(6 × 107)/[(3 nucleotides/covarion) × 28 covarions] = 7 × 10-9 replacement substitutions per variable nucleotide per year. This may be compared to the neutral rate, which, for cow and goat β hemoglobin pseudogenes (Li and Graur, 1991; p. 72) is 3.6 × 10-9 substitutions per nucleotide per year. Since only 3/4 of those substitutions lead to replacements, the neutral replacement substitution rate is 2.7 × 10-9. As the fitted replacement rate is 2.6 times the pseudogene replacement rate, it suggests that there may be positive selection occurring at the SOD locus.
In conclusion, the following observations may be derived from our analysis. First, a molecular clock (i.e., a particular gene or protein) may appear to be very unreliable, yet be fairly accurate. The apparent distortion may emerge because relevant components of the clock (such as covarions, persistence, and the number of alternatives allowed per site) have not been taken into account. As we have shown, the apparently erratic SOD becomes a fairly accurate clock when the appropriate components are taken into account.
The flip side of the observation just made is that inferences of divergence time between lineages derived from a particular clock cannot be assumed to be correct unless the relevant components of the clock have been ascertained. The apparent rate of SOD divergence observed among mammals or flies would yield grossly erroneous time estimates when simply extrapolated to the differences observed between the vertebrate classes (fish versus tetrapod or amphibian versus mammal) or between metazoans and fungi. The SOD ''clock" is a complex of several component parts subject to different constraints and that interact with each other.
Finally, divergence times inferred from a particular molecular clock are subject to the possibility of variations caused by natural selection or other extraneous factors. Thus, for example, some chloroplast genes evolve at rates that are similar in several grass lineages [such as maize and rice (Gaut et al., 1993)] but different from those observed in the tobacco lineage (Kwiatowski et al., 1991). Another example is the acceleration of the rate of lysozyme c evolution in ruminant lineages as lysozyme was recruited for a distinctive stomach function (Stewart and Wilson, 1987; Jolles et al., 1990). A conspicuous anomaly in the SOD data is that there are fewer amino acid differences between fish and tetrapods than between amphibians and mammals. Whether or not the differences are statistically significant, they would seem to support the wrong branching order among the corresponding lineages. The conclusion derived from this anomaly, as well as from the chloroplast, lysozyme, and other examples of uneven rates, is the obvious one that inferences based on a particular clock must be taken with caution, but they become