Analysis of the biases in the estimation of deleterious mutation parameters from natural populations at mutation-selection balance.

Indirect estimates of the genomic rate of deleterious mutations (lambda), their average homozygous effect (s) and their degree of dominance (h) can be obtained from genetic parameters of natural populations, assuming that the frequencies of the loci controlling a given fitness trait are at mutation-selection equilibrium. In 1996, H.-W. Deng and M. Lynch developed a general methodology for obtaining these estimates from inbreeding/outbreeding experiments. The prediction of the sign and magnitude of the biases incurred by these estimators is essential for a correct interpretation of the empirical results. However, the assessment of these biases has been tested so far under a rather limited model of the distribution of dominance effects. In this paper, the application of this method to outbred populations is evaluated, focusing on the level of variation in h values (sigma(h)(2) and the magnitude of the negative correlation (rs,h) between s and h. It is shown that the method produces upwardly biased estimates of lambda and downwardly biased estimates of the average s in the reference situation where rs,h=0, particularly for large values of sigma(h)(2), and biases of different sign depending on the magnitude of the correlation. A modification of the method, substituting the estimates of the average h for alternative ones, allows estimates to be obtained with little or no bias for the case of rs,h=0, but is otherwise biased. Information on rs,h and sigma(h)(2), gathered from mutation-accumulation experiments, suggests that sigma(h)(2) may be rather large and rs,h is usually negative but not higher than about -0.2, although the data are scarce and noisy, and should be used with caution.