Forebrain weight in 34 BXD RI lines and the two parental strains C57BL/6J and DBA/2J ranged from 260 to 352 mg, with a mean (SD) of 304 mg (20 mg). Forebrain weight adjusted for variation in body weight, age, sex, and brain weight other than forebrain (see Methods), was expectedly reduced in total variation (SD = 17.5 mg), but otherwise retained the approximate normal distribution of the unadjusted forebrain weight, and was well correlated with unadjusted weight (r = 0.959).

Simple interval mapping of adjusted forebrain weight revealed two QTLs on chromosome 1 (Fbrwt1) and 11 (Fbrwt11) with likelihood ratio statistics (LRSs) above a significance threshold determined by 10,000 permutations of the data (Figures 1, 2). Each C57BL/6J allele was estimated to increase adjusted forebrain weight by 10 mg, somewhat more than half a standard deviation in effect size. A pairwise scan using the DIRECT algorithm for epistasis detection 18 implemented at http://www.genenetwork.org did not discover significant interaction between these or other loci.

Although we were unable to verify differences in Tnni1 protein abundance (see Methods), we were able to discover a 273 bp insertion located 1080 bp upstream of the Tnni1 transcription start site, that was present in DBA/2J mice, but absent in C57BL/6J mice (Figure 7). C57BL/6J sequences confirmed those published in Genbank for the Tnni1 promoter. Bioinformatic analysis of the indel polymorphism suggests the presence of multiple transcription factor binding sites (see Additional file 4).

Using a five-filter protocol for QTL candidate gene nomination that combines classical trait QTL mapping with gene expression data in BXD mice, we provide convergent evidence that nominates troponin 1 (Tnni1, Chr 1) and ankyrin repeat and SOCS box-containing protein 3 (Asb3, Chr 11) as candidate genes for QTLs affecting mouse forebrain weight variation, with C57BL/6J alleles conferring increased weight over DBA/2J alleles. To nominate these candidate genes, we required candidate genes near to the forebrain weight QTL to have correlated mRNA abundance among the BXD lines as well as a difference between the parental lines C57BL/6J and DBA/2J. We further required verification of mRNA abundance by an independent method, and finally we required either differences in protein levels or confirmed DNA sequence differences. Tnni1 and Asb3 met each of these requirements.

Asb3 encodes a member of the ankyrin repeat and SOCS box containing proteins 19. Ankyrin repeats are a common motif involved in molecular recognition via protein-protein interactions. The SOCS box found in the Asb family of proteins targets suppressor of cytokine signaling (SOCS) proteins with the elongin B/C complex and can be involved in targeting for degredation. Asb3 is widely expressed, including high levels of expression in the brain and mediates ubiquitination and degredation of tumor necrosis factor receptor II (TNF-R2). Asb3 protein is thus also involved in inhibition of TNF-R2-mediated Jun N-terminal protein kinase (JNK) activation. JNK is involved in cortical neuron migration 20 and may be involved in signaling cell survival in developing forebrain 21 which suggests a role for Asb3 in forebrain development.

Tnni1 encodes the slow-twitch muscle isoform of Troponin I 22, the inhibitory subunit of the troponin complex. Although we could not verify the effect of Tnni1 transcript differences on protein levels, we did discover a 273 base pair indel polymorphism upstream of the transcription start site of Tnni1 in which we found bioinformatic evidence for transcription factor binding sites. Without testing a broader sample of inbred lines for the presence of the indel and Tnni1 expression, it remains unknown if this polymorphism is an interfering insertion, reducing DBA/2J Tnni1 transcription. Nonetheless, a working hypothesis is that the discovered indel is a Tnni1 promoter variant underlying the cis-eQTL for Tnni1 mRNA abundance and the QTL for forebrain weight. Tnni1 is typically expressed in skeletal muscle 2324, and functions to prevent actin-myosin interaction, but it is also considered a cytoskeletal element in some neurons 25. We demonstrated that Tnni1 mRNA is also found in the mouse brain, and we suggest it may have alternative functions in CNS tissue, perhaps during development. This is particularly interesting because of recent work on mutations in other cytoskeletal element genes and their roles in human brain size and microcephaly 26.

An earlier paper by Belknap et al., 27 reported two QTL affecting brain to body weight ratios on chromosomes 11 (9 cM) and 17 (25–40 cM) in a sample of 20 BXD RI lines. The QTL reported on Chr 11 is in the same proximal location as Fbrw11 reported above. As defined, our forebrain weights included midbrain and forebrain (telencephalon, diencephalon, and mesencephalon), but excluded olfactory bulbs and the brain stem (metencephalon and myelencephalon). We also used linear regression to control body weights rather than ratios 2728, and unlike Belknap et al., 27, we included both sexes, and used a wider range of ages. Nonetheless, the genetic correlation between our measures and those reported more than 15 years ago in Table 1 in 27 for total brain weight to body weight ratios for male mice is r = 0.58 (p = 0.0068) 29. It is therefore possible that the same genetic signal (candidate Asb3) underlies both QTLs. A more recent study on neocortex volume and remaining brain volume among BXD lines suggests that the genetic signal is mostly expressed in neocortex variation. Beatty and Laughlin 30 reported a QTL for neocortex volume in the same location as Fbrw11; noncortical brain volume did not reach statistical significance on proximal Chr 11 after genome-wide multiple test correction 30. The correlation between our forebrain weights and the neocortex volumes was r = 0.73 (p < 0.0001).

The literature is replete with reviews describing a status quo of relative ease of QTL identification and great difficulty discovering the underlying polymorphism (e.g., see 1). Indeed, without bioinformatic support, the purely experimental path to polymorphism identification is effortful, long, and without guaranteed success. A key pillar of the protocol we describe is use of the BXD genetic reference population and the availability of gene expression data on each BXD line. The list and specificity of traits for which BXD gene expression is available is growing rapidly, and already contains different organs as well as specific cell types http://www.genenetwork.org. Multiple conditions are also coming online (e.g., developmental time (cerebellum) or environmental manipulations (stress, alcohol)). It is not strictly necessary for the gene expression database to exactly match the measured trait for our five-filter protocol to be applied. Behavioral phenotypes could reasonably be correlated to brain expression databases (e.g., trait anxiety but perhaps not state anxiety). Certainly, the correlations would become more relevant with a more specific expression database (e.g., amygdala), and cases can be conjectured for which our approach would be uninterpretable (e.g., G × E QTLs using only baseline gene expression databases). In such cases, our suggested protocol is not flawed, but the available resources would need generating (an expensive proposition). In the present study, for example, we used adult gene expression to make inferences about development of forebrain weight. This will work in cases where developmental switches are left on (e.g., 3132) but will fail with temporally limited gene expression differences (e.g., 3334). More complete results will come from application of the five filters to a developmental forebrain gene expression database.

Some of the strengths of our protocol are also disadvantages. The use of BXD mice necessarily limits the number of genomes investigated to only two common inbred strains (C57BL/6J and DBA/2J). Unfortunately, there is a dearth of large mouse recombinant inbred line panels, although this may change in the future 12. Our results, for example, could be extended by validating brain weight as measured in the current BXD RI lines 35 to the more recently developed, advanced intercross derived BXD RI lines 8. Finally, nominating genes that act through gene expression also presents an obvious weakness, by missing gene variants that cause null or poorly trafficked proteins, for example. Despite these limitations, the methods we illustrate warrant careful consideration by those working with phenotypic differences between C57BL/6J and DBA/2J.

All experimental procedures were performed in accordance with (1) the Guidelines for the Care and Use of Laboratory Animals published by the National Institutes of Health (publication 86-23) and (2) the University of Tennessee Health Sciences Center Animal Care and Use Committee (protocol number 680).

Separate cohorts of 34 BXD recombinant inbred (RI) lines as well as inbred strains C57BL/6J (B6) and DBA/2J (D2) were used for QTL mapping and gene expression. BXD RI lines were generated by Taylor and colleagues 35 from C57BL/6J and DBA/2J parental strains in the mid-1970s (BXD1 through 32) and 1990s (BXD33 through 42); additional so-called "Williams" lines were generated recently by our group 8. RI strains are fully inbred lines derived from brother-sister matings starting from an F2 intercross. Although forebrain weight data are now available for both the original "Taylor" BXD mice and the "Williams" BXD mice, we focus only on the "Taylor" BXD mice. Not all of the "Williams" BXD mice are fully inbred, and some lines retain a small amount of heterozygosity that may affect forebrain weight though dominance mechanisms we do not account for in this paper.

Mice were housed at 20 to 24°C on a 14/10 h light/dark cycle in a specific pathogen-free (SPF) facility at the University of Tennessee. All animals were fed 5% fat Agway Prolab 3000 (Agway Inc., Syracuse, NY) rat and mouse chow. The average age of BXD/Ty animals at the time of sacrifice was 82 days with a range of 21–763; age did significantly predict forebrain weight and was considered along with other factors in our forebrain weight model below.

Mice were deeply anesthetized with Avertin (1.25% 2,2,2-tribromoethanol and 0.8% tert-pentyl alcohol in water, 0.5–1.0 intraperitoneal injection). Next, they were transcardially perfused with 0.1 M phosphate buffered saline followed by 4% paraformaldehyde in 0.1 M phosphate buffer. Tissues were stored in fixative thereafter.

The gene expression data set used for these analyses was selected from a larger set that has been previously described 11. There is a link to extensive metadata describing the samples and sample processing on GeneNetwork http://www.genenetwork.org/dbdoc/BR_U_1203_M.html. Briefly, tissues were dissected from BXD animals (both sexes, aged 8, 20, or 52 weeks) in 32 of the same strains (but different animals) as for the forebrain weight analysis, but using unfixed tissue. Total RNA was extracted and labeled according to Affymetrix protocols and hybridized with Affymetrix U74Av2 microarrays http://www.affymetrix.com/index.affx. A total of 2–4 littermates were dissected and equal amounts of tissue were combined (pooled) for hybridization to each array. A total of 100 arrays were used. Array data were normalized using the MAS5 algorithm from Affymetrix. For a subset of confirmation analyses, we accessed a newer Affymetrix M430 microarray dataset (described at http://www.genenetwork.org/dbdoc/IBR_M_0106_R.html).

The number of BXD mice used to collect forebrain weights was 386, with an average of 11 mice measured per strain (minimum, 25%, median, 75%, maximum: 3, 7, 10, 15, 21). Because our forebrain weight measurements were not taken from a population balanced for important covariates, we used multiple regression to fit effects of age, body weight, sex, and non-forebrain brain weight (weight of total brain after the weight of the forebrain was substracted). Residual forebrain weights are available on http://www.genenetwork.org (Trait 10701, standardized to the mean forebrain weight by addition of the average forebrain weight by strain) as are simple raw trait averages by strain (Trait 10699).

QTL and eQTL mapping was performed using GeneNetwork http://www.genenetwork.org and a standardized set of 3795 genotyped markers (mapping algorithm and genotypes described at http://www.genenetwork.org/dbdoc/BXDGeno.html; genotypes downloadable as a text file from http://www.genenetwork.org/genotypes/BXD.geno). Residuals from the model described above (Trait 10701) were simple interval mapped using a modified Haley-Knott algorithm 3637, weighted by the within strain variances. Genome-wide significance was calculated by comparing the best likelihood ratio statistic of the original data set with the distribution of highest LRS computed for 10,000 permutations.

eQTL mapping is QTL mapping of gene transcript abundance, generally measured by microarray. eQTLs can be classified as either cis-eQTLs, that map to the same location of gene encoding the transcript being mapped, or trans-eQTLs, that map to locations other than the gene encoding the mapped transcript. Cis-eQTLs are suggestive of a polymorphism in the gene promoter.

Selection criteria for candidate genes included five filters. The first filter required a candidate gene to be located near to the mapped forebrain QTL, within 10 million bases from the genetic marker with the highest LRS resulting from simple interval mapping. Physical locations of genes in the BXD are known, because the genomes of the parental inbred strains C57BL/6J and DBA/2J have been sequenced. Physical positions from the mm6 assembly of the mouse genome http://genome.ucsc.edu/cgi-bin/hgGateway were used with Genenetwork to generate lists of genes residing in or near to the QTLs. The second filter required a significant genetic correlation between forebrain microarray gene expression and forebrain weight among BXD strains. GeneNetwork can be used to rapidly estimate genetic correlation between BXD phenotypes 910. We used GeneNetwork to correlate forebrain weight and gene expression from a dataset of 100 microarrays on 32 BXD RI lines (Spearman rho, alpha = 0.05). The third filter required a significant difference in forebrain microarray gene expression between the BXD parental inbred strains C57BL/6J and DBA/2J. Unpaired, equal variances t-tests were used to compare 3 and 3 Affymetrix U74Av2 microarrays (alpha = 0.05). Because gene microarray technology platforms change, we also verified candidate gene differences between C57BL/6J and DBA/2J using a newer microarray data set, based on the Affymetrix M430 A and B chips http://www.genenetwork.org/dbdoc/IBR_M_0106_R.html. The fourth filter required verification of gene expression differences by reverse transcriptase PCR (RT-PCR) in C57BL/6J and DBA/2J, and in a two BXD RI lines that had low and high transcript abundance by microarray. The fifth filter required protein differences by Western blot on genes verified by RT-PCR. Together, these five filters strongly nominate genes for classical trait QTLs that act by differences in gene expression.

The third filter, requiring a gene expression difference in the parental lines of the BXD, may be conservative, because an absence of a difference in the parental lines doesn't necessarily preclude heritability in the BXD. Shockley and Churchill 38 found more gene expression differences between A/J:C57BL/6J consomic lines than between the A/J and C57BL/6J parental inbred strains. One interpretation is that A/J and C57BL/6J carry compensating (epistatic) increaser and decreaser alleles that are segregated in the consomic lines. This has also been described in the BXD, when the RI lines have a range in phenotypic scores than is greater than ("transgresses") the difference between the C57BL/6J and DBA/2J parental inbred strains (e.g., 39).

Microarrays have been shown capable of yielding quantitative estimates of RNA levels 4041. However, it is generally accepted that differences benefit from verification with independent samples and methods. Also, in the present application of the short probe Affymetrix U74Av2 platform, it is possible that expression differences on the chip (but not in vivo) could arise from polymorphism between C57BL/6J and DBA/2J in Affymetrix probe sequences, because these were designed from C57BL/6J sequence information. BXD RI lines inheriting C57BL/6J alleles at such a location could in theory exhibit stronger hybridization than lines inheriting DBA/2J alleles. We used real-time PCR to verify expression differences, using duplicate samples for the parental strains C57BL/6J and DBA/2J, as well as duplicate samples for a high and low expressing BXD RIL, BXD40 and BXD25. Total RNA was isolated from whole brains using TRIzol reagent (Invitrogen, Carlsbad, CA). RT-PCR was performed on a SmartCycler (Cepheid, Sunnyvale, CA) using the AccessQuick RT-PCR system (Promega, Madison, WI), and SYBR green I (Molecular Probes, Eugene, Oregon) according to the manufacturer's instructions. The primers used to target mouse genes were (F is the forward primer, and R is the reverse): Tnni1, F: CAC CAG AGA GAT CAA GGA CC, R: TGT GCT TAG AGC CCA GTA GG; Asb3, F: TTT CAT CCA TCA GTT GCC AC, R: GCC TTG CTG GTT TCT CCA TC. Reverse transcription was performed at 48°C for 45 min and RT-PCR cycling parameters were as follows: denaturation at 95°C for 2 min followed by 35 cycles of amplification (94°C, 30 sec; 62°C, 30 sec). Product size was initially monitored by agarose gel electrophoresis and melting curves were analyzed to control for specificity of PCR reactions. The data on the target genes was normalized to the expression of the housekeeping gene β-actin and the relative units were calculated from a standard curve, plotting 3 different concentrations against the PCR cycle number at which the measured intensity reaches a fixed value (with a 10 fold increment equivalent to ~3.1 cycles).

BXD25 and BXD40 mouse brains were lysed directly in radioimmunoprecipitation (RIPA) buffer for analysis of whole cell lysates. 50 μg protein, calculated using a BCA (Bicinchoninic Acid) Protein Assay Kit (Pierce, Rockford, IL), were subjected to SDS-PAGE. Proteins were transferred to nitrocellulose membranes, immunoblotted with Asb3 specific antibodies (Santa Cruz Biotechnology, Inc., Santa Cruz, CA; antibody sc-19932) and visualized by enhanced chemiluminescence using the SuperSignal western blotting detection system (Pierce, Rockford, IL). The average intensity of bands was calculated using ImageJ http://rsbweb.nih.gov/ij/. Unfortunately, the only available antibody for Tnni1 protein (Santa Cruz Biotechnology, Inc.) could not be made to work successfully in our lab, and we therefore only report results for Asb3 protein.

PCR fragments of Tnni1 were amplified from mice genomic DNA and subcloned into the pCR2.1 TA vector (Invitrogen). The sequences for primers P1 and P2 were: P1, 5' GAA TGG TAC CCC AGG TCG ACT TG 3' and P2, 5' AAG TCT GCT CTT CAC AGG TCA CA 3'. Sequencing was done by Macrogen (Rockville, MD).

The transcriptional start site (TSS) of Tnni1 was determined using the Database of Transcriptional Start Sites (DBTSS; http://dbtss.hgc.jp) 42. Potential transcription factor-binding sites (TFBSs) were then identified using the TRANSFAC database and P-Match software by screening the upstream region of the Tnni1 indel sequence. All sites were found by the P-Match using the default parameters 43.