Nowadays the tests are applied to molecular data as demonstrated by Takezaki et al.
(1995). Multiple nested relative rate tests are performed on different triplets within the
phylogenetic tree. This information is then combined in order to remove lineages which
have evolved at significantly different rates from the majority consensus of the group, defined
by the mean total branch length from root to tip over all lineages.
Trees built in this way are referred to as ‘linearized’ trees because they make the remaining
divergence times linear with distance (Takezaki et al. 1995; Hedges et al. 1996). However,
linearized tree methods do not give reliable divergence estimates when evolutionary rate
varies markedly between species groups, for example between orders or classes of
vertebrates (e.g. Cao et al. 1998; Nei et al. 2001). In the Cambrian-Precambrian case the
influence of such factors is particularly hard to assess.
Bromham et al. (2000) showed that relative rate tests have very limited power when used
on short sequences (400 or fewer sites free to vary) and are unlikely to detect moderate
levels of rate variation (where the rate of one lineage is 1.5–4 times the other). They found
that sequences of at least 1000 sites are required to detect three-fold rate differences and
above, and much longer sequences (>2000 sites) are required to detect two-fold rate
differences. The inability of relative rate tests to detect moderate to high levels of rate
variation on short sequences can provide misleading results in studies where relative rate
tests are performed on individual genes.
Datasets will also pass relative rate tests when rates of evolution have accelerated or
decelerated and affected all lineages equally (Marshall 1990) and, thus, caution must be
taken when using the test. This has particular resonance when studying radiations —where
the counter-argument to deep divergences is that some kind of universal rate acceleration
may have occurred. This has been claimed in ‘Snowball Earth’ (Hoffman et al. 1998) and
‘True Polar Wander’ (Kirschvink et al. 1997) theories. Both claim that severe stress led to
large amounts of genetic change in a short period of time.
Models of sequence substitutionAs the number of substitutions between two sequences increases, they become progressively
more saturated, as most of the sites changing have already changed previously. Surprisingly,
very few studies investigate the saturation levels of the sequences they use. Saturation brings
internal tree nodes artefactually closer to one another and can exacerbate long branch
attraction (Philippe and Adoutte 1994; Philippe et al. 1994).
This problem obviously needs to be addressed when investigating the divergence of
lineages as far back as the Cambrian. The saturation of sequences can be tested by plotting
the inferred substitutions between all species pairs, against the actual number of differences
recorded between the species pairs (Philippe et al. 1994). When the number of actual
counted differences levels off, while the inferred distance continues to rise, saturation has
occurred. In order to minimize the effects of saturation, there are a number of ‘distance
correction’ methods available. These convert observed distances into measures of actual
evolutionary distance, thus estimating the amount of evolutionary change that has been
‘overprinted’.
PHYLOGENETIC FUSES AND EVOLUTIONARY ‘EXPLOSIONS’ 55