CONFIDENCE IN EVOLUTIONARY TREES FROM BIOLOGICAL SEQUENCE DATA

THE reliable construction of evolutionary trees from nucleotide sequen ces often depends on randomization tests such as the boot-strap1 and P TP (cladistic permutation tail probability) tests2-6. The genomes of b acteria7, viruses8, animals7,9,10 and plants11, however, vary widely i n their nucleotide frequencies. Where genomes have independently acqui red similar G+C base compositions, signals in the data arise that caus e methods of evolutionary tree reconstruction to estimate the wrong tr ee by grouping together sequences with similar G+C content12-14. Under these conditions randomization tests can lead to both the rejection o f the correct evolutionary hypothesis and acceptance of an incorrect h ypothesis (such as with the contradictory inferences from the photosyn thetic rbcS and rbcL sequences14). We have proposed one approach to te sting for the G+C content problem15. Here we present a formalization o f this method, a frequency-dependent significance test, which has gene ral application.