Background

For many ocular diseases that result in the loss of vision, the death of retinal ganglion cells (RGCs) is the final common pathway. Glaucoma is one such ocular disease where the sporadic family history and the presence of significant risk factors in select populations suggest that the susceptibility of RGC death is a complex trait [1, 2]. For example, elevation in intraocular pressure (IOP) in open angle glaucoma is strongly associated with an increased likelihood of RGC death. Lowering IOP almost always has the beneficial effect of sparing RGCs. However, some patients with normal or even low IOP develop glaucoma with associated RGC death [3, 4]. The reverse is also true: selected populations of people have very high IOPs and yet do not develop glaucoma or lose RGCs [3]. The fact that some patients with low IOPs develop glaucoma while others with high IOPs do not has led to the hypothesis that critical genetic sequence variants segregating human populations influence the relative susceptibility or resistance to ganglion cell death [5]. One efficient way to measure the influence of sequence variants on complex traits is to compare different inbred strains of mice. For example, Nickells and colleagues [6] studied the differential survival of RGCs in 15 highly diverse strains of mice following optic nerve crush, finding that ganglion cells in some strains were highly susceptible whereas other strains were relatively resistant. This difference demonstrates the importance of genetic background on the complex process of ganglion cell death. Defining the genomic differences between these strains has the potential to lead to novel treatments to prevent ganglion cell loss and preserve vision.

One obvious approach to examining the molecular differences that underlie the susceptibility or resistance of ganglion cells to injury is to use microarray methods to profile the transcriptomes of inbred strains of mice. A considerable amount of published microarray data describes the retina's response to injury in different rodent strains. When one looks across all of these studies, there is a general agreement that changes in gene expression are classic responses of the central nervous system (CNS) to injury [7]. For example, genes that are associated with reactive gliosis, such as Gfap, are often upregulated, whereas neuronal marker genes such as Thy1 are often downregulated [7–13]. Some studies have gone further, focusing on the response of the inner retina to look at regional changes [14]; other studies have used laser-capture microdissection to examine expression profiles of isolated RGCs [15]. The common responses to injury can be observed in a variety of different types of insult to the eye. Earlier work in our laboratory [7] found that many of the changes resulting from mechanical injury to the eye are similar to changes in other models of retinal injury, including ischemia [16], elevated intraocular pressure [8], photocoagulation [17], and photo-oxidative stress [18]. These common responses across different rodent species with different types of insult are interesting, however since they are common to susceptible and resistant strains they are of little use in determining the underling susceptibility to insult.

In this study, we compared the response of the retina to optic nerve crush in C57BL/6J and DBA/2J mice. Both of these strains are widely used in vision research. There is an extensive catalogue of ocular phenotypes and genetic modifications for both strains [19]. Also, these two strains are the parents of the BXD recombinant inbred (RI) strain set, which has been used for more than a decade to study the genetic basis of variations in the structure of the eye, retina, and central visual system [20, 21]. We have comprehensive gene expression data for the eyes of most of these BXD strains [20] and therefore can use all of these array and phenotype datasets to map sequence variants that influence entire genetic networks regulating the response of the retina to injury.

Results and Discussion

To study molecular mechanisms underlying differential response of the retina to injury, we exploited two mouse strains, C57BL/6J and the DBA/2J. The first part of this section covers the differential effect of optic nerve crush on ganglion cell survival. For the anatomical studies the retinas were examined 30 days after the optic nerve crush. This extended period of time allowed us to have a clear picture of the long-term effects of the insult to the optic nerve. In the second part, we describe changes in the transcriptome at two time points after the crush. To define potential changes in the transcriptome that underlie the long-term effects of optic nerve crush, the microarray samples were taken at 2 and 5 days after optic nerve crush. We have supplemented the analysis by generating microarray data of cultured astrocytes from C57BL/6J and DBA/2J strains and by a meta-analysis of array data from existing public databases.

Retinal Ganglion Cell Loss After Optic Nerve Crush

To define difference in the response of the C57BL/6J and DBA/2J mice to optic nerve crush, we examined the retinas of normal mice and mice 30 days after optic nerve crush (Figure 1). The most obvious change was a dramatic decrease in the number of NeuN-stained cells in the retinas of both strains when the nerve was crushed. We counted the number of NeuN-labeled cells in the ganglion cell layers of control retinas and retinas from animals 30 days after optic nerve crush for both the C57BL/6J mouse and the DBA/2J mouse. We sampled between 14 and 18 fields from 5 retinas for each strain to define the density of NeuN positive cells per mm2 (see methods section). In the control C57BL/6J mice, the average number of NeuN-positive cells (ganglion cells plus the population of displaced amacrine cells) was 5012 cells/mm2 with a standard error of 317 cells/mm2. The control DBA/2J mice had 4295 cells/mm2 with a standard error of 187 cells/mm2. This difference in the density of NeuN-positive cells between the two strains is significant using the student t-test (p < 0.04).

Figure 1
figure 1

Effects of optic nerve crush on the retinas of C57BL/6 and DBA/2J mice. The effect of optic nerve crush on the survival of retinal ganglion cells is shown in 4 panels. C57BL/6 retinas are shown in A (control) and B (optic nerve crush); DBA/2J retinas appear in C (control) and D (optic nerve crush). Note the significant decrease in ganglion cell staining after optic nerve crush. Ganglion cells were stained with NeuN. All panels are at 40× magnification. The scale bar in panel D represents 50 μm.

Full size image

In our data, we see a significant difference in the number of NeuN-positive cells between the C57BL/6J and the DBA/2J strains. When one examines the literature for comparative numbers, Williams et al. [22] calculated the axon number to be a mean of 54,600 ganglion cells per retina for the C57BL/6J strain (pooled data from pigmented and coisoenic albino mice) and 63,400 ganglion cells per retina for the DBA/2J strain. If this is converted to cell density per mm2 the numbers are 2,884 cells/mm2 for the C57BL/6J strain and 3,254 cells/mm2 for the DBA/2J strain. In the study by Buckingham et al. [23] they used NeuN staining and cell counting similar to that used in the present study and they arrived at very similar numbers (see Figure 2 in Buckingham et al., 2008). Extrapolating from their graph they see 4,800 cells per mm2 for C57BL/6J and 4,200 cells per mm2 for the DBA/2J mouse [23]. These numbers are very similar to the ones from the present study. It is hard to say why there are differences between studies. The most parsimonious explanation is that differences seen from study to study may be related to the sampling method.

Figure 2
figure 2

Quantification of loss of retinal ganglion cells. The percent survival of retinal ganglion cells following optic nerve crush (ONC) is shown for two strains of mice, C57BL/6J (B6) and DBA/2J (D2). The data points represent the percent survival for each of the retinas that received a crush of the optic nerve. The means of each group are indicated by the horizontal lines. This number was calculated by dividing the density of NeuN-positive cells optic nerve crush retinas by the average density of the control retinas of the same strain. There was a statistically significant (t-test, p < 0.01) difference in the survival of NeuN-positive cells, with 54% survival in the C57BL/6J mouse and 62% survival in the DBA/2J mouse.

Full size image

After the optic nerve crush, the average density of NeuN-positive cells decreased to 2694 cells/mm2 in C57BL/6J retina and to 2677 cells/mm2 in DBA/2J retina. To determine if there is a difference in the response of the retinal ganglion cells to injury, we calculated the percent survival for each optic nerve crushed retina, dividing the density of retinal ganglion cells in that retina by the average density in the control retina of the same strain. The disparity in survival between the two strains is shown in Figure 2. There was a 54% survival of NeuN-positive cells in the C57BL/6J mouse, while the DBA/2J mice had a 62% survival of neurons. This difference in NeuN-labeled cell loss is significant using a student t-test (p < 0.01). Since we used NeuN to label cells in the ganglion cell layer, it is possible that displaced amacrine cells were included in our counts. The displaced amacrine cells represent a diverse population of cells [24, 25]. There are strain differences in the number of amacrine cells in the mouse [26, 27]. The strain differences include the number of ChaT-positive displaced amacrine cells [26]. In the C57BL/6 mouse the ChaT-positive amacrine cells are estimated to be 19% of the total population of displaced amacrine cells [28]. Thus, the neurons surviving in the ganglion cell layer may include both ganglion cells and some displaced amacrine cells. It is worth noting that Buckingham et al. [23] directly addressed this question and found that a consistent proportion of ChaT-positive displaced amacrine cells expressed detectible levels of NeuN.

When we compare our data to others, we find that Li et al. [6] did not observe a statistically significant difference in survival following optic nerve crush between the C57BL/6J mouse and the DBA/2J mouse. It is interesting to note that the DBA/2J trended higher than the C57BL/6J (see Figure 1 of Li et al.) [6]. In our study, we found a modest, but statistically significant differences between the two strains. This could be due to differences in techniques to crush the optic nerve, differences in staining, or the fact that we used control mice and Li et al. [6] used the contralateral eye as a control for optic nerve crush [6]. The small difference in survival we observe may have significant ramifications to our microarray studies and will allow us to map genetic networks using the BXD recombinant inbred mice along with the powerful bioinformatics tools in GeneNetwork http://genenetwork.org[29].

Changes in the Transcriptome After Optic Nerve Crush in the C57BL/6J and DBA/2J Strains

Microarrays were run on both the C57BL/6J and the DBA/2J strains with RNA isolated from control mice and mice 2 and 5 days after optic nerve crush. The first step in the data analysis was to select a set of genes for analysis. For this analysis we used Significance Analysis of Microarrays (SAM, Stanford University, http://www-stat.stanford.edu/~tibs/SAM). We made five biologically meaningful comparisons to define groups of genes that change in the two strains following optic nerve crush. The first comparison was to define genes that had significantly different levels of expression between the two normal retinas, control C57BL/6J retina and control DBA/2J retina (522 probes). The next four comparisons were done to select genes that were differentially expressed after optic nerve crush, comparing the control samples to the 2-day crush and 5-day crush samples of each strain for both the C57BL/6J retina and DBA/2J retina. This will define genes that are changing following optic nerve crush: C57BL/6J control versus 2 day after injury (49 probes) and versus 5 days after injury (1007 probes) and DBA/2J control versus 2 days after injury (0 probes) and 5 days after injury (52 probes). After the duplicate probes were scrubbed, a total of 1,580 probes from the Illumina array were selected with a false discovery rate less than 0.03. The next phase of the analysis was to define functional clusters of genes using a principal component analysis. Using the CLUSFAVOR v6.07 program [30], we identified clusters of genes based on their expression patterns (Figure 3). This analysis resulted in 5 eigenvectors dividing the dataset into 10 groups, with a positive and negative component for each vector (Figure 3). The 10 clusters account for all of the variability in the dataset. A number of different data-mining tools were used to extract biologically meaningful information from these clusters. The web-based software used included the National Center for Biotechnology Information web site for PubMed [31] and Entrez Gene databases [32], Chilibot for searching PubMed relationships [33], Genomatix for searching for possible transcription factors [34], Transfac 7.0 for transcription factor mining [35], UCSC Genome Bioinformatics for its genome browser [36], WebGestalt for its annotation abilities [37], Gensat [38], and GeneNetwork [29].

Figure 3
figure 3

Principal component expression patterns. The average gene expression profiles are shown for all 10 clusters. In PC1, the genes having a common expression profile in the C57BL/6J (B6) and DBA/2J (D2) strains are clustered. PC1 positive represents genes in both strains that have decreased expression after ONC. PC1 negative has the reverse of this pattern, with genes increasing in expression. PC2 appears to represent genes that are differentially expressed in the two strains. These gene cluster patterns are almost mirror images of each other. In the remaining clusters (PC3 to PC5), very different patterns appear to represent interactions between strain differences and the effects of ONC.

Full size image

To complement this work, we generated a microarray dataset for astrocytes isolated from the C57BL/6J and the DBA/2J cortices. This astrocyte data set was used to generate three lists of transcripts. We used two independent approaches to create the lists. The first approach compared the average expression of all of transcripts in the 6 (3 C57BL/6J and 3 DBA/2J) astrocyte micorarrays to the average of the gene expression in the 6 normal retina samples (3 C57BL/6J and 3 DBA/2J). The first list is a group of 325 genes, which were 20-fold enriched in the astrocyte sample relative to the retina. We considered this list of genes to be an astrocyte signature. Of these 325 genes, 36 were present in the selected genes of the optic nerve crush dataset (See additional file 1: Genes with ≥ 20 fold change between astrocyte and control retina). The second list of genes, totaling 379, contained transcripts that were 20-fold enriched in the retina relative to the astrocyte sample, was classified as a non-astrocyte signature. Of these 379 genes, 44 were present in the optic nerve crush dataset (See additional file 2: Genes with ≤ -20 fold change between astrocyte and control retina). The third list of genes, totaling 934, was created to identify transcripts that were differentially expressed in cultured C57BL/6J astrocytes relative to DBA/2J astrocytes. In this list, 321 genes were more highly expressed in C57BL/6J astrocytes and 613 genes were more highly expressed in DBA/2J astrocytes. Of the 934 genes, 186 were present in the optic nerve crush dataset (See additional file 3: C57BL/6J astrocyte vs. DBA/2J astrocyte fold change). The first two of these three lists were combined and considered to be differentially expressed genes between the two types of astrocytes. These lists of genes were used in our analysis of the differential response of the C57BL/6J and DBA/2J retina to optic nerve crush.

Of the 1,580 genes in our optic nerve crush data set, 9 were selected and examined by real-time qRT-PCR. These genes were Bcl2a1a, Chrna6, Cryaa, Crybb2, Crym, Egr1, Gfap, Sox11, and Thy1. (See additional file 4: Oligonucleotides used for Real-Time qRT-PCR). In general, the patterns of change for the genes monitored by real-time qRT-PCR were similar to those observed in the averaged microarray datasets. The levels of Gfap were similar in comparison between the control data versus 2-day injury data, as well as the comparison between the 2-day versus 5-day injury data. Other genes, such as Bcl2a1a, Chrna6, Cryaa, Crybb2, Crym, and Thy1, were similar in the overall trend from control to 5-day. However, the 2-day versus 5-day fold change was not always in the same direction, although the differences were small. In other cases, there were slight differences in the fold changes, as with the Egr1 gene, for which the real-time qRT-PCR indicated a +0.47-fold change between the C57BL/6J 2-day versus 5-day, the microarray result was -0.8 for the same condition. Egr1 had a -0.35-fold change in the DBA/2J between the 2-day versus 5-day, but the microarray data yielded +1.42. Nonetheless, the agreement between the microarray data and real-time qRT-PCR data was excellent; indicating that for the genes tested, the microarray provided an appropriate measure of transcript level. Once the quality of the dataset was proven, we clustered the data into groups and began our functional analysis.

Most genes are found in the first principal component (PC1), which contains 823 genes or 44% of the total variance in the dataset. The genes in PC1 change similarly in response to optic nerve crush in both C57BL/6J and DBA/2J mice. As shown in Figure 3, the expression level for genes in PC1 is approximately equal in control C57BL/6J mice as compared to the DBA/2J mice. For PC1 positive, the genes are downregulated 2-days after optic nerve crush and even further downregulated 5-days after nerve crush. A brief scan of this list of genes (See additional file 5: Genes of principal component 1 (PC1)) reveals that many genes, such as Thy1 and Chrna6, are genes associated with ganglion cell injury and death. To provide an alternative method of determining the cell types represented in PC1 positive, we examined the labeling patterns of the GENSAT project web site [38]. GENSAT describes a series of mouse strains that label sets of cells within the mouse brain that express trans-genes under labels of gene-specific promoter constructs. Of the 407 genes in PC1 positive, 28 were found in GENSAT, with 24 genes labeling neurons in the brain and 5 (Npc1, Pax6, Chrnb3, Htr1d, and Chrna6) specifically labeling axons in the optic tract. These data indicate that PC1 positive represents genes that are involved in the generalized decrease in neuronal markers, specifically ganglion cells, in response to optic nerve crush.

PC1 negative is also composed of genes with approximately equal expression in control C57BL/6J and DBA/2J mice. However, both sets of genes are upregulated 2-days after optic nerve crush and upregulated even further at 5-days after crush (Figure 3). Genes in this group are associated not only with reactive gliosis, but with neuronal genes in an attempt at abortive regeneration (See additional file 5: Genes of principal component 1 (PC1)). Glial fibrillary acidic protein (Gfap), the hallmark for reactive gliosis, is found in this group. Many of the genes in this list are also markers for astrocytes. These include Gfap, the penultimate astrocytic cytoskeletal marker, and Sox11 a known astrocyte transcription factor.

To further evaluate astrocytes marker genes, we examined our data from cultured astrocytes from C57BL/6J and DBA/2J mice. In one analysis, we identified genes that were enriched in astrocytes as compared to the normal retina. Of the 36 astrocyte signature genes in our dataset of 1,580 probes, 27 are present within PC1 negative, indicating that PC1 negative contains an astrocyte signature. We also observed genes involved in abortive regeneration, among them growing axon protein 43 (Gap43). In examining the labeling patterns on GENSAT, 21 of the 314 genes in PC1 negative were found. Four of the PC1 negative genes (Gfap, Stat3, Ank2, and Vcam1) labeled glial cells and two of the genes (Rax and Snap91) labeled axons in the optic nerve. The changes observed in PC1 are common to CNS injury and are associated with the experimental condition of axonal injury caused by optic nerve crush. This represents the most variance in the microarray dataset and, as expected, it is due to the experimental condition. When these results are compared to other microarray studies that examined the effects of injury in the retina, there is a surprisingly similar list of genes. These are injury response elements, many of which are found in the retina after a variety of insults: ischemia [16], elevated IOP [8], photocoagulation [17], photo-oxidative stress [18], or direct retinal injury [7].

A recent publication [39] demonstrated the importance of the mTOR pathway, not only in regulating axonal regeneration in the optic nerve, but also the survival of retinal ganglion cells. A functional knockout of Pten, the negative regulator of mTOR, promoted axon regeneration and the survival of RGCs after optic nerve injury [39]. In our dataset, we observed various changes in 25 genes within the PTEN/mTOR pathways. Furthermore, PTEN itself is significantly upregulated and is in PC1 negative. These findings clearly show that some of the molecular changes associated with the abortive regenerative response of the PTEN/mTOR pathways are associated with the changes we observed in our study. The presence of 14 of these genes in PC1 positive indicates that they have a generalized function in the abortive response of RGCs. Future studies will focus on the importance of this pathway following injury to the optic nerve.

The next principal component (PC2) appears to group genes having different levels of expression in the two strains of mice. PC2 contains 496 genes, which represents 23% of the variance in the dataset (See additional file 6: Genes of principal component 2 (PC2)). The most obvious difference is the level of gene expression between the C57BL/6J control group and the DBA/2J control group (Figure 3). In PC2 positive (223 probes), the expression level in C57BL/6J mice is low, while the DBA/2J mice have high levels of expression. Furthermore, the pattern of expression after optic nerve crush in the DBA/2J mouse is virtually a mirror image of the C57BL/6J, with expression decreasing 2-days after crush in the DBA/2J mice and increasing at 2-days after crush in the C57BL/6J mice. The PC2 negative (270 probes) is almost the reverse of the PC2 positive component, with the expression level in C57BL/6J mice being high and the expression level in DBA/2J mice being low. To investigate further, we rank-ordered the fold change between the C57BL/6J and DBA/2J control retinas. Among the genes in PC2 in this ranked list, 407 (82%) of the 496 probes were in the top 5% of the most variable genes in the dataset. The most parsimonious explanation is that the genes in PC2 are differentially expressed in the C57BL/J and DBA/2J retinas and are not differentially affected by optic nerve crush.

To test the hypothesis that PC2 represents genes that differ between the two strains, we examined the Hamilton Eye Institute Mouse Eye Database (HEIMED) on GeneNetwork [20] to determine whether any of the genes in PC2 had a significant quantitative trait locus (QTL) or were part of a gene network within the eye. Most of the genes had strong cis-QTLs and were not part of an overall genetic network. The large cis-QTL found in GeneNetwork also points to the fact that these are genes with differences in expression in C57BL/6J and DBA/2J mice, the parental strains of the BXD RI strain set. The gene network analysis and expression patterns led us to the same conclusion.

This being the case, one would predict that these differences are present in different tissues from these two strains. Thus, we examined the microarray databases generated from C57BL/6J and DBA/2J cultured astrocytes. A surprising number of genes are differentially expressed between the C57BL/6J and DBA/2J astrocytes and are also present in PC2. Of the 184 differentially expressed astrocyte genes found among the 1,580 genes of the Optic Nerve Crush dataset, 162 were in PC2. This indicates that the differential expression between the two strains is also in astrocytes as well as the retina. Furthermore, genes that are more highly expressed in DBA/2J astrocytes are highly expressed in the DBA/2J retina, while genes that are expressed at higher levels in C57BL/6J astrocytes are highly expressed in the C57BL/6J retina.

If PC2 represents genes that are differentially expressed, one would predict that these genes would be expressed in a variety of different cell types which is in fact, the case. The genes in PC2 are found in all cell types within the CNS. A total of 30 genes were found in the GENSAT database. In PC2 positive, 3 genes (Slc7a14, Cacng5, and Susp3) were expressed in optic nerve axons and neurons. In PC2 negative, one gene (Casp9) was observed labeling axons in the optic nerve. In addition, neurons were labeled in the brain by 9 additional genes, including Mtap1b, Lypd1, Cdon, D430039no5Rik, Dap3, Dcnq2, Rgs16, Tac2, and Tph2. Both PC2 negative and positive contained glial genes; 2 (Sf1 and H2-D1) were in PC2 positive and 4 (Dusp16, Fcer1g, Prom1, and Hdc) were in PC2 negative. Thus, PC2 appears to represent genes that are differentially expressed in the C57BL/6J and DBA/2J retinas. These genes do not appear to be related to any specific cell type or function.

The most intriguing functional components are within PC3, PC4 and PC5 (representing 33% of the variance in the dataset) (See additional file 7: Genes of principal component 3 through 5 (PC3-PC5)). These components represent interactions between the effects of optic nerve crush injury and the genetic background of the two mouse strains. If there is a signature for susceptibility or resistance to ganglion cell death, or if there is a signature of reactive gliosis, then it lies within these components. Accordingly, we used all of the bioinformatics tools at our disposal to identify a meaningful association between the genes within each component. First, to determine if there was a molecular signature of a specific cell type within any of the PCs, we searched cell-type-specific profiles in Cahoy et al. [http://www-stat.stanford.edu/~tibs/SAM) to calculate a list of transcripts we considered to be differentially expressed across our datasaet. We examined differences between the control samples and between the controls and either 2 days after injury or 5 days after injury for each strain. All of the transcripts had a false discovery rate below 0.03. The duplicates were scrubbed resulting in a final 1580 genes. The 1,580 SAM significant transcripts were entered into CLUSFAVOR 6.0 (Departments of Medicine, Molecular and Human Genetics, and Scott Department of Urology, Baylor College of Medicine, Houston, TX) for pattern analysis.

Microarray Confirmation Through Real-Time qRT-PCR

Selected genes from the microarray datasets, including Bcl2a1a, Chrna6, Cryaa, Cryba2, Crybb2, Crym, Egr1, Gfap, Lpin1, Rho, Sag, Sox11, and Thy1 as well as Actb as a control, were validated using real-time qRT-PCR, which was done on the Roche LightCycler 480 system (F. Hoffmann-La Roche, Switzerland). To normalize the data, Actb was run as a housekee** gene. Each gene was run under 5 concentrations in duplicate along with negative control, neg-RT, as well as water for controls. Assays were designed on the Roche web site [82]. The primers were synthesized by Integrated DNA Technologies (See additional file 4: Oligonucleotides used for Real-Time qRT-PCR).