Network analysis of human glaucomatous optic nerve head astrocytes
- Tatiana Nikolskaya†1,
- Yuri Nikolsky†2,
- Tatiana Serebryiskaya1,
- Svetlana Zvereva1,
- Eugene Sviridov1,
- Zoltan Dezso2,
- Eugene Rahkmatulin2,
- Richard J Brennan2,
- Nick Yankovsky1,
- Sanjoy K Bhattacharya3, 4,
- Olga Agapova5,
- M Rosario Hernandez6 and
- Valery I Shestopalov3, 7Email author
© Nikolskaya et al; licensee BioMed Central Ltd. 2009
Received: 25 April 2008
Accepted: 09 May 2009
Published: 09 May 2009
Astrocyte activation is a characteristic response to injury in the central nervous system, and can be either neurotoxic or neuroprotective, while the regulation of both roles remains elusive.
To decipher the regulatory elements controlling astrocyte-mediated neurotoxicity in glaucoma, we conducted a systems-level functional analysis of gene expression, proteomic and genetic data associated with reactive optic nerve head astrocytes (ONHAs).
Our reconstruction of the molecular interactions affected by glaucoma revealed multi-domain biological networks controlling activation of ONHAs at the level of intercellular stimuli, intracellular signaling and core effectors. The analysis revealed that synergistic action of the transcription factors AP-1, vitamin D receptor and Nuclear Factor-kappaB in cross-activation of multiple pathways, including inflammatory cytokines, complement, clusterin, ephrins, and multiple metabolic pathways. We found that the products of over two thirds of genes linked to glaucoma by genetic analysis can be functionally interconnected into one epistatic network via experimentally-validated interactions. Finally, we built and analyzed an integrative disease pathology network from a combined set of genes revealed in genetic studies, genes differentially expressed in glaucoma and closely connected genes/proteins in the interactome.
Our results suggest several key biological network modules that are involved in regulating neurotoxicity of reactive astrocytes in glaucoma, and comprise potential targets for cell-based therapy.
Astrocyte activation is a hallmark of various CNS injuries and pathologies, including stroke, trauma, tumor, infection, and neurodegenerative diseases [1–3]. Upon activation, astrocytes display altered metabolism and the ability to preserve CNS homeostasis and support neuronal function. Reactive astrocytes were shown to reduce damage during the acute phase of CNS insults . In contrast, progressive degenerative diseases, such as glaucoma, feature chronic astrocyte activation that exacerbates damage to neurons and impairs regeneration of their axons [5, 6]. Importantly, a prominent astrocyte reactivation in primary open angle glaucoma (POAG) is localized to the optic nerve head, which is also the site of primary damage to the retinal ganglion cells (RGCs) .
In common with many other complex, age-related diseases, neurodegeneration in POAG is associated with a homeostatic imbalance resulting from environmental factors and multiple genetic components interconnected within complex epistatic networks . Such imbalance is manifested at three interconnected functional levels: intercellular stimuli, intracellular signal transduction, and core effectors (i.e. endogenous metabolism, structural complexes, etc). Disease causes alterations at all three levels. These can be measured by high-content screens that include differential gene expression, proteomics and metabolomics, as well as in genetic linkage studies that connect genes or protein variants to disease onset [9, 10]. Recent progress in systems biology has allowed a quantification, cross-comparison and functional interpretation of heterogeneous datasets within the framework of human biological pathways, networks and processes, which are assembled from a knowledgebase of functional biological interactions [10, 11]. This systems level approach requires an understanding of connectivity between the genes and proteins affected in a given disease. Connectivity is defined by binary protein interactions with genes, proteins and biologically active compounds . The biological networks are scale-free but converge in regulatory nodes and modules, such as major transcription factors and receptors [13, 14]. Identification of such key topological elements [15, 16] on the networks derived from disease-related data may reveal potential therapeutic targets. This approach is particularly powerful for diseases of complex etiology, such as glaucoma.
Regulation of astrocyte activation, which is associated with increased neurotoxicity, involves differential activation of key cellular network modules . To perform in silico reconstructions of the cellular pathways affected during the development of glaucoma, however, data derived specifically from astrocytes must be used, rather than data derived from whole-tissue (retina or optic nerve) samples. It is feasible to suggest that the cell-specific data from interacting cell types, such as astrocytes and retinal ganglion cells, will allow us to analyze differences in trans-cellular crosstalk that are implicated in glaucoma. Here, we performed functional analysis of the signaling and effector networks  potentially implicated in the pathophysiology of glaucoma, using both small experimental and high-content differential data obtained from whole optic nerves and from reactive ONHAs isolated from the optic nerves of glaucomatous patients. We reasoned that the bulk of changes to the ONHA transcriptome would be disease-related, and subjected the data to meta-analysis at the level of affected biological processes, and at the level of biological network topology, using the MetaCore Analytical Suite [19, 20]. This suite of analytical tools, linked to a comprehensive interaction and biological function database, is available as a stand-alone application and can be linked via a "plugin" to other pathway tools such as Cytoscape . Biological networks were built starting from either gene expression data alone, or a combination of high content data sets reflecting multiple levels of information flow in the cell. Using a two-step procedure, we narrowed the set of differentially expressed genes, and merged it with genes associated in the literature with glaucoma, thus filling gaps in the fragmented, literature-derived genetic data. The final direct interaction network displayed a synergy between these two minimally-overlapping datasets, allowing broad characterization of pathological changes in reactive astrocytes, and defining network modules potentially implicated in the shift to neurotoxicity. Modules regulating such shift represent therapeutically valuable targets that can be further validated experimentally.
Functional characterization of gene expression data
Activated processes and pathways
Metabolic pathways activation revealed by gene expression data from glaucomatous ONHAs
Metabolic pathway maps
Enzyme EC #
common protein name
Alcohol dehydrogenase 1A
Alcohol dehydrogenase 1B
Alcohol dehydrogenase 1C
Aldehyde dehydrogenase 1A3
Aldehyde dehydrogenase 3A2
retylaldehyde dehydrogenase 1A3
aldo-keto reductase 1C1
cytochrome P450 family 2D6
cytochrome P450 2C9
cytochrome P450 family IA
cytochrome P450 family IIIA4
cytochrome P450 family 7A1
cytochrome P450C 27/25
leucine carboxyl methyltransferase 2
Distribution of Gene Ontology (GO) processes for the largest Direct Interactions (DI) network of genes up-regulated in the combined data set
number of DE genes
# genes in a pathway
upregulation, 196 nodes
Cellular physiological process
Regulation of transcription, DNA-dependent
Positive regulation of cell proliferation
Cytokine and chemokine signaling
downregulation, 50 nodes
Regulation of sodium ion transport
Norepinephrine-epinephrine regulation of blood pressure
Positive regulation of bone mineralization
Transmembrane receptor protein tyrosine kinase activation
Negative regulation of smooth muscle contraction
Norepinephrine-epinephrine vasodilation during regulation of blood pressure
Overall, the enrichment analysis of transcriptional changes in glaucomatous ONHAs revealed significantly more activated than down-regulated pathways, which is consistent with the "re-activation" status of these cells. Transcriptional activation orchestrated by the transcription factors AP-1/c-FOS, VDR, and NF-kB, was modulated by increased activity of TLR, AR and ephrin receptors. Pro-inflammatory processes in these cells were mediated by an increased activity of TOLL-like receptors, NF-kB, MAPKs, JNK3 and complement system. While increased signaling via VDR, AP-1, IL-6 and AR in the optic nerve can be broadly interpreted as a cytoprotective, activation of ephrin A4 receptor and multiple pro-inflammatory pathways represent potentially cytotoxic events, which are conducive to oxidative stress and possibly harmful to CNS neurons and their axons. Significantly, enrichment analysis of activated genes across metabolic maps revealed alterations in cholesterol metabolism that correlated with increased sensitization of neurons to glutamate in glia-neuron co-cultures . Similar changes in cholesterol metabolism were also detected in transcriptomic analysis of whole optic nerve head tissue in rat experimental glaucoma . We compared published transcriptomic and biochemical studies in retina and optic nerves from different experimental glaucoma models. Importantly, the published data showed multiple similarities with our data from human ONHAs, particularly in the activation of innate and adaptive immune responses. Consistent with our results, animal models of glaucoma showed robust activation of pathways mediated by AP-1/c-Fos/c-Jun [5, 37, 38], NF-κB , complement [36, 40–42], androgen receptor [39, 43] and ephrins . Oxidative stress response and signaling via NF-κB, AP-1 and TLRs have also been detected in transcriptomic studies of hydrostatic pressure-induced human ONHAs, and in cytokine-activated murine astrocytes [45, 46]. Gene expression analysis of EGFR-stimulated rat ONHAs revealed additional commonalities in the activation of genes related to cell migration, ECM reorganization and immune response . Major differences in the transcriptomic data derived from rodent astrocytes included a lack of activation of VDR-mediated signaling pathway. This effect may be specific to glaucomatous human ONHAs. These comparisons lead us to suggest that the signature profile of the glaucomatous process in reactive astrocytes includes chronic activation of AP-1, NF-kB, complement, AR, TLRs and cell adhesion pathways.
Down-regulated pathways and processes
Analysis of down-regulated genes revealed a fewer number of significantly affected pathways. Perturbed pathways included PDGF signaling via MAPK cascades, membrane trafficking and signal transduction via G-αI (Figure 1, also see Additional File 3, Supplement Figure S6). We identified several levels of inhibitory regulation in these processes, including structural proteins, signaling complexes, cellular matrix proteases and intracellular signal transduction proteins. Down-regulation of many integrins (a total of twelve in the two donor groups, see Additional File 1 Supplement Table S4), evidenced profound changes in the adhesion state and cell signaling in glaucoma. Interactome analysis of the mapped data was performed to analyze connectivity between down-regulated pathways using the MetaCore network building tools and gene content from maps (69 genes found on the 12 most significantly down-regulated maps). Analysis of global connectivity in the networks using the Direct Interactions (DI) algorithm (see Additional File 1 Supplement Table S3) revealed integrins, TGF-β, XIAP, kinase families PKC and PI3K as critical hubs interconnecting a compact network (see Additional File 3, Supplement Figure S7A). Concerted inhibition of adhesion complexes (see Additional File 3, Supplement Figure S7B) is a likely explanation for the increased ONHA motility observed in glaucoma, and may contribute to a change in the physiological role of astroglia upon activation [3, 48]. In good agreement with these findings, studies in an animal model of glaucoma also showed significant changes in the expression of ECM components , and a toxic decrease of PKC activity in retinal neurons . However, results of individual studies on TGF-β, and PDGF activation in glaucoma  varied significantly, likely due to differences in the design, species and cellular populations examined.
Networks for genes differentially regulated in glaucomatous ONHAs
Networks for genes with increased expression
Top hubs differentially activated in the DI network
Top differentially regulated membrane receptors
Galpha(q)-specific peptide GPCRs
Substance P receptor
Angiotensin II receptor, type-1
Beta-2 adrenergic receptor
Regulation of activated genes by transcription factors in the DI network
Networks for genes with decreased expression
Fifty out of a total of 301 genes down-regulated in the combined data set (2.5 fold threshold) were connected in MetaCore into a DI network (see Additional File 3, Supplement Figure S9) with a p-value < 0.1. This p-value is relatively high due to a smaller percentage of functionally interconnected genes in the subset of genes with decreased expression. The major hubs in the network are interferon gamma (IFN-γ, 19 edges), and transcription factor STAT1 (17 edges) (Table 3). STAT1 and INF-γ are the core hubs in the down-regulation network that are closely connected to the down-regulated modules c-Myb, PAI1, PDGF receptor, GPCR G alpha (q)-specific peptide, Substance P receptor, ATGTR1, TACR1 and CCR1 (Table 4). This network features a number of signaling protein ligands, including thrombopoetin (CCL5), lactoferrin, substance P, IFN-γ, neuromedin U, activin beta A, protein C, CCL8, and is consistent with a decrease in STAT1 and STAT4, a transcriptional regulators for most of these genes. STAT1 activation is pro-apoptotic , it is involved in the regulation of vital processes such as cell cycle, survival , and is implicated in ischemic response and inflammation. Down-regulation of STAT1 is in agreement with the suppression of its key regulators IFN-γ, EGFR and PDGF , and contributes to activation of inflammation-associated genes IRF-1, iNOS, and TNF-α. Decreased IFN-γ and STAT1 signaling, therefore, contributes to the survival of activated ONHAs in glaucoma.
Integration of proteomics data
Distribution of GO processes in glaucomatous ONHAs revealed by proteomics
Top 10 GO processes (for 35 proteins)
maps (for 20 mapped proteins)
Complement activation. Alternative pathway
CDC42 in cellular processes
Response to heat
Alternative complement pathway
Putative ubiquinin pathway
Role of ASK1 under oxydative stress
Role of IAP proteins in apoptosis
DNA DSB repair via homologous recombination
Glucocorticoid receptor signaling
Role of Akt in hypoxia
Parkin disorder under Parkinson's disease
Innate immune response
Role of Parkin in the ubiquitin-proteosomal pathway
Small GTPase mediated signal transduction
Comparison of the hub structure in the DI networks built from the three data sets and the combined set
Integration of genetic data on the networks
Distribution of GO processes and pathways maps for the "literature" G-set gene list
Top GO processes
Response to wounding
VDR in regulation of differentiation
Blood pressure regulation
IL1 signaling pathway
MIF in innate immune response
Integrin outside-out signaling
MAPK cascade. Nuclear function of p38-MAPK
Cytosol to ER transport
HGF signaling pathway
Positive regulation of I-kB/NF-kB cascade
The basic network statistics ("top tens") for the combined G/DE/DI network
Regulation of transcription from RNA Pol II promoter
BMP signaling pathway
Regulation of transcription, DNA-dependent
Positive regulation of progression through mitotic cell cycle
Mammalian eye development
Despite little overlap between the G-set and DE-set, more similarities were revealed at the network topology level by a functional proximity search for differentially expressed genes within the interactome "neighborhood" of the G-set. This approach establishes relationships between datasets of different types (meta-analysis) using the parameter of functional proximity, which is defined as one-step physical interactions between the gene products from different sets . We built a set of networks using the G-set gene content as input nodes and the Analyze Networks (AN) algorithm. We then mapped the expression data onto the AN networks. The AN algorithm prioritizes sub-networks based on relative enrichment with experimental (gene expression) data and canonical pathways (See Materials & Methods). This helped us to identify the proximal DE-set network objects and to fill the gaps in the fragmented disease association data (see Additional File 3, Supplement Figure S13). After analyzing the 50 top scored AN networks built from the G-set, we identified 37 up-regulated and 20 down-regulated genes (see Additional File 1, Supplement Table S8). The majority (32 out of 57) of genes identified in the interactome neighborhood of the genetic network were also present on the global DI network for activated genes and proteins (see Additional File 3, Supplement Figures S14 a, b).
We then merged the G-set with the 57 DE-set genes from close interactome neighborhood to identify possible connectivity between the networks originating from genetic and gene expression data. The combined data set of 119 genes included only nine genes common to both data sets. Differential activation of another four genes from the G-set was suggested by the proteomic data. GO distribution showed enrichment in the processes: regulation of transcription, BMP signaling pathway, cell differentiation and development, in the combined G- and DE-sets (Table 7). The stringent DI algorithm connected 102 out of 119 nodes into one combined G/DE network (see Additional File 3, Supplement Figures S14 a, b), with a p-value < 0.0001 (see Additional File 1, Supplement Table S9). It narrowed the interaction gaps between the literature-derived G-set data. An important feature preserved in both DE and G/DE networks is the synergy between the major transcription factors AP-1, SP1, VDR, AR and NF-κB. As the G-set data are not cell type- or tissue-specific, the conservative hub structure of the disease network may represent a core feature of the glaucomatous process across all eye tissues. Some major hubs present only in the G/DE network included P53, STAT1, TNFα, IL-1β, EGFR, and IFN-γ (Table 7), from which STAT1 and P53 hubs increased their relative scoring, suggesting that their potential role in astrocyte activation in glaucoma should be further investigated.
The ultimate goal of our functional analysis of glaucoma-related high content data was reconstruction of the condition-specific model at three levels: i) stimuli, i.e. the endogenous and exogenous (environmental) signals, which trigger the disease condition; ii) intracellular signaling activated in the disease; and iii) effectors – the cellular processes, structural complexes and metabolic pathways altered by aberrant signaling in the disease. Changes at the effector level should be consistent with the observed disease manifestations in molecular profile and phenotype. While complex exogenous signals are known to include genetics, age, mechanical damage, etc., our network analysis has revealed molecular modules that are activated intracellularly, and suggests plausible disease effectors in glaucomatous ONHAs. This analysis suggests that intracellular signaling in reactive ONHAs in response to glaucomatous conditions is mediated by a synergy of key transcription factors VDR, NF-κB, AP-1 and AR. These changes are paralleled by decreased signaling via integrins, PDGF and STAT1, which might be insulting to the retinal ganglion cells, since these molecules control adherence, while the motility of glaucomatous astrocytes is likely to be disruptive to their metabolic support functions [2, 3, 48]. Integration of the gene expression and proteomic data for ONHAs confirmed substantial activation of the innate immune response and suggests that this is one of the key effector events in activated astrocytes, which has been also detected in recent gene expression studies [42, 83]. At the level of metabolism, our analysis revealed coordinated activation of endogenous alcohol and aldehyde dehydrogenases, which is consistent with increased metabolic activity and oxidative stress response [69, 84]. Similar activation of these pathways has been shown in the brain of Alzheimer's sufferers, and is likely to play a protective role in detoxification of reactive intermediates of lipid peroxidation . Significant activation of phospholipase A2 and it's receptor PLA2R1, which are involved in a complex network linking receptor agonists, oxidative agents, and proinflammatory cytokines to the release of arachidonic acid , represents a pro-inflammatory effect of glial activation also observed in brain pathologies .
Our network analysis demonstrated activation of potentially neurotoxic and neuroprotective programs in ONHAs in response to glaucomatous ON injury. We found that NF-κB is significantly over-connected on the DI and AN networks, showing links to both programs. In contrast to the well known pro-survival role of this transcription factor, recent studies have demonstrated a profound negative impact of NF-κB activation on neuronal survival in various CNS pathologies [88–91]. Our analysis suggests that the activation of a NF-κB-controlled transcriptional program in glaucomatous ONHAs facilitates pro-inflammatory events and is, therefore, potentially neurotoxic. It has been demonstrated that in the glia reactivated with LPS, amyloid-beta, elevated hydrostatic pressure, cytokines or double-stranded RNA, NF-κB induction correlated with neurotoxic levels of NOS-2, MMPs, JNKs (MAPK8-10), complement, clusterin and cytokines [6, 45, 92–94]. Activation of several NF-κB subunits demonstrated at the levels of gene transcripts, proteins and over-connectivity with upstream regulators and downstream targets on the disease network, supports the notion that NF-κB is key convergence hub in this pathology. In contrast to systemic inhibition of this vital transcription factor that would have a negative impact on neuronal survival, as shown in recent studies utilizing sulfasalazine inhibitor  or a knockout mouse model , our experimental data strongly endorse astroglial NF-κB as a potential target for a cell-level inhibition in glaucoma therapy.
Direct networking of gene products via protein interactions can be used to identifying modules of epistatic interaction between disease related genes, as described for yeast metabolism . As we showed previously , major topological features of differential networks represent molecules affected by the disease process, and represent potential targets for therapeutic intervention. In this work, we took advantage of the functional analysis capabilities and database of molecular interactions in MetaCore and cross-referenced two experimental datasets, proteomic and genetic, which are otherwise poorly comparable. Using a two-step procedure, we narrowed the set of differentially expressed genes to fifty seven genes (DE-set) most relevant to the G-set of genes associated in the literature with glaucoma, effectively filling the gaps in the fragmented literature-derived genetic data. Our analysis of glaucomatous astrocytes indicated that the synergistic activation of the major hubs SP-1, VDR, NF-κB and AP-1 in response to oxidative stress and mild ischemia associated with glaucoma, orchestrates genome-wide changes in the transcriptional profile and chronic activation of ONHAs. In contrast to the pro-survival effects of VDR, AP-1and AR, the harmful effects of chronic activation of major downstream effectors such as NF-κB, clusterin, complement and ephrin A4 receptor are likely to be responsible for tipping the balance in the optic nerve microenvironment towards neurotoxicity and inhibition of regeneration. Combined, these glial factors could actively contribute to the retinal ganglion cell death in glaucoma by exacerbating inflammatory damage to challenged axons. Our results imply that the suppression of the major downstream effectors may represent feasible strategy for glia-targeted therapy of glaucoma, ensuing further experimental validation of these targets.
Optic nerve Astrocytes
Cultures of human ONHA were generated as previously described . Briefly, ONH were dissected and processed within 24 h of death. Four explants from each lamina cribrosa were dissected and placed into 25-cm2 Primaria tissue culture flasks (Falcon, Lincoln Park, NJ). Explants were maintained in DMEM-F12 supplemented with 10% FBS (Biowhittaker, Walkerswille, MD) and PSFM (10,000 U/ml penicillin, 10,000 μg/ml streptomycin and 25 μg/ml amphotericin B; Gibco/BRL, Gaithersburg, MD). Cells were kept in a 37°C, 5% CO2 incubator. After 2–4 weeks, primary cultures were purified by using modified two-step immunopanning procedure described by Mi and Barres . Purified cells were expanded after characterization by immunostaining for astrocyte markers GFAP and NCAM as described [98, 99]. Second passage cell cultures were stored in RPMI 1640 with 10% DMSO in liquid nitrogen until use. For each set of experiments, cells were thawed and cultured for one more passage so that sufficient cells from the same batch were available in each set of experiments. Astrocytes from both normal and glaucomatous eyes were cultivated similarly in parallel experiments. This ensured that differential transcriptomic data obtained from them have been normalized for cultivation-induced changes, and showed only disease-imprinted profile changes as reported by others [24, 25].
Seventeen pairs of normal human eyes from donors (age 58 ± 12) with no history of chronic CNS or eye disease were obtained from the National Disease Research Interchange (NDRI) and from Mid-America Eye Bank (St. Louis, MO) within 2–4 h of death (see Additional File 1, Supplement Table S1). A total of eight eyes from six donors with documented primary open angle glaucoma (POAG) (age 73 ± 9) were obtained for the analysis. Seven normal eyes were used to generate astrocytes for microarray analysis, the rest were used for real time RT-PCR. ONH were dissected and processed within 24 h of death to generate astrocyte cultures. To test whether the normal eyes in this study may have had hidden optic nerve disease, and to assess damage in samples from POAG. Samples of the myelinated nerves were fixed in 4% paraformaldehyde, post-fixed in osmium, embedded in epoxy resin, and stained with paraphenylendiamine to detect axon degeneration [100, 101].
Human ONHA microarray data
The microarray data for two separate donor groups, Group 1 and Group 2 (four or three normal and four eyes with glaucoma in each group, a total of fifteen human optic nerve samples) were obtained from primary human ONHAs, type 1B, derived from glaucomatous and from normal control eyes as described previously . The microarray analysis of Groups #1 and #2, which was separated by a two year time interval possessed significant difference in the mean threshold levels of signal intensities and, therefore, corresponding data were analyzed separately. In total, we analyzed three differential expression (DE) data files: donor Group 1, revealed 322 up-regulated and 152 down-regulated Affymetrix gene IDs (that correspond to 298 and 130 non-redundant genes by HUGO nomenclature, respectively); donor Group 2, (357/335 up- and 271/261 down-regulated AFFI/HUGO IDs); and combined dataset comprised of 461 up-regulated and 301 down-regulated AFFI IDs/genes (see Additional File 2). The latter set was obtained by pair-wise comparison of all disease vs. all normal samples. All original data were deposited in GEO (accession #GSE2378). A portion of these data (referred to as donor Group 1) has been published previously . The raw normalized data were statistically analyzed as described below and used to obtain the gene-specific ratios of differential expression. In total, we analyzed three data files composed of genes with fold change exceeding 2.5: donor Group 1, revealed 322 up-regulated and 152 down-regulated Affymetrix gene IDs (that correspond to 298 and 130 non-redundant genes by HUGO nomenclature, respectively); donor Group 2, (357/335 up- and 271/261 down-regulated AFFI/HUGO IDs). In addition, 484 up-regulated and 323 down-regulated genes comprised the combined non-redundant set of genes with known functions, in which 461 and 301, respectively, were recognized as "network objects" in the MetaCore database (see Additional File 2).
Human ONHA proteomic data
Proteomic data was generated by shotgun MS/MS analysis of human ONH tissue from eight normal and eight glaucomatous donor eyes as described previously. From the combined set of 248 proteins identified by this analysis we selected two subsets: proteins detected only in normal tissue (31 IDs) and those found only in glaucomatous eyes (67 IDs). Shotgun MS-identification of proteins has very limited ability in ruling out the absence of a protein, so for further analysis we only selected those proteins that were identified only in glaucomatous tissue. To focus on the data relevant to astrocytes, the selected protein IDs were filtered against proteins with known association with neuronal axons, which formed the bulk of the optic nerve head tissue. Of the remaining 50 IDs, 35 had links to MetaCore maps and networks and were used in our analysis. The expression of all 35 proteins in astrocytes was confirmed by microarray data.
Statistical analysis of gene expression data
Affymetrix microarray data contain intensity values and absent/present flags for each probe set. There were a total of sixteen sets of data used in this study, eight with glaucoma and eight with normal expression. The calculations and data analyses were done with R statistical software http://www.r-project.org. First, we tested whether or not the gene expression values from the glaucoma set were different overall from those of the normal set. This was done by comparing two statistical models:
yi = μei, i = 1,..., number of genes, grand mean, ei – error, and
yij = α i + eij, i = 1,2, j = 1,..., number of genes, α i – group mean, ei – error.
The first model ignores the presence of two different groups. In the second model, we differentiate between the means of the glaucoma and normal groups.
Next, we analyzed the expression of every gene in the glaucoma set with respect to the normal set. Since we had data from eight independent experiments, we could not obtain accurate results from pairing each set of glaucoma data with a unique set of normal data. Every possible glaucoma-normal combination was therefore considered for every gene. The number of such combinations is n*m, where n is the number of glaucoma values, and m is the number of normal values. When no expression values are missing for a specific gene in any of the experiments, there are 16 combinations for every of the two donor groups. The ratios of glaucoma to normal expression were then calculated. To allow the construction of a linear model from the data, base-ten logarithms were taken for each ratio.
The table of results consists of eight columns (see Additional File 1, Supplement Table S10). The first column is the gene; the second is the mean of the log ratios. To help visualize the results, the third column contains the ratio itself, such that the values for all over-expressed genes are glaucoma/normal ratios, and the values for under-expressed genes are negative normal/glaucoma ratios. The fourth column shows the standard error (se) for the mean log ratio of a gene, which is defined by , where is the sample variance, and n is the size of the sample.
Due to the very large number of degrees of freedom (189375), values from the standard normal distribution could also have been used instead of those from the Student t distribution. The seventh column has the "TRUE" value if the 95% confidence interval for that log ratio contains 0 and "FALSE" if it does not. That way it is easy to see which genes have glaucomatous expression levels different from the normal tissue expression level of the same genes.
Another way to interpret the results is to use p-values, i.e. the smallest levels of significance for which one can consider the corresponding log ratio means is to be different from zero. If a given p-value is smaller than the chosen level of significance for the test, then it can be concluded that the gene is expressed. A p-value is calculated by P(Z < (- gene/se)) + P(Z > ( gene/se)), Z ~ N(0, 1). A Bonferroni adjustment is then applied to this result to ensure small enough overall error. Tables were created from the results with genes that have expression ratios greater than a chosen ratio or smaller than the negative chosen ratio and p-values smaller than 0.05. Another pair of columns contains the average glaucoma and normal expression levels for every gene, the standard error for these values and the 95% confidence intervals to identify the ranges for the true averages. A present flag was given to every gene that is marked as present in at least three out of four experiments. Otherwise the flag is set to "Absent".
Functional analysis of the data
The functional analysis workflow consists of series of qualitative and quantitative procedures for parsing large datasets into smaller, functionally-meaningful subsets, such as linear signaling and metabolic pathways, and cellular and molecular processes. The overlap between genes/proteins within a functional category and the components of a high-content dataset identified as meaningfully altered can be given a p-value based on the likelihood of this overlap happening by random chance (see below). Multiple functional categories can be scored for each dataset, a procedure referred to as enrichment analysis. The distribution of categories reflects their relative relevance to the condition within the dataset. We used the content of gene ontologies from OnthoExpress  in MetaCore™ (GeneGo Inc, MI, http://www.genego.com) for enrichment analysis of genes differentially expressed in reactive vs. normal ONHAs.
Interactome analysis for relative connectivity
N – the number of proteins (protein-based network objects) in our global "interactome" extracted from Metacore
n – number of proteins derived from the sets of genes of interest
D – the degree of a given protein in the global "interactome" database
k – the degree of a given protein within the set of interest
The p-value calculated above gives the probability of observing k or more interactions of a given protein (with degree D in global network) by random chance within the set of interest (of size (n)).
The probability of observing "under-connected" connected proteins can be calculated by 1-p(k).
The input lists of genes were converted to protein-based network objects which have been used in our analysis. The resulting network objects sets were divided by subsets based on the molecular function (receptors, ligands, etc.).
Identification of differentially expressed genes in close proximity to the genetic network
Data of different origin can be connected on the same networks. For instance, genes implicated in the onset of the disease can be linked with differentially expressed genes via physical interactions between their corresponding protein products. The network analysis process does not require clustering or statistical analysis other than when defining the probability (p-value) of the assembly of a network of a certain size out of randomly selected relevant nodes (described in the Methods). Each network is, therefore, unique for the data set at the level of specific proteins, subunits and binary interactions. The Analyze Networks (AN) algorithm (see Additional File 1, Supplement Table S3) connects every node with all others by unidirectional shortest paths using the G-set as the input nodes. This algorithm builds sets of the overlapping small networks consecutively covering all root nodes from the input list. The essential difference between DI and AN networks is that the latter allow non-root objects to form connectors between root nodes. In each of the top-scoring 50 networks, 40 – 55% of the objects were root nodes from the G-set. Next, we mapped the differential expression data (DE-set) onto AN networks using the filter experiments tool in MetaCore, which mainly maps the non-root nodes. Between one and seven differentially expressed "close neighbor" genes were typically located on each of AN networks (see example in See Additional File 3, Supplement Figure S13). In our data, the majority (32 out of 57) of the interactome neighborhood genes were also present on the global DI network (See Additional File 3, Supplement Figure S14). Since the global DI network included only 277 out of 807 input genes (34%), one would expect only 19 out of 57 genes from the vicinity list to overlap with the DI network list with random distribution. The 56% overlap between two gene expression lists (p-value of such an event is 0.005) indicates a strong correlation between the genetic network and the gene expression network for glaucoma.
To validate the microarray data, total RNA was extracted from ONHAs that were obtained from normal and glaucomatous eyes using Qiagen RNAeasy mini kits (Qiagen, Valencia, CA). RNA was then purified and quantified by measuring absorbance at 260 nm and treated with RNase free DNase (Ambion). Single strand cDNA was prepared from 2 μg total RNA using SuperScript II reverse transcriptase (Invitrogen). 5 μl of 1:20 -1:80 diluted cDNA were used for reaction with 2× Bio-Rad SYBRGreen SuperMix (Bio-Rad Laboratories Inc, Hercules, CA) in 25 μl, and quantitative PCR was performed by monitoring in real time the increase of fluorescence of SYBR Green using the MyiQ (Bio-Rad Laboratories Inc). Custom primers were designed using the Primer Express program (PE Applied Biosystems) for 22 transcripts representing major network hubs as well as randomly picked DE genes. At least one primer crossed the exon-exon boundary to prevent genomic DNA amplification. Primer quality (lack of primer-dimer amplification) was confirmed by melting curve analysis. Sequences of primers are available upon request. Relative quantification of gene expression was performed using the standard curve method (User Bulletin 2 of the ABI Prism 7700 Sequence Detection System, PE Applied Biosystems, Forster City, CA). Serial dilutions (1:4: 1:16, 1:64, 1:256) of mixtures of all samples were used for standard curves. For each cell culture the relative amount of mRNA for a target gene was normalized to the relative amount of reference gene RNA (18S RNA). Then, the average of normalized relative expression levels for all glaucoma and average for all normal samples for each gene was calculated. The higher average (glaucoma or normal) was divided by the lower average (normal or glaucoma) to determine fold change for each gene. The fold increase in glaucoma (glaucoma higher than normal) was expressed by positive numbers and fold decrease (normal higher than glaucoma) by negative numbers (see Additional File 1, Supplement Table S11). Quantitative PCR data for validation of the expression of the three 3αHSD isoforms used isoform-specific probes and primers previously described . The 3 α HSD probes have full homology to the AKR1C3 gene sequence and recognized mostly AKR1C3. The sequence of the AKR1C1 probe has full homology to the AKR1C1 and AKR1C2 gene isoforms and recognizes both of them.
Network visualization and analysis
The set of genes differentially expressed at the cut-off value of 2.5-fold was uploaded into the MetaCore Analytical suite (GeneGo, Inc. St. Joseph, MI). MetaCore is a web-based computational platform primarily designed for the analysis of high-content experimental data in the context of human metabolic and regulatory networks and pathways. MetaCore includes a manually-curated database of human protein interactions, metabolism and bioactive compounds. Analysis was conducted in accordance with the application manual and has been described previously [102, 103].
Scoring and prioritization of networks according to the relevance of input data
Where q = R/N defines the ratio of marked objects.
It is essential that these equations are invariant in terms of exchange of n for R which means that the "subset" and "marked" are equivalent and symmetrical sets. Importantly, in the cases of r > n, r > R or r <R + n - N, P(r, n, R, N) = 0
Where N is the total number of nodes after filtration, R is the number of nodes in the input list or the nodes associated with experimental data, n is the number of nodes in the network,r is the number of the network's nodes associated with experimental data or included in the input list, and μ and σ are the mean and dispersion of the hypergeometric distribution described above.
P-value and evaluation of statistical significance of networks
Where: N-total number of nodes in MetaCore Database, R – number of network's objects corresponding to the genes and proteins in the list; n- total number of nodes in each map or network generated from the list; r- number of nodes with data in each map or small network generated from the list.
Statistical test for network non-randomness
If we consider a list of nodes corresponding to experimentally altered genes and proteins, this pre-selected list is then used for building the networks using one of the algorithms. The statistical significance of the resulting networks can be defined using the DI algorithm, which displays only the nodes from the input list connected to each other in one step. We can now evaluate the probability of random generation of an interaction cluster of equal or larger size than the number of nodes in the DI network. After defining the list of nodes, a random subset is selected from this list, the network is built with the same algorithm and settings, and the size of the largest network cluster is calculated. The procedure is repeated many times (e.g. 1000 or 10,000 trials) and statistics are accumulated on the number of times the largest cluster of a certain size is generated. The ratio of the number of random clusters the same as the original DI network or larger compared to the total number of trials is a parameter describing the significance of the input list. For example, we calculated these parameters in 10,000 trials using the gene list with unique identifiers from Affymetrix U95 human array as the initial set.
Among the total gene set recognized by MetaCore on the Affymetrix U95A array were 5117 network nodes; random samples of 50 to 500 nodes were selected as an input for network building using the DI algorithm with the same settings as for the original network. The mean cluster size and p-values are presented in see Additional File 1, Supplement Table S12. The calculated p-value of the large DI cluster of 196 out of 484 random nodes was < 0.0001. Calculation was based on statistical probability that 196 genes form a network by random chance.
Sagittal sections of four human normal, and four glaucomatous optic nerve heads, 6 μm thick, were used for immunodetection of glial fibrillary acid protein (GFAP), clusterin and complement protein C3c. Human anti-clusterin antibodies were purchased from Upstate (Lake Placid, NY) and applied at a dilution of 1:200; human anti-C3c sheep polyclonal antibodies (Abcam, Cambridge, UK) were used at a dilution of 1:250. For double immunofluorescence staining of reactive astrocytes we used rabbit polyclonal antibody against human GFAP (Sigma, St. Louis, MO) diluted 1:150. The distribution of the primary antibodies was visualized using anti-mouse IgG or anti-sheep IgG secondary antibody conjugated to the Alexa 568 fluorochrome (Molecular Probes, Eugene, OR). To visualize cell nuclei, DAPI stain (Molecular Probes) was applied at the dilution 1:10,000 in PBS buffer (Sigma). For negative controls, the primary antibody was replaced with the appropriate non-immune serum. To control for cross-reactivity in double immunofluorescence, sections were incubated with secondary antibody only. Sections of normal and glaucomatous eyes were stained simultaneously to control for individual variations in immunostaining. Stained slices were then analyzed by confocal microscopy using a Zeiss LSM 510 microscope equipped with argon, HeNe and UV lasers for multiple fluorophore excitation.
primary open angle glaucoma
optic nerve head astrocyte
retinal ganglion cell
nuclear factor kappaB
central nervous system
Shortest Path algorithm: DI: Direct Interactions algorithm
Analyze Network algorithm
We thank Andrei Bugrim, Dmitry Novikov, Eugene Kirillov, Vadim Brodianski, Dmitry Ivanov for their keen insight and help with the manuscript preparation; Robert Haselcorn, Thomas Lukas and Sujit Dike for critical reading of the manuscript, Galina Dvoriantchikova for the expert help with immunohistochemistry and Julia Shestopalov for illustrations.
This work was supported by a 2001 Michigan Life Science Corridor grant to GeneGo Inc., NIH EY06416 Grant and RPB Senior Investigator Award to MRH; NIH grant EY017991, RPB Career Development Award (VIS and SKB) and The Glaucoma Foundation grant to VIS, NIH EY06416 Grant and RPB Senior Investigator Award to MRH, NIH EY16112 grant (SKB), RPB unrestricted core grant and NIH Center Grant P30-EY014801 to the Department of Ophthalmology at the University of Miami Miller School of Medicine.
- Hernandez MR, Pena JD: The optic nerve head in glaucomatous optic neuropathy. Arch Ophthalmol. 1997, 115: 389-395.PubMedGoogle Scholar
- Swanson RA, Ying W, Kauppinen TM: Astrocyte influences on ischemic neuronal death. Curr Mol Med. 2004, 4: 193-205.PubMedGoogle Scholar
- Varela HJ, Hernandez MR: Astrocyte responses in human optic nerve head with primary open-angle glaucoma. J Glaucoma. 1997, 6: 303-313.PubMedGoogle Scholar
- Sofroniew MV: Reactive astrocytes in neural repair and protection. Neuroscientist. 2005, 11: 400-407.PubMedGoogle Scholar
- Hashimoto K, Parker A, Malone P, Gabelt BT, Rasmussen C, Kaufman PS, Hernandez MR: Long-term activation of c-Fos and c-Jun in optic nerve head astrocytes in experimental ocular hypertension in monkeys and after exposure to elevated pressure in vitro. Brain Res. 2005, 1054: 103-115.PubMedGoogle Scholar
- Liu B, Neufeld AH: Expression of nitric oxide synthase-2 (NOS-2) in reactive astrocytes of the human glaucomatous optic nerve head. Glia. 2000, 30: 178-186.PubMedGoogle Scholar
- Quigley HA: Neuronal death in glaucoma. Prog Retin Eye Res. 1999, 18: 39-57.PubMedGoogle Scholar
- Thornton-Wells TA, Moore JH, Haines JL: Genetics, statistics and human disease: analytical retooling for complexity. Trends Genet. 2004, 20: 640-647.PubMedGoogle Scholar
- Jenssen TK, Laegreid A, Komorowski J, Hovog E: A literature network of human genes for high-throughput analysis of gene expression. Nature Genetics. 2001, 28: 21-28.PubMedGoogle Scholar
- Nikolsky Y, Nikolskaya T, Bugrim A: Biological networks and analysis of experimental data in drug discovery. Drug Discov Today. 2005, 10: 653-662.PubMedGoogle Scholar
- Sharan R, Ideker T: Modeling cellular machinery through biological network comparison. Nat Biotechnol. 2006, 24: 427-433.PubMedGoogle Scholar
- Barabasi A-L, Oltvai ZN: Network biology: understanding the cell's functional organization. Nature Reviews Genetics. 2004, 5: 101-113.PubMedGoogle Scholar
- Jeong H, Tombor B, Albert R, Oltvai ZN, Barabasi AL: The large-scale organization of metabolic networks. Nature. 2000, 407: 651-654.PubMedGoogle Scholar
- Ravasz E, Somera AL, Mongru DA, Oltvai ZN, Barabasi AL: Hierarchical organization of modularity in metabolic networks. Science. 2002, 297: 1551-1555.PubMedGoogle Scholar
- Milo R, Shen-Orr S, Itzkovitz S, Kashtan N, Chklovskii D, Alon U: Network motifs: simple building blocks of complex networks. Science. 2002, 298: 824-827.PubMedGoogle Scholar
- Bortoluzzi S, Romualdi C, Bisognin A, Danieli GA: Disease genes and intracellular protein networks. Physiol Genomics. 2003, 15: 223-227.PubMedGoogle Scholar
- Kitano H: Computational systems biology. Nature. 2002, 420: 206-210.PubMedGoogle Scholar
- Csete ME, Doyle JC: Reverse engineering of biological complexity. Science. 2002, 295: 1664-1669.PubMedGoogle Scholar
- Ekins S, Bugrim A, Brovold L, Kirillov E, Nikolsky Y, Rakhmatulin E, Sorokina S, Ryabov A, Serebryiskaya T, Melnikov A, et al: Algorithms for network analysis in systems-ADME/Tox using the MetaCore and MetaDrug platforms. Xenobiotica. 2006, 36: 877-901.PubMedGoogle Scholar
- Ekins S, Kirillov E, Rakhmatulin EA, Nikolskaya T: A novel method for visualizing nuclear hormone receptor networks relevant to drug metabolism. Drug Metab Dispos. 2005, 33: 474-481.PubMedGoogle Scholar
- Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T: Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003, 13: 2498-2504.PubMedPubMed CentralGoogle Scholar
- Hernandez MR, Agapova OA, Yang P, Salvador-Silva M, Ricard CS, Aoi S: Differential gene expression in astrocytes from human normal and glaucomatous optic nerve head analyzed by cDNA microarray. Glia. 2002, 38: 45-64.PubMedGoogle Scholar
- Kobayashi S, Vidal I, Pena JD, Hernandez MR: Expression of neural cell adhesion molecule (NCAM) characterizes a subpopulation of type 1 astrocytes in human optic nerve head. Glia. 1997, 20: 262-273.PubMedGoogle Scholar
- Mense SM, Sengupta A, Zhou M, Lan C, Bentsman G, Volsky DJ, Zhang L: Gene expression profiling reveals the profound upregulation of hypoxia-responsive genes in primary human astrocytes. Physiol Genomics. 2006, 25: 435-449.PubMedGoogle Scholar
- Rozsa FW, Scott KM, Pawar H, Samples JR, Wirtz MK, Richards JE: Differential expression profile prioritization of positional candidate glaucoma genes: the GLC1C locus. Arch Ophthalmol. 2007, 125: 117-127.PubMedGoogle Scholar
- Van Beek J, Chan P, Bernaudin M, Petit E, MacKenzie ET, Fontaine M: Glial responses, clusterin, and complement in permanent focal cerebral ischemia in the mouse. Glia. 2000, 31: 39-50.PubMedGoogle Scholar
- Falsig J, Latta M, Leist M: Defined inflammatory states in astrocyte cultures: correlation with susceptibility towards CD95-driven apoptosis. J Neurochem. 2004, 88: 181-193.PubMedGoogle Scholar
- Ridet JL, Malhotra SK, Privat A, Gage FH: Reactive astrocytes: cellular and molecular cues to biological function. Trends Neurosci. 1997, 20: 570-577.PubMedGoogle Scholar
- Lang B, Liu HL, Liu R, Feng GD, Jiao XY, Ju G: Astrocytes in injured adult rat spinal cord may acquire the potential of neural stem cells. Neuroscience. 2004, 128: 775-783.PubMedGoogle Scholar
- Goldshmit Y, Galea MP, Wise G, Bartlett PF, Turnley AM: Axonal regeneration and lack of astrocytic gliosis in EphA4-deficient mice. J Neurosci. 2004, 24: 10064-10073.PubMedGoogle Scholar
- Sobel RA: Ephrin A receptors and ligands in lesions and normal-appearing white matter in multiple sclerosis. Brain Pathol. 2005, 15: 35-45.PubMedGoogle Scholar
- Sandvig A, Berry M, Barrett LB, Butt A, Logan A: Myelin-, reactive glia-, and scar-derived CNS axon growth inhibitors: expression, receptor signaling, and correlation with axon regeneration. Glia. 2004, 46: 225-251.PubMedGoogle Scholar
- Sappington RM, Chan M, Calkins DJ: Interleukin-6 protects retinal ganglion cells from pressure-induced death. Invest Ophthalmol Vis Sci. 2006, 47: 2932-2942.PubMedGoogle Scholar
- Schmidt JF, Agapova OA, Yang P, Kaufman PL, Hernandez MR: Expression of ephrinB1 and its receptor in glaucomatous optic neuropathy. Br J Ophthalmol. 2007, 91: 1219-1224.PubMedPubMed CentralGoogle Scholar
- Chou YC, Lin SB, Tsai LH, Tsai HI, Lin CM: Cholesterol deficiency increases the vulnerability of hippocampal glia in primary culture to glutamate-induced excitotoxicity. Neurochem Int. 2003, 43: 197-209.PubMedGoogle Scholar
- Johnson EC, Jia L, Cepurna WO, Doser TA, Morrison JC: Global changes in optic nerve head gene expression after exposure to elevated intraocular pressure in a rat glaucoma model. Invest Ophthalmol Vis Sci. 2007, 48: 3161-3177.PubMedPubMed CentralGoogle Scholar
- Wang X, Ng YK, Tay SS: Factors contributing to neuronal degeneration in retinas of experimental glaucomatous rats. J Neurosci Res. 2005, 82: 674-689.PubMedGoogle Scholar
- Levkovitch-Verbin H, Quigley HA, Martin KR, Harizman N, Valenta DF, Pease ME, Melamed S: The transcription factor c-jun is activated in retinal ganglion cells in experimental rat glaucoma. Exp Eye Res. 2005, 80: 663-670.PubMedGoogle Scholar
- Agapova OA, Kaufman PL, Hernandez MR: Androgen receptor and NFkB expression in human normal and glaucomatous optic nerve head astrocytes in vitro and in experimental glaucoma. Exp Eye Res. 2006, 82: 1053-1059.PubMedGoogle Scholar
- Ahmed F, Brown KM, Stephan DA, Morrison JC, Johnson EC, Tomarev SI: Microarray analysis of changes in mRNA levels in the rat retina after experimental elevation of intraocular pressure. Invest Ophthalmol Vis Sci. 2004, 45: 1247-1258.PubMedGoogle Scholar
- Steele MR, Inman DM, Calkins DJ, Horner PJ, Vetter ML: Microarray analysis of retinal gene expression in the DBA/2J model of glaucoma. Invest Ophthalmol Vis Sci. 2006, 47: 977-985.PubMedGoogle Scholar
- Kuehn MH, Kim CY, Ostojic J, Bellin M, Alward WL, Stone EM, Sakaguchi DS, Grozdanic SD, Kwon YH: Retinal synthesis and deposition of complement components induced by ocular hypertension. Exp Eye Res. 2006, 83: 620-628.PubMedGoogle Scholar
- Agapova OA, Yang P, Wang WH, Lane DA, Clark AF, Weinstein BI, Hernandez MR: Altered expression of 3 alpha-hydroxysteroid dehydrogenases in human glaucomatous optic nerve head astrocytes. Neurobiol Dis. 2003, 14: 63-73.PubMedGoogle Scholar
- Schmidt J, Agapova OA, Yang P, Kaufman PL, Hernandez MR: Expression of EphrinB1 and its Receptor in Glaucomatous Optic Neuropathy. Br J Ophthalmol. 2007, 91 (9): 1219-1224.PubMedPubMed CentralGoogle Scholar
- Falsig J, Porzgen P, Lund S, Schrattenholz A, Leist M: The inflammatory transcriptome of reactive murine astrocytes and implications for their innate immune function. J Neurochem. 2006, 96: 893-907.PubMedGoogle Scholar
- Yang P, Agapova O, Parker A, Shannon W, Pecen P, Duncan J, Salvador-Silva M, Hernandez MR: DNA microarray analysis of gene expression in human optic nerve head astrocytes in response to hydrostatic pressure. Physiol Genomics. 2004, 17: 157-169.PubMedGoogle Scholar
- Liu B, Chen H, Johns TG, Neufeld AH: Epidermal growth factor receptor activation: an upstream signal for transition of quiescent astrocytes into reactive astrocytes after neural injury. J Neurosci. 2006, 26: 7532-7540.PubMedGoogle Scholar
- Hernandez MR: The optic nerve head in glaucoma: role of astrocytes in tissue remodeling. Prog Retin Eye Res. 2000, 19: 297-321.PubMedGoogle Scholar
- Liang H, Baudouin C, Behar-Cohen F, Crisanti P, Omri B: Protein kinase C-zeta mediates retinal degeneration in response to TNF. J Neuroimmunol. 2007, 183: 104-110.PubMedGoogle Scholar
- Kirwan RP, Leonard MO, Murphy M, Clark AF, O'Brien CJ: Transforming growth factor-beta-regulated gene transcription and protein expression in human GFAP-negative lamina cribrosa cells. Glia. 2005, 52: 309-324.PubMedGoogle Scholar
- Yu H, Zhu X, Greenbaum D, Karro J, Gerstein M: TopNet: a tool for comparing biological sub-networks, correlating protein properties with topological statistics. Nucleic Acids Res. 2004, 32: 328-337.PubMedPubMed CentralGoogle Scholar
- Lukashin AV, Lukashev ME, Fuchs R: Topology of gene expression networks as revealed by data mining and modeling. Bioinformatics. 2003, 19: 1909-1916.PubMedGoogle Scholar
- Kirillov E: MetaCore: An Integrated Software Platform and Statistical Evaluation Method for Network Analysis and Mapping of Human High Throughput Data. Appl Bioinform. 2006.Google Scholar
- Rulten SL, Ripley TL, Hunt CL, Stephens DN, Mayne LV: Sp1 and NFkappaB pathways are regulated in brain in response to acute and chronic ethanol. Genes Brain Behav. 2006, 5: 257-273.PubMedGoogle Scholar
- Garcion E, Wion-Barbot N, Montero-Menei CN, Berger F, Wion D: New clues about vitamin D functions in the nervous system. Trends Endocrinol Metab. 2002, 13: 100-105.PubMedGoogle Scholar
- Magrassi L, Bono F, Milanesi G, Butti G: Vitamin D receptor expression in human brain tumors. J Neurosurg Sci. 1992, 36: 27-30.PubMedGoogle Scholar
- Wiseman H: Vitamin D is a membrane antioxidant. Ability to inhibit iron-dependent lipid peroxidation in liposomes compared to cholesterol, ergosterol and tamoxifen and relevance to anticancer action. FEBS Lett. 1993, 326: 285-288.PubMedGoogle Scholar
- Shaulian E, Karin M: AP-1 as a regulator of cell life and death. Nat Cell Biol. 2002, 4: E131-136.PubMedGoogle Scholar
- Jones BW, Means TK, Heldwein KA, Keen MA, Hill PJ, Belisle JT, Fenton MJ: Different Toll-like receptor agonists induce distinct macrophage responses. J Leukoc Biol. 2001, 69: 1036-1044.PubMedGoogle Scholar
- Kawai T, Akira S: TLR signaling. Cell Death Differ. 2006, 13: 816-825.PubMedGoogle Scholar
- Takeuchi O, Akira S: Toll-like receptors; their physiological role and signal transduction system. Int Immunopharmacol. 2001, 1: 625-635.PubMedGoogle Scholar
- Rivieccio MA, John GR, Song X, Suh HS, Zhao Y, Lee SC, Brosnan CF: The cytokine IL-1beta activates IFN response factor 3 in human fetal astrocytes in culture. J Immunol. 2005, 174: 3719-3726.PubMedGoogle Scholar
- Dhandapani KM, Hadman M, De Sevilla L, Wade MF, Mahesh VB, Brann DW: Astrocyte protection of neurons: role of transforming growth factor-beta signaling via a c-Jun-AP-1 protective pathway. J Biol Chem. 2003, 278: 43329-43339.PubMedGoogle Scholar
- Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ, Shen D, Boca SM, Barber T, Ptak J, et al: The genomic landscapes of human breast and colorectal cancers. Science. 2007, 318: 1108-1113.PubMedGoogle Scholar
- Delfino FJ, Boustead JN, Fix C, Walker WH: NF-kappaB and TNF-alpha stimulate androgen receptor expression in Sertoli cells. Mol Cell Endocrinol. 2003, 201: 1-12.PubMedGoogle Scholar
- Tam NN, Gao Y, Leung YK, Ho SM: Androgenic regulation of oxidative stress in the rat prostate: involvement of NAD(P)H oxidases and antioxidant defense machinery during prostatic involution and regrowth. Am J Pathol. 2003, 163: 2513-2522.PubMedPubMed CentralGoogle Scholar
- Shibata N, Asayama K, Hirano A, Kobayashi M: Immunohistochemical study on superoxide dismutases in spinal cords from autopsied patients with amyotrophic lateral sclerosis. Dev Neurosci. 1996, 18: 492-498.PubMedGoogle Scholar
- Agapova OA, Malone PE, Hernandez MR: A neuroactive steroid 5alpha-androstane-3alpha,17beta-diol regulates androgen receptor level in astrocytes. J Neurochem. 2006, 98: 355-363.PubMedGoogle Scholar
- Malone PE, Hernandez MR: 4-Hydroxynonenal, a product of oxidative stress, leads to an antioxidant response in optic nerve head astrocytes. Exp Eye Res. 2007, 84: 444-454.PubMedGoogle Scholar
- Funk JL, Trout CR, Wei H, Stafford G, Reichlin S: Parathyroid hormone-related protein (PTHrP) induction in reactive astrocytes following brain injury: a possible mediator of CNS inflammation. Brain Res. 2001, 915: 195-209.PubMedGoogle Scholar
- Cowell RM, Plane JM, Silverstein FS: Complement activation contributes to hypoxic-ischemic brain injury in neonatal rats. J Neurosci. 2003, 23: 9459-9468.PubMedGoogle Scholar
- Wiggins AK, Shen PJ, Gundlach AL: Delayed, but prolonged increases in astrocytic clusterin (ApoJ) mRNA expression following acute cortical spreading depression in the rat: evidence for a role of clusterin in ischemic tolerance. Brain Res Mol Brain Res. 2003, 114: 20-30.PubMedGoogle Scholar
- Edwards AO, Ritter R, Abel KJ, Manning A, Panhuysen C, Farrer LA: Complement factor H polymorphism and age-related macular degeneration. Science. 2005, 308: 421-424.PubMedGoogle Scholar
- Tornqvist E, Liu L, Aldskogius H, Holst HV, Svensson M: Complement and clusterin in the injured nervous system. Neurobiol Aging. 1996, 17: 695-705.PubMedGoogle Scholar
- Zenkel M, Kruse FE, Junemann AG, Naumann GO, Schlotzer-Schrehardt U: Clusterin deficiency in eyes with pseudoexfoliation syndrome may be implicated in the aggregation and deposition of pseudoexfoliative material. Invest Ophthalmol Vis Sci. 2006, 47: 1982-1990.PubMedGoogle Scholar
- Xie Z, Harris-White ME, Wals PA, Frautschy SA, Finch CE, Morgan TE: Apolipoprotein J (clusterin) activates rodent microglia in vivo and in vitro. J Neurochem. 2005, 93: 1038-1046.PubMedGoogle Scholar
- Han BH, DeMattos RB, Dugan LL, Kim-Han JS, Brendza RP, Fryer JD, Kierson M, Cirrito J, Quick K, Harmony JA, et al: Clusterin contributes to caspase-3-independent brain injury following neonatal hypoxia-ischemia. Nat Med. 2001, 7: 338-343.PubMedGoogle Scholar
- Stephanou A: Role of STAT-1 and STAT-3 in ischaemia/reperfusion injury. J Cell Mol Med. 2004, 8: 519-525.PubMedGoogle Scholar
- Calo V, Migliavacca M, Bazan V, Macaluso M, Buscemi M, Gebbia N, Russo A: STAT proteins: from normal control of cellular events to tumorigenesis. J Cell Physiol. 2003, 197: 157-168.PubMedGoogle Scholar
- Lee JH, Park EJ, Kim OS, Kim HY, Joe EH, Jou I: Double-stranded RNA-activated protein kinase is required for the LPS-induced activation of STAT1 inflammatory signaling in rat brain glial cells. Glia. 2005, 50: 66-79.PubMedGoogle Scholar
- Bhattacharya SK, Crabb JS, Bonilha VL, Gu X, Takahara H, Crabb JW: Proteomics implicates peptidyl arginine deiminase 2 and optic nerve citrullination in glaucoma pathogenesis. Invest Ophthalmol Vis Sci. 2006, 47: 2508-2514.PubMedGoogle Scholar
- Ekins SNY, Bugrim A, Kirillov E, Nikolskaya T: Pathway mapping tools for analysis of high content data. High Content Screening Handbook. 2007, Humana Press, 356: 319-350. [Methods Mol. Biol.]Google Scholar
- Stasi K, Nagel D, Yang X, Wang RF, Ren L, Podos SM, Mittag T, Danias J: Complement component 1Q (C1Q) upregulation in retina of murine, primate, and human glaucomatous eyes. Invest Ophthalmol Vis Sci. 2006, 47: 1024-1029.PubMedGoogle Scholar
- Tezel G, Yang X, Cai J: Proteomic identification of oxidatively modified retinal proteins in a chronic pressure-induced rat model of glaucoma. Invest Ophthalmol Vis Sci. 2005, 46: 3177-3187.PubMedGoogle Scholar
- Picklo MJ, Olson SJ, Markesbery WR, Montine TJ: Expression and activities of aldo-keto oxidoreductases in Alzheimer disease. J Neuropathol Exp Neurol. 2001, 60: 686-695.PubMedGoogle Scholar
- Arai K, Nishiyama N, Matsuki N, Ikegaya Y: Neuroprotective effects of lipoxygenase inhibitors against ischemic injury in rat hippocampal slice cultures. Brain Res. 2001, 904: 167-172.PubMedGoogle Scholar
- Sun GY, Xu J, Jensen MD, Yu S, Wood WG, Gonzalez FA, Simonyi A, Sun AY, Weisman GA: Phospholipase A2 in astrocytes: responses to oxidative stress, inflammation, and G protein-coupled receptor agonists. Mol Neurobiol. 2005, 31: 27-41.PubMedGoogle Scholar
- Herrmann O, Baumann B, de Lorenzi R, Muhammad S, Zhang W, Kleesiek J, Malfertheiner M, Kohrmann M, Potrovita I, Maegele I, et al: IKK mediates ischemia-induced neuronal death. Nat Med. 2005, 11: 1322-1329.PubMedGoogle Scholar
- Kitaoka Y, Kumai T, Kitaoka Y, Lam TT, Munemasa Y, Isenoumi K, Motoki M, Kuribayashi K, Kogo J, Kobayashi S, Ueno S: Nuclear factor-kappa B p65 in NMDA-induced retinal neurotoxicity. Brain Res Mol Brain Res. 2004, 131: 8-16.PubMedGoogle Scholar
- Pizzi M, Sarnico I, Boroni F, Benetti A, Benarese M, Spano PF: Inhibition of IkappaBalpha phosphorylation prevents glutamate-induced NF-kappaB activation and neuronal cell death. Acta Neurochir Suppl. 2005, 93: 59-63.PubMedGoogle Scholar
- Schneider A, Martin-Villalba A, Weih F, Vogel J, Wirth T, Schwaninger M: NF-kappaB is activated and promotes cell death in focal cerebral ischemia. Nat Med. 1999, 5: 554-559.PubMedGoogle Scholar
- Davies AM: Regulation of neuronal survival and death by extracellular signals during development. Embo J. 2003, 22: 2537-2545.PubMedPubMed CentralGoogle Scholar
- Huang Y, Krein PM, Muruve DA, Winston BW: Complement factor B gene regulation: synergistic effects of TNF-alpha and IFN-gamma in macrophages. J Immunol. 2002, 169: 2627-2635.PubMedGoogle Scholar
- Molina-Holgado E, Arevalo-Martin A, Castrillo A, Bosca L, Vela JM, Guaza C: Interleukin-4 and interleukin-10 modulate nuclear factor kappaB activity and nitric oxide synthase-2 expression in Theiler's virus-infected brain astrocytes. J Neurochem. 2002, 81: 1242-1252.PubMedGoogle Scholar
- Choi JS, Kim JA, Kim DH, Chun MH, Gwag BJ, Yoon SK, Joo CK: Failure to activate NF-kappaB promotes apoptosis of retinal ganglion cells following optic nerve transection. Brain Res. 2000, 883: 60-68.PubMedGoogle Scholar
- Takahashi Y, Katai N, Murata T, Taniguchi SI, Hayashi T: Development of spontaneous optic neuropathy in NF-kappaBetap50-deficient mice: requirement for NF-kappaBetap50 in ganglion cell survival. Neuropathol Appl Neurobiol. 2007, 33: 692-705.PubMedGoogle Scholar
- Segre D, Deluna A, Church GM, Kishony R: Modular epistasis in yeast metabolism. Nat Genet. 2005, 37: 77-83.PubMedGoogle Scholar
- Mi H, Barres BA: Purification and characterization of astrocyte precursor cells in the developing rat optic nerve. J Neurosci. 1999, 19: 1049-1061.PubMedGoogle Scholar
- Yang P, Hernandez MR: Purification of astrocytes from adult human optic nerve heads by immunopanning. Brain Res Brain Res Protoc. 2003, 12: 67-76.PubMedGoogle Scholar
- Pena JD, Netland PA, Vidal I, Dorr DA, Rasky A, Hernandez MR: Elastosis of the lamina cribrosa in glaucomatous optic neuropathy. Exp Eye Res. 1998, 67: 517-524.PubMedGoogle Scholar
- Yucel YH, Kalichman MW, Mizisin AP, Powell HC, Weinreb RN: Histomorphometric analysis of optic nerve changes in experimental glaucoma. J Glaucoma. 1999, 8: 38-45.PubMedGoogle Scholar
- Ekins S, Nikolsky Y, Nikolskaya T: Techniques: application of systems biology to absorption, distribution, metabolism, excretion and toxicity. Trends Pharmacol Sci. 2005, 26: 202-209.PubMedGoogle Scholar
- Nikolsky Y, Ekins S, Nikolskaya T, Bugrim A: A novel method for generation of signature networks as biomarkers from complex high throughput data. Toxicol Lett. 2005, 158: 20-29.PubMedGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1755-8794/2/24/prepub