QTL analyses of complex traits with cross validation, bootstrapping and other biometric methods

Abstract

With the development of molecular markers, dissection of complex quantitative traits by mapping the underlying genetic factors has become a major research area in plant breeding. Here, we report results from a vast QTL mapping experiment in maize with testcrosses of N= 976 F_4:5 lines evaluated in E= 16 environments. Although the number of detected QTL confirmed the infinitesimal model of quantitative genetics (e.g., 30 QTL detected with LOD ≥ 2.5 for plant height, explaining p= 61% of the genetic variance), cross validation (CV) still revealed an upward bias of about 10% in p. With smaller values of N (122, 244, 488) and E (2, 4), the number of detected QTL decreased, but the estimates of p remained almost the same due to a tremendous increase in the bias. This illustrates that QTL effects obtained from smaller sample sizes are usually highly inflated, leading to an overly optimistic assessment of the prospects of MAS. Moreover, inferences about the genetic architecture (number of QTL and their effects) of complex traits cannot be achieved reliably with smaller sample sizes. Based on simulations, we conclude that CV and one method of bootstrapping (BS) performed well with regard to yielding realistic estimates of p. In addition, we briefly review progress in new biometric methods and approaches to QTL mapping in plants including Bayesian methods that show great promise to overcome the present limitations of QTL mapping.