The PMBSF was examined in CO stained tissue in 140 mice from 45 strains (BXD = 42, parentals = 2, F1 = 1) (Table 1). The PMBSF consists of five well-defined anterior-posterior running rows (A row through E row) each representing one large mystacial vibrissae on the contralateral face. Rows A and B typically contain four barrels, while rows C-E contain five barrels. Four straddler barrels (α, β, γ, δ) form the posterior border of the PMBSF. Together, a total of 27 barrels form the PMBSF and the area defining these barrels was measured in the present study. It is worth noting that in a number of cases (n = 22) one barrel from either the A or B or both rows were found missing in one or both hemispheres, but the general overall PMBSF pattern was not disrupted. Furthermore, there was no significant difference (p = 0.86) between total PMBSF areas of hemispheres with asymmetrical number of barrels. Several examples of the PMBSF area are shown in Figure 1.

Broad-sense heritability (h²) of PMBSF area was calculated using raw data for all phenotyped cases. The computed value for PMBSF area variation was approximately 58% (h² = 0.58). This confirmed previous findings where similar heritability values (60%) were calculated for combined PMBSF/ALBSF using only 10 strains 8. This similarity in heritability further substantiates our previous study that the use of a small number of animals per strain (n = 2.90 in this study) is appropriate for mapping purposes given a large number of phenotyped strains. As expected, PMBSF area adjusted for brain weight showed a lower heritability of approximately 41% (h² = 0.41).

Brain and body weight values of all phenotyped cases correlate. Pearson's product-moment correlation showed that brain and body weights correlate at r = 0.44 (p < 0.0001). Group mean correlation of brain and body weights for all strains was r = 0.37 (p = 0.013).

PMBSF areas of DBA/2J and C57BL/6J were 2.08 ± 0.06 mm² and 2.00 ± 0.03 mm², respectively, making the DBA/2J larger by 4.24%. This difference was not statistically significant (p = 0.11). However, when the data were adjusted for brain weight, there was a significant (p = 0.008) difference between DBA/2J and C57BL/6J (2.14 ± 0.06 mm² and 1.96 ± 0.03 mm², respectively) making the DBA/2J larger by 8.50%. Total brain weight of DBA/2J animals was significantly (p < 0.0001) smaller than that of C57BL/6J animals (0.35 ± 0.01 g, 0.42 ± 0.01 g respectively). In comparison, body weights of the two strains did not vary with DBA/2J = 16.53 ± 0.99 g and C57BL/6J = 16.67 ± 0.95 g.

The majority of strains used contained at least one male and one female; however, same sex mice were used in seven strains (see Table 1). Animal sex did not show a strong relationship to PMBSF area and thus was not used as a predictor of PMBSF area. Using a Pearson's product-moment correlation, sex correlated with PMBSF area (r = 0.03). A simple regression analysis confirmed the significance of this relationship where we failed to reject the null hypothesis. The regression also showed that sex only accounted for approximately 0.1% of the variance seen in PMBSF area (F_1,138 = 0.15, p = 0.70).

Animals used in this study fell within a small age range: 40 – 61 days old. Simple correlation analysis among all phenotyped individuals showed age to be moderately correlated with PMBSF area (r = 0.26, p = 0.0016). Likewise, age and PMBSF area of group means correlated, albeit with less significance (r = 0.26, p = 0.08). A simple regression analysis revealed that age accounts for approximately 7.0% of total PMBSF area variability (F_1,138 = 10.4, p = 0.0016). However, this may be misleading given that age was also coincidently correlated with strain (r = 0.22, p = 0.0091). For this reason, age was excluded as a factor when modeling PMBSF area variability.

Body weight correlated significantly with total PMBSF area in all individual cases at a modest level (r = 0.20, p = 0.02). Approximately 3.8% of variance in PMBSF area may be accounted for by body weight (F_1,138 = 5.44). Similarly, brain weight also correlated significantly with PMBSF area (r = 0.25, p = 0.0034). However, unlike body weight, brain weight accounted for about 6.1% of the variance seen in PMBSF area (F_1,138 = 8.89).

Our results showed a 33% difference between PMBSF area means of the largest and smallest BXD RI strain measurements (see Table 1). We utilized direct correlations between PMBSF area and other variables to explore genetic relationships. A linear model was utilized to distinguish among the most significant factors for the prediction of PMBSF area variability.

Three factors were taken into consideration in the linear model: sex, body weight, and brain weight. Strain was excluded as a predictor because it is the genetic factor that we hypothesized to account for PMBSF area variability. Using all individual cases, we observed that sex was the least significant factor (p = 0.46) in predicting PMBSF area variability; sex was thus the first factor removed from the model. Next, body weight became the least significant factor (p = 0.24) and was therefore removed. Brain weight remained a significant factor both in the presence of sex and body weight and of course as the last remaining factor as well (p = 0.0034). This was consistent with the finding that brain weight accounts for approximately 5.4% of the variability in PMBSF area, the largest percentage of variance as predicted by other factors. We therefore adjusted PMBSF area for brain weight and used the adjusted values for QTL mapping.

In this study, all QTL mapping was completed using only BXD RI strains (n = 42). To further confirm that our PMBSF area QTLs are not due to variations in brain and body weights, we mapped the QTLs for these traits on WebQTL using raw data. Figure 2A shows a QTL map of body weight using raw data. Clearly, this map does not show any significant QTLs. It should be noted however, that there was a signal near the suggestive threshold on chromosome 4, however the location is remote from the PBMSF area QTL as discussed below. The QTL map of brain weight using raw data is shown in Figure 2B. Again, only a couple of near-suggestive outcomes were observed with one on chromosome 4. None of the detected signals, including the one on chromosome 4 for brain weight, overlapped with our proposed PMBSF area suggestive QTL.

We used our 42 BXD strains to map total PMBSF area QTL. Both raw data and adjusted values of PMBSF area were utilized for mapping in order to test whether our data modeling affected position of the identified PMBSF area QTL. By examining the simple regression QTL map of raw PMBSF area (Figure 3A), we immediately noticed two suggestive QTLs, one on the proximal arm of chromosome 4 and the other on chromosome 17. In addition, a couple of other signals (chromosomes 15 and X) were also observed to cross the suggestive threshold. When PMBSF area was adjusted for brain weight, the suggestive QTL on chromosome 4 remained largely unchanged, in terms of its significance and its location (Figure 3B). However, most of the other suggestive QTLs observed in the raw data map had diminished signals, in particular those on chromosomes 17 and X. Marker regression analysis using WebQTL revealed four loci with highly suggestive LRS values, all at 14.20 (suggestive threshold LRS = 10.68). The detected loci are: rs3674982, rs13477546, gnf04.004.855, rs13477551.

Interval regression mapping (Figure 3C) of chromosome 4 with 1000 permutations and bootstrap analysis shows a detailed view of the total PMBSF area suggestive QTL. Bootstrap analysis is a method of testing the reliability of the peak QTL signal by randomly producing a new sample from the original data set. The newly created sample set contains the same number of cases as the original set, however one or more of the cases may be duplicated and/or removed. The bootstrap analysis histogram at right in Figure 3C shows that the suggestive QTL is regenerated at the same location approximately 54% of the time, further suggesting the reliability of the signal within chromosome 4.

The QTL signal spans a relatively small region, from 5.5 to 9.0 Mb. Although there are no obvious multiple QTLs, we tested for interaction among different loci given that there are a number of suggestive loci. Both LRS Interaction and LRS full pair-scan analyses revealed no significant two-loci epistatic interactions.

We detected 9 genes that lie under the chromosome 4 interval of 5.5 to 9.0 Mb (Table 2). We utilized a 1.5-LOD support interval around the peak value of the suggestive QTL to derive the 5.5 to 9.0 Mb range. In order to increase the likelihood of detecting potential candidate genes, we first prioritized the genes by the number of SNPs present within each gene. Biological relevance was the next factor that went into consideration when identifying potential candidates. With these criteria in mind, we identified two genes of particular interest that are discussed below (Car8 and Rab2). Car8 is a cis-acting gene whereas Rab2 is a trans-acting gene.

Microarray data analysis is a powerful method that may be used in conjunction with QTL analysis to detect genes of interest that may be expressed in remote regions of the genome. We used RNA expression profiles of whole brain from one of the GeneNetwork databases (INIA Brain mRNA M430 (Jan06) PDNN) to detect genes of interest that are correlated with our adjusted PMBSF area phenotype. An initial correlation query resulted in the detection of 2000 transcripts from the entire genome that were correlated with adjusted PMBSF area. Although we phenotyped 42 BXD RI strains, the highest number of strains that was detected by the correlation analysis was n = 22.

After searching through the 2000 transcripts, we detected two genes of particular interest, adenylate cyclase 1 (Adcy1) and growth associated protein 43 (Gap43), that have a significant biological relationship to PMBSF area. The first gene, Adcy1, is correlated at r = 0.40 (p = 0.063) with a high mean expression of 16.28. Adcy1 is found on chromosome 11 of the mouse starting at 7.08 Mb. The second gene of interest, growth associated protein 43 (GAP43), is correlated with total PMBSF area (r = -0.55, p = 0.007) and is found on chromosome 16 starting at 42.18 Mb.

We took advantage of previously published phenotypes by searching for traits that correlated with Car8, one of our potential candidate genes. Using the online Allen Brain Atlas 16, we observed that Car8 was highly expressed in cerebellar Purkinje cells. Car8 has strong negative correlations with volume of CNS regions in BXD RI strains, including basolateral amygdala, hippocampus, cerebellum, and striatum.

This study was an extension of our previous work 8 in which the total PMBSF/ALBSF area was examined in four common inbred strains and 10 BXD RI strains of mice. In the present study, we collected data from a total of 42 BXD RI strains of mice, two parental strains, and one F1 hybrid. We examined a number of phenotypes including body weight, brain weight, and total PMBSF area in a total of 140 cases. This is the first time that such a wide array of BXD RI strains have been used for mapping a QTL of any cortical area. We found wide continuous variations of total PMBSF area among the BXD RI strains with a 33% difference between highest and lowest values. Using previously established methods of QTL analysis, we identified a suggestive QTL on the proximal arm of chromosome 4 linked to total PMBSF area. This locus contained two interesting candidate genes: Car8 and Rab2. In addition, correlation analysis with a whole brain microarray expression database revealed a number of highly correlated genes throughout the genome including adenylate cyclase 1 (Adcy1), a gene known to disrupt barrel organization of mouse somatosensory cortex 17, and Gap43, a gene previously reported to be involved in barrel formation during mouse development 18.

One of the most difficult aspects of examining the barrel field in the somatosensory cortex arises from the flattening process of the hemispheres. Only two experimenters performed the cortex flattening process (TJ and CL), with one (TJ) conducting the process in 32/42 (76%) of the BXD RI strains. All reconstructions and measurements were conducted by a single experimenter (TJ) to reduce measurement biases. We also selected the PMBSF to measure since this subfield is the largest, contains well-delineated barrels, and suffers less distortion in the flattening process. This is particularly important, because when the PMBSF and ALBSF were previously examined 8, the ALBSF was often distorted in the flattening process, since the most anterior barrels within the ALBSF were located over a major curvature in the cortex. Therefore, we focused exclusively on the PMBSF in the present study.

In our data set, there was a significant correlation between age and strain, and thus we were unable to test age as a predictor of total PMBSF area in our linear model. However, this should not affect the genetic analysis of total PMBSF area given that the age range in the dataset was fairly compact (40 – 61 days old) and no significant relationship between age and total PMBSF/ALBSF area was previously detected 8.

Brain and body weight measurements correlated significantly. Both also correlated significantly with total PMBSF area. However this correlation was modest at best. This suggests that total PMBSF area is not simply modulated by body and brain weights. In order to dissect these seemingly interrelated phenotypes, we used a linear model that essentially removed the effects of significant predictors of total PMBSF area. This led to the generation of adjusted total PMBSF area values that remove effects of brain weight, the most significant predictor. To gain a visual understanding and further confirm our total PMBSF area QTL, we mapped the raw data of both body and brain weights separately. As evident from body weight and brain weight QTL maps (Figure 2), there were no significant or suggestive QTLs at any locus on the genome. The body weight and brain weight QTL signals that were found on chromosome 4 for both phenotypes do not overlap with the suggestive QTL we detected for total PMBSF area.

As previously reported 819, the average brain weight of C57BL/6J is significantly larger than DBA/2J. It was also previously observed that the DBA/2J mice have a significantly larger ALBSF/PMBSF area than the C57BL/6J mice 8. In our study, we did not detect an absolute difference between the PMBSF areas of C57BL/6J and DBA/2J when using raw data. Nonetheless, the trend is still present in our data; the DBA/2J PMBSF area is larger than C57BL/6J PMBSF area by approximately 4%. However, after adjusting for brain weight, total PMBSF area of DBA/2J is significantly larger than that of C57BL/6J by approximately 8.5%.

We report a genome-wide suggestive QTL on the proximal arm of chromosome 4 that is associated with total PMBSF area. Mapping was conducted using total PMBSF area values adjusted for brain weight. As a comparison, raw value total PMBSF area was also mapped. There were no major differences between the two maps with regards to the signal on chromosome 4, however the most notable differences are the reduction of two suggestive QTLs on chromosomes 17 and X when adjusted total PMBSF area values are used. It is worth noting that following bootstrap analysis, we detected that one of our phenotyped strains (BXD73) had a large effect on the significance of our suggestive QTL. When BXD73 was removed, the suggestive QTL on chromosome 4 exceeded the significance level. However, we included BXD73 for completion. Furthermore, the shape of the QTL peak was nearly identical whether BXD73 was included or removed.

Using the marker regression tool in WebQTL, we observed four markers on chromosome 4 that were linked to adjusted total PMBSF area. Interestingly, we previously reported two markers (D8Mit145 and D10Mit3) that were associated with total PMBSF/ALBSF area 8. However, only 10 BXD RI strains were utilized for mapping in that study. The much larger power of 42 BXD RI strains for the detection of significant QTLs must be highlighted. In addition, different phenotypes were measured, where the previous study measured the entire representation of the whiskers of PMBSF and ALBSF area; here we focused exclusively on the total PMBSF area.

Given the small number of genes within the suggestive QTL on chromosome 4, we began exploring the genes that may be responsible for modulating total PMBSF area. In the absence of functional biological data, it is difficult to present a compelling case for any particular candidate gene present within a typically large QTL interval. In this study, we examined prospective genes within a fairly small interval spanning 5.5 to 9.0 Mb on chromosome 4. Two of these genes within this interval, (Car8 and Rab2), have a high number of SNPs, and particular biological relevance to total PMBSF area.

Carbonic anhydrase-related protein VIII (Car8 or CA-RP VIII) is a cis-acting gene that is located within our QTL interval starting at 8.07 Mb and contains 174 SNPs. Cis-acting genes are known to regulate their own expression and are considered to be strong candidates for complex traits 2021. Car8 is a member of the carbonic anhydrase gene family which contains 14 isoforms in humans 22, three of which lack enzymatic activity due to missing one or more of the three zinc-binding histidines critical for catalysis 23. Those without catalytic activity are known as carbonic anhydrase-related proteins that include Car8, Car10, and Car11 24. Murine Car8, Car10, and Car11 have high cDNA and amino acid sequence similarity (89.2–94.5% and 97–100%, respectively) with their human homologues, indicating a fundamental biological role for the carbonic anhydrase-related proteins 25. Car8 is expressed in the cell body of neurons in most parts of the CNS 23. Early in human gestation, Car8 is specifically expressed in neuroprogenitor cells in the subventricular zone and observed in neural cells migrating to the cortex 23.

A more recent study has implicated a role for Car8 in motor control in mice 26. The Car8 knockout, known as the waddles mutation, is characterized by wobbly side-to-side ataxic movement, typically seen 2 weeks after birth 26. Thus, Car8 appears to be an important gene that is necessary for CNS functioning.

Rab2 is also a gene of interest that starts at 8.46 Mb on chromosome 4 under the total PMBSF area QTL. This gene harbors a total of 176 SNPs in the BXD cross and has a trans-acing QTL on distal chromosome 1 (LRS > 15), in hippocampus, cerebellum, and whole brain transcriptomes. Rab2 belongs to the Rab subfamily of small GTP-binding proteins whose many members are thought to be involved in aiding in the targeting of cytosolic transport vesicles to their many destinations. Perhaps one of the most studied members of this family is the Ras gene. Rabs work with an array of proteins including GTPase Activating Proteins (GAPs) and Guanine Nucleotide Exchange Factors (GEFs) 27. The Rab2 protein is a resident of pre-Golgi intermediates and is required for protein transport from the endoplasmic reticulum to the Golgi complex and is essential for the maturation of pre-Golgi complexes 28.

Whole-brain mRNA expression profile correlation analysis revealed two genes that correlate significantly with total PMBSF area. The first, adenylate cyclase I (Adcy1), is a neurospecific membrane bound protein that catalyzes the formation of the crucial second-messenger cAMP 17. Expression of Adcy1 is known to be linked with regions of the brain associated with plasticity, including the hippocampus and cerebral cortex 29. Mice with knockout Adcy1 gene (barrelless mutation) do not develop the commonly observed barrel fields in the SI cortex 17. However, such mice are still known to have normal topological organization of the SI cortex 30. In addition to showing that disrupted Adcy1 displays a barrelless phenotype, for the first time, a second messenger was directly linked to cortical development.

A second gene of interest, Gap43, negatively correlated with total PMBSF area (r = -0.52, p = 0.02). Although not highly expressed, we report it for completion and more importantly, for its significant biological role. GAP43 expression following stroke is strikingly similar to staining of GAP43 in first and second postnatal weeks in the barrel cortex (Erzurumlu et al., 1990). GAP43 is shown to be involved in post-stroke axonal sprouting 31 and is important in critical stages of neurodevelopment of the barrel cortex 18. Erzurumlu and colleagues also showed that Gap43 expression pattern was similar to the barrel field pattern, including the PMBSF region 18. They reported the transient nature of GAP43, since almost no expression of GAP43 was observed after post-natal day 7.

A total of 140 mice from 45 strains was used in this study. Phenotypic data were collected from 122 mice from 42 BXD RI strains (n = 2.90 ± 0.22 per strain) ranging between 40 to 61 days of age (average 46.66 ± 0.44 days). All but six strains, BXD 55 (F18), BXD 56 (F14), BXD 71 (F16, F17), BXD 80 (F19), BXD 83 (F17), and BXD 84 (F17), had been inbred for at least 20 generations. The phenotypes of the two parental inbred strains (C57BL/6J, DBA/2J) and one F1 strain (B6D2F1) were also analyzed. A total of 13 animals was phenotyped from the parental strains and 5 animals were phenotyped from the F1 strain. The age range of these mice was similar to that of the BXD strains (average 45.06 ± 0.64 days). While all strains were not balanced evenly with respect to sex, our data contained at least one male and one female from 38 of the total 44 strains phenotyped.

All animals were maintained at a temperature of 22°C on a 12 hr light/dark cycle with 35–40% humidity in a specific pathogen-free (SPF) environment at the University of Tennessee Health Science Center Animal Facility. Animals were fed 5% fat Agway Prolab 3000 rat and mouse chow. All experimental procedures were carried out in accordance with Principles of Laboratory Animal Care (NIH publication No. 86–23, revised 1985) and were approved by the Institutional Animal Care and Use Committee at the University of Tennessee Health Science Center. The Animal Care Facility is AAALAC approved.

Tissue preparation was conducted essentially as previously described 8. Mice were deeply anesthetized with Nembutal (50 mg/kg) and perfused intracardially with 0.9% saline solution followed by 4% paraformaldehyde in 0.3 M sodium phosphate-buffered saline (NaPBS) at a pH of 7.4. The brain was removed from the skull, blocked, hemispheres separated and flattened between two Plexiglas plates, and flattened hemispheres were refrigerated overnight in 4% paraformaldehyde. Body weight and brain weight (with and without cerebellum and olfactory bulbs, cerebellum and brainstem weight, olfactory bulb, and left and right cortices) were measured prior to flattening. The following day, hemispheres were sectioned at 100 μm-thickness using a Vibratome and sections were placed in test tubes containing 0.01 M potassium phosphate-buffered saline (KPBS). Sections were washed in KPBS (3 × 10 min), stained with cytochrome oxidase (CO) to visualize the PMBSF, and mounted on gelatin-coated slides. Slides were left to air dry before they were coverslipped with Permount.

Slides containing PMBSF barrels were viewed under a light microscope (Nikon Optiphot) and photographed with a digital camera (Kodak Coolpix 9000). Photomicrographs containing the PMBSF were transferred and stored on a Macintosh (G5) computer, and the PMBSF reconstructed using Adobe Photoshop CS. In cases where the entire PMBSF did not appear on a single section, blood vessels were used as fiducials to align adjacent sections to produce a composite reconstruction. The total PMBSF was measured using NIH program ImageJ 1.33 u. The PMBSF typically consisted of four straddler barrels (α, β, γ, δ), the first four barrels of rows A and B, and the first five barrels of rows C through E, resulting in a total of 27 barrels. The total PMBSF area was defined by drawing a line around the 27 barrels to produce an outline that was measured using ImageJ. Figure 4 shows a section containing the entire barrel field. All reconstructions and measurements were done by a single experimenter (TJ) to minimize technical error. Measurements were stored in an Excel spreadsheet for subsequent analysis.

Left and right PMBSF areas of all mice were measured and compared against one another as a possible gauge of technical error; our assumption being that R-L asymmetry would be modest compared to technical sources of error. A paired t-test and Pearson's product-moment correlation were used to compare the total PMBSF area for each of the two hemispheres. No significant differences were observed between total PMBSF area of hemispheres (p > 0.60) and right and left PMBSF areas showed a high degree of correlation (r > 0.85). Therefore, we averaged the PMBSF areas for both hemispheres in subsequent analyses.

All basic statistical analyses and tests were carried out using Data Desk (6.1) or Excel. Standard error of means was calculated by dividing the standard deviations by the square root of the number of samples.

We modeled our data based on careful correlation examinations to determine what factors may be used as predictors of PMBSF area. Correlation significance was determined using simple regression p-values. A general linear model was utilized to determine predictors of PMBSF area. Age, sex, body weight, and brain weight were all considered as possible predictors of PMBSF area. The linear model (Data Desk 6.1) that was initially used contained all possible predictors of total PMBSF area. Non-significant predictors were then removed in a sequential manner until only significant predictors were left.

Broad-sense heritability (h²) was measured by comparing between-strain and total variances, using the Hegmann and Possidente method where h² = V_A/[V_A + 2V_E] (V_A = genetic variance and V_E = environmental variance) 32. Broad-sense heritability provides an estimate of the total variance due to genetic factors.

QTL analysis involves classifying strains based on their genotypes at discrete chromosomal markers and comparing these groups with a quantitative variable, total PMBSF area in the present study. If the variation in phenotype matches the differences in genotype, then a gene locus will be detected 3.

Raw and adjusted PMBSF data from BXD RI strains were used to map potential QTLs. QTL maps of brain and body weights were also generated. All QTL maps were generated using the online WebQTL 33. Markers are reported with genome-wide significance and those considered suggestive based on 1000 permutation tests. It is important to note, that a slight variability in significant and suggestive thresholds will be detected each time the maps are recomputed because of the random generation of the permutations. Three types of QTL mapping regressions were utilized in this study and they include: simple regression, interval regression, and marker regression. The likelihood ratio statistic (LRS) was used to assess genome-wide significance in linkage analysis. Logarithm of odds (LOD) values are obtained simply by dividing the LRS by 4.6.

GeneNetwork provides users with an array of analytical tools to compare a given trait with a number of data sets available from other experimenters. Microarray data of gene expression in the brain and data of other phenotypes are two such examples of possible tools. For this study, we correlated our PMBSF area phenotype with the mRNA expression level data "INIA Brain mRNA M430 (Jan 06) PDNN." Pearson's product-moment correlations were utilized. Unfortunately, we could not take advantage of the trait-to-trait correlations for this study given that many of the phenotyped strains here have not been available for a long period of time, thus experimenters have not yet collected a plethora of phenotypes for comparison. However, we did take advantage of other published phenotypes by correlating them with potential candidate genes, specifically, Car8.