1. Introduction
Glaucoma is a multifactorial disease characterized by degeneration of the neurons known as retinal ganglion cells (RGCs) and their long axons which transmit visual information from the retina to the brain [
1]. A variety of in vivo animal models have been used to unravel the mechanisms of both RGC degeneration and reorganization of the retina. The animals were subjected to either acute or chronic elevation of intraocular pressure, transection, or crush of the optic nerve, as well as the spontaneous glaucoma-like disease (on mice of the DBA/2J strain) [
2,
3,
4]. However, there has been little progress in developing efficient strategies for neuroprotection in glaucoma.
The retina is a complex, multilayered tissue of the central nervous system (CNS) composed of diverse neurons, glia, and vascular cells, as well as a rich extracellular matrix. Recent studies of single-cell RNA-sequencing disclosed cell type-specific molecular markers [
5]. Laser capture microdissection was also used to isolate individual cells from the retinal tissue and compare genes expressed in RGCs of either normal or glaucomatous rat retina [
6]. However, similar to other neurodegenerative conditions [
7], mechanisms involved in glaucomatous neurodegeneration appear to depend on a complex interplay of signaling pathways and often involve multicellular networks [
8,
9,
10]. The systemic character of such events suggests that an examination of the whole tissue transcriptional landscape may help guide both further investigations into the pathophysiology, as well as the development of novel therapeutic approaches to glaucoma.
Transcriptional changes following either acute or chronic optic nerve injury have been identified in animal models ranging from zebrafish to non-human primates [
11,
12,
13]. It is usually assumed that insights into functional changes caused by RGC injury in rodents may help elucidate the course of glaucomatous neurodegeneration in humans. Microarray-based transcriptome analyses are currently available for rodent models of glaucoma [
6,
14,
15], retinal and optic disc injury [
16,
17], and ischemic damage [
18].
Overall, the data suggest evolving stages associated with altered gene expression, roughly defined as follows: an acute phase, within hours, with transient upregulation of both immediate/early response genes and inflammatory responses [
10,
19,
20]; a sub-acute phase, between 1 and 3 days, characterized by the expression of cell-cycle and cell death genes [
21]; and a late chronic phase, at 5 to 7 days, characterized by the expression of genes involved in structural remodeling of neurons and glia [
15,
16,
22,
23,
24,
25]. In the earliest stage, genes related with triggering of apoptosis have been found to be upregulated in response to retinal injury [
22,
26,
27]. In the sub-acute phase, in line with apoptotic RGC death following optic nerve damage, studies identified upregulation of genes related with the execution phase of apoptosis, as well as downregulation of cytoprotective and anti-apoptotic genes [
28]. The upregulated pro-apoptotic genes include caspase 3,
tumor necrosis factor receptor type 1 associated death domain (
TRADD),
tumor necrosis factor receptor superfamily member 1a (
TNFR1a), and
BCL2 associated X apoptosis regulator (
Bax). Downregulated cytoprotective and anti-apoptotic genes include
X-linked inhibitor of apoptosis (
XIAP),
mitogen activated protein kinase 1 (
Mapk1),
Ras/Rac guanine nucleotide exchange factor 1 (
Sos1),
c-Raf-1, and
YY1 [
29]. In later stages, only elevated intraocular pressure and optic nerve transection models have been explored by microarray analysis, and retinal changes were found to be associated with altered expression of genes with roles in regeneration, synaptic plasticity, axonogenesis, neuronal projections, and neuron differentiation [
17,
20,
30,
31,
32].
In the present study, we used Agilent gene expression microarrays and deep computational approaches to profile the expression patterns of differentially expressed genes in the rat retina at 14 days after optic nerve crush (ONC) compared to control (CTR) sham operated animals. ONC is an acute, mechanical injury to the nerve that leads to gradual RGC apoptosis and has been widely utilized to examine glaucomatous disease pathophysiology [
33]. At 14 days after the lesion, a small number of RGCs remain in the retina, and thus the modified transcriptome is likely to reflect mainly the adaptation of the tissue to the loss of those [
34].
Enrichment analysis performed in accordance with previous studies [
35,
36] shows complement cascade and Notch signaling pathway components as major players in the retina 14 days after ONC and, to a lesser extent, also oxidative stress and kit receptor signaling pathways were also altered. We went beyond the enrichment analysis to extract more information of the transcriptome such as the relative expression variability (REV) and the derived gene expression stability (GES). Moreover, we analyzed the coordination of gene expressions within each enriched pathway and with the entire transcriptome to determine the ONC-induced remodeling of the associated genomic fabrics. Thus, for the 17,657 quantified genes, our study generated a total of 155,911,310 values to analyze, that is, 8830x more data per condition than a traditional transcriptomic analysis. The genomic fabric of a given pathway was defined as the transcriptome associated to the most interconnected and stably expressed gene network responsible for that pathway [
37]. These deep bioinformatic analyses provide additional insight into the predicted mechanisms for both RGC death and adaptation of the retinal tissue following ONC.
4. Discussion
In this study, we conducted a transcriptomic analysis of rat retina at 14 days after ONC, a time point when the population of RGCs was reduced to only 40% of the original population, as compared to controls. We identified differentially expressed genes, at a total of 127 genes significantly upregulated, and 66 downregulated. We examined changes in GES and found a reduction in the number of very stably expressed genes from 336 to 124. Using bioinformatic tools to propose protein interaction (STRING) and main altered molecular pathways (PathVisio3), we found that the complement cascade and Notch signaling pathway were the main affected pathways. Although protein–protein interaction determined with STRING (on the basis of data mining in genome-wide datasets) may be a good predictor of real interactions [
48], we deepened our analyses with additional bioinformatics and mathematical modeling of gene networks of ONC retinas, which has been shown by our group to be a consistent method for such predictions. Coordination analysis for both pathways showed an increase in the number of synergistically expressed gene pairs and of genes with similar coordination profile.
Our analysis was carried out on whole-retina extracts, and the transcriptome profiles analyzed represent the adaptations of the retinal tissue to the RGC degeneration promoted by the ONC. Although the observed regulation is triggered by the ONC, which leads to RGC death, changes at 14 days after injury likely involve the activation of the retinal glia (i.e., astrocytes, Müller cells, and microglial cells), and also other retinal neurons besides RGCs, as signs of retinal remodeling. These changes may impact potential treatment strategies to ameliorate the secondary degeneration associated with CNS insults [
34].
Unique among transcriptomic studies of the retina is that beyond the regulation of gene expression, we also disclosed REV, changes in GES, and coordination following ONC. GES score ranks the genes according to the strength of the cellular homeostatic mechanisms to keep their expression variability within narrow limits. GES, which includes the most stably expressed genes in the 100th percentile, actually identifies the genes most critical for the survival and phenotypic expression of the tissue. By contrast, the most unstably expressed genes (GES in the first percentile) most likely empower the cell to respond/adapt to the environmental fluctuations [
45,
46]. Moreover, the expression coordination analysis refines the functional pathways by the identification of gene pairs whose expression fluctuations among biological replicas are positively (synergistically), negatively (antagonistically), or neutrally (independently) correlated. In previous papers [
43,
45,
46], we speculated that genes whose encoded proteins are linked in functional pathways should coordinate their expression to optimize a kind of “transcriptomic stoichiometry”. Changes in the coordination profiles indicate remodeling of the functional pathways [
37,
42,
43,
44,
45,
46,
53,
54,
55].
Expression of individual genes depends on local conditions that, although similar, are not identical among biological replicas. We assume that expression of key genes is kept by the cellular homeostatic mechanisms within narrow intervals while that of non-key genes is less restrained to readily adapt to environmental changes. In our results, we found a higher variability in the REV profile of genes with ONC (
Figure S1A), as expected in pathological conditions, suggesting that CTR mechanisms are also affected.
Thus, by analyzing GES changes following the ONC, we found clues about retina cells changing priorities in controlling the expression level of certain genes. Our results identified, at 14 days after ONC, that only a few genes presented an altered profile in their GES, with enrichment in genes of eye structure (
Table S1). The
crystallin genes were found to have relevant changes in their GES, thus acquiring a more unstable expression profile. Crystallins have chaperone-like properties involved in an increase of cellular resistance to stress-induced apoptosis [
56]. Upregulation and downregulation of crystallin expression has been reported upon different cellular stresses, and those alterations are viewed as a cellular response against environmental and metabolic insults [
57,
58,
59]. Crystallins have been related with both RGCs [
60] and amacrine cells [
61], supporting its active participation in the remodeling of retinal tissue after injury, possibly through a decrease in resistance to stress, as well as leading to tissue stabilization. Curiously, two other genes presented a drastic change in GES:
Adad2 (GES (CTR) = 79 to GES (ONC) = 5), a gene related with RNA editing [
62], and
RGD1308160 (GES (CTR) = 61 to GES (ONC) = 10), a gene that, according to BlastP, has a loose homology with the
Caenorhabditis elegans gene
smc-4, involved in chromosome stabilization [
63]. These data suggest that both such mechanisms are very active at 14 days after ONC.
Typical RGC-expressed genes, underrepresented in ONC retinas compared to CTR (
Figure 1 D,E), reflect RGC death after optic nerve damage. Among the 66 downregulated genes, 20% have been described as expressed by ganglion cells [
5,
6,
20]. Enrichment analysis of the downregulated genes showed cytoskeletal organization as the main biological process, with genes of intermediary filaments and of proteins that interact with them. Neurofilaments are particularly abundant in axons, and essential for their growth, maintenance of caliber, transmission of electrical impulses, velocity of impulse conduction, as well as regulation of enzyme function and the structure of linker proteins [
64,
65,
66].
Additional biological roles have been attributed to neurofilaments such as synaptic plasticity. Marked decrease in the expression of the neurofilament genes was first shown in brain tissue from Alzheimer’s disease patients as compared to CTRs [
67]. It is generally accepted that CNS disorders are divided between early and late changes, associated with dysfunction of neuronal activity caused by perturbations of synapses [
68]. Neural reorganization after stroke is initiated as cellular reactions to degeneration. While neurons die in an ischemic region, axons and synapses degenerate in further brain regions, promoting regenerative responses that lead the growth of new connections among surviving neurons [
69]. RGC dendritic arbors also showed significant reduction after 2 weeks of elevation of intraocular pressure, a common model of chronic glaucoma [
70]. In a primate glaucoma model, a correlation was established between abnormalities of parasol RGC dendrite morphology and function [
71]. Morphological changes of RGC dendritic arbors have been detected before axon thinning or soma shrinkage, suggesting that dendritic abnormalities may precede degeneration of other ganglion cell domains. Alterations of neurofilament expression in various neurological disorders not only underlie axonopathy, but also synaptic remodeling because of its vital roles in synaptic function [
72,
73]. Since at 14 days after ONC the retinal tissue still harbors 40% of the RGCs, the decreased expression of genes related with cytoskeletal organization, as intermediary filaments, may be representative of retinal remodeling, not restricted to the injured axon but also at the level of dendrites. This was further reinforced by the predicted protein–protein interactions (STRING) analysis that showed “axon” and “neuron projection” cellular components were significantly altered. PPI along with Pearson correlation analyses performed herein took advantage of the concept that expression of individual genes is linked to each other so that the protein amounts would respect the “stoichiometry” of the biochemical reactions [
74].
Many components of the innate immune system are expressed intrinsically by retinal cells (Retinal Database, GeneNetwork.org), and innate immune networks are activated by various types of insults to the retina, such as ONC, partial transection of the optic nerve, and DBA2J mouse models of glaucoma [
6,
10,
14,
15,
20,
75,
76]. In the present study, enrichment analysis disclosed the complement cascade as the most prominent biological process among differentially expressed genes after ONC (
Table S2), which was further confirmed by predicted PPI as “innate immune response” and “immune system process”. Complement activation was shown to be upregulated in the retina as early as 2 days after ONC [
27]. Such activation, specifically including
C1q,
C3, and
Cfi [
19], suggests that the innate immune system plays an important role in retinal immune response through glia cells, and specifically in the response of the retina to injuries. It has been suggested that, at an early stage of a mouse glaucoma model, the complement cascade mediates synapse loss, in particular
C3 [
77,
78,
79,
80]. At late-stage glaucoma, the complement cascade is thought to play its traditional role of opsonizing and removing the cellular debris from widespread RGC death [
81]. Recently, it was described that C1q marks a subset of RGC in the embryonic retina and that knockout of C3 receptor, which is only found in microglia, resulted in increased RGC numbers, indicating a direct relationship with microglia-mediated RGC elimination [
82].
Our data showed that ONC led to an increase in synergistic coordination, most evident in the classic pathway of complement cascade, which indicated a stronger gene network in this pathway. It is particularly interesting that the
serping1 gene showed increased synergistic coordination at 14 days after ONC (
Figure 4). Serping1 is a C1 inhibitor known to block the initiation of the complement cascade through the classical pathway, therefore limiting inflammation. It is indirectly involved in the production of bradykinin, a peptide that promotes inflammation by increasing the permeability of blood vessel walls and also inhibits leukocyte recruitment into ischemic cardiac tissue, an effect independent of protease inhibition [
83]. Therefore, the upregulation of
serping1 in ONC retinas may help control the inflammatory response after injury, both by modulation of the complement cascade, as well as preventing infiltration of peripheral inflammatory cells through the blood–retina barrier.
Our enrichment analysis also showed upregulation of the Notch signaling pathway at 14 days after ONC. Notch signaling is known to play important roles during retinal development, in both RGC differentiation [
84] and proliferation of Müller glia in acutely damaged chick retina [
85]. Notch is expressed in most Müller glia cells at low levels in undamaged retina, and blockade of Notch activity prior to damage is protective to amacrine, bipolar, and ganglion cells [
86]. Moreover, overexpression of Notch1 has been observed in several neurodegenerative diseases [
87,
88]. Besides Notch, the
deltex2 gene was also upregulated (
Figure 5). A recent study showed that Dtx together with its interacting partner, Hrp48, downregulated Notch signaling and induced cell death during
Drosophila development [
89]. At 14 days after ONC, the upregulation of both
Notch and
deltex2 suggests that Notch signaling may act upon retinal Müller cells via a non-canonical pathway, thereby positively influencing RGC degeneration. Notch signaling mediates many different intercellular communication events, influencing retinal Müller cells, ganglion cells, and microglial cells. It is known that Notch-1 signaling in neurons during ischemic conditions may enhance apoptotic signaling cascades, while activation of Notch-1 in microglial cells and leukocytes may exacerbate inflammatory processes that contribute to neuronal death [
90]. Moreover, several studies found that Notch signaling serves important functions in the regulation of neurite outgrowth and maintenance. Both Sestan et al. [
91] and Berezovska et al. [
92] found that Notch signaling influences dendritic morphology—while activation inhibits neurite outgrowth or causes their retraction, inhibition promotes neurite extension. These data, together with our results, suggest that Notch signaling may confer a positive effect on RGC degeneration through an exacerbation of inflammatory processes as well as promoting retinal tissue remodeling by altering dendritic morphology.
Oxidative stress pathway was also found to be enriched in our analyses, and this was shown to occur due to
Cyba and
Nfix upregulation in ONC retinas. In the CNS, normal NAD(P)H oxidase function appears to be required for processes including neuronal signaling, but overproduction of reactive oxygen species (ROS) contributes to neurotoxicity and neurodegeneration. Conversely, RGCs were found to express NAD(P)H oxidase subunits under physiologic conditions and after ischemia [
93]. NOX proteins are homologs of NAD(P)H oxidases that generate reactive oxygen species [
94]. NOX proteins interact with and are stabilized by Cyba (or p22phox) in intracellular vesicles, which, when activated by cytosolic p47phox, form a protein complex that transports electrons from cytoplasmic NADPH to extracellular oxygen to generate superoxide (O2−). Cyba was also found to be increased in a rat model of diabetic retinopathy [
95] and in retinal pigmented epithelial cells under angiotensin-II-induced oxidative stress [
96].
NFI transcription factors a/b/x were shown to be temporally expressed during retinal development by retinal progenitor cells, bipolar cells, and Müller glia [
97]. Moreover,
Nfix is known to have a role on CNS development, and
Nfix-null mice display major brain malformations in the cortex, cerebellum, and hippocampus ([
98] for reviews). In the skeletal muscle, Nfix can also regulate myogenesis [
99,
100], however, deletion of Nfix leads to decreased muscle regeneration and has been implicated in the pathogenesis of muscular dystrophies [
101]. The knockdown of
Nfix in a myoblast cell line was protective against oxidative stress [
102], thus indicating that this transcription factor may have a versatile role in health and disease. In the context of retinal degeneration, the upregulation of Cyba and Nfix transcripts reinforces the hypothesis that oxidative stress takes place and accelerates RGC death in ONC models [
103].
Oxidative stress and inflammation are closely related pathophysiological processes, one of which can be easily induced by another. Thus, both processes are simultaneously found in many chronical pathological conditions, such as glaucoma [
104]. The immune activity is a necessary intrinsic response to promote tissue cleaning, healing, and regeneration process after injury [
105]. However, unbalanced or sustained immune response turns the beneficial potential into potentiation of the injury process. The mechanical injury together with hypoxia, both promoted by the ONC, could be the triggers for the activation of inflammation, oxidative stress, and metabolism alterations. Notch-1, c-Kit, and complement signaling are known as linkers between inflammation, metabolism, oxidative stress, and dendritic plasticity, acting into RGC and Müller cells [
88,
106,
107,
108,
109,
110]. Our results support the hypothesis that not only the events intrinsic to RGCs, but also environmental signals from other cells, such as Müller cells, are critical for cell death stimuli, and that RGC–glia interactions are critically important for different aspects of RGC neurodegeneration. Improvement of the current understanding of the role of RGC–glia interaction, mitochondria, and the immune system in neurodegeneration will potentiate the development of innovative treatment methods for neuroprotection.