A population of 196 F2 mice underwent optic nerve crush and cell loss phenotyping. The distribution of phenotypes in this population (Fig. 1) matched well with a smaller population of F2 mice examined previously 20, indicating that the phenotype was consistently inherited. For genome wide mapping, we selected mice from the F2 population that had 54% or less cells remaining (the most susceptible mice) and 64% or more cells remaining (the most resistant mice). These animals are indicated by the filled bars in the phenotype frequency histograph of the F2 population (Fig. 1).

Quantitative trait linkage analysis using 65 microsatellite markers, with and without sex as a covariate, showed a single significant peak on chromosome 5 between markers D5Mit254 (34 cM) and D5Mit338 (59 cM) (Fig. 2). The maximum LOD score predicted by interval analysis was 5.825 at Chr5.loc38 (inset Fig. 2), which was the only locus that exceeded the significance level of P = 0.05 (P = 0.0011 for the peak). The next highest LOD score in the analysis of all the samples modeled without interactive covariates was 3.045 for D6Mit83 on chromosome 6 (Chr6.loc3.5). It is notable that evaluation of larger numbers of animals from the F2 population may have not only increased the significance of the locus on chromosome 5, but also increased the LOD score at Chr6.loc3.5 to above the threshold value. The QTL dataset was also analyzed for interacting loci, again including sex as a covariate. No significant associations were detected at the resolution of this genome-wide screen, thus a single locus on chromosome 5 accounts for 11.61% of the phenotypic variance of this trait. This locus has been designated as Retinal ganglion cell susceptible 1 (Rgcs1). For subsequent analyses, we have conservatively defined this locus by the nearest markers used for mapping. This 58 Mb region includes an approximation of the 95% confidence interval as suggested by a 1.5 LOD drop from the peak 21.

We also examined the inheritance pattern of phenotype as a function of marker D5Mit254, which lies closest to the region with the statistical maximum LOD score (Fig. 3). Mice inheriting 2 BALB/c alleles have significantly fewer cells remaining than mice inheriting 1 or 2 DBA/2J alleles (t-test, P = 4.4 × 10^-7). Conversely, mice heterozygous for the DBA/2J locus exhibited a statistically equivalent phenotype to mice homozygous at this locus (P = 0.169). Together, these data are supportive of our original observation of the Mendelian-dominant nature of the Rgcs1 locus obtained from reciprocal breeding experiments 20.

Genomic sequence information, available for multiple different strains of mice (please see Availability & requirements for more information), was examined to identify candidate genes. The region of Chr5.loc34-59 was found to be relatively gene poor, with a total of 578 known genes and/or expressed sequence tags (ESTs) identified. Based on existing sequence data, we used in silico analysis to find single nucleotide polymorphisms (SNPs) creating non-synonymous amino acid changes between the DBA/2J and BALB/cByJ strains. This search identified 36 amino acid changes in 25 genes (Table 1). Of these changes, 12 were conservative changes and 24 were non-conservative, including 2 that change a hydrophobic amino acid in DBA/2J mice to a proline in BALB/c mice (genes Tlr1 and Ugt2a3).

We also examined the list of polymorphisms for similarities to other mouse strains with cell death phenotypes similar to either the resistant or susceptible strains. Of the 25 genes identified, 2 genes clearly exhibited strain clustering of a polymorphism, while 7 other genes (totaling 8 SNPs) showed possible clustering. The basis for this latter definition was that only partial sequence data was available for reference strains or intermediate strains (see Methods). Of these 9 potential genes, 7 exhibited non-conservative amino acid changes.

A second layer of analysis was then conducted to identify genes in this region that were highly expressed or enriched in cells of the ganglion cell layer, including ganglion cells themselves, by screening available microarray data. Four genes in the Rgcs1 region were found as being either highly enriched, or highly expressed, in the ganglion cell layer and optic nerve head, including Pcdh7, Uchl1, Sparcl1, and Cplx1 (Table 2). Of these 4, Sparcl1 appeared on both lists of SNP-containing genes and genes expressed in the ganglion cell layer. All 4 of these candidate genes showed a decrease in expression level in arrays of axotomized or glaucomatous rat retinas 22.

Finally, a search of the 578 genes was made to identify candidates that had already been identified as susceptibility alleles for other neurodegenerative diseases. This search yielded 2 genes, HspB8 and Uchl1. The latter gene was also identified in the search of genes present in the retinal ganglion cell layer. Collectively, using the criteria of expression in the retina, ganglion cell layer, or other CNS neurons; and/or the presence of strain-specific non-conserved amino acid changes; and/or the involvement of genes in neurodegeneration, we identified 7 genes as possible candidate genes that could affect retinal ganglion cell death after lesion to the optic nerve (Table 2).

The 7 candidate genes (Table 2) are distributed in two main clusters, each at either end of the 25 cM interval currently defining the Rgcs1 locus (Fig. 4). In response to a lesion of the optic nerve, each candidate gene could realistically play a role in the process of retinal ganglion cell death. However, this does not preclude the possibility that a different gene (or genes) in this region may be responsible for the QTL. Also, it is important to note that our evaluation of candidate genes did not address the possibility that the causative gene could be an expression QTL, where polymorphisms in the promoter region could affect expression levels. Alternatively, non-coding polymorphisms could also create alternative splice variants in some genes that can affect protein structure or transcript stability. Lastly, our in silico analysis did not address the possibilities of differences in non-coding RNAs (such as microRNAs) that may exist between the two strains, but which could also affect gene expression and cellular processes. Still, the filters we used for selection criteria were biased towards assessing relevant features that could dramatically account for a QTL between the two strains. A summary of each gene follows:

Although Rgcs1 affects ganglion cell death after acute optic nerve crush, the role of the causative gene in this region as a susceptibility allele in glaucoma is not yet resolved. Acute nerve lesion and elevated IOP appear to activate a similar intrisinc apoptotic program, associated with Bax-dependent mitochondrial changes 141547 and the activation of the caspase cascade 1748495051. Additionally, genome wide microarray studies conducted on early, intermediate, and late retinal gene expression changes from both axotomy injured rat eyes and eyes with experimental glaucoma showed significant overlap in the genes that were differentially expressed in both conditions 22. These changes included early injury response genes (i.e., cJun and Junb), and stress response genes (i.e., Hsp27 and Ceruloplasmin). Additionally, ganglion cell specific genes, such as Thy1 and Sncg, were similarly down-regulated. There were specific differences in the pattern of differentially regulated genes, however, that may reflect different mechanisms of ganglion cell pathology between the two modalities of injury. Most prevalent was the up-regulation of genes involved in neuroinflammation, such as components of the complement cascade, in the experimental glaucoma retinas. Thus, neurinflammation may play a greater role in ganglion cell death during a chronic neurodegenerative process, than in an acute paradigm.

Interestingly, DBA/2J mice, which carry the resistant Rgcs1 genotype, exhibit a glaucoma phenotype characterized by progressive ganglion cell loss 525354. Since the dominant Rgcs1 allele does not completely abrogate cell death, we would not expect this strain to be completely resistant to glaucoma. Instead, we speculate that DBA/2J mice would exhibit a more severe phenotype if they carried the BALB/cByJ allele and experiments to create such a congenic strain are underway.

The role of Rgcs1 in glaucoma may also be inferred by comparison of mouse locus with the syntenic region in the human genome. The interval of interest on mouse Chr5 is principally represented on human Chr4 between 4p15 and 4q21 and a short segment of human Chr12, centered at 12q24 (Fig. 4). To date, no linkage studies examining forms of glaucoma with Mendelian inheritance patterns have identified genes or loci in these regions 855. This is not surprising since these studies have depended on relatively rare forms of glaucoma, while most forms of primary open angle glaucoma exhibit much higher levels of genetic complexity 8. We predict that the retinal ganglion cell susceptibility allele, if it does influence cell loss in glaucoma, would be just one of several genes that contribute to a more complex genetic disorder.

Interestingly, markers on human Chr4 that map near the region of interest (Fig. 4), showed promising results in a genome-wide screen of DNA from 113 sib-pairs affected with primary open angle glaucoma 56. In particular, marker D4S2397 (4p15.2) was initially positive in 3 independent analyses, including model-dependent LOD scores and sib-pair maximum LOD scores. Marker D4S400 (4q21.22) was positive in 2 of 3 analyses. These markers on Chr4, however, failed to continue to show positive results when this analysis was expanded to include a second cohort of affected sib-pairs. Given the small size of this sib-pair screen, it may not yet be possible to interpret the relative importance of this region in human glaucoma, but it is encouraging that it has shown some promising results in linkage studies.

All mice were handled in accordance with the Association for Research in Vision and Ophthalmology statement for the use of animals for research and experimental protocols were approved by the Animal Care and Use Committee of the University of Wisconsin. DBA/2J and BALB/cByJ parental mice were purchased from the Jackson Laboratory (Bar Harbor, ME). These mice were bred to generate F1 offspring, which were then interbred by reciprocal crosses to generate F2 mice for analysis. Mice were housed in microisolator cages and kept on a 12 hr light/dark cycle. They were maintained on a 4% fat diet (8604 M/R, Harland Teklad, Madison, WI).

To estimate the size of the mapping population required to successfully identify loci associated with the cell death phenotype, we conducted power calculations using phenotype and inheritance information acquired from earlier studies of smaller populations of N2 backcrossed and F2 mice 20. The mouse genome size was approximated as 20 chromosomes of 100 cM in length and the proposed mapping interval was 20 cM. Since parental strain data provided no information regarding the dominance effect (δ), power calculations were made using the assumptions that δ = α (where α = the additive effect of phenotype) or δ = 0 (δ had no additive effect). Heritability of the locus of interest (h²) was then calculated using the information provided from our preliminary data sets (h² = 32.7% for δ = α, and 24.5% for δ = 0). Power calculations were performed using the formula and parameters described previously 57. Within the heritability range of our existing population studies, these calculations suggested that a population of 100 F2 mice would have sufficient power to detect a single locus that associates with phenotype using this mapping strategy. To further increase the likelihood of detecting a QTL, we also biased the genotyping of the population to the most extreme ends of the distribution 58. Consequently, 196 mice were phenotyped and from this, 60 were selected as the most resistant and 49 were selected as the most susceptible.

Consistent with our previous studies on other inbred mice, we aged the F2 pups to 8 weeks, at which time they underwent the optic nerve crush protocol. Optic nerve crush was performed on the left eye of each mouse using an intraorbital approach 2059. Previously, mice were euthanized 2 weeks after this procedure. However, time course analysis of cell loss over a 3 week period showed the same difference in cell loss between the 2 parental strains at both 2 and 3 weeks (data not shown). Since 3 weeks represented a more end-stage point of the cell death process, we opted to analyze the mapping population at this time after crush. Retinas were processed for quantification as described previously 20. Approximately 10% of the entire population of neurons in the ganglion cell layer was sampled from each retina. Cell loss for each mouse was recorded as the percentage of cells remaining in the experimental retina relative to the fellow control retina. Previous studies have verified that DBA/2J and BALB/cByJ mice have an equal percentage of ganglion cells making up the neuronal population of the ganglion cell layer 20, and we have made the assumption that this relationship holds true to F2 mice generated from these parental strains.

More than 100 informative microsatellite markers between DBA/2J and BALB/cByJ mice were identified from the Mouse Genome Informatics (MGI) website of the Jackson Laboratory (please see Availability & requirements for more information). These markers were spaced roughly 20 cM apart and exhibited at least a 5 bp difference between the strains. Primer pairs for each marker were individually tested on genomic DNA isolated from spleens of parental mice. PCR conditions were empirically optimized for each primer pair, but typically primers were annealed at 55°C and reactions were run for a minimum of 40 cycles. Bands were evaluated on standard 1–3% agarose gels stained with ethidium bromide. After empirical testing, we selected 65 markers for genotyping. This included a minimum of 3 markers per chromosome and an approximate genome-wide scan radius of 15 cM with each marker separated by approximately 30 cM. At this distance, the proximal ends of chromosomes 1, 9, and 13, and the distal ends of chromosomes 3, 5, 11, and 19 were not fully covered by the markers used. Genotyping was performed using spleen DNA isolated from 60 resistant and 49 susceptible mice. Data were scored by an observer (JAD) masked to the phenotype and entered into an Excel (Microsoft, Redmond, WA) spreadsheet. Discrepancies in genotype reading were resolved by regenotyping.

The resulting data set (of phenotype and genotype for all 109 mice) was analyzed using the R/qtl statistical software package (please see Availability & requirements for more information) 60. Simple interval mapping adjusting for possible sex by genotype interaction was performed across the genome with 2cM steps. A 5% permutation threshold 61 was estimated as 4.15 based on 10000 sets of permuted genotypes.

Chromosomal regions of interest identified from interval mapping analysis were analyzed further for potential candidate genes using an in silico approach. All known genes and ESTs present in the region were identified from mouse genomic sequence data present in the MGI database (please see Availability & requirements for more information). This database also included supplementary information on localization and known function of gene products. Candidate genes were identified using three levels of criteria.