The antibiotic robenidine exhibits guanabenz-like cytoprotective properties by a mechanism independent of protein phosphatase PP1:PPP1R15A

The aminoguanidine compound robenidine is widely used as an antibiotic for the control of coccidiosis, a protozoal infection in poultry and rabbits. Interestingly, robenidine is structurally similar to guanabenz (analogs), which are currently undergoing clinical trials as cytoprotective agents for the management of neurodegenerative diseases. Here we show that robenidine and guanabenz protect cells from a tunicamycin-induced unfolded protein response to a similar degree. Both compounds also reduced the tumor necrosis factor α–induced activation of NF-κB. The cytoprotective effects of guanabenz (analogs) have been explained previously by their ability to maintain eIF2α phosphorylation by allosterically inhibiting protein phosphatase PP1:PPP1R15A. However, using a novel split-luciferase–based protein–protein interaction assay, we demonstrate here that neither robenidine nor guanabenz disrupt the interaction between PPP1R15A and either PP1 or eIF2α in intact cells. Moreover, both drugs also inhibited the unfolded protein response in cells that expressed a nonphosphorylatable mutant (S51A) of eIF2α. Our results identify robenidine as a PP1:PPP1R15A-independent cytoprotective compound that holds potential for the management of protein misfolding–associated diseases.

The unfolded protein response (UPR) 3 is a cellular reaction to the accumulation of unfolded proteins in the endoplasmic reticulum (ER). This stress response aims to restore homeostasis in the ER by lowering protein translation rates, augmenting the number of folding chaperones, and increasing the degradation of misfolded proteins. This is achieved through different pathways originating from three different ER-stress sensors that are activated by unfolded proteins in the ER (1). One sensor is inositol-responsive enzyme 1 (IRE1), which catalyzes splicing of transcription factor X-box-binding protein 1 (XBP1) mRNA, enabling translation of functional XBP1 and up-regulation of target genes encoding ER chaperones and components of the ER-associated protein degradation pathway (2). A second sensor is activating transcription factor 6 (ATF6), which induces various stress-related genes following its translocation to the Golgi and subsequent proteolytic activation (3). The third sensor is protein kinase R-like endoplasmic reticulum kinase, which phosphorylates the ␣ subunit of eIF2␣ at Ser-51 to reduce global translation while enhancing the translation of specific stress response genes such as ATF4, CHOP, and PPP1R15A (4). The latter branch of the UPR is also part of the more general integrated stress response (ISR) pathway. Importantly, the UPR and ISR are both regulated by a negative feedback loop that involves the delayed expression of PPP1R15A, also known as growth arrest and DNA damage-induced protein 34 (GADD34) or R15A, which forms a trimeric complex with protein phosphatase 1 (PP1) and G-actin (5,6). This PP1 holoenzyme counteracts UPR and ISR signaling through dephosphorylation of eIF2␣ at Ser-51. Inhibition of eIF2␣ phosphatase(s) is generally considered an attractive therapeutic strategy for prolonging UPR signaling, as it gives cells more time to cope with the detrimental effects of unfolded proteins, a hallmark of several neurodegenerative diseases (7,8).
Guanabenz is an aminoguanidine-type ␣ 2 adrenoreceptor agonist developed during the 1980s as an anti-hypertension drug but was later identified in a yeast-based screen as a compound with anti-prion activity (9). Subsequently, several groups demonstrated cytoprotective effects of guanabenz (analogs) in murine models of neurodegenerative diseases, such as amyotrophic lateral sclerosis (10), multiple sclerosis (11), and vanishing white matter disease (12). The cytoprotective effects of guanabenz (analogs) have been explained by its ability to maintain eIF2␣ phosphorylation through inhibition of PP1:R15A/B (13). However, the molecular mechanism underlying the cytoprotective action of guanabenz (analogs) is controversial (14 -16). Indeed, the proposed effects of guanabenz (analogs) on PP1: R15 complexes include such diverse mechanisms as disruption of the PP1:R15A interaction (10,13), interference with the recruitment of eIF2␣ (17), and targeting of R15B (a constitutively expressed R15 variant) for proteolytic degradation (18). Moreover, guanabenz (analogs) had no effect on the composition or activity of an in vitro reconstituted PP1:R15A:G-actin holoenzyme (16). Furthermore, guanabenz retained its cytoprotective effects in cells and organisms that lacked R15A or only expressed nonphosphorylatable eIF2␣ (S51A) (15). Despite the uncertainties concerning their mechanism of action, guanabenz (analogs) remain attractive therapeutics. Guanabenz itself has rapidly progressed to clinical trials using a drug-repurposing approach. A phase I clinical trial to evaluate the pharmacokinetics of guanabenz in multiple sclerosis patients has been completed (NCT02423083), and a phase II clinical trial for ALS (EudraCT no. 2014-005367-32) and earlychildhood onset vanishing white matter disease (EudraCT no. 2017-001438-25) are ongoing.
Robenidine (1,2-bis[(E)-(4-chlorophenyl)methylideneamino]guanidine) is an aminoguanidine that has been used since the early 1970s to prevent coccidian infections in rabbits, chickens, and turkeys (19,20). Often, animals are fed with robenidine-supplemented pellets (50 mg/kg of food) for their entire life, except for a washout period before slaughter. The chronic and widespread use of robenidine as a food supplement indicates that it is safe and well-tolerated, making it a suitable candidate therapeutic agent for humans. Furthermore, robenidine belongs to the aminoguanidine compound class, which represents a diverse group of bioactive compounds used for the treatment of a broad variety of diseases ranging from bacterial infections to cancer and diabetes, and many of these compounds are approved for human use (21). Despite its long history of use in animals, there have been no in-depth examinations of robenidine as a potential therapeutic agent for the treatment of human diseases. Here we show that robenidine and guanabenz show similar cytoprotective effects in stressed cells. Furthermore, both compounds reduced the expression of UPR and ISR markers in stressed cells. Finally, using mutant cell lines and split-luciferase sensors, we demonstrate that the cytoprotective effects of robenidine do not involve changes in eIF2␣ phosphorylation or the eIF2␣ phosphatase PP1:R15A. Our data identify robenidine as a novel aminoguanidine with therapeutic potential for the treatment of protein misfolding diseases.

Robenidine is a structural and functional analog of guanabenz
Using the PubChem search engine, we noticed structural similarities between robenidine and guanabenz (Fig. 1A). In contrast to guanabenz, robenidine has two substituted phenyl rings attached to its central aminoguanidine scaffold. In addition, these phenyl rings are para-chloro-substituted, whereas the phenyl ring of guanabenz has two ortho-chloro substitutions. Further mining of the PubChem BioAssay database revealed that guanabenz and robenidine were both active in a high-throughput screen against the malaria-causing protozoan Plasmodium falciparum (Fig. 1B) (National Center for Biotechnology Information, PubChem BioAssay Database, AID 504834, https://pubchem.ncbi.nlm.nih.gov/bioassay/504834, accessed July 28, 2018). Several anti-malaria drugs target the apicoplast (a nonphotosynthetic plastid) through interference with its translational machinery (23). Interestingly, a similar mechanism of action was proposed for the cytoprotective prop-erties of guanabenz in mammalian cells undergoing unfolded protein stress (24). These findings prompted us to explore whether robenidine has cytoprotective properties similar to those of guanabenz.

Cytoprotective effects of robenidine
The ER stress-induced UPR hampers cell cycle progression and proliferation. Indeed, treatment of HeLa cells with tunicamycin, an inhibitor of protein glycosylation that induces accumulation of misfolded proteins and activates the UPR, caused a dose-dependent decrease in proliferation, as derived from confluency assays with live-cell microscopy ( Fig. 2A). This effect of tunicamycin can be rescued by guanabenz (13,17). Likewise, 5-20 M robenidine reversed the proliferation defect of tunicamycin-treated HeLa cells in a dose-dependent manner (Fig.  2B). However, at higher concentrations of robenidine, this cytoprotective effect was lost. We selected 15 M robenidine as an optimal concentration for treatment, as it produced nearly maximal cytoprotective effects but was sufficiently separated from the higher, toxic concentrations. The cytoprotective effect of 15 M robenidine was detected at relatively narrow tunicamycin concentrations of 400 -800 ng/ml (Fig. 2C), similar to what has been reported for guanabenz (15). Guanabenz and robenidine were similarly cytoprotective at a concentration of 15 M (Fig. 2D).
To evaluate whether drug-drug interactions exist between robenidine and guanabenz, we performed an isobologram analysis (25). For this purpose, we first generated dose-response curves to accurately determine their EC 50 values (Fig. 2E). These studies indicated that, on average, guanabenz is 10 -15 times more potent than robenidine. Next, the isobologram was drawn by connecting the two axis intersection points (i.e. the EC 50 for each drug alone), according to the "Loewe additivity" principle. Robenidine and guanabenz drug combinations were tested at 10:1 and 15:1 ratios, and the corresponding EC 50 values of these combination regimens were plotted against the isobologram (Fig. 2F). The 95% confidence interval of the combination treatments overlapped with the Loewe additivity line, indicating that there are no synergistic or antagonistic interactions between robenidine and guanabenz.

Cytoprotective effects of robenidine
Pyrimethamine shares structural features with robenidine, as it also has a central guanidine scaffold and a para-chlorosubstituted phenyl ring (Fig. S1A). Moreover, pyrimethamine also displays potent anti-malaria activity (26). However, pyrimethamine was lethal at 10 M (data not shown) and did not promote survival of tunicamycin-treated HeLa cells at a sublethal dose of 3 M (Fig. S1B).

Robenidine inhibits TNF␣-induced NF-B activation
Guanabenz inhibits the TNF␣-induced activation of the transcription factor NF-B (27). However, the concentration of guanabenz (200 M) that was used in the latter study is toxic and far exceeds the concentrations that are routinely used to study its cytoprotective effects in cultured cells. Therefore, we tested whether robenidine and guanabenz affect NF-B activation at a more optimal concentration of 15 M (Fig. 2). To this end, HEK293 cells were transfected with an NF-B-driven luciferase reporter plasmid and treated with TNF␣ concentrations of 0.1-10 ng/ml in the absence (CTRL) or presence of 15 M robenidine (Fig. 3A). Robenidine significantly inhibited NF-B activation by TNF␣ concentrations between 0.1 and 0.3 ng/ml, with a maximal effect (inhibition of ϳ25%) at 0.3 ng/ml TNF␣. A similar level of inhibition was observed with 15 M guanabenz (Fig. 3B). We ruled out that the observed reduction in bioluminescence was due to loss of viable cells, as detected by 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assays (Fig. S2).

Robenidine does not affect the interaction between R15A and PP1 or eIF2␣
The cytoprotective effects of guanabenz (analogs) have been explained by an inhibitory effect on the eIF2␣ phosphatase PP1: R15A, resulting from disruption of either the PP1:R15A interaction (13) or recruitment of the substrate eIF2␣ (17). To exam- The data were analyzed with one-way ANOVA using Tukey's post hoc test. E, dose-response relationship of the cytoprotective effect of robenidine or guanabenz on HeLa cells treated with 400 ng/ml tunicamycin for 3 days. Curves were fitted using a four-parameter model with variable slope. For each concentration, the average of three technical repeats is presented. F, isobologram analysis of robenidine (R) and guanabenz (G) combination treatments at the indicated dose ratios. EC 50 values of dose-response experiments are plotted. Black open and closed dots represent the mean with 95% confidence intervals of 15:1 and 10:1 combination treatments, respectively. The Loewe additivity line (black line) represents all theoretical robenidine-guanabenz combinations resulting in a 50% cytoprotective effect, assuming that both drugs interact in an additive manner. Datapoints to the left of the curve are synergistic combinations, whereas datapoints to the right are antagonistic. All panels except F show one of at least three independent experiments. A-D show individual datapoints of at least three technical replicates, with bar graphs indicating the mean and error bars the standard deviation (n.s., not significant, p Ͼ 0.05; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001).

Cytoprotective effects of robenidine
ine whether guanabenz and robenidine interfere with the binding of PP1 or eIF2␣ to R15A, we designed NanoBiT splitluciferase sensors that report on PP1:R15A and eIF2␣:R15A interactions in live cells (Fig. 4A). We fused the 19-kDa "large bit" (LgBiT) luciferase fragment to the N terminus of PP1 and eIF2␣. The "small bit" (SmBiT) luciferase fragment of 11 residues was inserted between the eIF2␣-binding PEST repeat region and the PP1-binding KVRF sequence of R15A. This insertion is unlikely to interfere with the activity or structure of R15A, as the protein is almost completely unstructured in its native form. We envisaged that binding of LgBiT-PP1 or LgBiT-eIF2␣ to R15A-SmBiT would bring the split-luciferase fragments into close proximity, resulting in complementation and active luciferase (Fig. 4B). Indeed, transfection of HEK293T cells with LgBiT-PP1 ϩ R15A-SmBiT or LgBiT-eIF2␣ ϩ R15A-SmBiT resulted in luminescence, which was largely lost after deletion of the PEST region and mutation of the KVRF sequence to KARA (Fig. 4, C and D). Co-immunoprecipitation analysis confirmed that FLAG-tagged R15A immunoprecipitated endogenous eIF2␣ and PP1, but this interaction was lost after deletion of the PEST region in combination with a mutation of the PP1-binding site (Fig. 4E). Together, these results indicated that the adopted split-luciferase sensors allowed monitoring of the interaction of R15A with eIF2␣ and PP1 in live cells. We subsequently used these splitluciferase sensors to test whether guanabenz or robenidine affects the binding of LgBiT-eIF2␣ or LgBiT-PP1 to R15A-SmBiT in live cells. HEK293 cells transfected with the sensors were treated with 15 M guanabenz or robenidine for 24 h. However, we observed no reduction in luminescence signal (Fig. 4F). Furthermore, HEK293 cells transfected with a constitutively active split luciferase showed no effect of the compounds on cell viability or luciferase activity. Together, these results indicate that cytoprotective concentrations of guanabenz and robenidine do not measurably affect PP1: R15A or eIF2␣:R15A interactions, as measured with transient, cytomegalovirus-driven expression of split-luciferase sensors in live cells.

Robenidine inhibits the UPR and ISR
The cytoprotective effects of guanabenz under conditions of ER stress have been attributed to its inhibitory effect on the UPR and ISR (10). To investigate whether robenidine acts through the same mechanism, we used CHO cells containing reporters for both the ISR (CHOP::GFP) and the UPR branch that involves IRE1 (XBP1s::Turquoise) (15). Flow cytometry analysis disclosed clear induction of both genes following addition of 175 ng/ml tunicamycin for 55 h (Fig. 5A). Cotreatment of these cells with 15 M robenidine or guanabenz decreased the expression of CHOP and XBP1s by 40 -60% (Fig. 5B).
A previous study showed that guanabenz (analogs) retain their cytoprotective activity in cells that express a nonphosphorylatable eIF2␣ mutant (15). This finding contradicts the proposal that guanabenz's cytoprotective effects are mediated by an increased phosphorylation of eIF2␣ at Ser-51 (8,24). To examine whether inhibition of the expression of CHOP and XBP1s by robenidine and guanabenz is mediated by increased eIF2␣ phosphorylation, we used a variant CHO reporter cell line that expresses nonphosphorylatable eIF2␣ (S51A). Consistent with ISR-dependent CHOP activation, the activation of the CHOP::GFP reporter was less pronounced in the cell line expressing eIF2␣ S51A (compare Fig. 5, A and C). Nonetheless, both robenidine and guanabenz inhibited the induction of CHOP::GFP and XBP1s::Turquoise in the mutant cell line upon treatment with tunicamycin (Fig. 5D). These results demonstrate that robenidine, like guanabenz, does not exert its cytoprotective effect by increasing the phosphorylation of eIF2␣ at Ser-51.

Cytoprotective effects of robenidine
benz could not be explained by disruption of R15A:PP1 or R15A:eIF2␣ and were not mediated by changes in the phosphorylation state of eIF2␣ (Fig. 5). These results therefore demonstrate that the phenotypic effects of these drugs cannot be mediated by inhibition of the R15A-PP1-actin holoenzyme or altered phosphorylation of eIF2␣ at Ser-51. Similar conclusions were recently drawn for guanabenz and sephin1 (15,28) but are at variance with reports hinting at inhibitory effects of guanabenz (analogs) on PP1:R15 assembly and its recruitment of eIF2␣ as a substrate (10,17).
Other targets of guanabenz (analogs) that have been proposed to underly their cytoprotective function are cholesterol 25-hydroxylase (29), rRNA (30), and lipid membranes (31). It is currently uncertain whether the antibacterial, antiinflammatory, and cytoprotective properties of robenidine and guanabenz (analogs) arise from the same cellular target(s), but this seems likely, as both compounds had very similar effects at high nanomolar to low micromolar concentrations. Identification of the cellular target(s) of robenidine and guanabenz (analogs) remains an important goal for fur-

Cytoprotective effects of robenidine
ther exploration, as it will aid the development of analogs with higher potency.
We show here, for the first time, that guanabenz and robenidine significantly inhibit the activation of NF-B (Fig. 3). However, this effect was rather small (ϳ25% inhibition) and there-fore unlikely to contribute significantly to the cytoprotective properties of these compounds. At present, we cannot exclude that the inhibitory effects of robenidine and guanabenz in the NF-B and CHOP/XBP1s reporter cell lines (Figs. 3 and 5) are (partially) explained by a general inhibitory effect on protein

Cytoprotective effects of robenidine
translation, as the assays rely on the production of luciferase and fluorescent protein for signal readout.
Guanabenz, sephin1, and raphin1 have shown promising results in various murine models of neurodegenerative diseases (10,11,18). We speculate that robenidine has similar therapeutic potential, in particular because its widespread industrial use for nearly 50 years as a food additive and coccidostat documents its safety and tolerance, at least in animals. Pharmacokinetic studies in rabbits showed that robenidine is orally available and has a moderate half-life of around 12 h (32). Micromolar serum concentrations are reached with a single oral dose, which are the concentrations needed for cytoprotective effects in vitro (Fig. 2). Our isobolographic analysis suggests that there is no benefit of combining robenidine and guanabenz (analogs). Nonetheless, our studies indicate that robenidine is an attractive candidate for preclinical testing in animal models of neurodegenerative diseases.

PubChem database searching
The guanabenz entry was accessed (https://pubchem.ncbi. nlm.nih.gov/) under PubChem CID 5702063. Robenidine (CID 9570438) was identified through the "related record" search function, focusing on substituted aminoguanidines. Highthroughput assays that tested both compounds were identified manually under the "BioAssay results" section. Subsequently, the BioAssay assay identification numbers were used to query the obtained potency for each compound. MarvinSketch was used for drawing and exporting chemical structures (Marvin-Sketch, 17.21.0, ChemAxon).

Cytoprotection experiments
HeLa cells were plated at ϳ10% confluency in transparent 96-well plates. Three technical replicates were used per condition. Treatment with compounds started at the time of seeding. All compounds were dissolved in DMSO. The final concentration of DMSO was the same under all conditions of each exper-iment and never exceeded 0.5%. Proliferation was measured using an IncuCyte ZOOM system and analyzed using IncuCyte ZOOM software (Essen BioScience).

Isobologram analysis
Cytoprotection experiments were performed as described above. The EC 50 isobole was generated by performing three dose-response experiments for both robenidine and guanabenz. From these data, the average EC 50 values for both drugs were calculated and used as intercepts of the isobologram. Next, dose-response experiments were performed for the indicated robenidine-guanabenz combination ratios, and the calculated EC 50 values were plotted individually on the isobolograph. Calculation of EC 50 values was performed using the four-parameter model with variable slope with GraphPad Prism V5.

NF-B reporter assay
HEK293 cells were transfected with an NF-B-driven firefly luciferase reporter plasmid (Addgene, 14886) for 24 h. Subsequently, the cells were seeded in white opaque 96-well plates at ϳ50% confluency and treated with the indicated concentrations of TNF␣, guanabenz, and/or robenidine. A minimum of four technical replicates were used per condition. The final concentration of DMSO was identical under all experimental conditions and never exceeded 0.5%. One day after seeding and treatment of the cells, firefly activity was measured on a Luminoskan Ascent (Thermo Scientific) after replacement of the cell culture medium with homemade firefly assay buffer (0.5% Triton X-100, 50 mM Tris (pH 7.4), 5 mM MgCl 2 , 20 M pyrophosphate, 500 M ATP, and 150 g/ml D-luciferin). As firefly luciferase exhibits flash-type kinetics, the signal was measured in the stable phase, which typically occurred 3-4 min after incubation with firefly assay buffer. To account for possible variations in cell proliferation between the different conditions, a parallel plate of HEK293 cells was seeded and treated identically to the firefly assay plate. At the time of luciferase measurement, cell proliferation was measured in the parallel plate using 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assays. For this, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide dissolved in Tris buffer (pH 7.4) was added at a final concentration of 0.5 mg/ml to the assay medium and incubated overnight. Formazan crystals were dissolved in DMSO and transferred to a transparent 96-well plate, and the absorbance was measured at 550 nm using a BioTek ELx808 microplate reader.

Flow cytometry with CHO reporter cell lines
CHO cells were plated in 6-well plates at ϳ20% confluency and treated with the indicated concentration of compounds for 55 h. Three technical replicates were included per condition. The final concentration of DMSO (Ͻ0.5%) was the same under all conditions. Cell culture medium and compounds were refreshed halfway through the incubation period to prevent medium evaporation and stressing the cells. Subsequently, cells were washed, harvested in PBS, and used for flow cytometry. GFP and Turquoise expression were measured on a BD FACSCanto II HTS system using the following fil-Cytoprotective effects of robenidine ters: GFP: excitation, 488 nm; emission, 500 -560 nm; Turquoise: excitation, 405 nm; emission, 400 -500 nm. At least 10,000 cells were recorded per sample. Dead cells were excluded based on the scatter profile of a calibration sample containing 50% dead and 50% live cells. Data were processed using FlowJo software (Tree Star, Inc.).

NanoBiT split-luciferase assay
HEK293 cells were plated at ϳ20% confluency and transfected with the indicated split-luciferase sensors for 48 h. To achieve sufficient sensor activity, sensors were cloned in vectors containing the cytomegalovirus promoter and not in the weaker TK promoter-driven vectors of the NanoBiT kit. Subsequently, the cells were harvested and resuspended in PBS. The luciferase substrate coelenterazine was added to the cell suspension at a final concentration of 25 M, and cells were dispensed at ϳ10,000 cells/well in white 96-well plates. Luciferase activity was measured on a Luminoskan Ascent (Thermo Scientific). When the cells were treated with compounds, they were reseeded in opaque 96-well plates after the 48-h transfection period. A minimum of three technical replicates were used per condition. To correct for effects of the compounds on cell proliferation or expression of the split-luciferase sensors, a third batch of cells, expressing a constitutively active NanoBiT luciferase (see "Materials"), was seeded and treated in parallel.

Biochemical techniques
HEK293 cells were lysed with lysis buffer (50 mM Tris-HCl (pH 7.4), 0.01% saponin, 150 mM NaCl, and 10% glycerol) supplemented with protease inhibitors (0.5 mM phenylmethanesulfonyl fluoride, 0.5 mM benzamidine, and 5 M leupeptin). Subsequently, lysates were sonicated with three 10-s cycles at high intensity and cleared by centrifugation at 10,000 ϫ g. For FLAG traps, cleared cell lysates were incubated with 20 l of FLAG affinity beads (Sigma, catalog no. P4333) for 2-3 h at 4°C. The beads were then washed once with PBS and subjected to immunoblotting. SDS-PAGE gel electrophoresis was performed with 10% or 4%-12% BisTris (NuPAGE, Invitrogen). Immunoblots were visualized using ECL reagent (PerkinElmer Life Sciences) in an ImageQuant LAS4000 imaging system (GE Healthcare). The signals were quantified using ImageJ (National Institutes of Health).