Demographic characteristics of the normal AA and CA donors used in this study are detailed in Additional data file 2. Demographic and clinical data for AA donors with glaucoma (AAGs) and CA donors with glaucoma (CAGs) included in the microarray analyses and other assays are detailed in Additional data file 1. Twelve eyes from ten CAG donors and six eyes from AAG donors were used in this study. Glaucoma drug treatment history was available for some POAG donors. None of the drug treatments are known to affect astrocytes in the ONH. The degree of glaucomatous damage in donors with POAG was assessed using histories when available and by evaluating axon degeneration in cross-sections of the myelinated optic nerve (Additional data file 1). A limitation of this study is that only six eyes from three AAG donors were available due to the extreme rarity of these samples. Consequently, we used all six eyes to generate primary cultures for all experiments in our study. Primary cultures of samples from AAG and CAG donors were fully characterized as ONH astrocytes as described in detail earlier 11.

For the comparisons amongst the four groups, our primary focus was to establish the differentially expressed genes between AAG and CAG donors (Additional data file 7); our secondary focus was the comparison between normal and glaucomatous astrocytes and our tertiary focus was to identify differentially expressed genes within each population: AAG versus AA and CAG versus CA.

The comparisons allowed us to identify the unique gene expression profile in AAG astrocytes compared to CAG astrocytes and AAG compared to AA (Additional data file 8). In addition, we identified a common group of genes that exhibit a similar gene expression pattern in both AAG and CAG compared to normal AA and CA astrocytes, which we named common glaucoma-related genes (Tables 1 and 2).

Eight eyes from six CAG donors were used to generate astrocytes for eight Hu95v2 chips. Six eyes from three AAG were used to generate astrocytes for six Hu95Av2 chips and six Hu133A 2.0 chips. Eighteen Hu133 2.0 chips from nine normal AA and nine normal CA donors, and seven Hu95v2 chips from six normal CA donors were used for comparisons within the appropriate platform. All microarray data have been deposited in the NCBI GEO database under the series accession number GSE9963.

The data measured by the two types of chips were normalized separately by RMA normalization as described in Materials and methods. Differentially expressed genes required an up or down fold-change of more than 1.5-fold (p < 0.01, false discovery rate < 0.05). A total of 618 genes were differentially expressed in AAG-CAG comparisons, 484 upregulated and 134 downregulated (Additional data file 7); 509 genes were differentially expressed in AAG compared to normal AA astrocytes, 167 upregulated and 342 downregulated (Additional data file 5); and 195 genes were differentially expressed in the CAG-CA comparison, 132 upregulated and 63 downregulated (Additional data file 6). We used empirical Bayesian methods to identify differentially expressed genes; both our results (not shown) and previous studies 1213 have suggested that the empirical Bayesian method has performance similar to statistical analysis of microarrays (SAM). To reduce batch effects, we added fold-change criteria because genes with larger fold-change are less likely to be affected by such effects.

Gene Ontology (GO) analysis of differential expression in glaucomatous astrocytes was done with GoMiner 14. There were 33 significant categories for CAG-CA, 80 for AAG-AA, and 67 for AAG-CAG comparisons (p < 0.01). The significant genes in selected categories were mined using GOstats in Bioconductor (Additional data file 9). The phosphorylation category (GoID: 16310) was significant in the three datasets. The percent distribution of the genes common to all of the datasets in this category was determined (Additional data file 10). For example, the genes encoding myosin light chain kinase (MYLK) and calcium/calmodulin-dependent serine protein kinase (CASK1) were found in all three glaucoma comparisons. Those encoding the regulatory subunit of phosphatidylinositol-3-kinase (PIK3R1), transforming growth factor (TGF)β-receptor 2 (TGFBR2), ERBB2, and Ephrin receptor A5 were some of the genes found in two datasets (AAG-CAG and AAG-AA). Similarly, another category with overlaps between the datasets was cell-cell signaling (Additional data file 10). Some of the genes in this category include those encoding latent transforming growth factor beta binding protein 4 (LTBP4), the glutamate receptor subunit (GRIK2), and parathyroid hormone-like protein (PTHLH). As we show below, expansion of these and other GO categories using network-protein interaction software yielded three networks that include differentially expressed GTPases, protein kinases, transmembrane receptors, and proteins involved in trafficking at cellular membranes. Altogether, the GO analysis suggests that alterations in the signaling networks that regulate cell motility, polarity, adhesion, and trafficking are present in glaucomatous astrocytes. Moreover, the overlap among the datasets in multiple categories suggests that there is a spectrum of changes in gene expression in glaucoma.

Three detailed network maps were constructed from the differential gene expression data. We focused mainly on the differences between AAG and CAG as this difference represents the maximal differential expression group (Additional data file 7). The networks include regulation of myosin, actin, TGFβ signaling and protein trafficking. For the myosin network, the initial node was myosin light chain kinase (MYLK) (Figure 1b). The actin regulatory networks were initiated using the TGFβ receptors (Figure 2a), and the protein trafficking networks were initiated using GOLGA3, catenin beta1 (CTNNB1) and RAB4A as nodes (Figure 3a). These were expanded using the BioGrid database for protein-protein interactions. In each network graph, the differentially expressed genes are shown by large nodes and font (red for increased, blue for decreased expression), while the connecting genes that are not differentially expressed are shown by black smaller nodes and font. Expression data for network nodes that are differentially expressed in the AAG-CAG comparison (Additional data file 7) are included in Table 3. Some network nodes were also selected from differentially expressed genes in AAG-AA (Additional data file 5) and in common AAG-AA and CAG-CA comparisons (Tables 1 and 2). In the description of each network, we present selected experimental data that verify changes in gene expression and effects on function.

From the microarray and quantitative RT-PCR (qRT-PCR) data, the following genes related to myosin regulation were differentially expressed in AAG: MYLK, MYPT1, RAC2, CALM1, RPS6KA3, MYH10, and PIK3R1. Shown in Figure 1b is the network of proteins associated with the phosphorylation of the regulatory light chain of myosin II and activation of myosin-ATPase (MYH10). Two network nodes are critical for the regulation of myosin. These include MYLK, a calmodulin-activated protein kinase that phosphorylates Ser19 on the myosin regulatory light chain and MYPT1, the regulatory subunit of myosin-light chain phosphatase, which dephosphorylates the myosin light chain. We found that both genes were expressed in AAG astrocytes at significantly higher levels than in CAG astrocytes (Table 3). Similarly, calmodulin (CALM1), the activator of MYLK is also upregulated in AAG astrocytes (Table 3)

The upregulation of MYLK suggests that the myosin regulatory system may exhibit increased responsiveness towards modulation by various cellular second messenger signaling systems such as Ca²⁺, diacylglycerol, and cyclic nucleotides 21. Similarly, changes in expression of RAC2 indicate that other members of the Rho-family signaling network are altered in AAG astrocytes (Figure 1c). These changes allow us to predict that the myosin-regulated motility may be sensitized to signals from Ca2⁺, Rho GTPase, and growth/trophic factors coupled to the activation of phosphoinositides. Within the phosphoinositide pathway, PIK3R1 is upregulated in AAG astrocytes (Figure 1c). The PIK3R1 pathway is important for the motility of ONH astrocytes 22 and their responses to increased hydrostatic pressure 20. PIK3R1 is the regulatory subunit of the lipid kinase that transforms phosphoinositide (4,5) biphosphate (PIP2) into the triphosphate (PIP3). PIP3 in turn mediates activation of several of the Rho GTPases as well as selected protein kinases. Thus, in AAG astrocytes, lipid-activated pathways that modulate astrocyte motility are altered.

ERK1 potentiates MYLK activity through phosphorylation 23 and interacts with PEA15 (Phosphoprotein enriched in astrocytes) 24. The increased expression of the S6-family kinase (RPS6KA3) may compete with ERK1 for binding to the phosphoprotein PEA15 25, potentially increasing the pool of active ERK1. Consistent with this finding, we have shown that ERK1 is activated in normal CA ONH astrocytes, under increased hydrostatic pressure and in experimental glaucoma in primates 26. Thus, myosin-based motility may be influenced by changes in MYLK expression and potentiation through ERK1 activation under hydrostatic pressure.

Co-localization of MYLK and glial acidic fibrillar protein (GFAP) by immunohistochemistry indicates that ONH astrocytes in tissue sections in the lamina cribrosa of normal AA and AAG expressed visibly higher levels of MYLK protein in situ (Figure 4a).

The MYLK gene has multiple genes within its locus 27. In some tissues up to three transcripts are expressed, including for long and short forms of the kinase and a protein identical to the carboxyl-terminal sequence 27. ONH astrocytes express both the 130 kDa (MYLK-130) and 210 kDa (MYLK-210) kinase isoforms and we quantified changes in both using standard densitometry measurements. Western blots (Figure 4b) show that the fraction of MYLK-210 in ONH astrocytes is higher in AAG and CAG compared to normal astrocytes, while the fraction of the MYLK-130 isoform decreases (Figure 4b). These differences were quantified using densitometry (Figure 4c, d). Thus, in glaucoma there appears to be MYLK isoform switching towards the larger protein. The difference between the two proteins is the presence of an amino-terminal extension in the 210 kDa species that contains additional actin binding domains. Other studies have shown that MYLK-210 displays enhanced interaction with the actin cytoskeleton compared to the 130 kDa isoform 2829. These results are consistent with the enhanced migration of ONH astrocytes mediated in part by increased expression of MYLK-210.

MYLK variants have been found to confer risk of lung injury 30, asthma or sepsis 31, particularly in African Americans 32. Some of the common polymorphisms in MYLK affect its expression 31. Therefore, in some populations, it is possible that the effects of increased expression of MYLK may be further modified by genetic polymorphisms.

From the microarray and qRT-PCR data the following genes were differentially expressed in AAG: TGFBR2, TGFBR1, SMAD3, NCK1, PTPN11, ARHGEF7, PDLIM1, LM04, and PLEC1. Figure 2a shows several signal transduction networks that participate in the regulation of actin. Remodeling or redistribution of actin at cellular edges is an essential part of establishing cell polarity 33 and the formation of processes in astrocytes 34. Actomyosin interactions and actin polymerization are regulated by intracellular proteins such as α-actinin (ACTN4) and the ARP protein complex (ACTR2, WASP: Figure 2a). These networks involve the Rho GTPase signaling pathway. Therefore, we used a pull-down Rho activation assay to measure activated Rho in cell lysates. ONH astrocytes from CAG and AAG donors exhibited significantly higher Rho activity compared to those from normal AA and CA donors (Figure 2b, c), consistent with the differential expression of Rho regulatory components. Rho activity was also increased in astrocytes exposed to elevated hydrostatic pressure 35. Thus, increased Rho activity is another contributor towards increased migration of AAG astrocytes. We suspect that Rho activity may be altered by changes in the signaling proteins included in these networks. For example, RAC2 and ARGEF7 are upregulated in AAG. The Rho-family GTPase, RAC2, is downstream of TGFβ signaling 36 and ARHGEF7 stimulates guanine nucleotide exchange on Rho family GTP-binding proteins. We further elaborated changes in TGFβ signaling as a driver to changes in Rho activity.

TGFβ1 and TGFβ2 act via TGFBR1 and TGFBR2 receptors. Using qRT-PCR we confirmed that TGFBR2 and the downstream signaling protein SMAD3 are up-regulated in AAG astrocytes, suggesting increased responsiveness (Figure 5a). TGFBR1 is down-regulated in AAG compared to CAG (Figure 5a). SMAD proteins not only function as transcriptional regulators in ONH astrocytes 37 and other cells in the central nervous system 38, but also participate in the regulation of cell polarity. SMAD3 was also upregulated in ONH astrocytes exposed to hydrostatic pressure in vitro, suggesting that pressure activates the TGFβ pathway 35. In addition, LM04, a LIM domain protein that modulates SMAD3 transcriptional activity 39, is upregulated in glaucomatous astrocytes in both populations (Table 1). One path that limits SMAD3 signaling is ubiquitin-linked degradation by SMURF2. Although SMURF2 expression is not altered in glaucomatous astrocytes, SMURF2 is downregulated by an increase in hydrostatic pressure 35. Thus, there may be additional potentiation of TGFβ signaling in AAG astrocytes with changes in intraocular pressure, which may be a susceptibility factor to glaucomatous changes in the AA population.

TGFβ regulates cellular motility through two components. One is through the expression of extracellular matrix (ECM) proteins, which will be discussed in detail below. Contractile forces are transmitted to the ECM through actin-based stress fibers via focal adhesions, which are assemblies of ECM proteins, transmembrane receptors, and cytoplasmic structural and signaling proteins, such as integrins. TGFβ modulates integrin-mediated cellular migration, where FYN is one of the primary signal transducing proteins. A second component of TGFβ signaling is the regulation of cell polarity. For example, PARD3 and PARD6 are part of a multi-component polarity complex that controls polarized cell migration 40. These complexes involve the Rho, CDC42, and RAC signaling pathways, which provide the means to remodel actin during migration 3341

As shown in Figure 2a, NCK1 was upregulated in AAG (Table 3). The Nck1 SH2/SH3 adaptor couples phosphotyrosine signals to the actin cytoskeleton and receptor signaling to the regulatory machinery of the cytoskeleton 42. The enigma family member PDLIM1 was upregulated in AAG astrocytes (Table 3) and functions by allowing interactions among cytoskeletal proteins through PDZ and amino LIM domains 4344. Downregulation of other actin binding proteins such as PLEC1 (Table 3) may alter actin dynamics with respect to cytoskeletal changes induced by Rho-GTPase, phospholipids, and tyrosine kinase (Src) mediated signaling 45.

Figure 5b illustrates immunohistochemistry of the TGFBR2 on astrocytes in normal and glaucomatous ONH tissue. GFAP positive astrocytes in the lamina cribrosa of AAG exhibit higher expression of TGFBR2 compared with astrocytes in normal ONH tissue. Consistent with these findings, western blots of lysates of ONH astrocytes from AAG indicate higher levels of TGFBR2 protein compared to the normal tissue and CAG (Figure 5c).

To further investigate alterations in TGFβ signaling in ONH astrocytes, we examined the production of TGFβ1 and TGFβ2. As seen in Figure 5d, TGFβ2 is the primary form of TGFβ produced by ONH astrocytes 46. There are significantly increased levels of secreted TGFβ1 in AA compared to CA astrocyte supernatants but the increases in AAG and CAG astrocytes were not significant compared to normal astrocytes. These data suggest that most of the changes in TGFβ signaling are due to alterations at the level of TGFβ receptors in astrocytes from AAG.

Mutations in TGFBR2 are associated with Marfan syndrome type 2 474849. Ocular abnormalities, including glaucoma, are associated with Marfan syndrome type 1 in which there are mutations in the gene for fibrillin (FBN1) 50. However, it has not been established that mutations of TGFBR2 are associated with ocular problems in Marfan syndrome type 2 4849.

The secondary and tertiary comparisons identified genes that were differentially expressed in AAG compared to AA and in CAG compared to CA, including BMP1, LTBP1, AMIGO2, SLIT2, GPC1, and OLR1 (Tables 1 and 2). Selected genes were confirmed by qRT-PCR (Figure 7).

In this list we found that specific cell-surface-associated proteins are downregulated in glaucoma. These include BMP1, which activates cleavage of LTBP1 proteins that release nascent TGFβ1 65, and AMIGO2, a type I transmembrane protein that regulates axon extension 66. Down-regulation of BMP1 may reduce the levels of free TGFβ1 and thus unbalance signaling between TGFβ isoforms. A decrease in AMIGO2 might negatively impact axon survival.

Two differentially expressed genes that are involved in reactive astrocyte responses to neuronal injury are SLIT2 and GPC1 (glypican-1). SLIT2 serves as a chemorepellant for multiple types of axons 67, while GPC1 is a proteoglycan that binds SLIT2 67. Upregulation of expression of SLIT2 and a reduction of GPC1 by glaucomatous astrocytes suggest an inhibitory microenvironment for RGC axons in the ONH. These data are consistent with the idea that the enhanced migratory properties of glaucomatous astrocytes coupled with the release of factors that negatively impact upon axon survival are part of the pathophysiology of the disease.

Finally, lectin-like oxidized-LDL receptor (OLR1; also known as LOX-1) is highly upregulated (Additional data files 5 and 11) in AAG astrocytes. OLR1 expression is induced by TGFβ1 signaling and is known to be a component of the fluid shear stress response of endothelial cells in early atherosclerotic lesions 68. These data are further confirmation of enhanced TGFβ signaling in AAG astrocytes as suggested by the differential receptor expression described earlier.

Two genes directly involved in Ca²⁺ homeostasis are differentially regulated in ONH astrocytes of AAG (Additional data files 5 and 7). CACNB4 encodes a beta subunit of the voltage-dependent calcium channel complex. CACNB4 plays an important role in calcium channel function by modulating G protein inhibition, increasing peak calcium current, controlling the alpha-1 subunit targeting to the membrane and shifting the voltage dependence of activation and inactivation. The second gene, ATP2C1 (Additional data file 7), encodes a protein that belongs to the family of P-type primary ion transport ATPases, which pump Ca²⁺ into the endoplasmic reticulum.

Transcripts encoding the calcium/calmodulin-related signaling proteins calmodulin 1 (CALM1) and Ca²⁺/calmodulin-dependent membrane-associated kinase (CASK1) are differentially expressed in one or more glaucoma groups. CALM1 was increased in AAG compared to AA donors (Additional data file 5), while CASK1 was increased in glaucomatous astrocytes from both AA and CA donors (Table 1, Figure 7). Calmodulin is the Ca²⁺ sensor of key signaling molecules, such as adenylyl cyclase, CAMKII, CAMKIV, and MYLK discussed above. CASK1 is a member of the membrane-associated guanylate kinase proteins (MAGUKs), a prominent family of scaffolding molecules associated with intercellular junctions. CASK1 targets Ca²⁺ and K⁺ channels 69 and/or the Ca²⁺ pump 4b/CI 70 to the plasma membrane, interacts with liprins 71 and regulates transcription by interacting with transcription factors in the nucleus 72. Interestingly, CASK is a candidate gene for X-linked optic atrophy 73. The differential expression of genes in Ca²⁺ signaling pathways could be a common theme in glaucomatous astrocytes that may have a higher impact in optic nerves from AA donors due to increased sensitivity to elevated IOP in these donors.

The first step in the analysis of the microarray data was to determine which genes to consider 'present' or 'absent'. We estimated the probe-set present/absent calls by using the Wilcoxon signed rank-based algorithm. In order to reduce false positives, we removed the probe-sets with all samples as 'absent' (Additional data file 3).

As the experiments were done at different times, two types of Affymetrix microarrays (Human Genome U95Av2 array and Human Genome U133A 2.0 array) were used. Samples from eight CAG donors, seven CA normal donors and six AAG samples were measured using a Human Genome U95Av2 array. Eighteen CA samples, eighteen AA samples and six AAG samples were measured using a Human Genome U133A 2.0 array. The data measured by two types of arrays were normalized separately by RMA normalization 7475. We defined common glaucoma-related genes as genes differentially expressed in both CAG versus CA and AAG versus AA, and did comparisons of CAG versus CA and AAG versus AA separately. The differentially expressed genes were identified by the empirical Bayesian shrinkage moderated t-statistics in the limma Bioconductor package 76. A mixed effects model was used to account for the effect of technical replicates. Genes exhibiting a fold-change >1.5 and p-value < 0.01 were considered significant. To reduce false positives because the AAG has only three biological replicates, we applied the Benjamini and Hochberg false discovery rate multiple testing correction with a false discovery rate of 0.05 (AAG versus CAG and AAG versus AA).

To compare the significant gene list based on two types of microarray platforms, the Affymetrix probeset IDs were transferred as Entrez IDs based on the Bioconductor library. Genes whose Entrez IDs appear in both the differentially expressed gene list from CAG versus CA (using the Human Genome U95Av2 array) and AAG versus AA comparisons (using the Human Genome U133A 2.0 array) and change in the same direction were considered as common glaucoma-related genes. Genes that are differentially expressed for AAG versus CAG (using the Human Genome U95Av2 array), but without significant changes for AA versus CA (using Human Genome U133A 2.0 array), were considered as the glaucoma race-related genes. Here we considered a p-value > 0.05 as indicative of changes that were not significant.

GO analysis of differential expression in glaucomatous astrocytes was done with GoMiner 14. Briefly, gene lists of up- and downregulated genes (p < 0.01 as described above) were normalized to 1 and -1, respectively, for genes that exhibited at least a 1.5-fold change in either direction. These lists were then loaded into GoMiner using local GO databases accessed using the 'Derby' module. GoMiner output was analyzed with a significance cutoff of p < 0.01 and at least four genes per category.