Relationships between major groupsMore disturbing, perhaps, is the lack of consensus on phylogenetic relationships using
either morphological or molecular data, between some of the major groups involved in
the Cambrian ‘explosion’ (Figure 3.2). Considering arthropods, for example, despite
previous enormous databases and total evidence analyses (e.g. Wheeler 1998), a recent
edition of Nature featured two papers back to back (Giribet et al. 2001; Hwang et al.
2001) which claimed myriapods as a sister group of ‘pancrustacea’ and chelicerates,
respectively. This kind of phylogenetic uncertainty directly affects the estimates of upper
and lower bounds on divergence times, as differing opinions on the affinities of fossil taxa
can lead to very different divergence times (Marshall 1990). Similar debates exist as to the
sister group of Brachiopoda, or relationships within the Mollusca. At a deeper level, the
concept of a clade Ecdysozoa (Aguinaldo et al. 1997) to embrace all moulting animals has
been disputed by others (Erwin 2001).
Unresolved nodes between major taxa involved in the Cambrian ‘explosion’ have been
used in support of the evolutionary ‘explosion’ hypothesis (Field et al. 1988). Philippe et al.
(1994) reported a method to estimate the number of nucleotides that would be required
to resolve uncertain nodes confidently. They showed that an ‘experimentally unfeasible’
number of nucleotides is required to resolve many of these uncertain nodes. Although this
method provides an indication of the amount of resolution we can obtain from a gene, its
utility is limited by inconsistent substitution rates between genes and lineages. This
analysis was based on 18S rRNA sequences. The lack of tree resolution is likely to be at
least partially due to both very short and very long internal branches being present, which
can cause long branch attraction (LBA) (Philippe and Germot 2000). LBA can cause the
false placement of distantly related taxa together in a phylogeny (Philippe and Adoutte
1998; Philippe and Germot 2000) and may thus obscure or ‘blur’ phylogenetic signal.
Causes of LBA include differing G+C contents, rate heterogeneity and increased
proportions of variable positions and other biases within ingroup taxa in comparison with
those taxa in crown-groups (Philippe and Germot 2000). LBA can apply to both
nucleotide and amino acid data. It is important to eliminate LBA as much as possible when
tree building, by testing for nucleotide and amino acid skew (Foster and Hickey 1999;
Penny et al. 1999). Where inequality exists, corrections can sometimes be applied such as
RY coding (Phillips et al. 2001) and non-stationary modelling (Galtier and Gouy 1998).
Furthermore, the addition of sequence data from a greater number of species breaks up
long branches, thus reducing LBA (Hillis et al. 1996).
Molecular estimates of divergence times can be confounded by phylogenetic
uncertainty. The placement of both nematode and molluscan clades has not been clearly
resolved despite analysis involving many datasets (Aguinaldo et al. 1997; Adoutte et al.
2000) (see Figures 3.2, 3.5). In cases where the true phylogeny is unresolved, estimates
of divergence time and their confidence intervals should be calculated not only for the
consensus tree (Cutler 2000) but also for those trees not significantly worse fitting. This
allows the calculation of a total interval estimate for all likely phylogenies rather than only
the constrained tree which is assumed to be the correct one.
As new sequence data become available, particularly from slowly evolving genes, it is
expected that the resolution of the metazoan phylogeny will improve. Although data from
protein-coding nuclear genes are likely to provide an improvement over mitochondrial
48 RICHARD A.FORTEY ET AL.