Reliability of statistical analyses for estimating relative specificity in quantitative resistance in a model host-pathogen system.

The reliability of analyses of variance for evaluating host cultivar x pathogen isolate specificity in resistance controlled by polygenes with additive effects was tested with combinations of hypothetical host and pathogen genotypes in a model system. In each test, varying numbers of host and pathogen genotypes were combined in all combinations, the resulting disease severities were calculated according to the model, and those data were subjected to analysis of variance. The percentage of total variance accounted for by host x pathogen interaction decreased with increasing numbers of host and pathogen genotypes per test. Simulated selection for virulence among randomly generated pathogen genotypes increased the percentage of variance attributable to host x pathogen genotype interaction, but simulated selection for resistance among host genotypes decreased it. The percentage of variance accounted for by interaction was greatest when selection of resistant host genotypes was followed by selection of the most virulent pathogen genotype on each selected host genotype. When gene frequencies were varied in the model, the interaction variance was greatest at low frequencies of resistance genes and high frequencies of virulence genes, but the number of matches between genes for specific virulence and specific resistance was greatest for high frequencies of both resistance and virulence genes. A simplified method of analysis was developed to estimate the amount of specific resistance in a set of host genotypes inoculated in all combinations with a set of pathogen genotypes. This method, based on the variance of disease severity adjusted to remove general virulence, proved consistently accurate with varying numbers of genotypes in the set, varying numbers of loci for resistance and virulence, and varying frequencies of genes for resistance and virulence. The variance method is of comparable accuracy and is much simpler than the previously proposed methods based on regression analysis. Simulated selection for resistance in the host and for virulence in the pathogen population increased the accuracy of both the variance method and the regression method.