Brief Summary of Minutes of Annual Meeting

ACCOMPLISHMENTS AND IMPACTS Objective 1: Develop and compare statistical methodology to map genes Xu (UC Riverside) continued developing new statistical methods for QTL mapping in pedigrees. He finished the Monte Carlo method for calculating IBD matrices of pedigrees with arbitrary complexity (Mao and Xu, 2005). The method was further modified to fit pedigrees containing all inbred lines. He analyzed a pedigree containing 404 inbred lines of maize with known genealogical relationship and identified several major QTL responsible for the genetic variance of flowering time (Zhang et al., 2005). Xu has developed also a Bayesian shrinkage method for estimating QTL parameters. The method can simultaneously estimate the effects and positions of several hundred QTL with a single model (Wang et al., 2005). The model assumes that the maximum number of QTL is p. The positions of these QTL along the genome are disjoint and vary based on Metropolis-Hastings rule. The method can handle extremely high marker density. In addition, Xu has been studying statistical methods for clustering expressed genes based on their association with a quantitative trait. He first examined the linear association (Jia and Xu, 2005) and then higher order association using orthogonal polynomials (Qu and Xu, 2005). Both methods have been applied to data collected from 31 subjects in a microarray experiment for Alzheimer disease. He detected many genes that are associated with the disease phenotype MMSE. Wang (Michigan State U.) reported on results of three studies aiming the genetic mapping of quantitative trait loci underlying agronomic traits in soybean. The first experiment referred to identification of putative QTL for yield in interspecific soybean backcross populations. The second study was related to genetic mapping of QTLs that condition waterlogging tolerance in soybean. The third trial aimed the genetic mapping of genes underlying partial resistance to Sclerotinia Stem Rot in Soybean PI 391589B. Misztal (U. Georgia) reported on results of a study aiming the estimation of competitive effects for average daily gain in swine. There were 4,946 records from 2,409 litters and 362 pen-groups. Pen size ranged from 12 to 16. Models included the effects of contemporary group (farm-barn-batch), birth litter, pen-group and two additive genetic effects: direct and associative. The additive genetic variance was a function of the number of competitors in a group, the additive relationships between the animal performing the record and its pen mates, and the additive relationships between pen mates. To partially account for differences in pen size and in relationships among members of the pen a covariable was added to the associative genetic effect. Estimates of direct and associative heritability were 0.15 and 0.03, respectively. Misztal mentioned that the magnitude of competition effects may be larger in commercial populations, where housing is denser and food is limited. Jannink (Iowa State U.) reported on research on selective phenotyping to accurately map QTL. The marker genotypes of the progeny should allow the number of recombination events they carry to be determined such that the most useful progeny could be phenotyped, in a procedure termed selective phenotyping. Two methods to select genotypes for their usefulness in mapping were evaluated, one that maximizes the overall mapping information content in the selected progeny, and one that seeks to maximize both overall mapping information and the uniformity of its distribution across the genome. Simulations showed that both methods successfully decreased the mean squared error for QTL position. Average mean squared errors were similar for the two methods and variability of mean squared error was slightly lower for the latter relative to the former method. Simulations indicated that a large fraction of the decrease in the mean squared error achievable by selective phenotyping could be obtained by genotyping twice the number of progeny than were ultimately phenotyped, though further decreases in the mean squared error were observed when up to sixteen times more progeny were genotyped than phenotyped. The procedure appears to most improve the accuracy of QTL mapping for QTL of small effect or when available markers do not allow marker spacing below 10 cM. Rosa (Michigan State U.) continued development of alternative techniques for the statistical integration of potentially miscoded genotypes in linkage analysis and QTL mapping studies in line crosses and outbred populations. The same ideas are being implemented in the context of paternity assignment with uncertain pedigrees, and on mark-recapture applications using molecular markers. Rosa reported also on research being conducted on linear mixed models suitable for the analysis of either log ratios or log intensity values of microarray data in the presence of multiple sources of variability. These models have been used also to compare the power and efficiency of different microarray experimental designs within a hierarchical replication context. A third area of research reported by Rosa referred to environmental risk assessment, by extending the net fitness components model of Muir and Howard (1999, 2001, 2002) to estimate environmental risk of genetically modified organisms (GMO) by predicting the fate of transgenes introduced into wild populations by escaped GMO. Dekkers (Iowa State U.) discussed a theoretical analysis of alternative measures of LD based on multi-allelic microsattelite markers. Effectiveness of marker-assisted selection (MAS) using population-wide LD depends on the extent of marker-to-QTL (M-Q) LD. To evaluate alternative measures of observable (marker-to-marker) LD as predictors of M-Q LD, LD among 4-allele markers and a biallelic QTL was simulated by 100 generations of random mating of 100 parents. Using 100 individuals in generation 100, M-Q LD was quantified by the R2 of regression of QTL allele on alleles at a single marker. Observable LD was evaluated using: Lewontins D2; r2=pooled square of correlations between alleles weighted by the product of allele frequencies; c2=Chi-square statistic for association between alleles; and a standardized c22=c2/[N(n-1)], where N=number of haplotypes and n=smallest number of alleles across the 2 markers. Extensive M-Q LD existed at short distances but declined rapidly with distance. Observable LD showed similar declines for r2, c2 and c22, but D2 was strongly inflated. Correlations of mean D2, r2, c2 and c22 between markers with mean M-Q LD at corresponding distances (d20cM) were 0.85, 0.96, 0.96 and 0.96. Correlations of means for different population sizes (10 ~ 200) of D2, r2, c2 and c22 at 2cM with means of M-Q LD at 1cM were 0.80, 0.80, -0.66 and 0.75. Corresponding correlations for means at different generations (0 ~ 200; population size of 100) were 0.56, 0.70, 0.33 and 0.74. Although r2 and c22 both correlated well with M-Q LD, c22 is preferred because it ranges from 0 to 1, while r2<1 for multi-allelic markers in complete LD. To assess the decline in LD with distance, LDd=1/(1+4²d) was fitted to data on 100 individuals in generation 100 where d is distance in Morgans, and ² is related to effective population size. Estimates for ² were 51.4 and 63.8 for c22 and M-Q LD, resulting in very similar lines. In conclusion, c22 is a good predictor of LD between markers and QTL when LD is generated by drift alone. Objective 2: Examine the efficiency of incorporating molecular tools in breeding programs through theoretical modeling, computer simulations, and biological testing in actual breeding populations. Muir (Purdue) examined the efficiency of incorporating molecular tools in breeding programs through theoretical modeling, computer simulations, and biological testing in actual breeding populations. For traits of low heritability, initial theoretical examination showed that MAS could increase response to selection by as much as 500%. However, a decade of experimentation and simulations has since demonstrated a much more moderate response. These shortcomings were found to be due a critical assumption: that the quantitative trait loci (QTL, or closely linked makers) affecting such traits were known. In actuality, these QTL associations are found by statistical estimation and hypothesis testing based upon similar data breeders would use to make selection decisions, i.e. have the same limitations of a high environmental variance. Thus QTLs for traits of low heritability are difficult or impossible to locate. A gene level simulation was used to compare results of genome wide MAS (GMAS) with that using conventional methods of BLUP estimation of genetic based on pedigrees, as well as study the number of generations of training needed to accurately estimate the breeding values just from genotypes, and for how many generations after phenotypes were no longer collected the predictions were still good. Results showed that as expected for GMAS, the more generations of training you do, the better it is at prediction. For a trait of heritability of .5, the accuracy of selection with GMAS reached about 88%, whereas the BLUP line only reaches 82% accuracy. The accuracy of BLUP dropped off rapidly in generations where the genotype is predicted based only on ancestors information whereas GMAS continued at a higher accuracy. Bill stressed that the power of genomics is much higher for traits of low heritability. For example, a trait with a heritability of .1, the accuracy of selection was up to almost 70% with GMAS and 3 generations of training while BLUP was only hitting about 60%. Bills sees this as the future of genomics in animal breeding, the only real issue is if we can get the price down. Objective 3: Use molecular genetics to test hypotheses generated from the fundamental theories of population, quantitative, molecular and evolutionary genetics. No station reports included results relevant to this objective. Submitted April 15, 2005 Guilherme J. M. Rosa (Secretary)