Optimizing detection of QTLs retarding aging: choice of statistical model and animal requirements.

Document Type


Publication Date



Animals, Computer-Simulation, Genotype, Life-Expectancy, Mice, Models-Genetic, Models-Statistical, Phenotype, Quantitative-Trait-Heritable

JAX Source

Mech Ageing Dev 2002 Jan; 123(2-3):131-44.


Quantitative trait locus (QTL) analysis makes no assumptions about the identity of genes involved in regulating aging. Moreover, it may be used as the first step in identifying such genes and, thus QTL analysis may be instrumental in formulating new hypotheses about aging. Genetic experiments, however, require hundreds to thousands of animals and are very expensive in mammals. Statistical power to detect longevity genes could be improved by excluding accidental, unrelated to aging mortality. While many early deaths are probably accidental, excluding early mortality altogether eliminates the age-related component, too. We used computer simulations to assess the effect of excluding early age-related, mortality on the statistical power of several common tests, such as t-test, Mann-Whitney and chi(2). Surprisingly, even the age-related, Gompertz component of early mortality reduces the statistical power of the t- and Mann-Whitney tests. For example, in a backcross design, to detect a gene slowing down the rate of aging and increasing mouse life span by 10% (P=0.0001; power=0.8), a regular t-test will require 640 mice, all kept for the entire life span and genotyped. If life spans of only 25% of the longest-lived animals from each of the two groups, carrying a putative longevity allele and not carrying it, are compared, population size can be reduced by two-fold, to about 300, and genotyping by seven-fold, to 90. Confirming simulation results, the significance of the effect of caloric restriction on life span increased from P=3.4x10(-5) to 1.1x10(-7), when life spans of only 40% of the longest-lived mice from each of the two groups, ad libitum fed and calorie restricted, were compared. Finally, finding the optimal combination of statistical test, the number of phenotyped and the number of genotyped animals, which would minimize experimental costs was addressed.