conclude that the paraphylies of most of the
notothenioid families revealed by the recent molecular studies are not
artefactual. First of all, both the paraphyly of the Bovichtidae (with
Pseudaphritis as the sister group of the rest of the nonbovichtid
notothenioids) and the paraphyly of the Nototheniidae are supported both
by the nuclear 28S rDNA [10] and the mitochondrial 12S-16S rDNA