It is frequently true that molecular sequences do not evolve in a strictly
clocklike manner. Instead, substitution rate may vary for a number of reaso
ns, including changes in selection pressure and effective population size,
as well as changes in mean generation time. Here we present two new methods
for estimating stepwise changes in substitution rates when serially sample
d molecular sequences are available. These methods are based on multiple ra
tes with dated tips (MRDT) models and allow different rates to be estimated
for different intervals of time. These intervals may correspond to the sam
pling intervals or to a priori-defined intervals that are not coincident wi
th the times the serial samples are obtained. Two methods for obtaining est
imates of multiple rates are described. The first is an extension of the ph
ylogeny-based maximum-likelihood estimation procedure introduced by Rambaut
. The second is a new parameterization of the pairwise distance least-squar
es procedure used by Drummond and Rodrigo. The utility of these methods is
demonstrated on a genealogy of HIV sequences obtained at five different sam
pling times from a single patient over a period of 34 months.