Differential gene expression in mouse retina related to regional differences in vulnerability to hyperoxia

Animals and hyperoxic damage

C57BL/6J mice, a hyperoxia-vulnerable strain [21], were used for this study. Mice were raised in dim cyclic illumination (12 h at 5 lx and 12 h in dark) and maintained in this level of lighting for the duration of the experiment. At the postnatal age (P) of P83–P90 days, some mice were exposed to constant hyperoxia (75% O₂) for 14 days. During exposure, the mice were given free access to food and water and were housed in plexiglass chambers, in which the oxygen level was controlled by a feedback device (OxyCycler; Reming Bioinstruments, Redfield, NY).

RNA isolation and purification

Oxygen-exposed animals were euthanized at the end of the period of hyperoxia; controls were euthanized at the same age. Retinas were removed and divided into superior and inferior halves. Retinal samples from two animals (one male and one female) were pooled for each condition (superior and inferior retina from the controls, superior and inferior retina exposed). RNA extraction was performed using TRIzol Reagent (Invitrogen™ Life Technologies, Carlsbad, CA) and an RNAqueous-Micro Kit (Ambion, Foster City, CA). TRIzol was used to isolate the RNA and the RNAqueous-Micro Kit was used to purify and DNase-treat the RNA. Retinas were quickly placed into a 1.5 ml tube containing 200 µl of TRIzol. Following the homogenization of retinas on ice, a further 660 µl of TRIzol and 160 µl of chloroform were added to the tube. The tube was vortexed for 20 s and allowed to stand for 7 min at room temperature. The tubes were centrifuged at 13,000× g for 10 min at 4 °C. The supernatant was then removed and placed into a clean 1.5 ml tube with half its volume of 100% ethanol. The tube was vortexed briefly before the contents underwent purification and DNase treatment, as detailed in the RNAqueous-Micro Kit manual. Purified DNAase-treated RNA was analyzed using a ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE) and a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA), to determine the quantity and quality of the sample. RNA samples were used only if the A₂₆₀/A₂₈₀ ratio was above 1.9 and the RIN (RNA integrity number) was greater than 8.5.

Affymetrix microarray analysis

The Affymetrix (Santa Clara, CA) Mouse Gene 1.0 ST Array was used; this is a whole transcript-based array that interrogates 28,853 well annotated genes. Staining, hybridization, washing, and scanning of the array were performed at the Biomolecular Resource Facility at the John Curtin School of Medical Research, Australian National University, following manufacturers’ protocols. Two experimental replicates were run independently, providing duplicates for each condition. Gene expression data were stored in .cel format files that could be interpreted by array analysis software.

Quality assessment and hierarchical clustering

Affymetrix^® Expression Console™ software was used to supply probe set summarization and calculate quality assessment metrics. Gene expression data were uploaded into the Expression Console™ software to run a gene level core analysis. Once the analysis was complete, the numeric values of quality control metrics were summarized in a report and visualized in line graphs. Hierarchical clustering of data sets was performed using the heat map function in the Partek Genomics Suite (Partek Inc., St. Louis, MO). It uses log₂ signal values. The following parameter settings were used: Method - average linkage; Dissimilarity - Euclidean; Clustering method-agglomerative.

Statistical analysis

Gene expression files in .cel format were imported into the Partek Genomics Suite, with the following normalization of their configuration: Probes to import - interrogating probes; Probe filtering - include core; Background correction - RMA background; Log base - 2; Probe set summarization - Median Polish. Statistical differences between experimental groups were examined using a t-test (post-hoc contrasts).

Gene expression in the inferior and superior regions of control retinas

Gene expression was compared between the superior and inferior regions of the control retinas using statistical and fold change analysis. In this comparison, there was no experimental procedure applied; therefore, the conventional term “fold change” seems inappropriate. Instead we use “fold difference.” Fold difference (FD) is then:

FD=E2/E1 when expression is greater in the second of two experimental (E) samples, FD=- E1/E2, when expression in the second sample is less.

In these equations, E1 is the level of expression in the first (usually the control) sample, and E2 is the level in the second (usually treated) sample. To apply this convention to the present samples, we chose the inferior retina as sample 2 (E2) and the superior retina as sample 1 (E1).

Data were then analyzed using two-sample t-tests for significant differences (increases and decreases) between the inferior and superior retina in the expression of all probes.

Gene expression in the inferior and superior regions of hyperoxia-exposed retinas

To identify such genes, we analyzed the four samples arising from the two variables in this part of the study: treatment (hyperoxia versus control) and site (inferior versus superior). Each sample was available in duplicate. A 2-way ANOVA test applied to the eight data sets provided three p values for each gene, one giving the probability that the gene was not regulated by hyperoxia in the superior retina, the second giving the probability that the gene was not regulated in the inferior retina, and the third (the “combination” p value) giving the probability that the gene was not regulated in either region. For each gene, there were also available twofold ratio values, one that indicated the change in its expression induced by hyperoxia in the inferior retina, and a second that indicated the change in expression in the superior retina.

To identify genes regulated differently by hyperoxia in the two regions, we selected genes for which the 2-way ANOVA test gave a combination p value <0.05. Then we developed a specific measure of expression difference, called “fold margin” (FM).

We calculated a FM value for each gene, FM=FRi – FRs where FM is the fold margin, FRi is the fold ratio for that gene in the inferior retina, and FRs are the fold ratios for that gene in the superior retina.

Functional analysis

Analysis of gene ontology and molecular pathways was performed using the bioinformatics resource Database for Annotation, Visualization and Integrated Discovery (DAVID) and Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems Inc., Redwood City, CA). Using the Ingenuity Pathways Knowledge Base, IPA Core Analysis tools were used to assign genes to the most significant known molecular networks and biologic processes.

Quantitative polymerase chain reaction

Quantitative PCR (qPCR) was used to validate the expression changes of selected genes identified in the microarray experiment. Three biologic replicates were used for each condition; RNA was processed as for the microarray experiments. For each sample, 1 µg retinal RNA was reverse transcribed into cDNA using the Superscript III Reverse Transcriptase protocol. The resulting cDNA sample served as a template for real time qPCR using TaqMan^® probes and accompanying Master Mix (Applied Biosystems, Foster City, CA). A typical 20 µl reaction is composed of 10 µl 2× TaqMan Gene Expression Master Mix, 1 µl TaqMan Gene Expression Assay, 0.4 µl cDNA templates and RNase-free water. All reactions were set up in duplicate and negative controls without any template or probe were included. The qPCR was performed on a Rotor-Gene 3000 and analyzed using the Rotor-Gene 6 software (Corbett Robotics, Mortlake, NSW, Australia). A typical qPCR program was used, including 2 min incubation at 50 °C and 10 min enzyme activation at 95 °C, followed by 40 cycles of 15 s denaturation at 95 °C plus 1 min probe annealing at 60 °C. The C_t (cycle threshold) means, representing the quantitation of the amount of cDNA in the original sample, were used to calculate fold change using the Pfaffl Equation [22]. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was employed as a reference gene against which relative expression values of other genes were calculated.

Clustering analysis

Hierarchical clustering analysis was performed using the heat map function in the Partek Genomics Suite (Figure 1). The replicates of each group clustered most closely to each other (GEO accession number GSE21246) and were therefore averaged; Figure 1 therefore shows the clustering of the averaged samples. Superior retina samples before and after hyperoxia clustered more closely to each other than to other groups, indicating that hyperoxia had little effect on the superior retina. Gene expression in the inferior retina diverged from the superior retina, with hyperoxia increasing the separation.

Control retina: genes differently expressed in inferior and superior regions

Comparison of gene expression levels in inferior and superior regions of control retinas identified expression changes in a limited cohort of sequences.

Identification

The FD analysis described in the Methods section identified only 31 sequences with an FD >1.5. Of these, 27 (listed in Table 1) are known genes. Approximately half of the genes were expressed more highly in the inferior retina, and the other half were expressed more highly in the superior retina. The FD was greater than two for only two of the 27 genes (T box 5 [Tbx5], membrane metalloendopeptidase [Mme]).

Functional analysis

The list of 27 genes was subjected to gene ontology analysis, using the bioinformatics resource DAVID, which organized the genes into several functional groups, including Anatomic structure development, Sensory perception, and Catalytic activity (Table 1). Genes related to retinal development formed the largest functional group; these relate to the organization of the retina’s projection to brain centers. A superior-inferior gradient is consistent with previous work on these genes [23-25]. Five of the 27 genes were in the Sensory perception group. Two genes are for short- and medium-wavelength cones; short-wave-sensitive cone opsin1 (Opsn1sw) is more strongly expressed in the inferior retina, and medium-wave-sensitive cone opsin1 (Opsn1mw) in the superior retina. These differences in expression match known regional differences in the retinal distributions of short- and medium-length cones [26-28], and confirm the present separation of inferior and superior retina samples.

Quantitative PCR validation

Six genes were selected for confirmation of their differential expression in control retinas, using qPCR (Figure 2). The gene Tbx5 was chosen as an example of the Anatomic structure group, Opn1sw and Opn1mw were chosen as retina-relevant examples from the Sensory perception group, Mme and somatostatin (Sst) were chosen as a functionally inter-related pair, and rhodopsin (Rho) was chosen as a non-regulated, retina-relevant gene. Quantitative PCR confirmed the even expression of Rho and the differential regulation of Mme, Sst, Tbx5, and Opn1sw. For one gene, Opn1mw (encoding M-opsin), the FD assessed by qPCR was less than the threshold of 1.5 fold (−1.33 by PCR versus −1.64 by microarray). However, the difference was in the same direction in the qPCR and microarray assessments.

Hyperoxia-exposed retina: genes differently regulated in inferior and superior regions

Genes whose regulation by hyperoxia differed significantly between the inferior and superior retina could include genes regulated in one region but not the other, genes upregulated in one region and downregulated in the other, or genes regulated in the same direction in the two regions, but much more strongly in one region.

Identification

Hyperoxia induced significant changes in gene expression in both the inferior (hyperoxia-vulnerable) and superior (hyperoxia-resistant) retina, but the number of hyperoxia-regulated genes was much higher in the inferior retina (Figure 3). For example, using the criterion p<0.01 on the 2-way ANOVA test, the expression of 281 genes and 110 ncRNAs changed significantly in the superior retina, while the expression of 618 genes and 231 ncRNAs changed in the inferior retina. Using p≤0.05 and requiring a fold ratio > 2 or < 0.5, the total number of genes differentially expressed was over six times greater in the inferior retina than in the superior retina (58 versus 9). Overall, more genes were upregulated than downregulated. This imbalance was particularly marked after applying both selection criteria; for example, 56 genes were upregulated in the inferior retina, compared to 2 that were downregulated.

Combining the criteria of p<0.05 and abs (FM) >0.5 (Table 2), a total of 112 sequences with Affymetrix probe IDs were identified as differently regulated by hyperoxia between the inferior and superior regions. Of these 112, 97 matched known genes (Table 2). The gene with the most significant FM was Edn2 (endothelin 2), a possible stress response gene. The expression of gene Edn2 increased 7.02 fold in the inferior retina, but increased just 1.62 fold in the superior retina after being exposed to hyperoxia.

Functional analysis

Ingenuity Pathway Analysis software was used to identify biologic functions, diseases, and well characterized pathways that are most relevant to the genes of interest. The program compared the genes identified as differentially regulated between the inferior and superior retina with a database of genes involved in particular functions, diseases, and pathways. It then calculated the probability that each function, disease, or pathway is not involved in the differential regulation. Using the criteria p<0.001 and a minimum involvement of five genes, the processes and diseases listed in Table 3 were identified. Some clusters of differentially regulated genes that possibly impact photoreceptor vulnerability and survival are listed in Table 4, arranged according to function, disease, or pathway.

Interestingly, four of the biologic processes identified in Table 3 are strongly related to the function of the immune system. Moreover, “immunological disorders” is the disease category most strongly related to the gene set identified as differentially regulated by hyperoxia between the inferior and superior retina.

Quantitative PCR validation

Expression changes of 11 genes were selected for confirmation by qPCR. From these, eight genes were chosen from Table 2: Edn2, glial fibrillary acidic protein (Gfap), B-cell leukemia/lymphoma 3 (Bcl-3), nuclear receptor subfamily 2, group E, member 3 (Nr2e3), wingless-related MMTV integration site 8A (Wnt8a), lin-7 homolog B (Lin7a), complement component 3 (C3), and growth arrest and DNA-damage-inducible 45 beta (Gadd45b). Three other genes were also chosen for validation: cyclic nucleotide gated channel alpha 2 (Cnga2), which has a p-value <0.05 but an FM <0.5, hypoxia inducible factor 1 alpha subunit (Hif1α), which has shown responses to hypoxia in previous studies [29,30], and medium-wave-sensitive cone opsin (Opn1mw), which is a gene known to be expressed by the retina and that is not hyperoxia-regulated [14].

The regulation changes of these genes observed by qPCR and in the microarray are shown in Figure 4. The direction of the change was confirmed in six of the eight genes that differentially regulate between the inferior and superior retina; the exceptions were Wnt8a and Nr2e3. The magnitude of the FM between the superior and inferior retina was more dramatic for genes Edn2, Gfap, and Bcl3. For Cnga2, the upregulation in the superior and downregulation in the inferior retina were more noticeable; however, the FM between them was < 0.5, which was consistent with the microarray data. For Hif1α and Opn1mw, which showed no significant change in the microarray, upregulation of approximately 30%–50% was observed by qPCR. However, such upregulation showed no regional differences. Upregulation of Hif1α by hyperoxia has been reported previously [14], but the upregulation was time-dependent, being prominent after three days of exposure, but not at the 14-day exposure used here.

Gene expression before hyperoxia

Comparing gene expression between the superior and inferior retina under normoxic (control) conditions, two genes showed an FD in expression of greater than two, and 27 showed an FD of greater than 1.5. These differences in gene expression are relatively limited, but may be significant in determining the relative invulnerability of the superior retina to hyperoxic damage and, conversely, the vulnerability of the inferior retina.

Seven out of the 27 genes (listed in Table 1) are involved in anatomic structure development (gene ontology GO-term 0009653). Vax2, Tbx5, and ephrin B2 (Efnb2) are important in eye morphogenesis and are topographic determinants of the visual projections between the retina and tectum. In vertebrate development, the polarity of the dorsal-ventral (dorsal-superior, ventral-inferior) eye axis is determined by spatially restricted expression of transcription factors Vax2 in the inferior retina and Tbx5 in the superior retina [24,25]. Efnb2 is an axon guidance molecule expressed in a low-to-high superior-inferior gradient in the ganglion cell layer [31]. The expression of Efnb2 and its receptor is regulated by Vax2 and Tbx5, which in turn control the dorsal-ventral polarity of retinotectal projection [23]. These developmental patterns were detected in the adult retina in the present experiments, providing evidence that our separation of inferior-superior retina was accurate.

Five genes are associated with sensory perception (GO-term 0007600). They belong to the rhodopsin-like superfamily of G-protein-coupled receptors, although they are involved in distinct sensory systems. Opn1mw encodes the medium-wave-sensitive cone opsins (expressed in M-cones) and Opn1sw encodes the short-wave-sensitive cone opsins (S-cones or “blue” cones). The uneven distribution of cone receptors, with S-cones more numerous in the inferior retina and M-cones in superior retina, has been described in mice [27,28] and cats [26]. The expression differences observed (1.76 for Opn1sw in an inferior-superior comparison and −1.64 for Opn1mw) confirm the differences in cone topography detected by immunohistochemistry.

The gene showing the largest FD between the superior and inferior regions of the normoxic retina is Mme, which is expressed more than twofold in the inferior retina. The product of this gene is a membrane-bound metalloendopeptidase, commonly known as neutral endopeptidase or neprilysin. It has been found in various cell lines, including endothelial cells and vascular smooth muscle cells. Mme belongs to a family of zinc metalloproteases [32].

Another member of this family is endothelin converting enzyme-1 (ECE-1), which shares 58% amino acid homology with Mme [33]. ECE-1 catalyzes big endothelin to peptide endothelin and plays a key role in proteolytic endothelin activation [34]. However, the conversion of big endothelins to their respective endothelin peptides may not solely depend on ECE-1. In guinea pig lung parenchyma, Mme was suggested to play a role in the conversion to endothelin-1 and endothelin-2 [35]. Given their structural homology, it is not surprising that they share functions. As reported previously [14], endothelin-2 is the most upregulated gene in the whole retina after two weeks of hyperoxia. In the present study, endothelin-2 is the most upregulated gene in the vulnerable inferior retina after hyperoxic exposure. This correlates with the overexpression of Mme in the inferior retina. Perhaps Mme is responsible for or at least represents one pathway of the biosynthesis of endothelin-2 in the inferior retina after hyperoxic exposure.

Somatostatin is uniquely capable of stimulating Mme activity in primary cortical neurons [36]. Most interestingly, the gene encoding somatostatin (Sst) is also expressed more in the inferior than the superior retina (Table 1). The relatively high expression of both Mme and Sst may provide some insight into the vulnerability of the inferior mouse retina. One possibility is that there is an unknown regulatory peptide that protects photoreceptors against oxidative damage and is also, like amyloid beta (Aβ), a substrate of Mme. Upregulation of Mme might degrade this peptide, reduce its protective action, and cause the vulnerability of the inferior retina to hyperoxic stress. Other possibilities include regional variations in retinal metabolism, which in turn may be observed in regional variations in retinal blood supply. In the rabbit and monkey, for example, regional variations in the choroidal circulation (the major source of oxygen supply to photoreceptors) have been described, matching retinal specializations [37]. An inferior-superior difference in the retinal circulation has also been observed in humans [38].

Another established difference between the superior and inferior retina is environmentally determined. When animals are raised (as in the present experiments) in light from sources in the ceiling of a room, the inferior retina is more exposed to ambient light and is more resistant to subsequent light stress [17]. Exposure to modest ambient light can “pre-condition” the retina, making it resistant to subsequent hyperoxic stress [39]. However, the relative vulnerability of the inferior retina clearly cannot be explained by such a mechanism. The current results provide a starting point for further detailed study and are a first step in describing the mechanisms involved.

Gene expression after hyperoxia

Endothelin-2

The gene Edn2 that encodes endothelin-2 is the most strongly regulated gene in the inferior retina after hyperoxic exposure and is also the most differentially regulated gene between the inferior and superior retina after hyperoxic exposure (Table 2). Its expression increased sevenfold in the inferior retina compared to 1.6 fold in the superior retina after exposure to hyperoxia.

Endothelin-2 belongs to a family of peptides comprising endothelin-1, endothelin-2, and endothelin-3 [41]. The endothelins are stress-responsive regulators that work in paracrine and autocrine fashion in a variety of organs, with either beneficial or detrimental results. For example, endothelin-1 is a potent vasoconstrictor in blood vessels. It also acts as a mitogen on the vascular smooth muscle and has been implicated in several vascular diseases, such as hypertension and atherosclerosis [42].

Endothelin-2 has been found to be produced in endothelial cells, the heart, and the kidneys [43,44]. In the retina, the expression of Edn2 is highly induced in models of retinitis pigmentosa, retinal detachment, and light damage and the increased production of Edn2 is localized to photoreceptors [45]. Photoreceptor-derived Edn2 is suggested to serve as a stress signal in the signaling cascade between the photoreceptor and Müller glia cells. In the presence of damaging stress such as excessive light exposure and gene mutation, Müller glia cells sense the photoreceptor damage and react by upregulating leukemia inhibitory factor (LIF) [46]. An increase in leukemia inhibitory factor in turn induces the expression of Edn2 in photoreceptors. The signal is then passed back to Müller glia cells by binding Edn2 to endothelin receptor B (Ednrb), which is expressed on Müller cells. This positive feedback loop then stimulates the production of survival factors, such as fibroblast growth factor 2 (FGF2), in Müller cells and/or photoreceptors to protect visual cells from further damage [45,46].

Involvement of the immune system

Misregulation of the innate immune response and other immune-mediated processes has been suggested to play a role in the development and progression of retinal degenerative diseases, such as age-related macular degeneration (AMD), in human patients [47,48]. The involvement of inflammation and immune-mediated processes has been suggested, for example, in the formation of drusen and inflammation triggered by the debris generated by dying retinal pigment epithelial (RPE) cells, as a factor in photoreceptor cell death [48].

Complement cascade proteins have been shown to be upregulated in the AMD retina, showing a strong presence along the RPE-choroid interface and in drusen, the hallmark of AMD, suggesting that the activation of the complement pathway has a potential role in drusen biogenesis and the etiology of AMD [47]. A murine model of laser-induced choroidal neovascularization (CNV) in C57BL/6J mice revealed the deposition of C3 and membrane attack complex in the neovascularizing tissue. Complement depletion in C3^−/− mice resulted in the inhibition of CNV and a marked reduction of angiogenic factors [49]. C3a is the bioactive fragment of complement component C3; a blockade of the C3a receptors (C3aR) also reduces CNV [50]. In addition, in the degenerative mouse model rd1, the gene ontology groups, “defense response,” “complement activity,” and “stress response,” were significantly over-represented in the differentially expressed genes in the degenerative retina. The classical complement pathway was found to be upregulated in the rd1 mouse [51]. The complement pathway has also been implicated in the development of glaucoma and in a model of ocular hypertension [52,53].

Immune-related factors and pathways may thus be important in the process of photoreceptor cell death induced by hyperoxic stress. The present data increase previous understanding of this involvement, but are arguably still inadequate to determine whether immune misregulation causes or results from cell death induced by hyperoxia.

Retinal disease relevant genes

Seven genes listed in Table 2 were identified as associated with ophthalmic diseases and three of them correspond to known retinal disease loci listed on the RetNet website (C3, versican [Vcan], and Nr2e3). Previous studies reported a significant association between a polymorphism in the C3 gene and susceptibility to AMD [54,55]. Wagner syndrome can be caused by a mutation in the human VCAN gene, which encodes versican, a proteoglycan present in the vitreous body [56]. The syndrome features peculiar lesions of the vitreous and retina, retinal detachment, and progressive chorioretinal dystrophy [56,57]. Versican is widely distributed in the extracellular matrix of a variety of tissues and plays an important role in reducing oxidant injury by enhancing cell-matrix interaction [58]. Mutations in the human NR2E3 gene, which encodes a retina-specific nuclear receptor, lead to enhanced S-cone syndrome and retinitis pigmentosa [59].

Antioxidant genes

Metallothioneins (MTs) are a class of low molecular weight, cysteine-rich proteins that bind with copper and zinc. The metallothionein family is composed of four separately encoded isoforms, known as MT-I, -II, -III, and -IV [60]. They serve as important regulators of metal homeostasis and provide cytoprotection in tissues subjected to oxidative stress as a result of their free radical scavenging ability [61]. Increased expression of MT has been reported in various induced models of retinal degeneration, including light [40], hyperoxia [14], and hypoxia [62]. Metallothioneins have been shown to protect RPE cells against toxic levels of cadmium, heme- and iron-induced oxidation, and UV light-induced apoptosis [63]. As endogenous antioxidants, they also protect ganglion cells from oxidative stress caused by the glutamate analog N-methyl-D-aspartic acid (NMDA) [64]. In this study, expression of Mt1 and Mt2 is more elevated in the inferior retina that has more tissue damage. The prominence of antioxidants in the retina’s response to hyperoxia indicates the importance of free radicals in the damage caused [14]. Conversely, the vulnerability of the inferior retina does not seem to be due to lack of antioxidants in that region. However, the upregulation of Mt1 and Mt2 are not sufficient for protecting the inferior retina against hyperoxia-induced cell death. This is consistent with previous findings that Mt1 and Mt2 alone are insufficient for protecting photoreceptors against hyperbaric oxygen-induced cell death [16].