Massively Parallel QTL Mapping using Microarrays Combined with RI and RIX Lines
Authors: Robert W. Williams1, Lu Lu1, Siming Shou1, Yanhua Qu1, David Threadgill2, John Mountz3, Hui-Chen Hsu3, Kenneth Manly4

 

Much of the individual variation in message levels measured using microarrays is due to technical error and environmental factors, but a significant fraction is generated by sequence variants in cis elements and trans-acting genes. We have begun to use arrays to map large sets of QTLs and to simultaneously uncover transcriptional networks. For this project we are exploiting recombinant inbred strains (BXD, CXB, and AXB) and a set of their RIX F1 progeny. These lines make it possible to reduce non-genetic variance by pooling and resampling while they also make it possible to analyze different stages of development and gene-by-environment effects. In initial experiments we assessed the reliability of expression data generated using the Affymetrix U74Av2 GeneChip. This array consists of ~390,000 25-mer oligonucleotide probes that collectively test the expression of ~9200 genes. Correlations of replicates arrays are ~0.92 and the median coefficient of error using two chips is 4-6%. Triplicate samples provide molecular phenotypes with adequate reliability for complex trait analysis.

Arrays were hybridized with RNA extracted from brain (forebrain plus midbrain minus olfactory bulbs) of sets of 3 isogenic animals of the same sex and same approximate age. Triplicate arrays of each isogenic line were run but at different ages (1 chip at 5-6 weeks, 10-12 weeks, and 4-6 months; giving 3 pools each of 3 animals at 3 ages). The array mapping panel consists of 21 BXD strains, parental lines, and the F1 intercross. We have developed procedures and code to analyze variation in expression level as conventional quantitative traits. Trait variance is being mapped in several stages: 1. Simple interval mapping using an implementation of the Haley and Knott regression equations.  2. We compute probabilities that variation in transcript expression is controlled by a cis-acting element.  3. Composite interval mapping with control for the transcript interval itself to detect trans acting modulators of expression.
          Distribution patterns and actions of QTLs should reveal common modulatory networks. Transcript type and shared promoter elements will also be valuable clues in assembling forebrain gene expression networks. There are significant statistical challenges but even more significant opportunities. Much larger and high precision mapping panels of RI strains and their RIX derivative (>100 lines) would greatly improve the ability to uncover networks of interactions among genes, and would also provide enough power to expose epistatic interactions.

 Acknowledgments:
This work supported by the Informatics Center for Mouse Neurogenetics, a Human Brain Project/Neuroinformatics program funded jointly by the National Institute of Mental Health, National Institute on Drug Abuse, and the National Science Foundation (P20-MH 62009).  We thank Dr. Divyen Patel (www..genome-explorations.com) for help in generating microarray data.