Large drug screen finds brain-protecting treatments in fish and mouse models of retinitis pigmentosa

We initiated our study using a robotics-automated large-scale in vivo screening system to identify compounds that promoted rod photoreceptor survival in a transgenic zebrafish line enabling targeted ablation of rod photoreceptors, Tg(rho:YFP-Eco.NfsB)gmc500, hereafter, rho:YFP-NTR (Walker et al., 2012; White et al., 2017). In this line, a 3.7 kb rhodopsin (rho) promoter fragment (Hamaoka et al., 2002) drives transgene expression exclusively in rod photoreceptor cells (Figure 1A). The transgene is a fusion protein linking a yellow fluorescent protein (YFP) reporter to a nitroreductase prodrug converting enzyme (NTR, encoded by the nfsB gene from Escherichia coli). NTR expression enables pro-drug inducible targeted cell ablation (Curado et al., 2007; Pisharath et al., 2007). Exposing rho:YFP-NTR fish to the prodrug metronidazole (Mtz) leads to the selective death of rod photoreceptors and concomitant loss of YFP (Figure 1A–C), physiologically mimicking the onset of RP (Hamel, 2006). An immunohistological analysis of rod and cone photoreceptor markers was performed on 7 days post-fertilization (dpf) zebrafish retinal sections to test if Mtz-induced ablation was specific to rod cells. In non-ablated controls, rod outer segment labeling was well correlated with YFP expression (Figure 1—figure supplement 1A,B; -Mtz, arrows). In Mtz-treated retinas, rod outer segment labeling and YFP expression were markedly diminished (Figure 1—figure supplement 1A,B; +Mtz). Cone photoreceptor labeling showed no overlap with YFP expression, and cone cell labeling appeared similar in non-ablated and Mtz-treated retinas (Figure 1—figure supplement 1C), suggesting cone cells were not affected by Mtz exposure or by acute rod cell loss.

Prior studies suggest NTR/Mtz-mediated ablation occurs by apoptosis (Chang et al., 1993; Doonan et al., 2003; Portera-Cailliau et al., 1994; Zeiss et al., 2004) and loss of rods in RP has also been linked to apoptosis (Chang et al., 1993; Doonan et al., 2003; Portera-Cailliau et al., 1994; Zeiss et al., 2004). Thus, we reasoned rho:YFP-NTR fish could be used to identify compounds that protected rods from apoptosis. To identify neuroprotective compounds with this model (i.e. drugs that sustain YFP expression after Mtz exposure), we used our established plate reader-based ARQiv-HTS assay (Figure 1D; Walker et al., 2012; White et al., 2016).

We first determined optimal conditions for inducing rod cell loss while maintaining larval health in a 96-well plate format. Major aspects of retinal cytogenesis are largely complete by five dpf in zebrafish (Schmitt and Dowling, 1999; Stenkamp, 2011). Reporter expression in rho:YFP-NTR larvae has also stabilized by this time point (Unal Eroglu et al., 2018), consistent with rho expression (Raymond et al., 1995). We therefore chose 5 dpf to initiate Mtz-induced rod cell ablation. We previously determined that rod cell loss reached a nadir at 7 dpf following a 24 hr pulse of 10 mM Mtz at 5 dpf (Walker et al., 2012). We reasoned that a 48 hr Mtz exposure initiating at 5 dpf would maximize the signal window to test for neuroprotective effects. Concluding the experiment by 7 dpf also avoids challenges associated with feeding, as zebrafish can subsist on their yolk sac up to that time point (Hernandez et al., 2018; Jardine and Litvak, 2003). However, 10 mM Mtz treatments extending beyond 24 hr become increasingly toxic (Mathias et al., 2014) and removing Mtz from microtiter plates after a 24 hr pulse could not be easily automated. We therefore sought a 48 hr Mtz treatment regimen sufficient for inducing maximal rod cell loss by 7 dpf that showed no evidence of general toxicity.

Five concentrations of Mtz were tested across a twofold dilution series from 10 mM to 625 μM. YFP reporter signals were quantified daily from 5 to 8 dpf using ARQiv. Changes in YFP levels were calculated as percentages normalized to non-ablated YFP controls. The data showed concentration-dependent reductions in YFP with maximal loss observed at 7 dpf (Figure 1B). Mtz exposures at 0.625, 1.25, 2.5, 5, and 10 mM led to a 55, 79, 87, 87, and 83% decrease in YFP detection, respectively (Supplementary file 1a). Although no lethality was observed for any condition, signs of distress were evident for 10 mM Mtz exposures (i.e. reduced motility). At ≤5 mM Mtz, however, no signs of stress were observed. As 2.5 mM Mtz was the lowest concentration producing maximal loss of YFP detection, and confocal imaging verified YFP loss following a 2-day (5–7 dpf) 2.5 mM Mtz treatment (Figure 1C), this condition was selected as the treatment regimen for the large-scale screen.

Previously established power analysis methods using ablated and non-ablated controls (White et al., 2016) determined that a sample size of nine larvae was sufficient to detect a 50% neuroprotective effect. For ease of dispensing, microtiter plate formatting, and to account for larval dispensing errors, we increased the sample size to 16 larvae per condition for the primary screen. We also reasoned that this would also allow us to detect subtler neuroprotective effects. The strictly standardized mean difference quality control (SSMD QC) score was 1.67, indicating the assay was of sufficient quality to justify a large-scale screening effort (Zhang, 2011).

To establish a positive control, we tested 17 compounds and one compound ‘cocktail’ previously implicated as neuroprotectants in mammalian RP models (). Unfortunately, none sustained YFP expression at the concentrations tested (4 μM to 125 nM). However, none have proven effective in RP patients either, suggesting a lack of conservation of neuroprotective effects across species. Our screen was designed to address this issue by testing for neuroprotective effects of these compounds across species and between different RP models. Fortunately, a compound identified as a retinal cell neuroprotectant by the Zack lab (manuscript in preparation) did show-dose-dependent effects on YFP levels. This compound was therefore used as a positive control (POS) for assay performance throughout the large-scale screen. Supplementary file 2a

The Johns Hopkins Drug Library (JHDL) was chosen for the large-scale screen. The JHDL is comprised of nearly 3000 compounds, most being human-approved drugs (Shim and Liu, 2014). To minimize false discovery rates, all compounds were tested using qHTS principles (Inglese et al., 2006) – that is, across six concentrations (4 μM - 125 nM) using a twofold dilution series. The screen largely followed published ARQiv-HTS methodologies (Wang et al., 2015b; White et al., 2016), with additional assay-specific details (Figure 1D, steps 1–8). In all, 2934 compounds were screened and more than 350,000 transgenic zebrafish larvae evaluated. Real-time data analysis was performed as previously detailed (White et al., 2016) to generate: (1) a plot of YFP signal levels, (2) a plot of SSMD scores across all tested concentrations, (3) a signal intensity heat map of each plate, and (4) an SSMD score table (Figure 1D, step 7). Compounds producing SSMD scores of ≥one were considered potential hits and flagged for visual inspection to assess fluorescence and general morphology using a stereo fluorescence microscope. This step facilitated elimination of false-positive compounds producing aberrant fluorescence due to larval toxicity (six drugs) or compound autofluorescence (27 drugs; Figure 1C, step 8; Supplementary file 2b). Additionally, this allowed visual confirmation of sustained YFP expression within the retina. At the conclusion of the primary screen, 113 compounds were identified as hits (Supplementary file 2c). Hit compounds were classified according to the highest SSMD score achieved across all concentrations tested, and whether concentration-dependent effects were observed. SSMD scores suggested one drug produced a strong effect (SSMD of 2–3); three had semi-strong effects (1.645–2), 21 showed moderate effects (1.28–1.645), and 88 had fairly moderate effects (1–1.28) (Supplementary file 2c). Forty-two drugs showed concentration-dependent effects, while 71 exhibited discontinuous or singular concentration effects. ‘On label’ mechanism of action (MOA) information available for the 113 hits implicated more than 50 targets/pathways in rod cell neuroprotection (Supplementary file 2d).

We next performed a series of confirmatory and orthogonal assays to evaluate a subset of 42 hit compounds prioritized by SSMD score, dose-response profile, and/or implicated MOA (Supplementary file 2c, highlighted compounds). Having extensive MOA data is a key advantage of testing human-approved compounds which we leveraged in a previous large-scale zebrafish PDD screen (Wang et al., 2015b). Similarly here, as studies have suggested inflammation plays a key role in retinal degeneration and regeneration (Hollyfield et al., 2008; Mitchell et al., 2018; White et al., 2017; Yoshida et al., 2013), hits implicated as modulators of inflammatory signaling were included as part of the prioritization scheme. In addition, several compounds that did not produce concentration-dependent effects were selected to test whether this criterion was useful in predicting reproducibility. All compounds were obtained from new sources to ensure reagent authenticity. To confirm survival promoting effects, three repeats of the primary screening assay (Figure 2A) were conducted, but using a wider concentration range to account for differences in reagent quality (from 100 μM to 1.28 nM, fivefold dilution series). If toxicity was observed at higher concentrations, dilution series were initiated at 10 µM or 1 µM. Using this strategy, 11 of the 42 prioritized hit compounds were confirmed as lead candidates (26%). Estimated effects on rod cell survival – that is, YFP signal levels relative to +Mtz/0.1% DMSO controls – ranged from 38 to 9% (Figure 2B; lead drug candidate names, abbreviations, PubChem CID, and chemical structures are provided in Table 1).

We next asked whether there was a correlation between SSMD scores and/or concentration-dependent effects and confirmation rates. Among 19 selected hit compounds with higher SSMD scores (≥1.3), seven (37%) were confirmed; among 23 with lower SSMD scores (1–1.28), four (17%) were confirmed. Of 27 selected hit compounds with a concentration-dependent trend, eight (30%) were confirmed. Conversely, of 15 compounds that did not show a concentration-dependent trend, 3 (20%) were confirmed. Among the 11 confirmed leads, 8 (73%) showed dose-dependent effects and 7 (64%) had higher SSMD scores (>1.28). These results suggest that prioritizing hit compounds by both relative SSMD scores and dose-dependent trends provides predictive value for confirming activity, consistent with qHTS principles (Inglese et al., 2006).

As rod cell death is induced by NTR-mediated reduction of the prodrug Mtz in our model, it is possible that some lead compounds simply suppressed NTR enzymatic activity. To test this, NTR activity was evaluated in the presence of each lead compound by assaying the reduction kinetics of the prodrug CB1954 in vitro (Prosser et al., 2010). To ensure any potential for NTR inhibition was accounted for, all compounds were tested at 300 μM (except MIC and CHL which precipitated at higher concentrations and were tested at 50 µM) that is, ~100-fold greater than neuroprotective concentrations. Compounds were deemed potential inhibitors if NTR activity was less than 75% of the NTR alone control. Seven compounds showed no evidence of NTR inhibition by this criterion, but four did: warfarin (WAR), ciclopirox olamine (CPO), calcimycin (CAL), and sulindac (SUL; Figure 2—figure supplement 1A,B). However, IC₅₀ measures ranged from 150 µM (for CPO) to 350 µM (for SUL, Figure 2—figure supplement 1B), approximately one to two orders of magnitude higher than observed neuroprotective concentrations (i.e. 0.4–20 µM). When NTR reduction of Mtz was tested for WAR, CPO, and CAL at 300 μM, similarly weak inhibitory effects were observed (Figure 2—figure supplement 1B). The differences in concentrations between neuroprotective and NTR inhibitory activities diminish the likelihood that these leads act directly on NTR. To test this further, lead compounds were assayed for neuroprotective effects in NTR-independent mouse RP models (see below).

To control for the possibility that lead candidates promoted rod photoreceptor development, rather than neuroprotection, YFP levels were quantified in rho:YFP-NTR larvae exposed solely to lead drugs from 5 to 7 dpf (Figure 2—figure supplement 2A). Retinoic acid (RA, 1.25 μM) was used as a positive control as it promotes rod fates during development in zebrafish (Hyatt et al., 1996). RA-treated fish displayed statistically significant increases in YFP levels compared to untreated controls (Figure 2—figure supplement 2B). In contrast, none of the retinas treated with lead compounds exhibited increased YFP expression, suggesting they do not promote rod photoreceptor cell fate. In fact, three lead compounds cloxyquin (CLO), cortexolone (COR) and CPO reduced YFP signals relative to the untreated control (Figure 2—figure supplement 2B, asterisks), suggesting negative effects on rod cell development.

It is well known that the zebrafish retina regenerates (Gorsuch and Hyde, 2014; Lenkowski and Raymond, 2014; Wan and Goldman, 2016). Therefore, to determine whether lead compounds acted by stimulating regeneration, we used a previously described ARQiv assay designed to detect changes in rod cell replacement kinetics (Walker et al., 2012; White et al., 2017). Briefly, rho:YFP-NTR larvae were first treated with 10 mM Mtz at 5 dpf for 24 hr to induce rod cell loss. At 6 dpf, Mtz was washed out and larvae were treated with lead compounds at concentrations corresponding to maximal neuroprotective effects and YFP levels quantified at 9 dpf (Figure 2—figure supplement 3A). Dexamethasone, which accelerates rod cell regeneration kinetics (White et al., 2017), was used as a positive control. The results showed that none of the compounds increased rod cell regeneration rates (Figure 2—figure supplement 3B), while four compounds, dihydroartemisinin (DHA), CLO, CPO and MIC inhibited regeneration. These data suggest that lead compounds do not increase YFP levels by promoting rod cell regeneration.

To test if lead compounds simply increased YFP signal intensity, rather than promoted rod cell survival, intravital microscopy was used to image Mtz-treated retinas ±lead compounds and non-ablated (-Mtz) controls. Mtz-treated retinas had reduced numbers of YFP-NTR-expressing rods (Figure 3, +Mtz). Conversely, rods in control retinas displayed robust YFP signal throughout the retina and elongated morphologies suggestive of healthy outer segments (Figure 3, -Mtz). In retinas exposed to Mtz and lead compounds, YFP signal loss was attenuated and rods typically displayed elongated morphologies (Figure 3). However, for some compounds, cells appeared rounded, suggesting degeneration was not fully inhibited (e.g. miconazole, MIC; Figure 3). To confirm lead compounds increased rod cell survival, confocal stacks of YFP-expressing rods were 3D-rendered and fluorescence volume quantified using Imaris software-based automated image analysis (White et al., 2017). The data showed a statistically significant increase in YFP volumes for all drug-treated groups (Figure 3—figure supplement 1), confirming that lead compounds promoted increased rod cell numbers and/or preserved outer segment morphology.

To further investigate if lead compounds preserved rod outer segments, we used a newly developed zebrafish transgenic line, Tg(rho:GAP-YFP-2A-nfsB_Vv F70A/F108Y)jh405 (hereafter rho:YFP-NTR2.0). In this line, rods co-express membrane-tagged YFP and an improved NTR (‘NTR 2.0’). The membrane-tagged YFP facilitates improved imaging of rod outer segments, while NTR 2.0 enables cell ablation at a reduced concentration of Mtz (e.g. 10 µM versus 2.5 mM; Sharrock et al., 2020). To assess if lead compounds preserved outer segment morphologies, intravital confocal imaging was performed as above but using the rho:YFP-NTR2.0 line and 10 µM Mtz treatments. Qualitative image analysis suggested the majority, six of seven tested lead compounds, preserved rod outer segments (Figure 3—figure supplement 2A). Reasoning that preservation of outer segments would equate to increased rod cell size, relative to rounded morphologies of dead or stressed cells, average rod cell sizes were calculated as the total YFP volume divided by the number of rod cells per each 3D confocal stack (i.e. average YFP volume per cell). Compared to rod ablated controls, five of seven lead compounds tested promoted statistically significant increases in rod cell volumes (Figure 3—figure supplement 2B). These data suggest some lead compounds were able to maintain rod outer segment morphology, thus potentially maintaining rod cell function.

To identify new therapeutics for RP patients, we reasoned compounds producing neuroprotective effects across fish and mammalian RP models would likely target conserved mechanisms, and thus be more likely to translate successfully in the clinic. We therefore tested the efficacy of lead compounds in a series of mouse models of retinal photoreceptor degeneration.

First, we tested compounds for the capacity to protect mouse primary photoreceptor cells from stressor-induced cell death in culture. Photoreceptors were isolated from postnatal day four (P4) QRX mice, a transgenic line in which GFP expression is restricted to photoreceptors (Wang et al., 2004), and grown as previously described (Fuller et al., 2014). To induce photoreceptor cell death, thapsigargin (0.25 µM) or tunicamycin (0.6 µg/mL) were added. These ‘stressor’ compounds deplete endoplasmic reticulum (ER) calcium levels (Thastrup et al., 1990) and inhibit protein glycosylation (Fliesler et al., 1984), respectively (Lai et al., 2007b). In turn, they elicit an unfolded protein response (UPR) (Wang et al., 2015b) and related ER stress (Oslowski and Urano, 2011; Zhang et al., 2014), both implicated in the etiology of RP (Griciuc et al., 2011; Rana et al., 2014). To test for survival effects, all eleven lead compounds were screened at seven concentrations across a threefold dilution series (from 30 µM to 40 nM). After 48 hr in stressor ± lead compounds, cells were imaged using an automated high-content screening system. Photoreceptor survival was assessed by automated quantification of GFP-expressing cells (Figure 4A). Nine of 11 lead compounds, proved protective in the thapsigargin-induced cell death assay (Figure 4B). MIC, CLO and SUL also protected photoreceptors from tunicamycin-induced cell death (Figure 4—figure supplement 1). Thus, nine of 11 lead compounds promoted the survival of mouse primary photoreceptor cells in at least one stressor-induced cell death assay, and three were neuroprotective in both assays.

We next tested lead compounds for survival effects in rd1 mouse retinal explant cultures. The rd1 mouse model of RP exhibits early onset rod cell degeneration caused by a mutation in the Pde6b gene (Danciger et al., 1990; Pittler and Baehr, 1991; Sidman and Green, 1965), which is an ortholog of the human RP-associated PDE6B (Khramtsov et al., 1993; McLaughlin et al., 1993). In rd1 mice, photoreceptor degeneration begins around P10 and progresses rapidly. By P21, only a single row of photoreceptor cells remain in the outer nuclear layer (ONL; Tansley, 1951) making it an excellent system for screening potential neuroprotectants (Beeson et al., 2016). Here, retinal explants from P10 rd1 mice were isolated and cultured ex vivo (Bandyopadhyay and Rohrer, 2010). Eight lead compounds, all but CAL, SUL, and Zinc pyrithione (ZPT), were tested for survival effects at three concentrations across a fivefold dilution series centered on the concentration most effective in fish RP models. After 11 days in culture, explants were fixed, stained and the number of photoreceptor rows counted (Figure 5A). Neuroprotective effects were defined as a concentration-dependent increase in the number of photoreceptor rows remaining in the ONL relative to untreated controls (p≤0.05). An average of 1.2 ±0.19 rows of photoreceptors remained in the ONL of control explants. Three lead drugs, CPO, DHA, and ART, increased the number of surviving photoreceptor layers (Figure 5B). Representative images of control (DMSO and POS) and lead-treated explants demonstrate photoreceptor layer protection (Figure 5C). CPO, DHA, and ART thus promoted rod cell survival in the fish RP model and two mouse cell culture RP models. However, high concentration CPO treatments (15 µM) led to disruption of retinal histology due to induction of proliferation in the inner nuclear layer (INL) and ONL. Therefore, as DHA is the active metabolite of ART-related compounds, we proceeded with testing DHA in an in vivo mouse RP model.

To test the potential of DHA as a ‘pan-disease’ therapeutic, that is, whether it was effective across multiple genetic RP models, it was tested for photoreceptor survival effects in the rd10 mouse model of RP (Pde6b^rd10). The rd10 line was selected for in vivo experiments due to its slower rate of photoreceptor degeneration relative to other rd mutants, thus allowing a prolonged window for pharmacological intervention. DHA was selected based on its superior performance in rd1 retinal explant assays (Figure 5) and because it is the active metabolite of a second lead drug candidate, ART, recently shown to protect photoreceptors from light damage (Lu and Xie, 2019). To create a long-release formulation, DHA was encapsulated in PLGA polymers (Poly(D,L-lactide-co-glycolide)). Release kinetics assays in PBS/0.1% DMSO in vitro suggested DHA would reach maximal concentrations after 30 days, and remain stable for at least 20 days thereafter (Figure 6A). We noted that release was minimal in the absence of DMSO, but were reluctant to include DMSO for injections due to the potential for deleterious effects on the retina (Tsai et al., 2009). PLGA-DHA was injected into the vitreous of one eye of rd10 mice at P14, with the contralateral eye serving as a vehicle injection control. At P32, 18 days after injection, and when DHA levels were predicted to reach 80% of maximal concentration based on in vitro releases kinetics, rd10 mouse eyes were processed for immunohistochemistry (Figure 6B) and ONL thickness quantified (Figure 6C). Despite initial promising results in pilot assays, no reproducible neuroprotective effects were observed. It is possible that in the absence of DMSO release kinetics may have been limiting in vivo. We plan to address this possibility in future studies.

Previously, we used ‘on label’ information to explore MOA of hit compounds identified during an in vivo phenotypic screen of the JHDL (Wang et al., 2015b). Recently, we have become interested in additional advantages afforded by HTS-based MOA data and whole-organism phenotypic screening, such as polypharmacology (Dar et al., 2012; Rennekamp and Peterson, 2015; Rihel et al., 2010). Accordingly, we applied a target-agnostic MOA analysis process by evaluating lead compound performance in HTS and ultra-HTS studies archived on PubChem (https://pubchem.ncbi.nlm.nih.gov/↗). Many lead compounds exhibited shared target/pathway activities, suggesting in-common MOA (Table 2). Among shared targets, Tyrosyl-DNA Phosphodiesterase 1 (TDP1), a DNA repair enzyme, was the most popular (eight leads). To test whether TDP1 inhibition promoted rod photoreceptor survival, we tested three TDP1 inhibitors in rod:YFP-NTR fish: paromomycin (PM), thiostrepton (ThS), and methyl-3,4-dephostatin (MD) (Huang et al., 2011; Liao et al., 2006). Both PM and ThS showed neuroprotective effects (Table 3, Figure 7).

This result was surprising given that NTR reduction of Mtz is thought to cause DNA damage-induced cell death (Curado et al., 2007). Thus, inhibition of a DNA repair enzyme would be expected to enhance, not inhibit, NTR/Mtz-mediated cell death. However, an alternative means of disrupting TDP1 activity is by inhibiting Poly (ADP-ribose) Polymerases (PARPs). Indeed, in a comprehensive 400,000 compound qHTS assay designed to identify TDP1 inhibitors, all five hits turned out to be PARP inhibitors (Murai et al., 2014). PARPs also mediate DNA repair but, interestingly, hyperactivation of PARP1 leads to a specific form of DNA damage-induced cell death, termed parthanatos (Fatokun et al., 2014; Wang et al., 2016). PARP is also involved in a cGMP-dependent cell death pathway implicated in RP (Power et al., 2020).

We therefore assayed PARP inhibitors for the capacity to promote rod cell survival in rod:YFP-NTR fish. All eight PARP inhibitors tested had neuroprotective activity, ranging from 9% to 20% (Figure 7). To account for other cell death mechanisms implicated in neurodegeneration, we tested an inhibitor of necroptosis (necrostatin-1; NEC), and three inhibitors of apoptosis (Ac-DEVD-CHO, CASI, and CASVII, Table 3). Surprisingly, NEC promoted rod cell survival, while the apoptosis inhibitors did not (Figure 7, Table 3). Interestingly, paired testing of PARP and necroptosis inhibitors resulted in additive effects (Figure 7—figure supplement 1, Supplementary file 1b), suggesting either differential responses to NTR/Mtz-induced DNA damage among rod cells, or that inhibiting both pathways keeps cells from activating the other mechanism of cell death when only one pathway is blocked.

To further dissect cell death pathways mediating NTR/Mtz-induced rod cell ablation, we used a ‘redundant’ CRISPR/Cas9 gene targeting approach (i.e. four guide RNA targeting), which enables phenotyping in injected zebrafish embryos/larvae (Wu et al., 2018). Key factors for apoptosis (casp3a and casp3b), PARP-dependent cell death pathways parthanatos and/or cGMP-dependent cell death (parp1), and necroptosis (ripk1l) were targeted. In addition, tdp1 was targeted as a DNA repair control. For each targeted gene, four gRNAs and Cas9 protein were co-injected into one-cell stage embryos. Injected and uninjected (control) larvae were treated ±2.5 mM Mtz at five dpf and rod-YFP levels quantified at six dpf by ARQiv. For Mtz-ablated larvae, knockdown of parp1 markedly increased rod cell survival (39% compared to 20% in the uninjected control, p=4E-12), and ripk1l knockdown modestly protected rods (26% compared to 22% in the uninjected control, p=0.05). However, no improvement in survival was observed for either casp3a/b or tdp1 knockdown (Figure 8A). In no Mtz controls, knockdown of parp1, ripk1l or tdp1 had no effect on rod cell numbers. Conversely, casp3a/3b double knockdown increased rod cell numbers (Figure 8B) suggesting inhibition of developmental apoptosis and confirming efficacy of casp3a/3b knockdown. Confocal imaging confirmed parp1 knockdown effects on rod cell survival and casp3a/3b knockdown effects on rod cell development (Figure 8—figure supplement 1A). Reduced gene expression was verified by qPCR (Figure 8—figure supplement 1B, Supplementary file 3). These results are consistent with the NTR/Mtz system eliciting DNA damage-associated cell death mediated by parp1, that is, parthanatos, in rod cells.

To further test for activation of PARP signaling during NTR/Mtz-induced rod cell death, accumulation of polymerized poly (ADP-ribose) (PAR), a downstream effector of PARP1 activation (Kam et al., 2018), was analyzed by western blot. five dpf rho:YFP-NTR larvae were treated ±2.5 mM Mtz and protein samples collected from ~30 fish at 3, 6, 12, 24, and 48 hr post-Mtz treatment initiation. Western blots showed clear accumulation of PAR in Mtz-treated fish at 24 and 48 hr post-Mtz (Figure 8—figure supplement 1C) which was verified quantitatively (Figure 8—figure supplement 1D). Collectively, results of cell death pathway analyses are concordant across chemical inhibition, genetic knockdown, and biochemical assays, strongly suggesting NTR/Mtz-induced cell ablation is mediated by PARP-dependent cell death (e.g. parthanatos, cGMP-dependent cell death), a target/pathway implicated across multiple mammalian models of RP (Arango-Gonzalez et al., 2014).

To determine if lead compound effects were mediated by inhibition of PARP signaling, parp1 expression was knocked down prior to testing for effects on rod cell survival in rho:YFP-NTR larvae. Four leads predicted to be PARP inhibitors (MIC, CLO, CPO, DHA), one predicted to be a non-PARP inhibitor (WAR), and a known PARP inhibitor control (BMN) were tested. Two lead compounds, MIC and DHA and BMN (the PARP inhibitor control) produced no statistically significant enhanced neuroprotective effects over parp1 knockdown. Conversely, WAR, CLO and CPO increased rod cell survival in parp1 knockdown fish (Figure 9A). qPCR confirmed mRNA knockdown (Figure 9B) and controls showed no effects of parp1 knockdown on rod cell development (Figure 9C, -Mtz) and enhanced rod cell survival upon Mtz treatment (Figure 9C, +Mtz). These data confirm that PARP inhibition mediates the neuroprotective effects of some lead compounds (e.g. MIC and DHA) while others are PARP-independent. The latter result suggests lead compounds may act via complementary neuroprotective mechanisms. Combined, these findings are consistent with rod cell loss in our fish RP model occurring, at least partially, by PARP-dependent cell death.

PubChem data also suggested potential complementary MOA, that is, multiple independent targets across lead compounds (Table 2). We therefore hypothesized that combining lead compounds may produce enhanced, additive, or even synergistic effects. To investigate this, combinatorial lead drug treatments were tested in primary mouse photoreceptor cell cultures. As above, photoreceptors were isolated from QRX mouse retinas and used in stressor-treated primary cell culture assays. All 11 lead compounds were tested at a selected concentration individually and in pairs and the survival of GFP-labeled photoreceptors analyzed by high-content imaging and automated cell counting. The results showed no wholly additive effects, that is, survival percentages equaling or excelling the summed effect of individual compound assays. However, enhanced effects – i.e., better median survival than for either compound alone – were observed for 30 of 55 pairs in thapsigargin-treated cultures and for 20 of 55 pairs in tunicamycin-treated cultures (Figure 4—figure supplement 2). Finally, seven lead compounds were tested alone and in pairs in rho:YFP-NTR zebrafish using optimal effective concentrations. Pairs producing survival effects equal to or greater than the sum of their individual values (±10%) were considered additive. By this criterion, 10 of 19 viable pairs (two pairs proved lethal) exhibited additive effects (Figure 10, bolded; Supplementary file 1c) and three pairs produced supra-additive effects (i.e. ≥25% greater than summed effects) suggesting possible synergy (Figure 10, asterisks). The maximal average rod cell survival effect was 58% (WAR + CPO). Several compounds showed broadly additive effects, for example, WAR was additive with all six drugs and COR with four of five pairs (one pair proving lethal). Additive effects suggest multiple signaling pathways are involved in NTR/Mtz-induced photoreceptor degeneration, thus that combinatorial drug regimens may be required to achieve maximal therapeutic benefits for RP patients.

All animal studies described herein were performed in accordance with both the Association for Research in Vision and Ophthalmology (ARVO) statement on the ‘Use of Animals in Ophthalmic and Vision Research’ and the National Institutes of Health (NIH) Office of Laboratory Animal Welfare (OLAW) policies regarding studies conducted in vertebrate species. Animal protocols were approved by the Animal Care and Use Committees of the Johns Hopkins University School of Medicine (protocol # FI19M489 and #MO20M253) and Medical University of South Carolina (protocol #2018–00399).

Zebrafish were maintained using established temperature and light cycle conditions (28.5°C, 14 hr of light/10 hr of dark). A previously generated zebrafish transgenic line, Tg(rho:YFP-Eco.NfsB)gmc500 (hereafter, rho:YFP-NTR) expresses a fusion protein of enhanced yellow fluorescent protein (YFP) and a bacterial nitroreductase (NTR) enzyme (encoded by the E. coli nfsB gene) selectively in rod photoreceptors (Walker et al., 2012). When rho:YFP-NTR fish are exposed to the prodrug metronidazole (Mtz), NTR reduces Mtz to a DNA damage-inducing cytotoxic derivative, resulting in the death of NTR-expressing cells (Chen et al., 2011; Curado et al., 2007). A newly generated transgenic line, Tg(rho:GAP-YFP-2A-nfsB_Vv F70A/F108Y)jh405 (hereafter rho:YFP-NTR2.0) expresses a membrane tagged YFP and an improved version of NTR which enables induction of cell death at much lower concentrations of Mtz (Sharrock et al., 2020). These lines were propagated in a pigmentation mutant with reduced iridophore numbers, roy^a9 (roy), to facilitate YFP reporter signal detection in vivo. Non-transgenic roy fish were used to define reporter signal cutoff values for fluorescent microplate reader assays. For each large-scale drug screening assays, 6000–12,000 eggs were collected by group breeding ~300 adult rho:YFP-NTR fish (White et al., 2016).

rho:YFP-NTR larvae were treated with 2.5 mM Mtz or 0.1% DMSO at 5 dpf for 2 days (10 fish per condition), and collected at 7 dpf for immunostaining using previously reported methods (Unal Eroglu et al., 2018). Briefly, larvae were fixed in 4% PFA at 4°C overnight, infiltrated with 30% sucrose, and embedded in OCT. Cryosections of 10 µm thickness were made for immunofluorescence staining. Each section was blocked with 1x PBS containing 0.5% Triton X-100% and 5% goat serum for one hour followed by primary antibody staining overnight. The next day, each section was washed and stained with the secondary antibody for 2 hr. All slides were mounted and underwent confocal imaging. The primary antibodies used in this study were zpr-1 (anti-Arrestin3a; 1:200 dilution; from ZIRC), 1d1 (anti-Rhodopsin; 1:50; gift from Dr. James M. Fadool) and 4C12 (uncharacterized rod cell antigen, 1:100 dilution; gift from Dr. James M. Fadool). The secondary antibody used was goat anti-mouse Alexa 647 (1:1000; Invitrogen).

rho:YFP-NTR larvae were separated into six groups of ~23 larvae per group. Each group was treated with varying Mtz concentrations (10, 5, 2.5, 1.25, 0.625, 0 mM), from 5 to 8 days post-fertilization (dpf). YFP signals were quantified daily by fluorescence microplate reader (TECAN Infinite M1000 PRO; excitation 514 nm, bandwidth 5 nm; emission 538 nm, bandwidth 10 nm) to track changes in rod photoreceptor cell numbers relative to Mtz concentration (Walker et al., 2012; White et al., 2016). Two technical repeats were performed and results aggregated across experiments by normalizing to non-ablated controls.

To calculate sample size (N) and evaluate assay quality, two 96-well plates of larvae were treated with 2.5 mM Mtz/0.1% DMSO (ablated ‘+Mtz’ control) or 0.1% DMSO only (non-ablated ‘-Mtz’ control) from 5 to 7 dpf (196 fish per condition). YFP signals were then quantified at seven dpf (due to the large sample size, a single technical replicate was performed). Power calculations were used to determine sample sizes across a range of error rates and effect sizes (White et al., 2016). This analysis suggested a sample size of nine per condition tested would sufficiently minimize false-discoveries (i.e. type I and type II error rates of 0.05 and 0.05, respectively) and account for drug effects reaching 50% of signal window of the positive (non-ablated) control. However, to account for plating errors and other confounding variables, we chose to increase the sample size to 16.

A schematic of the primary screening process is presented in Figure 1C. For the primary screen, 2934 compounds from John Hopkins Drug Library (JHDL) were screened across six concentrations (4 μM-125 nM using a twofold dilution series). The JHDL consists of ~2200 drugs approved for use in humans (e.g. FDA approved) with the remainder approved for clinical trials (Chong et al., 2006). The ARQiv screening process has been detailed previously (White et al., 2016) and was adapted here for large-scale quantification of YFP-expressing rod photoreceptors (Walker et al., 2012; White et al., 2016). rho:YFP-NTR embryos were collected and raised in zebrafish E3 embryo media (5 mM NaCl; 0.17 mM KCl; 0.33 mM CaCl; 0.33 mM MgSO₄). At 16 hr post fertilization (hpf), N-phenylthiourea (PTU) was added to E3 media (E3/PTU) at a final concentration of 0.2 mM to promote ocular transparency by inhibiting melanosome maturation in the retinal pigment epithelium. At 4 dpf, visual screens were performed to remove larvae with abnormal morphology or low YFP expression levels. Stock drug and DMSO (negative control) solutions were automatically dispensed and diluted across a 96-well plate containing E3/PTU using a robotic liquid handling system (Hudson Robotics). At 5 dpf, a COPAS-XL (Complex Object Parametric Analyzer and Sorter, Union Biometrica) was used to dispense single larvae into individual wells containing either drug or DMSO; the final DMSO concentration was 0.1% across all conditions. After a 4 hr pre-exposure to test drugs or DMSO alone, larvae were treated with 2.5 mM Mtz to induce rod photoreceptor death. Larvae were maintained under these conditions for 2 days until 7n dpf and then anesthetized with clove oil (50 ppm final concentration); YFP signals were measured as described above. Larvae exposed to 2.5 mM Mtz/0.1% DMSO without any tested drug served as controls (‘+Mtz’) for maximal rod cell ablation. Larvae treated solely with 0.1% DMSO served as non-ablated controls (‘-Mtz’) to calculate maximal YFP signal levels. Non-transgenic larvae were used to establish a signal cutoff value, as previously described (White et al., 2016). A customized R code program was applied for real-time data analysis of compound performance relative to controls, including dose-response curves and strictly standardized mean difference (SSMD) scores (https://github.com/mummlab/ARQiv2↗; Ding and Zhang, 2021). Compound concentrations producing a SSMD score ≥1 were considered potential ‘hits’ and evaluated visually using fluorescence microscopy to eliminate non-specific fluorescence; e.g., dead larvae or autofluorescent compounds.

After the initial screen, 42 top-performing ‘hit’ compounds were selected for confirmatory and orthogonal assays. Hit drugs were obtained from new sources and tested across a wider range of concentrations (fivefold dilution series, eight total concentrations, n=30 fish/condition). Based on the toxicity profile of each drug, the starting concentration was either 100, 10, or 1 μM. For validation assays, YFP signals were measured by fluorescence microplate reader with at least three experimental repeats conducted and SSMD scores calculated. All experimental results were normalized and pooled to calculate effect sizes, confidence intervals, sample sizes (N) and p-values. p Values were calculated using Student’s t-test followed by Bonferroni correction for multiple comparisons resulting in an adjusted significance level of 0.005 (α=0.005).

For in vivo confocal imaging assays, five dpf rho:YFP-NTR larvae were treated with drug plus 2.5 mM Mtz/0.1% DMSO (ablated +Mtz controls), or 0.1% DMSO (non-ablated -Mtz controls) for 48 hr and processed for intravital imaging at 7 dpf. Similar treatment strategy was applied to five dpf rho:YFP-NTR2.0 larvae except the Mtz treatment was reduced to 10 μM. All compounds were tested at the maximal effective concentration in 0.1% DMSO. Larvae from each group were anesthetized in 0.016% tricaine and embedded on their sides in 1% low melt agarose gel. An Olympus Fluoview FV1000 confocal microscope with a 20x water immersion objective (0.95 NA) was used to collect 30–40 z-stack images of YFP-expressing rod cells across the whole retina at 4 μm intervals. These image slices were stacked into a single maximal intensity projection image using Image J. To provide greater detail of rod photoreceptor morphology, a region in the dorsal-nasal quadrant was imaged using a 60× water immersion objective (1.10 NA) at 4.18 μm intervals. Three contiguous image slices were then stacked into a single maximal intensity projection image using Image J. As a means of assessing rod photoreceptor survival, intravital confocal retinal images were collected of YFP-expressing rod cells per condition and processed for volumetric quantification of YFP signal volume using automated image analysis software (Imaris 3.9.1; see White et al., 2017). Briefly, YFP-expressing rod cells volumes were automatically rendered using the same parameters across all treatment groups with the sum of all volumes used to estimate rod photoreceptor numbers per each retina. For images from rho:YFP-NTR2.0 larvae, the sum of rod cell YFP volumes in each retina was normalized to the number of YFP-expressing cells to estimate the average rod photoreceptor cell size per each condition. The average number of cells quantified per retina was 18.52, with sample sizes ranging from 9 to 19 fish per condition.

The eleven lead compounds were tested to eliminate false-positives that directly inhibited E. coli NTR enzymatic activity (from the nfsB_Ec gene) using an established anticancer prodrug (CB1954) reduction assay (Prosser et al., 2010). The highest soluble concentration of the test drug (up to 300 μM maximum) or 0.1% DMSO (rate control) was added to a master mix of 250 μM CB1954/1 μM NTR/250 μM NADPH/10 mM Tris (pH 7.0) to determine any inhibitory effect of the compounds on NTR-mediated reduction of CB1954. Reactions were conducted in 100 μl volume in a 96-well plate format. CB1954 reduction kinetics were assayed at 420 nm unless the test compound exhibited confounding absorbance at 420 nm, in which case NADPH depletion was monitored at 340 nm. The reaction rates for CB1954 reduction in the presence of tested compounds were compared to the DMSO control. Compounds with NTR activity <75% of controls were deemed potentially inhibitory and IC₅₀ values (the concentration required to restrict NTR activity to 50% of the DMSO control) determined.

Compounds were also evaluated for the ability to inhibit NTR-mediated reduction of Mtz. For this assay, the highest soluble concentration of the test drug (up to 300 μM maximum) or DMSO (rate control) was added to a master mix of 500 μM Mtz/17 μM NTR/100 μM NADPH/0.55 μM glucose dehydrogenase/5 mM glucose/50 mM sodium phosphate buffer (pH 7.0; the glucose dehydrogenase used for cofactor regeneration, maintaining a steady-state of NADPH to allow Mtz reduction to be monitored directly at 340 nm without interference from NADPH oxidation). Reactions were conducted in 100 μl volume and assayed at 340 nm using 1 mm quartz cuvettes (Rich et al., 2018). IC₅₀ values for potentially inhibitory compounds were measured as with CB1954. All kinetic assays were conducted in triplicate with a single purified batch of NfsB_Ec protein.

To evaluate whether lead compounds promoted rod photoreceptor development, rho:YFP-NTR larvae were handled as described for the primary screen with the following exceptions: at five dpf, larvae were exposed solely to test compounds; larvae were not treated with Mtz. YFP reporter signals were quantified by fluorescence microplate reader at 7 dpf and compared to non-ablated ‘-Mtz’ controls treated only with 0.1% DMSO. Sample sizes ranged from 58 to 88 across at least two experimental repeats.

To determine whether lead compounds stimulated retinal regeneration, rho:YFP-NTR larvae were handled as described for the primary screen with the following exceptions: at 5 dpf, larvae were incubated with either 10 mM Mtz or 0.1% DMSO for 24 hr. At 6 dpf, larvae were then placed in new 0.1% DMSO/E3/PTU media containing test compounds (or DMSO alone) for 3 days. YFP signal intensity was measured at 9 dpf. Sample sizes ranged from 31 to 83 across at least two experimental repeats.

A transgenic mouse line with an IRES-GFP cassette integrated at the QRX locus (QRX-IRES-EGFP) has GFP reporter expression restricted to retinal photoreceptors, and was used here to isolate primary photoreceptor cells. Mice were housed with 12 hr light/12 hr dark cycles, at 22°C, 30–70% relative humidity and food/water ad libitum. Primary mouse retinal photoreceptor cells were isolated and prepared for culture as previously described (Fuller et al., 2014). Briefly, murine retinas were isolated at postnatal day four (P4). Retinal tissue was dissociated into a single-cell suspension by incubating whole tissue in activated papain in Hibernate-E without Calcium (BrainBits) for 15 min at 37°C. Cells were resuspended in culture media (Neurobasal, 2% B-27, 0.5 mM L-Glutamine and 1x final Penicillin/streptomycin; all Life Technologies) and seeded onto poly-D-lysine coated 384 well tissue culture plates. To test lead compounds for survival effects, thapsigargin and tunicamycin (Sigma) and were used as stressor compounds to induce photoreceptor cell death. Stressors and test compounds were added at the time of primary cell seeding. Eleven lead compounds were tested at seven concentrations across a threefold dilution series (from 30 µM to 40 nM). For paired tests, each compound was applied at a single concentration, WAR (3.33 µM), CPO (0.04 µM), DHA (1.11 µM), ART (10 µM), ZPT (0.12 µM), CAL (0.04 µM), COR (3.33 µM), CHL (3.33 µM), SUL (3.33 µM), MIC (10 µM), and CLO (1.11 µM). After a 48 hr treatment, cells were stained with Hoescht and ethidium homodimer; nine field images were acquired via an automated imager (Cellomics Vti; ThermoFisher) using a 20x objective and images analyzed. Photoreceptor number and the percent of photoreceptor to retinal cell population per well was determined by automated quantification of the number of live Hoechst 33342-stained, GFP-expressing cells using a custom algorithm (Neuronal Profiling software package; ThermoFisher). A minimum of two experimental repeats was performed for each test with individual assays consisting of six biological replicates per condition tested.

Mice were generated from retinal degeneration 1 (Pde6b^rd1, hereafter rd1) breeding pairs (gift from Dr. Debora Farber; UCLA) and housed under a 12:12 light:dark cycle with access to food and water ad libitum. All chemicals used for organ cultures were tissue culture grade and purchased from Invitrogen unless otherwise noted. Cultures of retina with attached retinal pigment epithelium (RPE) were grown according to published protocols (Ogilvie et al., 1999; Pinzón-Duarte et al., 2000; Rohrer and Ogilvie, 2003) with modifications (Bandyopadhyay and Rohrer, 2010). P10 pups were decapitated and heads were rinsed in 70% ethanol; whole eyes were dissected and placed in ice-cold Hanks balanced salt solution plus glucose (6.5 g/L). Eyes were first incubated in 1 mL of high-glucose Hanks balanced salt solution/0.5 mg/ml proteinase K (37°C, 7 min) and then in Neurobasal medium (Life Technologies) plus 10% fetal calf serum to stop enzymatic activity. The retina with attached RPE was dissected free from the choroid and sclera after removing the anterior chamber, lens and vitreous. Relaxing cuts were made to flatten the tissue prior to transfer to the upper compartment of a Costar Transwell chamber (RPE layer faced-down) using a drop of Neurobasal medium. Neurobasal media with B-27 supplement (Life Technologies) was placed in the lower compartment. For each of the cohorts, three to fourindividual P10 retina/RPE explants were placed in culture. The cultures were kept in an incubator (5% CO₂, balanced air, 100% humidity, 37°C) and the lower compartment media changed every 2 days (neither antimitotics nor antibiotics were required). Test compounds were added to the culture media and refreshed every 2 days for 11 days. At completion, explants were fixed in 4% PFA, sectioned 14 µm thick, and stained with toluidine blue (Bandyopadhyay and Rohrer, 2010). For each culture, 10 counts of rows of photoreceptors were performed in the center of the explant; the average of this cell counts provides the mean for each retina, with the average of all retinas providing the mean ± SEM per culture condition. Cell counts were compared by repeated measure ANOVA to assess dose-response treatment effects followed by a Bonferroni posthoc analysis to determine if differences between control and lead compounds compared across the entire dose range were statistically significant. An ANOVA across groups but within a particular dose, using a Bonferroni/Dunn test to reduce type I errors, was used to determine if lead compounds promoted a statistically significantly increase in cell numbers compared to control media at low, middle, and/or high concentrations.

PLGA-DHA microparticles were prepared using a single emulsion solvent evaporation method. Briefly, 200 mg PLGA (Resomer RG 502 hr, Poly(D),L-lactide-co-glycolide; Evonik Corporation) was dissolved in 1 mL of dichloromethane (DCM, Sigma-Aldrich), and mixing with 40 mg DHA (TCI, Tokyo, Japan) dissolved in 0.125 ml dimethyl sulfoxide (DMSO, Sigma-Aldrich). The mixture was homogenized (L4RT, Silverson Machines) at 7000 RPM for 1 min. The homogenized mixture was then poured into a solution containing 1% polyvinyl alcohol (25 kDa, 88% hydrolys, Polysciences, Warrington, PA) in water under continuous stirring. Particles were hardened by allowing solvent to evaporate while stirring at room temperature for 2 hr. Particles were collected via centrifugation (International Equipment Co) at 1000 x g for 5 min, and washed three times with HyPure cell culture grade water (endotoxin-free, HyClone, Logan, UT) and re-collected by centrifugation three times. The washed particles were then lyophilized and stored frozen until use. Microparticles were resuspended at the desired concentration prior to injection in a sodium hyaluronate solution (Healon) diluted fivefold with endotoxin-free HyPure water at the desired concentration prior to injection.

Particle size distribution was determined using a Coulter Multisizer 4 (Beckman Coulter, Inc, Miami, FL). Particles were resuspended in double distilled water and added dropwise to 100 ml of ISOTON II solution until the coincidence of the particles was between 8% and 10%. At least 100,000 particles were sized for each batch of particles to determine the mean particle size and size distribution. To determine the drug loading, microparticles were dissolved in DMSO and the total drug content was calculated by measuring the UV absorbance at 289 nm after react with NaOH in triplicate (C.-S. Lai et al., 2007a). The average microparticle size was 14.2 ± 1.9 µm and the DHA loading was 20.3% ± 0.7 (w/w) across three experimental replicates.

Release kinetics were obtained by resuspending microparticles in 1 ml phosphate buffered saline (PBS containing 0.1% DMSO, pH 7.4) and incubating at 37°C on a platform shaker (140 RPM). Supernatant was collected at predetermined intervals by centrifugation at 2000 x g for 5 min. Drug-containing supernatant was collected and particles were resuspended in 1 ml of fresh 1xPBS containing 0.1% DMSO. DHA concentration in the collected supernatant was assayed via absorbance at 289 nm in triplicate for each sample (Figure 4—figure supplement 1).

B6.CXB1-Pde6b^rd10/J animals (Jackson Laboratories stock #004297) were maintained as homozygotes. On P14, animals were anesthetized with ketamine/xylazine (115 mg/kg and 5 mg/kg, respectively), and placed on a heating pad to maintain body temperature throughout injection and recovery. Glass needles were pulled and a bore size of approximately 75 µm was made to allow for the movement of the microspheres into and out of the needle. A PLI-100 picospritzer (Harvard Apparatus) was used for intravitreal injections with settings of 350 ms injection time at 30 psi, which yielded the injection of approximately 500 nL. One eye was randomly selected for injection with vehicle and the other was used for microsphere injections. Following injection, GenTeal gel (Novartis) was applied to the eyes to prevent corneal drying. Animals were monitored until they were awake and moving normally. A total of 17 mice was used across three technical repeats.

Eighteen days post-injection, animals were anesthetized with isoflurane and decapitated; the eyes were removed and placed in 4% PFA for 30 min. After removing the cornea and lens, the eye cups were placed back into fixative for 2 hr on ice. After thorough washing with 1× PBS, eyecups were soaked in 15% sucrose (w/v) in 1× PBS for 3 hr, then incubated in 30% sucrose (w/v)/1× PBS overnight. Samples were embedded in OCT compound (TissueTek) and frozen on dry ice. Cryosections of 12–14 µm thickness were cut on a Leica CM3050S cryostat and were mounted on Superfrost Plus slides (Fisher Scientific). Slides were rehydrated with 1× PBS and then incubated in blocking buffer (10% Normal donkey serum/ 0.3% Triton X-100/1× PBS) for 2 hr, and were then incubated for 30 hr at 4°C with primary antibodies diluted in blocking buffer. Sections were washed in 1× PBS and incubated for 2 hr at room temperature with secondary antibodies diluted in blocking buffer. Following additional washes, sections were incubated with DAPI and mounted with FluoroGel (Electron Microscopy Sciences).

Sections were examined using a Zeiss Axioplan two epifluorescence microscope fitted with an Axioscope camera. Following acquisition with a 40× air objective, images were analyzed in ImageJ using the measure tool to examine Outer Nuclear Layer (ONL) thickness. Four to 5 measurements were taken at 75 µm intervals for 200–500 µm both superior and inferior from the optic nerve head. The measured values from multiple sections were averaged for each eye. Data were analyzed and statistical tests were performed in GraphPad Prism 7. An Olympus FV1000 confocal microscope was also used to image sections stained with multiple dyes.

To evaluate tdp1 and parp1 as a potential shared target across multiple leads, as well as cell death pathways implicated in photoreceptor degeneration, eight PARP inhibitors, one necroptosis inhibitor, three apoptosis inhibitors, and three Tdp1 inhibitors were selected for testing using the confirmation screening protocol (Figure 2A). Based on the toxicity profile of each inhibitor, upper concentrations were adjusted to 64, 32, or 8 µM and a twofold dilution series was used to test a total of 10 concentrations for rod cell survival effects in rho:YFP-NTR larvae. Sample sizes varied from 27 to 173 across two to six experimental replicates for each tested compound. YFP signal levels were measured by fluorescence microplate reader (i.e. ARQiv assay). Results across conditions were normalized to non-ablated controls and pooled across experimental repeats to calculate effect sizes, confidence intervals, and p-values. P values were calculated by performing Student’s t-tests. Bonferroni correction for multiple comparisons resulted in an adjusted significance level of 0.003 (α=0.003). Note: assessments of TDP1 inhibitors followed the fivefold dilution series protocol used for secondary confirmation assays described above. For paired inhibitor tests, three concentrations of BMN (parp inhibitor) and three concentrations of NEC (necroptosis inhibitor) were combined. YFP quantification and statistical analysis of rod cell survival in Mtz-treated rho:YFP-NTR fish were performed as above.

Five genes, parp1, ripk1l, casp3a, casp3b, tdp1, were selected for CRISPR/Cas9-based knockdown in rho:YFP-NTR zebrafish using an established ‘redundant targeting’ methodology (Wu et al., 2018). Four sgRNAs were designed to target each gene. The oligonucleotide for generating each sgRNA was either those suggested by Wu et al., 2018, Table S2 or designed using CRISPRscan (https://www.crisprscan.org↗, Supplementary file 3; Moreno-Mateos et al., 2015). Preparation of sgRNAs followed the published method (Wu et al., 2018). Briefly, four sgRNAs of each gene were in vitro transcribed using the HiScribe T7 High Yield RNA Synthesis kit (New England BioLabs) and purified using the NEB Monarch RNA Cleanup kit. A mixture of four sgRNAs (1 ng in total) and Cas9 protein (2.5 µM, IDT) was injected into rho:YFP-NTR embryos at the one-cell stage for targeting parp1, ripk1l and tdp1. Four sgRNAs each for casp3a and casp3b were pooled to co-target both genes. Injected and uninjected embryos were treated ±Mtz (2.5 mM) at 5 dpf. Rod cell survival was evaluated by fluorescence microplate reader-based quantification of YFP at 6 dpf. Student’s t-test was used to compare rod survival between injected and uninjected (control) fish. Each gene knockdown experiment was conducted in triplicate or quadruplicate. A subset of lead compounds (and a PARP inhibitor control, BMN) was also tested in parp1 knockdown fish to determine if Parp1 inhibition was required for survival effects. Leads were applied to larvae one hour before Mtz administration and YFP quantification performed as above. Each drug test was conducted in duplicate or triplicate. To calculate p-values relative to +Mtz controls, Student’s t-tests were performed. Bonferroni correction for multiple comparisons resulted in an adjusted significance level of 0.006 (α=0.006).

To quantify gene expression levels in rho:YFP-NTR larvae ±CRISPR/Cas9-based gene knockdown, 16–24 larvae from each condition were pooled for mRNA extraction at six dpf. RNA was purified using NEB Monarch RNA Cleanup kit and reverse transcribed to cDNA using qScript cDNA synthesis kit (QuantaBio) according to the instruction from the manufacturer. Quantitative PCR was conducted using designed primers (Supplementary file 3) and PowerUp SYBR Green Master Mix (ABI) in QuantaStudio (ABI). Delta delta CT analysis was performed to calculate relative fold change in gene expression levels between knockdown larvae and controls. Each experiment was performed in triplicate or quadruplicate.

To further evaluate Parp-dependent signaling downstream of Mtz treatments in rho:YFP-NTR larvae, PAR accumulation, a vetted measure of Parp activity, was quantified by western blot following the published method (Kam et al., 2018). Briefly, ~30 rho:YFP-NTR fish were treated ±2.5 mM Mtz at five dpf were collected at 3, 6, 12, 24, 48 hr post treatment (hpt). Larvae were homogenized and lysed using RIPA buffer (Sigma) supplemented with protease inhibitors (Roche). Protein concentrations were quantified using Pierce BCA kit (ThermoScientific) following the manufacturer’s instruction. Protein samples were separated on Tris-Glycine gel and then transferred onto nitrocellulose membranes. The membrane was blocked with 5% non-fat milk in TBST (Tris-buffered saline with 0.1% Tween 20) at room temperature for 1 hr and then blotted with the primary antibody, anti-human-PAR (1:2000, Dowson lab reagent), at 4°C overnight. After rinsing in TBST for three times, the secondary antibody, anti-Human IgG-HRP (1:10,000, Abcam) was applied for 1 hr at room temperature. Anti-beta-actin-HRP antibody (1:20,000) was used as a loading control. Immunolabeled bands were visualized by ECL substrate and quantified in Fiji. 24 hr post-Mtz PAR accumulation assays were performed in triplicate. Mann-Whitney U test was performed and a p-value of ≤0.05 used to indicate statistical significance.

To test for additive neuroprotective effects in zebrafish, seven lead compounds were tested at optimal concentrations either individually or in pairs using the primary screen protocol. All experimental results were normalized to controls and pooled per compound tested to calculate effect sizes, confidence intervals, and p-values using Student’s t-test. Bonferroni correction for multiple comparisons resulted in an adjusted significance level of 0.002 (α=0.002). Sample sizes for each paired condition varied from 86 to 160 across at least four experimental replicates.

For the primary screening, under the R environment, a customized ARQiv data analysis package was used to calculate sample size (n), quality control strictly standardized mean difference (SSMD) and SSMD scores as previously described (White et al., 2016). The following results of each drug were derived: (1) a plot of signal to background ratio at all tested concentrations, (2) a plot and a table of SSMD scores, and (3) a signal intensity heat map of each drug plate (96-well plate view). To combine and analyze the data from different experiments, data normalization was conducted by (S_i -X-_neg)/(X-_Pos -X-_neg). S_i is the signal of i^th reading, X-_Pos is mean of positive controls, and X-_Neg is mean of negative controls. To compare between experimental conditions and controls, Student’s t-tests were performed followed by Bonferroni correction for multiple comparisons (α/n). Relative effects, 95% confidence intervals and p-values were calculated. For mouse photoreceptor cell culture assays and volumetric confocal image analyses, adjusted p-values were calculated by performing Mann-Whitney U tests followed by false discovery rate correction for multiple comparisons.