**4. Discussion**

The key defining characteristic of glaucomatous optic neuropathy is the progressive loss and dysfunction of RGCs and is one of the only shared features across the pathophysiological spectrum in human glaucomatous disease and animal models of glaucoma. The pathophysiology of RGC degeneration in different animal models is likely to vary. There may be shared intrinsic RGC degenerative mechanisms across all glaucoma models as well as human glaucoma, which can be identified and explored. The purpose of this study was to identify commonalities between RGC injury models (rather than to individually analyze distinct models of glaucoma) with which to identify neuroprotective treatments that may be applicable across the heterogeneity of glaucoma (in animal models and human glaucoma patients).

In this study, we used three publicly available RNA-sequencing datasets of animal models of glaucoma. The DBA/2J mouse is one of the most frequently used glaucoma models in research. In DBA/2J mice, mutations in two genes (*GpnmbR150X* and *Tyrp1b*) drive an iris disease with features of human iris atrophy and pigment dispersion. Pigment disperses from the iris and induces damage in the drainage structures of the eye. This inhibits aqueous humor outflow and leads to an increase in IOP. By 9 months of age, IOP is high in the majority of eyes and transcriptomic and metabolic datasets demonstrate mitochondrial and metabolic dysfunction in RGCs [12]. In the mouse controlled optic nerve crush model, axonal injury is induced mechanically by a temporary compression of the distal optic nerve, resulting in RGC death [3]. Similarly, in the optic nerve transection model, axonal injury triggers axon degeneration, leading to a rapid RGC degeneration [4].

With the increased prevalence of -omics technologies, there is a wealth of open data within the field of glaucoma for researchers to explore and utilize. The aim of the present study was to unbiasedly test a conserved gene set between three publicly available RNA-sequencing datasets. Typically, pathway analysis and ranking of changed genes/proteins/metabolites reveal multiple potential mechanistic avenues for exploration, but practical (funding and publication limits) and narrative/hypothesis limitations leave many of these unexplored. Comparison of datasets to identify common changes can be a powerful method to identify conserved pathological processes in optic nerve injury, as has been demonstrated by previous comprehensive comparisons including the DBA/2J mouse model, optic nerve crush/axotomy models across multiple species, and other CNS neurodegenerations [13,14]. These studies identified commonality and enrichment predominantly within neuroinflammatory and innate immune responses. With the growth of this approach, the field can form data-driven hypotheses to identify and test new mechanisms of neurodegeneration/neuroprotection.

In this study, we compared common gene changes in transcriptomic datasets from RGC injury models and screened these against the Comparative Toxicogenomics Database to identify novel therapeutics. Using a retinal explant model to rapidly test a number of these compounds, we demonstrate the therapeutic potential of drugs identified in this way. Multiple drugs identified had established literature support for RGC and neuronal protection, further supporting this method of identifying potential therapeutics. The ex vivo retinal explant model is ideal as a first-pass model for assessing neurodegenerative events and testing neurodegenerative or neuroprotective drug candidates. In this model, RGCs are axotomized, leading to RGC degeneration [9,11]. As 100% of RGCs are axotomized, the insult is controlled, and maintenance in tissue culture removes the influence of systemic events (e.g., infiltration of myeloid-derived cells) and tightly controls tissue conditions. Only small amounts of a drug is required for testing as it can be directly dissolved in the media. This overcomes any need to test or assess systemic metabolism and bioavailability in the first round of studies, thus keeping the cost of drug and animals low. Further to this, in many research-centric countries (e.g., EU, USA, Australia), additional animal ethical permits are not required to perform this model, increasing its availability to labs that might not have significant animal housing and testing resources.

Screening the shared DE genes between the three glaucoma models against the Comparative Toxicogenomics Database identified valproic acid as the drug that interacted with the most gene products (11/12; the exception was *Chrna6*), followed by SB-431542, progesterone, estradiol, choline, folic acid, WY-14643, and rosiglitazone (see Supplementary Table S3). Valproic acid is an FDA-approved anti-epileptic and migraine drug. Valproic acid's mechanism of action is proposed to be through modulating histone deacetylases activity, which has a well-established literature of limiting RGC degeneration in models of normal-tension and ocular-hypertensive glaucoma [15,16].

Progesterone and estradiol are both endogenous steroids and sex hormones. An extensive literature exists on their possible potential as neuroprotectors and multiple different mechanisms of action are proposed (e.g., regulation of mitochondrial calcium and *Bcl-2* expression, affecting the phosphatidylinositol 3-kinase/Akt signal pathway or preventing caspase-3 activation). A number of studies have been performed to explore the potential of progesterone and estradiol as neuroprotectors in animal models of photoreceptor and retinal ganglion cell loss [17–25]. Given the evidence that RGCs express estrogen receptors, lower estradiol levels are linked with primary open-angle glaucoma (POAG) [26], and that β-estradiol can protect from RGC death [23,27,28], an estrogen analogue-based therapy may be of benefit in many glaucoma patients.

Choline is a quaternary amine, the precursor of many cell components and signaling molecules (e.g., acetylcholine), and an essential nutrient for humans. It is further processed to citicholine and phosphatidylcholine in vivo. Both choline and citicoline are thought to have a neuroprotective effect, possibly by preserving sphingomyelin and cardiolipin and by promoting glutathione synthesis. The literature contains a number of studies suggesting a favorable effect on retinal cell survival, as assessed in various neurodegenerative animal models [15–19].

Folic acid is an essential nutrient and B vitamin which plays a crucial role in the biosynthesis of DNA and RNA. The vitamin is also indispensable for erythrocyte maturation. Clinically, folic acid is mainly used to prevent neural tube defects in the developing fetus. Yet, it has been shown to be protective against certain dysplasia and to reduce gingival inflammation [29]. A number of studies examined the effect of folic acid on microglia and astrocytes in animal models of cellular stress responses, and reported positive results [30,31]. However, to the best of our knowledge, a possible neuroprotective effect in glaucoma has not been previously investigated.

As an established antidiabetic drug of the thiazolidinedione class, Rosiglitazone functions as an insulin sensitizer. The drug binds to the peroxisome proliferator-activated receptors in adipocytes and renders the cells more responsive to insulin. The neuroprotective effect of Rosiglitazone and other thiazolidinediones has been studied extensively and a beneficial effect has been suggested through inhibition of inflammatory responses, pro-apoptotic cascades, and mitochondrial metamorphosis [32–34].

The TGF-β type 1 receptor inhibitor SB-431542 is a drug candidate proposed for the treatment of osteosarcomas in humans. The agent acts through binding the activin receptorlike kinase (ALK) receptors ALK5, ALK4, and ALK7. It has also been shown to promote the transformation of astrocytes into neurons [35]. Two studies found evidence for a possible protective effect from NMDA-induced retinal degeneration in rats and from N-methyl-Nnitrosourea (MNU)-induced rod photoreceptor degeneration in zebrafish [36,37].

Pirinixic acid, also known as WY-14643, is a synthetic drug candidate for prevention of severe cardiac dysfunction, cardiomyopathy, and heart failure as a result of lipid accumulation within cardiac myocytes. Its mechanism of action is through binding the peroxisome proliferator-activated receptor alpha (PPARα), thereby affecting cell proliferation and differentiation, lipid metabolism, and inflammatory signaling cascades. WY-14463 has the potential to protect neurons by modulating mitochondrial fusion and fission [38]. However, the drug s potential neuroprotective effect has not previously been explored in glaucoma models. In summary, our screen identified a number of neuroprotective

compounds novel to RGC survival. Valproic acid, folic acid, and WY-14643 performed best in terms of neuroprotection, while treatment with SB-431542 led to variable results.

As a further test of this drug discovery method, we sought to also test identified drugs that, based on literature, would not produce a neuroprotective effect. Genistein is a naturally occurring phytoestrogen and isoflavone. The compound is found in soybeans, flava beans, coffee beans, and others. Genistein is known to inhibit protein-tyrosine kinase and topoisomerase-II, affecting the process of cell differentiation and proliferation, and promoting DNA fragmentation and apoptosis [39,40]. It has been hypothesized that genistein could be used to treat different types of cancer through its antioxidant and antiangiogenetic effects. However, in various animal neurodegenerative models, genistein has been shown to block the neuroprotective effect of agents such as carbamylcholine, forskolin, and veratridine [41–43]. In the explant model, genistein had no effect on RGC survival, thus supporting the applicability of this method for identifying disease modifying compounds.

## **5. Conclusions**

Glaucoma is a complex neurodegenerative disease in which the only shared and clinically defined feature is the progressive dysfunction and degeneration of RGCs (although age, genetic risk, and high IOP are all common risk factors, glaucoma can still occur in the absence of one or more of these risk factors). Many models of glaucoma-related stress have been developed in a wide array of animal species, which recapitulate some of the features of human glaucoma. Comparison of these models to identify common changes can be a powerful method to identify conserved pathological processes in RGC injury. Publicly available -omics datasets such as those from RNA-sequencing represent a data-rich resource for identifying these potential critical pathogenic changes. We identified gene changes common to three modes of RGC injury and screened these against the Comparative Toxicogenomics Database to identify novel therapeutic agents for testing. We demonstrated the validity of this approach by testing the identified compounds using another independent model of RGC injury.

Our drug discovery and analysis platform used only publicly available tools and datasets and an ex vivo model widely amenable to the majority of research-intensive countries that does not require additional animal ethical permits, while keeping drug treatment costs low. This platform is a practical means to utilize the increasing wealth of open access -omics data generated by the glaucoma field in order to move forward towards identifying new therapeutics.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/jcm10173938/s1, Table S1: Shared differentially expressed genes between glaucoma RNAsequencing datasets, Table S2: Roles of shared differentially expressed genes, Table S3: Compounds with known interactions with differentially expressed genes common to all three RNAsequencing datasets.

**Author Contributions:** T.J.E. performed experiments, analyzed data, and wrote the manuscript; J.R.T. designed and performed experiments, analyzed data, and wrote the manuscript; P.A.W. conceived and designed experiments, analyzed data, and wrote the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Vetenskapsrådet 2018-02124, Ögonfonden (P.A.W.), and Ögonfonden (J.R.T.). The APC was funded by Vetenskapsrådet 2018-02124. Pete Williams is supported by Karolinska Institutet in the form of a Board of Research Faculty Funded Career Position and by St. Erik Eye Hospital philanthropic donations.

**Institutional Review Board Statement:** All breeding and experimental procedures were performed according to the tenets of the Association for Research for Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Research. Individual study protocols were approved by Stockholm's Committee for Ethical Animal Research (10389-2018).

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All data generated or analyzed during this study are included in this published article.

**Acknowledgments:** The authors would like to thank Diana Rydholm for assistance with animal husbandry and maintenance; Melissa Jöe, Elizabeth Kastanaki, and Amin Otmani for assistance with immunofluorescent labelling and imaging; St. Erik Eye Hospital for financial support for research space and animal facilities.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
