Telling the Evolutionary Time: Molecular Clocks and the Fossil Record

Methodology appropriate for test cases

From the previous discussions it is apparent that there are still many problems in
estimating divergence times. In terms of methodology currently available the following
paragraphs summarize the combination of methods that may lead to the most realistic results
—which are not necessarily those with the least range of uncertainty. It is inevitable that
the levels of uncertainty will increase when divergences as old as the Cambrian
‘explosion’ are the subject of investigation.

Calibration

Fossil dates should be used to provide lower and upper boundaries for divergence times
rather than as single calibration points (Sanderson 1997; Cutler 2000).

Number and type of loci

A range of single copy nuclear genes should be used in preference to 18S or mitochondrial
genes when estimating deep divergences (as in Nei et al. 2001). Additionally, sequences
should be lengthy. While being unsaturated at the level of divergence under scrutiny, an
ideal sequence would contain as many informative sites as possible (at least 2 kb worth) to
improve the power of relative rates tests (Bromham et al. 2000) and the resolution of the
tree (Philippe et al. 1994).

Branch lengths and phylogeny estimation

ML methods are preferable to pairwise distance methods (Sanderson 1997; Rambaut and
Bromham 1998; Cutler 2000).

Model choice

Models that best fit the dataset (using LRT) should be used. These can be selected using
model selection programs such as Posada and Crandall (2001). Tests for data skew should
be performed to determine whether to apply appropriate corrections. If using a
divergence estimation method that does not allow rates to vary within the tree, genes
should be scrutinized using saturation plots, relative rates tests, and LRT.

Phylogenetic uncertainty

In case of phylogenetic uncertainty, divergence estimates and their confidence intervals
should be calculated for not only the ‘best’ tree but for all those not significantly worse-
fitting in order to get a total interval estimate for all likely phylogenies (as in Cutler
2000).

PHYLOGENETIC FUSES AND EVOLUTIONARY ‘EXPLOSIONS’ 61