is that the majority of fossil-based divergence times are underestimates. If supposedly
‘accurate’ calibration points are used instead (those points very close to the true
divergence date), it is argued that this will provide a more realistic estimate than
calibration based on an average of many calibration points (Wang et al. 1998). In most
studies a fossil calibration point of 310 Ma, an estimate of the mammal-bird split, is used
(Hedges et al. 1996; Kumar 1996; Gu 1998; Wang et al. 1998; Nei et al. 2001). Despite
such widespread use, Lee (1999) argued that 310 Ma is an overestimate and stated that the
first tetrapods to be confidently assigned to either the mammal or bird lineage are
approximately 288 million years old. Using this date reduces inferred dates calibrated by
the mammal-bird divergence to 93 per cent of their reported value (Lee 1999). Yoder and
Yang (2000) obtained conflicting divergence estimates within the mammalian clade using
different calibration points. Strikingly, Lee (1999) pointed out that the second calibration
point of the primate-rodent divergence of 100 Ma, used by Hedges et al. (1996) and Gu
(1998) is a molecular clock estimate which is also based on the 310 Ma mammal-bird
divergence estimate. This is clearly not a second, independent, calibration point.
It should be noted that while these errors would all tend to cause underestimation of
divergence, the opposite is often seen to occur in metazoan divergence analysis, with
molecular dates pushing back fossil-based dates. It has been proposed that the smaller
body sizes and faster generation times of many Cambrian taxa may correlate with
relatively fast rates of evolution (Martin and Palumbi 1993), compared with more recent
taxa. When calibration points taken from recent taxa with larger body sizes and/or
generation times (e.g. the synapsid-diapsid split) are used to estimate divergence times,
they will cause an overestimation of the true date of divergence.
Rather than using fossils to ‘assign’ dates to nodes within a phylogeny, a number of
authors (Sanderson 1997; Rambaut and Bromham 1998; Cutler 2000) have allowed fossils
to provide constraints on divergence estimates, using a ML-based approach. This is a more
realistic approach as we can never be certain when two lineages diverged. Furthermore,
the addition of fossils can only improve estimates when using this method.
If possible, calibration point estimates should not be applied to distantly related reference
taxa. This is because rates of molecular evolution are well known to differ markedly
between groups (Hillis et al. 1996) and relative rate tests cannot detect changes shared
across long branches. However, calibration points can only be as good as the fossil record
from which they are drawn and are often unavailable or unreliable for the group in
question (especially when small and soft bodied). Accurately dated, well-characterized
fossils from just above and below the lineage divergence to be dated are ideal, although
admittedly also rare (e.g. Doolittle et al. 1996; Cooper and Penny 1997; Sanderson 1997;
Bromham et al. 1998). If applied well, the use of multiple calibration points can provide
good constrained estimates of absolute rates of molecular evolution (Marshall 1990).
Another problem with using multiple calibration points is the availability of calibrating
sequences for less frequently studied taxa. Often, the only sequences available for many
genes are those of human, mouse, and chicken (Wang et al. 1998) although it is likely that
sequence data for a wider range of taxa will become rapidly available in the coming years.
PHYLOGENETIC FUSES AND EVOLUTIONARY ‘EXPLOSIONS’ 51