Interest in DNA hybridization data has led to the development of statistical methods for critically evaluating the strength of phylogenetic inferences made from distances (Lanyon, 1985; Templeton, 1985, 1986; Fitch, 1986; Felsenstein, 1987). A recent addition to this set of statistical approaches (Krajewski and Dickerman, 1990) demonstrated the use of bootstrap resampling of replicate distance measures in assessing distance-matrix robustness using, as an example, the uncorrected distance measures presented by Sibley et al. (1990). This analysis was based on the simplistic assumption that variation within a cell of the distance matrix reflected random measurement error alone, with no systematic effects due to experimental design. Here the raw hominoid Tm's (not altered by the various corrections imposed by Sibley et al. [1990]) are reviewed by treating different experimental runs as covariates, resulting in a less biased portrayal of mean distances and a better estimate of random measurement error. Tree fitting (both with and without contemporaneous tips) performed on multiple bootstrap pseudoreplicate T,Tm matrices is used to assess stability in tree topology. Felsenstein (1987), in dissecting variance components in the DNA hybridization distances among hominoids (Sibley and Ahlquist, 1987), found that the only identifiable source of experimental error was that attributable to variation among experimental runs of the thermal chromatography machine. Such among-run variation encompasses both the vagaries of operational parameters of the thermal chromatography apparatus (e.g., temperature overshoots) as well as differences in tracer preparations such as the tracer-length variation treated by Caccone and Powell (1989). The thermal stability measures of Sibley et al. (1990:218-220) differed greatly among experiments in melting temperature of the homologous hybrid as well as in respective AXTm values, corresponding to differences in both intercept and slope of the bivariate comparisons illustrated by Blackstone and Sheldon (1991). For example, experiment 828 had much lower AXTm values than did experiment 864 for the same species comparisons. Pooling replicate distances drawn from disparate experimental runs could result in obscuring relative distances between driver taxa and the tracer species. In an extreme hypothetical case, species A may be represented solely in runs with depressed ATm's, whereas species B may be represented solely in runs with exaggerated ATm's. As an alternative to pooling essentially incomparable data, it should be possible to estimate deviations of whole runs from the average and to correct for these systematic errors before pooling data from different runs for use in phylogenetic reconstruction. A simple method for doing so is applied to the uncorrected Tm's of Sibley et al. (1990:218-220) with the addition of Tm's with Pongo tracer (experiments 679 and 822) calculated by Krajewski and Dickerman (1990) from raw data supplied by Charles Sibley (pers. comm.). Other aspects of the experimental design of DNA hybridization studies can and should be treated, as with multiway analysis of variance of the sort demonstrated by Felsenstein (1987). Here, I discuss only among-run variation, the single experimental factor explicitly identified in the presentation of replicate measures by Sibley et al. (1990). However, Sibley et al. (1990:217) discounted five experiments
Read full abstract