Introduction
Brain mRNA expression is modulated by numerous genetic factors and often varies substantially between strains of mice that have been reared in a standard laboratory environment. Examples include members of the NMDA receptor family that are critical in learning and memory, and genes involved in synaptic vesicle trafficking. Molecular variation of this type is often heritable and is produced by genetic polymorphisms at many locations across the genome. Differences in both alleles and mRNA levels will often produce significant behavioral, pharmacological and neuroanatomical variants1. Over the past several years, with support from the NIH Human Brain Project, we have assembled a suite of databases and web-based analysis software called WebQTL (www.webqtl.org). WebQTL is a freely accessible system that exploits sophisticated gene mapping methods2, 3 to rapidly perform whole-genome analysis at many levels—from differences in NR2B mRNA levels to differences in open-field activity levels.
WebQTL has three major applications: exploring variation in gene expression using a panel of more than 30 recombinant inbred strains and several different tissues (for example, forebrain, cerebellum, hematopoietic stem cells); mapping upstream gene loci that modulate transcript levels; and studying networks of genetic correlations among 
100,000 transcript assays and 650 published phenotypes. Additional features include tools for the simultaneous analysis of groups of traits, custom annotation of Affymetrix probes and probe sets, and external links to the Gene Ontology Machine (http://genereg.ornl.gov/gotm), the Gene Expression Atlas (http://expression.gnf.org), NCBI (www.ncbi.nlm.nih.gov) and the Genome Browser (http://genome.ucsc.edu). The integration of diverse data types provides a powerful resource for exploratory systems biology.
Data in WebQTL have been acquired from two common progenitor strains, C57BL/6J (B) and DBA/2J (D), their F1 hybrid, and a set of different BXD recombinant inbred (RI) strains. The two progenitor strains, B and D, have both been sequenced and are known to differ at roughly 1.8 million single-nucleotide polymorphisms (SNPs) across the mouse genome. This amounts to an average of one SNP every 1,500 base pairs. Each of the BXD strains is a unique 'mosaic' of chromosomal segments inherited from either the B or D progenitor strain4. About 34 BXD strains are available from The Jackson Laboratory, and an additional 45 strains will soon be available from The University of Tennessee. A wide range of phenotypes seen in the BXD reference population are also incorporated in WebQTL (see WebQTL's Published Phenotypes database). WebQTL also includes high-density marker maps based on 779 microsatellites5 and SNPs. By testing the association of genetic markers with variation in transcript levels and other traits, WebQTL maps the quantitative trait loci (QTLs) that are likely to contain modulators of these complex phenotypes. The value of this BXD reference population to the research community grows multiplicatively as additional phenotypes are collected and integrated into WebQTL.
The gene encoding the NMDA NR2B receptor subunit Grin2b provides an example of the type of analysis possible using WebQTL. The abundance of Grin2b mRNA transcript in several forebrain data sets varies approximately two-fold across 35 strains. Half of the variation in expression is heritable; and this makes it practical to map the responsible QTLs. Grin2b has two major QTLs, one on chromosome 8 (Fig. 1), and another on chromosome 6 near the transcript itself—probably associated with one or more of the 556 SNPs in this gene. The QTL on chromosome 8 is particularly intriguing, but it would be a project in its own right to discover the single correct gene associated with this QTL among 50 candidates.
Figure 1: A map of quantitative trait loci on chromosomes 6 and 8 regulating the expression of the ionotropic glutamate receptor gene Grin2b.
All mouse chromosomes, with the exception of chromosome Y, are plotted along the x-axis. The orange triangle is the transcript location; the solid blue trace is the likelihood ratio statistic for association of the phenotype with genotype across the genome. The horizontal lines indicate permutation significance thresholds for significant (P < 0.05, blue) and suggestive (P < 0.67, green) loci. Yellow bars indicate the most likely location of QTL peaks by bootstrap analysis.
Full size image (32 KB)WebQTL allows hundreds of covariates with high correlations to Grin2b to be rapidly extracted, analyzed and graphed. These include many ethanol-related phenotypes, as well as measures of locomotor activity, anxiety, maze learning, neuron cell numbers, hippocampal and cerebellar volumes and adult neurogenesis. Similarly, hundreds of transcripts with expression differences that covary with Grin2b expression can be extracted. These include Mpdz, which encodes a protein involved in the clustering and endocytosis of NMDA receptors; Ag1g1, which encodes an adaptor protein complex that is part of the clathrin coated pit; Inpp4a, which is involved in the cycling of clathrin to the Golgi; and Nfm, which encodes a neurofilament protein. At least five members of the kinesin family of motor proteins, essential for transport of the endocytic vesicle through the axoplasm, also have expression levels that correlate positively with Grin2b. What is intriguing about this example is that genetic variation underlying receptor expression may simultaneously influence expression of several critical components of synaptic receptor cycling. Several members of this extended family of genes also have known relations to alcohol-related phenotypes6. This example is just one of many query paths that can be navigated rapidly in WebQTL to generate and test hypotheses using this reference set of RI strains.
