RESEARCH ARTICLE

Drug screens of NGLY1 deficiency in worm and fly models revealcatecholamine, NRF2 and anti-inflammatory-pathway activationas potential clinical approachesSangeetha Iyer, Joshua D. Mast, Hillary Tsang, Tamy P. Rodriguez, Nina DiPrimio, Madeleine Prangley,Feba S. Sam, Zachary Parton and Ethan O. Perlstein*

ABSTRACTN-glycanase 1 (NGLY1) deficiency is an ultra-rare and complexmonogenic glycosylation disorder that affects fewer than 40 patientsglobally. NGLY1 deficiency has been studied in model organismssuch as yeast, worms, flies and mice. Proteasomal and mitochondrialhomeostasis gene networks are controlled by the evolutionarilyconserved transcriptional regulator NRF1, whose activity requiresdeglycosylation by NGLY1. Hypersensitivity to the proteasomeinhibitor bortezomib is a common phenotype observed in whole-animal and cellular models of NGLY1 deficiency. Here, we describeunbiased phenotypic drug screens to identify FDA-approved drugsthat are generally recognized as safe natural products, and novelchemical entities, that rescue growth and development of NGLY1-deficient worm and fly larvae treated with a toxic dose of bortezomib.We used image-based larval size and number assays for use inscreens of a 2560-member drug-repurposing library and a 20,240-member lead-discovery library. A total of 91 validated hit compoundsfrom primary invertebrate screens were tested in a human cell line inan NRF2 activity assay. NRF2 is a transcriptional regulator thatregulates cellular redox homeostasis, and it can compensate for lossof NRF1. Plant-based polyphenols make up the largest class of hitcompounds and NRF2 inducers. Catecholamines and catecholaminereceptor activators make up the second largest class of hits. Steroidaland non-steroidal anti-inflammatory drugs make up the third largestclass. Only one compound was active in all assays and species: theatypical antipsychotic and dopamine receptor agonist aripiprazole.Worm and fly models of NGLY1 deficiency validate therapeuticrationales for activation of NRF2 and anti-inflammatory pathwaysbased on results in mice and human cell models, and suggest a noveltherapeutic rationale for boosting catecholamine levels and/orsignaling in the brain.

KEY WORDS: N-glycanase 1 deficiency, NGLY1, SKN-1, Pngl,Congenital disorder of deglycosylation, Diseasemodel, Aripiprazole

INTRODUCTIONNGLY1 deficiency was the first congenital disorder of deglycosylation(CDDG) described in the biomedical literature (Lam et al., 2017; Ennset al., 2014). NGLY1 deficiency has multi-organ presentation andclinical features in patients, such as global developmental delay, acomplex hyperkinetic movement disorder, small body size, seizuresand alacrimia. NGLY1 is an ancient gene encoding a cytosolic enzymecalled N-glycanase 1 – also referred to as PNGase – which catalyzesthe hydrolysis and release of N-glycans from N-glycosylated proteins(Suzuki et al., 2016). NGLY1 is thought to function primarily in anevolutionarily conserved protein-surveillance and -disposal pathwaycalled ERAD, or endoplasmic-reticulum-associated degradation(Suzuki et al., 2016). NGLY1 also regulates the activity of specificglycoproteins by using deglycosylation as a post-translational on/offswitch. Cellular models of NGLY1 deficiency have shown that thetranscriptional regulator NRF1 is a specific deglycosylation target ofNGLY1, and that knocking out NGLY1 phenocopies knocking outNRF1 (Tomlin et al., 2017). Only deglycosylated NRF1 can beproteolytically processed into the mature nuclear-active form. Once inthe nucleus, NRF1 controls the expression of proteasomal subunitgenes in response to protein-folding stress in mammalian cells(Radhakrishnan et al., 2014), worms (Lehrbach and Ruvkun, 2016)and flies (Grimberg et al., 2011). In flies, NGLY1 regulates theglycosylation status of the ortholog of a bone morphogenetic protein(BMP) signaling ligand (Galeone et al., 2017). Demonstrating thecomplexity of how loss-of-function mutations in the gene lead topathophysiology in humans, NGLY1 regulates mitochondrialphysiology in human and mouse fibroblasts and in worms throughmechanisms that are still under investigation (Kong et al., 2018).Interestingly, mitophagy defects caused by loss of NRF1 function canbe rescued by activation of the related transcriptional regulator NRF2,which controls the expression of genes involved in antioxidant andredox-stress responses (Yang et al., 2018).

In the 5 years since the publication of the first NGLY1 deficiencydiagnostic cohort of eight patients (Enns et al., 2014), multipleresearch groups have contributed to our understanding of disease-causing and loss-of-function mutations in the NGLY1 gene and itsorthologs by generating and characterizing small and large animalmodels as well as patient-derived cell models. From thismarketplace of disease models, a common phenotype emerged:hypersensitivity to proteasome inhibition by bortezomib (Fenteanyet al., 1995). In worms, hypersensitivity to bortezomib toxicity wasobserved in an otherwise normally developing PNG-1/NGLY1-nullmutant, which has a half-maximal growth inhibitory concentration(IC50) several hundred times lower than wild-type worms (Lehrbachand Ruvkun, 2016). Using a chemically related proteasomeinhibitor lacking the reactive boronic acid group, it was shownthat mouse embryonic fibroblasts derived from NGLY1-knockoutReceived 17 May 2019; Accepted 9 September 2019

Perlara PBC, 2625 Alcatraz Ave, #435, Berkeley, CA 94705, USA.

*Author for correspondence (ethan@perlara.com)

S.I., 0000-0002-0575-1976; J.D.M., 0000-0002-2246-0580; H.T., 0000-0001-6461-3402; T.P.R., 0000-0003-3764-2587; N.D., 0000-0003-3461-6826; M.P.,0000-0001-6978-7516; F.S.S., 0000-0002-2029-1496; Z.P., 0000-0003-3261-7841;E.O.P., 0000-0002-4734-4391

This is an Open Access article distributed under the terms of the Creative Commons AttributionLicense (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use,distribution and reproduction in any medium provided that the original work is properly attributed.

1

© 2019. Published by The Company of Biologists Ltd | Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

mice and NGLY1-knockdown human cell lines are several-foldmore sensitive to carfilzomib toxicity compared to controls (Tomlinet al., 2017).In an effort to phenotype and screen a homozygous loss-of-

function Pngl/NGLY1 fly modeling the patient-derived C-terminalpremature stop codon allele R401X, we showed that fly Pngl−/−

homozygous mutant larvae are 25-fold more sensitive thanheterozygotes to the toxic effects of bortezomib (Rodriguez et al.,2018). Another group reported that a Pngl RNAi-knockdown flymodel of NGLY1 deficiency has constitutively reduced expressionof NRF1-dependent proteasomal subunit genes, consistent withfindings of hypersensitivity to bortezomib toxicity in the othermodels (Owings et al., 2018). Loss of NGLY1 causes intolerance tobortezomib that is as evolutionarily conserved as the underlyingNRF1-dependent proteasome bounce-back response because theygo hand in hand. The prediction that has been confirmed so far inmammalian cells and in nematodes (Lehrbach et al., 2019) is thatNGLY1 and its orthologs deglycosylate NRF1 and its orthologs.Wereason that small-molecule suppressors of bortezomib will safelyactivate bypass pathways that rescue or compensate for loss ofNGLY1 in a whole animal and will have a higher probability ofexhibiting a favorable therapeutic index in mammals.We used our Pngl−/− fly model in a drug-repurposing screen to

identify compounds that rescue larval developmental delay(Rodriguez et al., 2018). Homozygote flies fail to thrive in theabsence of any exogenous stressor or enhancer such as bortezomib.There are two limitations of our previous screen. First, flieshomozygous for a patient-derived nonsense allele are extremely sickand only one fly-specific validated hit was identified – the insectmolting hormone 20-hydroxyecdysone. Second, flies homozygousfor a patient-derived nonsense allele are further sickened by theorganic solvent dimethyl sulfoxide (DMSO) in which all testcompounds are solubilized, and all stock solutions prepared.Therefore, drug screens were conducted at the limit of detectionand a high false-negative rate was due to under-dosing of testcompounds.Here, we addressed the shortcomings of the initial fly-only drug-

repurposing campaign. First, we discovered that the fly Pngl+/−

heterozygote is two-fold more sensitive to bortezomib toxicity thanwild-type animals, it tolerates DMSO, and it has no detectabledevelopmental delay. We hypothesize that it will be easier for asmall molecule to suppress larval developmental delay inbortezomib-treated healthy heterozygotes versus constitutivelysick homozygotes because otherwise healthy heterozygotestolerate higher levels DMSO and thus allow for all testcompounds to be screened at a five-fold higher concentration.Second, we extended the bortezomib intolerance results ofLehrbach and Ruvkun (Lehrbach and Ruvkun, 2016) to a liquid-based, 384-well-plate quantitative worm larval growth anddevelopment assay for drug screening. We hypothesize that atwo-species bortezomib-suppressor screen conducted in parallelwould reveal hit compounds that target evolutionarily conserveddisease modifiers and disease-modifying pathways. Third, we notonly cross-tested all hit compounds from the primary worm screenin the fly assay, and vice versa, but we also cross-tested all hitcompounds in two different worm assay paradigms (bortezomibtreatment versus carfilzomib treatment), in two different fly assayparadigms (bortezomib treatment of heterozygotes versushomozygotes) and in a human cell NRF2 transcriptional activityreporter assay. This cross-validation scheme generated a short list ofgeneric drugs and nutriceuticals in which we had the highest degreeof confidence of reproducibility and clinical trial actionability.

Finally, we screened not only the same repurposing library asdescribed in Rodriguez et al. (2018) but also a ten-fold larger lead-discovery library. We hypothesize that, in some instances, the sameknown mechanisms of action of repurposable drugs will be targetedby novel chemical entities, which serve as starting points for leadoptimization toward a best-in-class clinical candidate.

RESULTSThere are currently no FDA-approved treatments for NGLY1deficiency despite the unmet medical need. Drug repurposinginvolves finding new uses for old drugs and is the shortest path to atherapy for ultra-rare disease communities with limited financialresources and few dedicated researchers (Pushpakom et al., 2019).We used a model-organism-based disease-modeling andphenotypic drug-screening approach, which is enabling precisionmedicine to bridge bench to bedside (Li et al., 2019).

Developing high-throughput bortezomib-modifier assays fornematode and fly larvaeThe nematode ortholog of NGLY1 is PNG-1. The strain used here isok1654, which contains an 800-base-pair deletion in the PNG-1open reading frame, resulting in a null mutant. We confirmed thatok1654 does not have an intrinsic growth defect but is markedlyhypersensitive to bortezomib toxicity (Fig. 1A). Bortezomibexacerbates the proteasomal stress that NGLY1-deficient wormsare already experiencing because of the concomitant loss of NRF1activity, and results in an exaggerated disease phenotype, in thiscase toxicity in the form of early larval growth arrest (Lehrbach andRuvkun, 2016). Because png-1 homozygous mutant worms do nothave a constitutive growth or developmental defect, we decided touse bortezomib to induce larval arrest and screen for compoundsthat restore normal growth and development as measured by the sizeand number of worms in each well of a 384-well plate. From thebortezomib dose-response data, we established that a worm png-1-null mutant is sensitive to bortezomib toxicity down to the lownanomolar range. After further assay optimization studies in whichwe tested additional doses between 100 nM and 1 µM, we selected205 nM bortezomib as the concentration for the primary screens andfor secondary retest and cross-test experiments to validate hitcompounds because, at this dose, the mean size of png-1 mutantworms is reduced by approximately 85% compared to untreatedcontrol worms (Fig. S1).

The ortholog of NGLY1 in flies is Pngl. We previously developeda Pngl−/− fly model based on a recurrent C-terminal nonsense allelefound in NGLY1 patients, optimized a high-throughput larval sizeassay, and screened a 2560-member drug-repurposing library on flyPngl−/− larvae (Rodriguez et al., 2018). That effort culminated in asingle validated hit compound, 20-hydroxyecdysone (20E). 20E isan insect-specific sterol-derived molting hormone. In order toidentify clinically actionable hits, we increased the overall hit rateand robustness of fly screens. We generated bortezomib toxicitydose-response data for fly Pngl+/− heterozygote larvae anddetermined that these animals are two-fold more sensitive thanwild-type animals (Fig. 1B). After further assay optimization studieswith doses between 5 µM and 10 µM, we selected 9 µM bortezomibas the concentration for the primary screens and for secondaryretests and cross-tests to validate hit compounds.

Worm repurposing screen and hit validationIn the presence of 205 nM bortezomib, png-1-null mutant wormsarrest as L1 larvae while wild-type animals treated with the samedose of bortezomib develop normally into progeny-bearing adults.

2

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

A worm repurposing hit is defined as a compound that rescuedbortezomib-treated png-1-null worms such that their size andnumber were indistinguishable from control wells containingvehicle-treated png-1-null worms. All test compounds werescreened at a final concentration of 25 μM. Images of arepresentative positive control well and a representative negativecontrol well, and two examples of wells containing suppressors andenhancers/toxic compounds, are shown in Fig. 2A. The Venndiagram in Fig. 2B summarizes the overlap of screening positivesbetween three independent replicates. A total of 63 suppressors haveZ-scores greater than two in all three replicates. The hit rate forworm suppressors is 63/2560 or 2.5%. A total of 51 enhancers haveZ-scores less than two in all three replicates. The hit rate forenhancers/toxic compounds is 51/2560 or 2%. Enhancers/toxiccompounds were not further investigated in this study. In total,60/63 worm suppressors were available for reorder as fresh powderstocks, retested in the primary bortezomib assay paradigm, and thenscored in a secondary non-bortezomib assay paradigm.Some compounds can directly inactivate bortezomib, leading to a

false-positive result. For example, polyphenols and, in particular,

polyphenols containing multiple catechols are known to formcovalent adducts with bortezomib by boronate-catecholcomplexation (Glynn et al., 2015; Golden et al., 2009). In orderto filter out those compounds, we tested the 60 worm suppressors onthe worm png-1-null mutant treated with the chemically relatedproteasome inhibitor carfilzomib, which lacks the reactive boronicacid that renders bortezomib vulnerable to covalent attack. Weestablished a carfilzomib dose-response curve for png-1-null mutantworms (Fig. S1). We directly compared these data to worms treatedwith 205 nM bortezomib in order to determine at whichconcentration we observe comparable larval growth arrest. Weretested all 60 worm suppressors in the presence of 27.3 µMcarfilzomib, meaning that carfilzomib is significantly less potentthan bortezomib.

A total of 48/60 (80%) worm suppressors retested in thebortezomib assay paradigm. In total, 15/60 (25%) wormsuppressors scored positively in the carfilzomib assay paradigm.Overall, 15/60 (25%) worm suppressors rescued in both worm assayparadigms, i.e. in the presence of either bortezomib or carfilzomib.Those 15 compounds are the worm repurposing hits (Table 1):

Fig. 1. Determining a half-maximal effective concentration (EC50) for bortezomib in NGLY1-deficient worms and flies. (A) Wild-type N2 worms (top row)and png-1 homozygous mutant worms (bottom row) were grown in liquid media in the presence of increasing concentrations of bortezomib (left to right).(B) Wild-type and Pngl+/− heterozygous mutant fly larvae were grown on solid media in the presence of either 5 µM or 10 µM bortezomib. Fly larvae were imagedand their sizes were plotted in arbitrary units (AU).

3

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

aripiprazole, benserazide, ellagic acid, epicatechin monogallate,epigallocatechin-3-monogallate, ethylnorepinephrine, gossypetin,koparin, phenylbutazone, pomiferin, purpurogallin-4-carboxylicacid, quercetin, theaflavin monogallate, triamcinolone and 3,4-didesmethyl-5-deshydroxy-3′-ethoxyscleroin.A total of 10/15 (66%) worm repurposing hits are plant-based

polyphenols. Epicatechin monogallate and epigallocatechin-3-monogallate are structural analogs, the latter containing an

additional hydroxyl group. They are both abundant in tea leaves.The two plant flavonols quercetin and gossypetin are structuralanalogs, the latter containing an additional hydroxyl group. Quercetinis found in many foods and gossypetin is found in Hibiscus flowers.Aripiprazole, benserazide and ethylnorepinephrine act oncatecholamine levels and/or signaling in the brain. Phenylbutazoneis a decades-old first-generation non-steroidal anti-inflammatorydrug, or NSAID, but is not currently approved and marketed in the

Fig. 2. 2560-compound drug-repurposing screens of png-1 homozygous mutant worms and Pngl heterozygous mutant flies. (A) 15 L1 png-1 mutantlarvaewere sorted into each well, and plates were incubated for 5 days at 20°Cwhile shaking. Worm screen images of a representative positive control well (A01),a representative negative control well (C23), two presumptive suppressors (K12, K13), and two presumptive enhancers/toxic compounds (C18, N14). Wormswere pseudo-colored blue during image processing and analysis. (B) Venn diagram of overlapping hits from three replicate screens. (C) Three fly first-instar Pnglheterozygous mutant larvae were sorted into each well, and plates were incubated for 3 days at room temperature. Z-score plot of three replicates of flyrepurposing screen positive control (red circles) versus negative control (cyan circles) wells. (D) Z-score plot of three replicates of fly repurposing screen testcompounds (red circles) versus negative (cyan circles) control wells.

4

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

United States. Triamcinolone is a synthetic corticosteroid and generictopical anti-inflammatory drug. These worm data suggest theexistence of at least three mechanistic classes of suppressors: (1)polyphenolic antioxidants; (2) catecholamine pathway modulators;and (3) anti-inflammatories.

Fly repurposing screen and hit validationIn the presence of 9 µMbortezomib,Pngl+/− flies are developmentallydelayed as first-instar larvae. A fly repurposing hit is defined as anycompound that in triplicate rescued bortezomib-treated fly Pngl+/−

larval growth and size such that they were indistinguishable fromcontrol wells containing DMSO-treated heterozygotes. All testcompounds were screened at a final concentration of 32 μM.Separation between positive and negative control wells is shown inFig. 2C. As expected, most test-compound wells do not affectbortezomib-treated heterozygotes as determined by quantifying larvalarea per well and remain close to the negative-control Z-score(Fig. 2D). We identified 31 fly suppressors that rescued larval sizewith Z-score greater than 2.5 in at least two out of three replicates,yielding a hit rate of 1.2%, or half the hit rate of the worm repurposingscreen (Fig. S2) but 30 times the hit rate of the previously published flyPngl−/− drug repurposing screen (Rodriguez et al., 2018).In order to eliminate suppressors that directly inactivate

bortezomib or that fail to rescue in the absence of bortezomib-induced proteasomal stress, we retested suppressors in a 12-well Petridish assay at 0, 10 µM, 50 µM or 100 µM of compound on either flyPngl+/− larvae in the presence of bortezomib or fly Pngl−/− larvaewithout bortezomib. We could not retest suppressors in a carfilzomibassay paradigm because fly Pngl+/− larvae were resistant tocarfilzomib toxicity at the highest concentration tested. Note thatwe might not observe rescue at 100 µM in the DMSO-hypersensitivehomozygous animals because of the increased amount of DMSO(0.2%) needed to test that concentration of compound. Although wedid not confirm by chemical analysis, any compound that strongly

rescued only in the bortezomib assay paradigm is likely a bortezomibinactivator. The top candidate bortezomib inactivators are tannic acidand gossypetin (Fig. S2). Gossypetin contains two catechols permolecule and tannic acid contains five catechols per molecule. Thesetwo compounds are the only overlapping primary screening hitsbetween the worm and fly repurposing screens.

A total of 30/31 fly suppressors were reordered as fresh powderstocks (one compound failed to dissolve at 10 mM as stocksolution). In total, 21/30 (70%) fly suppressors retested in theprimary heterozygote assay paradigm, which is comparable to theretest rate of worm suppressors. A total of 20/30 (66%) flysuppressors scored positively in the secondary homozygote assayparadigm; 13/30 (43%) suppressors validated in both homozygoteand heterozygote assay paradigms, which is higher than the wormcross-validation rate. Those 13 compounds are the fly repurposinghits (Table 1): aminacrine, aripiprazole, β-amyrin, butopyronoxyl,clindamycin, dyphylline, L-glutamine, ketorolac, lumefantrine,promazine, pyrimethamine, quinizarin and thioguanosine.

The fly repurposing hits span a range of mechanisms, includingan overlap with the worm repurposing hits that fall into the threeaforementioned pharmacological classes. Only one of the flyrepurposing hits appears to be fly specific, i.e. the insect repellentbutopyronoxyl. Ketorolac is a first-generation NSAID anti-inflammatory drug but is structurally distinct from the wormrepurposing hit and NSAID phenylbutazone. β-amyrin is atriterpene natural product with anti-inflammatory effects. Asstated above, aripiprazole is a catecholamine receptor modulator.The remaining fly repurposing hits appear to be mechanisticsingletons and do not have obvious functional or structural analogswith worm repurposing hits. There are two antimalarial compounds,pyrimethamine and lumefantrine, and two DNA-damaging cancerdrugs, aminacrine and thioguanosine. Dyphylline is a xanthinederivative with bronchodilator and vasodilator effects. Clindamycinis a lincosamide antibiotic used for a range of bacterial infections.

Table 1. Summary description of worm and fly repurposing hits

Compound name Pharmacological class Screening model FDA approved?

Lumefantrine Antimalarial Fly YesThioguanosine Antimetabolite antineoplastic Fly YesClindamycin Bacterial protein synthesis inhibitor (antibiotic) Fly YesDyphylline Bronchodilator Fly YesPyrimethamine Dihydrofolate reductase inhibitor (antimalarial) Fly YesKetorolac Nonsteroidal anti-inflammatory Fly YesPhenylbutazone Nonsteroidal anti-inflammatory Fly YesPromazine Phenothiazine antipsychotic Fly YesL-glutamine Amino acid supplement Fly NoQuinizarin Dye Fly NoButopyronoxyl Insect repellant Fly Noβ-amyrin Steroidal anti-inflammatory Fly NoAminacrine Topical antiseptic Fly NoAripiprazole Catecholamine receptor modulator Fly+worm YesGossypetin Polyphenol Fly+worm NoEthylnorepinephrine Adrenergic receptor agonist Worm YesBenserazide Aromatic L-amino acid decarboxylase inhibitor Worm YesTriamcinolone Synthetic corticosteroid anti-inflammatory Worm Yes3,4-didesmethyl-5-deshydroxy-3′-ethoxyscleroin Polyphenol Worm NoEllagic acid Polyphenol Worm NoEpicatechin monogallate Polyphenol Worm NoEpigallocatechin-3-monogallate Polyphenol Worm NoKoparin Polyphenol Worm NoPomiferin Polyphenol Worm NoPurpurogallin-4-carboxylic acid Polyphenol Worm NoQuercetin Polyphenol Worm NoTheaflavin monogallate Polyphenol Worm No

5

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

Cross-validation of worm and fly repurposing hitsIn order to quantify the overlap between worm and fly repurposinghits, we selected 11 worm repurposing hits for cross-testing in bothfly assay paradigms, and 19 fly repurposing hits for cross-testing inboth worm assay paradigms. The fraction of worm repurposing hitsthat rescue in flies is higher than the fraction of fly repurposinghits that rescue in worms. All [11/11 (100%)] worm repurposing hitsscored positively in the fly heterozygote assay paradigm. Six (55%)of these compounds cross-validated in both fly assay paradigms.There may be several reasons why a compound scored positively ina secondary retest in one organism but was not a hit in the primaryscreen in the same organism. This is especially true in flies, becausefly primary screens have more variance than primary screens inworms due to fewer animals per well, greater complexity of foodsource and increased developmental stochasticity.However, 3/19 (16%) fly repurposing hits scored positively in the

worm bortezomib paradigm; 3/19 (16%) fly repurposing hits scoredpositively in the worm carfilzomib paradigm. Only 1/19 (5%) flyrepurposing hits cross-validated in both worm assay paradigms:

aripiprazole. Of the 30 repurposing hits that were cross-tested, sevencompounds (23%) are active in both worm assay paradigms and inboth fly assay paradigms: aripiprazole, benserazide, phenylbutazone,pomiferin, quercetin, theaflavin monogallate and 3,4-didesmethyl-5-deshydroxy-3’-ethoxyscleroin. Their chemical structures are shownin Fig. 3.

Aripiprazole is an atypical antipsychotic drug with broadpolypharmacology at catecholamine receptors in the brain, includingagonist effects on dopamine receptors. It is approved for use alone orin combination in adults and children for the treatment of symptomsof many CNS diseases, including schizophrenia, autism spectrumdisorder and bipolar disorder (Shapiro et al., 2003; Kikuchi et al.,1995). Aripiprazole has been available in generic form since 2015. Asshown in Table 1, aripriprazole is the only FDA-approved drugindependently discovered in both worm and fly drug repurposingscreens, and cross-validated in all four secondary retesting paradigms.Aripiprazole fully rescued fly Pngl−/− homozygous mutant larvae at10 µM with dose-limiting toxicity at 50 µM at 100 µM (Fig. 4A). Atthe same time, 25 µM aripiprazole fully rescued growth and

Fig. 3. Chemical structures of cross-validated drug repurposing hits.(A) Theaflavin monogallate; (B) pomiferin;(C) quercetin; (D) 3,4-didesmethyl-5-deshydroxy-3′-ethoxyscleroin;(E) aripiprazole; (F) phenylbutazone;(G) benserazide.

6

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

development of worm png-1 homozygous mutant larvae treated witheither bortezomib or carfilzomib (Fig. 4B).Benserazide is an aromatic L-amino-acid decarboxylase inhibitor

that has been co-administered with levodopa (L-Dopa) for decadesto boost dopamine levels in the brain in the treatment of Parkinsondisease. As mentioned above, phenylbutazone is a decades-oldNSAID and quercetin is a plant flavonol with complexpharmacology, including anti-inflammatory effects. Theaflavinmonogallate is a polyphenol found in black tea leaves. Pomiferinis a prenylated isoflavone found in osage orange trees.

Drug discovery screens and cross-validation of novelchemotypesUsing the same screening conditions and hit-calling analysis as theworm repurposing screen, we scaled up efforts with worm png-1-null mutant larvae and a 20,240-member lead-discovery library. Weidentified 28 novel worm suppressors and six novel wormenhancers/toxic compounds with a combined hit rate of 0.143%(Fig. 5A; Fig. S3). Enhancers/toxic compounds were not furtherinvestigated in this study. In total, 12/28 (43%) novel wormsuppressors retested in the bortezomib assay paradigm; 10/28 (36%)novel worm suppressors scored positively in the carfilzomib assayparadigm. A total of five novel worm suppressors (18%) validatedin both paradigms, which is comparable to the cross-validation rateobserved for worm repurposing hits. Consistent with the results ofthe repurposing screens, the fraction of novel worm suppressors thatare active in flies is higher than the fraction of novel fly suppressorsthat are active in worms. A total of 4/5 (80%) of novel wormsuppressors cross-validated in both fly secondary assays (Table 2).These hits represent novel chemotypes and are starting points formechanism-of-action studies in human cellular models of NGLY1deficiency.Using the same screening conditions and hit-calling analysis for

the fly repurposing screen, we scaled up efforts with fly Pngl+/−

larvae and a 20,240-member lead-discovery library. We identified16 novel fly suppressors, resulting in a hit rate of 0.07% (Fig. 5B;Fig. S4). In total, 15/16 novel fly suppressors were reordered asfresh powder stocks; 10/15 (67%) novel fly suppressors retested inthe primary heterozygote assay paradigm; and 13/15 (87%) novelfly suppressors scored positively in the secondary homozygoteassay paradigm. A total of ten novel fly suppressors (67%) validatedin both paradigms. Of the 13 novel fly suppressors, 12 were retestedin both worm assay paradigms. Surprisingly, none of the novel flysuppressors scored positively in the bortezomib assay paradigm, but8/12 (66%) compounds scored positively in the worm carfilzomibassay paradigm.

That result is similar to what we observed in the drug-repurposingcross-validation experiments, namely the likelihood of a worm novelhit cross-validating in flies is higher than the likelihood of a fly novelhit cross-validating in worms. A combination of evolutionarydivergence and experimental assay differences between worms andflies likely explain the failure of a given compound to cross-validatein both species. For example, the knownNRF2 activator sulforaphaneat 50 µM did not rescue in either worm assay, potentially due tocompound solubility differences in worm versus fly media, but it didscore positively in the fly Pngl−/− homozygote assay paradigm,rescuing this mutant to 100% of the untreated homozygote positivecontrol (Fig. S5).

In the most striking example of overlap between the repurposinglibrary and discovery library, we identified two novel fly suppressorsthat are structural analogs of pyrimethamine, sharing adiaminopteridine group present in folic acid analog inhibitors ofDNA synthesis enzymes, such as methotrexate. Both pyrimethamineand one of the novel fly suppressors rescue homozygotes to near100% of the heterozygote control. A second novel fly suppressor alsorescues homozygotes to near 100% of the heterozygote control. Thechemical structures of those and all drug discovery hits aresummarized in the supplementary material.

Fig. 4. Aripiprazole worm and fly cross-validation data.(A) Fly Pngl−/− homozygote larvae treated with 10 µM, 50 µMand 100 µM aripiprazole compared to untreated controls.(B) Worm png-1 homozygote larvae treated with 25 µMaripiprazole in the presence of 205 nM bortezomib (Bzb) or27.5 µM carfilzomib (Czb).

7

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

Keap1-NRF2 activation assay in human cellsAlthough cross-validation of hits in worms and flies indicates that thetarget and mechanism of action of the test compound are conservedbetween these two species, it does not guarantee that the said targetand said mechanism of action are also conserved in humans. In anattempt to winnow down the list of hits to at least one FDA-approvedclinical candidate forNGLY1 deficiency, a total of 91 repurposing hitsand novel suppressors from both worm and fly screens were cross-tested in a Keap1-NRF2 activation assay in the U2OS osteosarcomacell line. A total of 7/91 (8%) compounds are active with a half-maximal effective concentration (EC50) between 2-30 µM (Table 3),or 10- to 100-fold less potent than the positive-control NRF2activators, e.g. sulforophane. The remaining 84 compounds haveEC50 values above 50 µM and were not considered further for lack ofpotency. Listed here in order of percentage of maximal responsenormalized to the positive control CDDO methyl ester: aripiprazole(78%)=pyrogallin (78%)>fisetin (59%)>purpurogallin-4-carboxylicacid (53%)=3-methoxycatechol (50%)>rhamnetin (43%)>pyrimethamine(27%). Of these seven, only fisetin has been previously shown toactivate NRF2 target gene expression (Smirnova et al., 2011). Toincrease confidence in the assay, we also tested three additionalknown NRF2 inducers: sulforaphane, omaveloxolone and dimethylfumarate. None of the novel chemotypes in Table 2 are NRF2inducers.

DISCUSSIONIn summary, we demonstrated the therapeutic relevance of unbiasedphenotypic drug screens of worm and fly models of NGLY1deficiency. We addressed the major shortcoming of our previousdrug screening effort by increasing the hit rate and reproducibilitywith the more permissive fly Pngl+/− heterozygote assay as theprimary screen followed by a secondary retest in the more restrictivefly Pngl−/− homozygote assay paradigm, which selected for thesubset of hit compounds that specifically rescues loss of NGLY1.We addressed the other significant shortcoming of our previousdrug screening effort by including worms as a second species in theprimary screening stage, which increased the total number andmechanistic diversity of hits. Additionally, by including humancells as a third species in the hit validation stage, we selected for hitcompounds that act on conserved targets and pathways. Weidentified three clinic-ready therapeutic approaches by focusingon the subset of hit compounds that rescue larval growth anddevelopment in both worms and flies and in the presence andabsence of bortezomib: (1) NRF2 activators/inducers; (2)catecholamine-boosting drugs; and (3) anti-inflammatory drugs.Only one compound was found to be active in all three species(worm, fly and human cells) and in all assay paradigms (plusbortezomib, plus carfilzomib, and NRF2 reporter): the atypicalantipsychotic aripiprazole.

Fig. 5. 20,240-compound novel lead-discovery screens of png-1 homozygousmutant worms andPngl+/− heterozygousmutant flies. (A)Worm screen Z-score plot of 20,240 test compounds in triplicate. Replicate 1 is shown as red circles. Replicate 2 is shown as red triangles. Replicate 3 is shown as red squares.Themean negative control Z-score is indicated by the red line. Themean positive control Z-score is indicated by the blue line. (B) Fly screen Z-score plot of 20,240test compounds in triplicate. Replicate 1 is the left panel. Replicate 2 is the center panel. Replicate 3 is the right panel.

8

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

Before discussing why we observed those three pharmacologicalclasses and the evidence that supports their testing in clinical trials, weconsider reasons why any given hit compound might be active in oneor a combination of the species and assay paradigms tested herein.Some reasons have to do with compound stability, compoundsolubility, bioavailability, drug metabolism and pharmacokinetics,not to mention the differences between worm screens versus flyscreens. The following is by no means an exhaustive list of salientfactors. Worm screens were all performed in simple liquid media thatis a buffered salt solution plus essential nutrients and containsbacteria as a food source. By contrast, the fly screens were performedin a nutrient-rich solid media (‘fly food’) with a molasses base. Theworm screen is a 6-day assay that spans all stages of the worm lifecycle, while the fly screen is a 3-day assay that only spans the larvalstages of fly development. The worm screen was performed with205 nM bortezomib while the fly screen was performed with 9 µMbortezomib. Primary drug screens with flies involve as few as threeanimals per test well, while primary drug screens with worms involvea minimum of 15 animals per test well. Secondary retest assaysinvolvedmulti-well plates with larger well volumes andmore animalsper well than primary screening conditions, which could explaininstances of false negatives. For example, the four novel chemotypesdescribed in Table 2 rescued fly Pngl+/− heterozygous larvae in thesecondary assay format but not in the primary drug screen.

Other reasons why a hit compound is active in one species but notthe other may have to do with pharmacodynamics; that is, the drugtargets are not evolutionarily conserved or the drug-binding sites ofshared drug targets are too divergent between worm and fly orthologs.Resolving which pharmacokinetic and pharmacodynamic variablesand their relative contributions explain the failure of a hit compound inany given assay or species paradigm is a very important researchquestion that is beyond the scope of the present study. However, whatwe can conclude from this study is that NRF2, anti-inflammatory andcatecholamine pathways appear to be evolutionarily conservedbetween worms, flies and human cells.

Indeed, NRF2 pathway activators and anti-inflammatory drugshave been proposed as clinical approaches to address the underlyingdefects in mitochondrial physiology observed in NGLY1-deficientcells, specifically the defect inmitophagy and excessivemitochondrialfragmentation (Yang et al., 2018). In addition to its more-well-studiedrole as a transcriptional regulator of the proteasome bounce-backresponse, NRF1 also controls the expression of mitochondrialhomeostasis and mitophagy genes not only in human cells but alsoin worms (Paek et al., 2012) and flies (Tsakiri et al., 2013). A unifyingtheory of NGLY1 deficiency is that disease phenotypes are primary orsecondary consequences of the loss of conserved NRF1-dependentTa

ble2.

Worm

andfly

cros

s-va

lidated

nove

lche

motyp

es

Struc

ture

SMILES

Sup

plier

Sup

plierID

Worm+

bortez

omib

Worm+

carfilzom

ibFlyho

moz

ygote

Flyhe

t.+bo

rtez

omib

CC1=

CC=C(O

C2=

CC=C(N

C(=O)N

C3=

CC=CC(C

l)=C3)C=C2)N=N1

Che

mBrid

ge93

1399

2Yes

Yes

Yes

Yes

FC1=

CC=C(C

=C1)C1=

NOC(=C1)C(=O)N

1CCCC1C

1=CC=CS1

Che

mBrid

ge92

8312

0Yes

Yes

Yes

Yes

CC(N

C1=

C2C

=CC=CC2=

NC(=N1)N

(C)C

)C1=

CC=CC=C1

Che

mBrid

ge92

6878

7Yes

Yes

Yes

Yes

O=C(N

1CCC2=

CC=CC=C2C

1)C1=

CC=C(C

N2C

CCCCC2)C=C1

Che

mBrid

ge92

2357

3Yes

Yes

Yes

Yes

SMILES,s

implified

molec

ular

inpu

tlineen

trysystem

;het.,he

terozygo

te.

Table 3. Keap1-NRF2 transcriptional activity reporter data

Compound name EC50 (µM) Hill Max. response

CDDO methyl ester 0.01389697 1.556 100.31Sulforaphane 0.297622 1.5274 106.9Omaveloxolone (RTA 408) 0.2115237 3.6273 90.953Pyrogallin 2.471992 1.0653 77.874Fisetin 2.503583 1.6383 58.956Rhamnetin 2.380603 1.0183 43.274Dimethyl fumarate 4.78086 1.5699 119.38Purpurogallin-4-carboxylic acid 5.528483 1.7406 53.143-methoxycatechol 15.76957 1.6907 50.248Aripiprazole 30.33891 5.0303 78.294Pyrimethamine 29.82909 3.7568 27.003

Hill, Hill coefficient; Max. response, percent of the CDDO methyl ester positivecontrol.

9

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

gene expression programs in cell types and tissues that are vulnerableto proteotoxic, oxidative or redox stress. A recent n-of-1 clinical studyof an NGLY1-CDDG patient who died of complications due toadrenal insufficiency suggests that steroidogenic secretory tissues fitthe criteria of a vulnerable cell type (van Keulen et al., 2019).The Pngl−/− fly has defects in its physiologically analogousneuroendocrine tissues, which fail to produce and secrete thecholesterol-derived hormone 20-hydroxyecdysone (Rodriguez et al.,2018). That said, there is only a single ortholog of both NRF1 andNRF2 in worms and flies (SKN-1 and cnc, respectively), so thefunctions of worm SKN-1 and fly cnc may not recapitulate allfunctionality of mammalian NRF1 and NRF2.Why do anti-inflammatory drugs rescue worm and fly models of

NGLY1 deficiency? Defects in mitophagy lead to the release ofmitochondrial genomic DNA and mitochondrial-encoded RNAfrom fragmented and damaged mitochondria into the cytoplasm,which drives constitutive cGAS-STING and related innateimmunity responses in NGLY1−/− mice and cell lines (Yang et al.,2018). These innate immunity pathways are conserved in flies(Martin et al., 2018) and worms (Wu et al., 2014). We speculate thatmitochondrial fragmentation and mitophagy defects are present incells and tissues of png-1-null mutant worms and Pngl−/− mutantflies. Mitophagy defects and oxidative stress were observed inSKN-1 mutant worms (Palikaras et al., 2015). We thereforehypothesize that the steroidal and nonsteroidal anti-inflammatorydrugs that rescue worm and fly models of NGLY1 deficiency, e.g.phenylbutazone, are suppressing basally elevated innate immuneresponses. There is evidence for this inflammation hypothesis in therecent NGLY1 literature. SKN-1 is required for pathogen resistancein worms (Papp et al., 2012). Transcriptional profiling of Pngl-knockdown flies revealed that genes involved in the innate immuneresponse are upregulated (Owings et al., 2018). The expression ofinterferon genes is constitutively upregulated in Ngly1−/− murineembryonic fibroblasts (Yang et al., 2018). It is tempting to speculatefurther that hyperactive innate immunity responses contribute tolarval growth arrest and developmental delay in NGLY1-deficientanimals, and that these anti-inflammatory responses are dampenedby inhibitors of DNA synthesis, specifically purine biosyntheticenzymes targeted by the fly repurposing hits thioguanosine and thedihydrofolate-reductase-like inhibitor pyrimethamine (Dziekanet al., 2019), and possibly the diaminopteridine-containing novellead compounds that resemble pyrimethamine. Even more sobecause the same mechanism of action appears to have beenrevealed by parallel repurposing and discovery screens.At this time, the use of catecholamine boosters are supported by

measurements of reduced catecholamine precursors in the cerebralspinal fluid of NGLY1-CDDG patients (Lam et al., 2017), as well asthe fact that an adult and wheelchair-bound NGLY1-CDDG patienthas received the dopamine precursor L-Dopa. Interestingly, there isevidence for dopamine insufficiency from a Pngl RNAi-knockdownfly model of NGLY1 deficiency, where the expression of genesinvolved in dopamine biosynthesis (e.g. tyrosine hydroxylase and thedopamine transporter DAT) are constitutively downregulated(Owings et al., 2018). Future natural history studies of NGLY1deficiency should focus on catecholamine insufficiency as a potentialdriver or axis of pathophysiology. The effects of aripiprazole inworms depends on catecholamine pathway genes (Osuna-Lugueet al., 2018). In flies, aripiprazolewas shown to reduce the levels of anaggregated polyglutamine-expanded mutant protein in a model ofMachado-Joseph disease, or spinocerebellar ataxia type 3 (Costaet al., 2016). Themechanism of rescue appears to rely on activation ofproteotoxic and antioxidant stress responses. It is tempting to

conjecture that aripiprazole achieved its rescue effects in theMachado-Joseph fly model and in our NGLY1-deficiency fly modelby activating both NRF2 and catecholamine pathways, and, further,that the unique polypharmacology of aripiprazole combinescatecholamine pathway activation and NRF2 pathway activation.

MATERIALS AND METHODSStrains and compound librariesThe png-1 deletion mutant ok1654 was previously described (Lehrbach andRuvkun, 2016) and is available from the CGC Stock Center (strain ID:RB1452). The nonsense allele Pngl fly was previously described(Rodriguez et al., 2018) and is available from the Bloomington StockCenter. The 2560-compound Microsource Spectrum Collection wasavailable for purchase before the distributor went out of business. The20,240-compound lead-discovery library was purchased from ChembridgeCorporation. All compounds were dissolved in DMSO and stored at −80°Cin Labcyte-compatible LDV plates until use.

High-throughput larval growth assays in wormsPositive controls for all screens are png-1mutants+DMSO (vehicle control)and the negative controls are png-1 mutants+DMSO+205 nM bortezomib.Both drug libraries were screened in triplicate with controls on each 384-well drug-screening plate. The final DMSO concentration in each well was0.27%. Worms tolerated up to 1% DMSO before showing signs of toxicity.Using the Echo550 liquid handler (Labcyte Inc.), bortezomib wasacoustically dispensed into the destination plates the day before the wormlarvae sort. In total, 5 µl of HB101 bacteria were dispensed into 384-wellplates containing S-medium (prepared in-house) in each well. Using theBioSorter (Union Biometrica), 15 L1 png-1 mutant larvae were sorted intoeach well, and plates were incubated for 5 days at 20°C while shaking. After5 days of benchtop incubation, 15 µl of 8 mM sodium azide was added toeach well to immobilize worms prior to imaging in a custom worm imager.Finally, automated image processing was run on each plate.

High-throughput fly larval growth assaysDrug screens of Pngl heterozygous flies were conducted in the presence of9 µM bortezomib. The negative controls in the screen are Pngl heterozygousflies+DMSO+bortezomib and the positive controls are Pngl heterozygousflies+DMSO. Both drug libraries were screened in triplicate. The finalDMSO concentration in each well was 0.1%. Flies tolerated up to 1%DMSO before showing signs of toxicity. Using the Echo550 liquid handler(Labcyte Inc.), test compound was acoustically dispensed and standard flyfood media (molasses, agar, yeast, propionic acid) lacking cornmeal, butcarrying 0.025% Bromophenol Blue, was dispensed using a Multi-Flo (Bio-Tek Instruments) into each well of a 96-well plate for drug screens or a 12-wellplate for hit retests. Then, the BioSorter was used to dispense fly Pnglheterozygous larvae, three per well. At 3 days post-incubation with thecompounds, the plates were scored for larval size rescue in a custom fly imager.

High-throughput drug screening and analysisWe used the statistical program R (https://www.r-project.org/) in all Z-scorecalculations described herein. Hit compounds rescued worm developmentto adulthood, i.e. increased the number of worms (total area taken up byworms) in the well. An increase in the total area occupied by worms over thecourse of the 5-day liquid growth assay is caused by the 15 L1 worm larvaegrowing in size to adults as well as by these adults producing progeny.Therefore, the output of the image processing was worm area per well. Eachplate contained 32 wells of positive controls (worms raised in the absence ofbortezomib) and 32 wells of negative controls (worms raised in the presenceof bortezomib). The average area occupied by worms in positive andnegative control wells as well standard deviation (s.d.) of control wells werecalculated. Outlier elimination was performed by identifying thosewells thatwere greater than 1.5 times the interquartile range (1.5× IQR). Z-scores fortest wells were calculated by normalizing by mean area s.d. of negativecontrol wells. For each replicate, a test well that had a Z-score ≥2 wascounted as a hit. Quality control was performed to ensure that the controlwell grew as expected, the worm png-1 mutants were sensitive to

10

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

bortezomib as expected, there were no obvious plate effects, and dispenseerrors were eliminated. Because image artifacts can often confound areameasurements, we also manually verified each well by visual analysis afterthe unbiased quantitative analysis to make sure no false positives werecounted as true hits. We count wells as hits that have a Z-score greater thantwo across all three replicates of the screen and were free of image artifacts.

Fly screening conditions were previously described (Rodriguez et al.,2018). Briefly, we imaged the plates using a custom fly imager and analysistool to determine the area that the larvae occupy per well. Then, we countedthe number of larvae per well to normalize the total-area-per-well data andidentified suppressors of the phenotype. We also manually scored the wellsfor contaminants that may give false positives. Similar to the statisticalanalysis for the nematode assay, test wells were assigned Z-scores relative tomean negative controls. Any well that had a Z-score >2.5 in at least two outof three replicates was counted as a hit.

Keap1-NRF2 activation luciferase reporter assay in U2OS cellsWe tested compounds using the PathHunter® eXpress Keap1-NRF2 NuclearTranslocation Assay (DiscoverX, San Diego). Briefly, PathHunter cells areplated and incubated for 24 h at 37°C. Ten µl of test compoundwas added andcells were incubated with compound for 6 h at room temperature. Workingdetection reagent solution was added and plates were incubated for 60 min atroom temperature. Chemiluminescence signal was read by a SpectraMaxM3.

AcknowledgementsWe thank Dr Matthew Might for feedback on the manuscript. We also thank theanonymous reviewers for their comments, which improved the readability and rigorof the manuscript.

Competing interestsThe authors declare no competing or financial interests.

Author contributionsConceptualization: S.I., J.D.M., N.D., E.O.P.; Methodology: S.I., J.D.M., H.T., T.P.R.,N.D., M.P., F.S.S., Z.P.; Software: H.T., M.P., Z.P.; Formal analysis: N.D., F.S.S.,Z.P., E.O.P.; Investigation: H.T., T.P.R., M.P., F.S.S., Z.P., E.O.P.; Data curation:Z.P.; Writing - original draft: E.O.P.; Writing - review & editing: S.I., J.D.M., N.D.;Supervision: E.O.P.; Project administration: S.I., J.D.M., E.O.P.; Fundingacquisition: E.O.P.

FundingWe acknowledge the Grace Science Foundation as a funding source.

Data availabilityHigh-throughput screening datasets are publically available as ‘Perlara ArkBase v1’on Collaborative Drug Discovery (https://www.collaborativedrug.com/; freeregistration required at https://app.collaborativedrug.com/register).

Supplementary informationSupplementary information available online athttp://dmm.biologists.org/lookup/doi/10.1242/dmm.040576.supplemental

ReferencesCosta, M. D. C., Ashraf, N. S., Fischer, S., Yang, Y., Schapka, E., Joshi, G.,McQuade, T. J., Dharia, R.M., Dulchavsky,M.,Ouyang,M. et al. (2016).Unbiasedscreen identifies aripiprazole as a modulator of abundance of the polyglutaminedisease protein, ataxin-3. Brain 139, 2891-2908. doi:10.1093/brain/aww228

Dziekan, J. M., Yu, H., Chen, D., Dai, L., Wirjanata, G., Larsson, A., Prabhu, N.,Sobota, R. M., Bozdech, Z. and Nordlund, P. (2019). Identifying purinenucleoside phosphorylase as the target of quinine using cellular thermal shiftassay. Sci. Transl. Med. 11, eaau3174. doi:10.1126/scitranslmed.aau3174

Enns, G. M., Shashi, V., Bainbridge, M., Gambello, M. J., Zahir, F. R., Bast, T.,Crimian, R., Schoch, K., Platt, J., Cox, R. et al. (2014). Mutations in NGLY1cause an inherited disorder of the endoplasmic reticulum-associated degradationpathway. Genet. Med. 16, 751-758. doi:10.1038/gim.2014.22

Fenteany, G., Standaert, R. F., Lane, W. S., Choi, S., Corey, E. J. and Schreiber,S. L. (1995). Inhibition of proteasome activities and subunit-specific amino-terminal threonine modification by lactacystin. Science 268, 726-731. doi:10.1126/science.7732382

Galeone, A., Han, S. Y., Huang, C., Hosomi, A., Suzuki, T. and Jafar-Nejad, H.(2017). Tissue-specific regulation of BMP signaling by Drosophila N-glycanase 1.eLife 6, e27612. doi:10.7554/eLife.27612

Glynn, S. J., Gaffney, K. J., Sainz, M. A., Louie, S. G. and Petasis, N. A. (2015).Molecular characterization of the boron adducts of the proteasome inhibitorbortezomib with epigallocatechin-3-gallate and related polyphenols. Org. Biomol.Chem. 13, 3887-3899. doi:10.1039/C4OB02512A

Golden, E. B., Lam, P. Y., Kardosh, A., Gaffney, K. J., Cadenas, E., Louie, S. G.,Petasis, N. A., Chen, T. C. and Schonthal, A. H. (2009). Green tea polyphenolsblock the anticancer effects of bortezomib and other boronic acid-basedproteasome inhibitors. Blood 113, 5927-5937. doi:10.1182/blood-2008-07-171389

Grimberg, K. B., Beskow, A., Lundin, D., Davis, M. M. and Young, P. (2011).Basic leucine zipper protein Cnc-C is a substrate and transcriptional regulator ofthe Drosophila 26S proteasome. Mol. Cell Biol. 31, 897-909. doi:10.1128/MCB.00799-10

Kikuchi, T., Tottori, K., Uwahodo, Y., Hirose, T., Miwa, T., Oshiro, Y. and Morita,S. (1995). 7-(4-[4-(2,3-Dichlorophenyl)-1-piperazinyl]butyloxy)-3,4-dihydro-2(1H)-quinolinone (OPC-14597), a new putative antipsychotic drug with bothpresynaptic dopamine autoreceptor agonistic activity and postsynaptic D2receptor antagonistic activity. J. Pharmacol. Exp. Ther. 274, 329-336.

Kong, J., Peng, M., Ostrovsky, J., Kwon, Y. J., Oretsky, O., McCormick, E. M.,He, M., Argon, Y. and Falk, M. J. (2018). Mitochondrial function requires NGLY1.Mitochondrion. 38, 6-16. doi:10.1016/j.mito.2017.07.008

Lam, C., Ferreira, C., Krasnewich, D., Toro, C., Latham, L., Zein,W.M., Lehky, T.,Brewer, C., Baker, E. H., Thurm, A. et al. (2017). Prospective phenotyping ofNGLY1-CDDG, the first congenital disorder of deglycosylation. Genet. Med. 19,160-168. doi:10.1038/gim.2016.75

Lehrbach, N. J. andRuvkun, G. (2016). Proteasome dysfunction triggers activationof SKN-1A/Nfr1 by the aspartic protease DDI-1. eLife 5, e17721. doi:10.7554/eLife.17721

Lehrbach, N. J., Breen, P. C. and Ruvkun, G. (2019). Protein sequence editing ofSKN-1A/NRF1 by Peptide:N-Glycanase controls proteasome gene expression.Cell. 177, 737-750.e15. doi:10.1016/j.cell.2019.03.035

Li, D., March, M. E., Gutierrez-Uzquiza, A., Kao, C., Seiler, C., Pinto, E.,Matsuoka, L. S., Battig, M. R., Bhoj, E. J., Wenger, T. L. et al. (2019). ARAFrecurrent mutation causes central conducting lymphatic anomaly treatable with aMEK inhibitor. Nat. Med. 25, 1116-1122. doi:10.1038/s41591-019-0479-2

Martin, M., Hiroyasu, A., Guzman, R. M., Roberts, S. A. and Goodman, A. G.(2018). Analysis of Drosophila STING reveals an evolutionarily conservedantimicrobial function. Cell Rep. 23, 3537-3550.e6. doi:10.1016/j.celrep.2018.05.029

Osuna-Lugue, J., Rodriguez-Ramos, A., Gamez-Del-Estal, M. D. M. and Ruiz-Rubio, M. (2018). Behavioral mechanisms that depend on dopamine andserotonin in Caenorhabditis elegans interact with the antipsychotics risperidoneand aripiprazole. J. Exp. Neurosci. 12, 1179069518798628. doi:10.1177/1179069518798628

Owings, K. G., Lowry, J. B., Bi, Y., Might, M. and Chow, C. Y. (2018).Transcriptome and functional analysis in a Drosophila model of NGLY1 deficiencyprovides insight into therapeutic approaches. Hum. Mol Genet. 27, 1055-1066.doi:10.1093/hmg/ddy026

Paek, J., Lo, J. Y., Narasimhan, S. D., Nguyen, T. N., Glover-Cutter, K., Robida-Stubbs, S., Suzuki, T., Yamamoto, M., Blackwell, T. K. and Curran, S. P.(2012). Mitochondrial SKN-1/Nrf mediates a conserved starvation response. CellMetab. 16, 526-537. doi:10.1016/j.cmet.2012.09.007

Palikaras, K., Lionaki, E. and Tavernarakis, N. (2015). Coordination of mitophagyand mitochondrial biogenesis during ageing in C. elegans. Nature 521, 525-528.doi:10.1038/nature14300

Papp, D., Czermely, P. and Soti, C. (2012). A role for SKN-1/NRF1 in pathogenresistance and immunosenescence in Caenorhabditis elegans. PLoS Pathog. 8,e1002673. doi:10.1371/journal.ppat.1002673

Pushpakom, S., Iorio, F., Eyers, P. A., Escott, K. J., Hopper, S., Wells, A., Doig,A., Guilliams, T., Latimer, J., McNamee, C. et al. (2019). Drug repurposing:progress, challenges and recommendations. Nat. Rev. Drug. Discov. 18, 41-58.doi:10.1038/nrd.2018.168

Radhakrishnan, S. K., den Besten, W. and Deshaies, R. J. (2014). p97-dependent retrotranslocation and proteolytic processing govern formation ofactive NRF1 upon proteasome inhibition. eLife 3, e01856. doi:10.7554/eLife.01856

Rodriguez, T. P., Mast, J. D., Hart, T., Lee, T., Sand, P. and Perlstein, E. O. (2018).Defects in the neuroendocrine axis contribute to global development delay in aDrosophila model of NGLY1 Deficiency. G3 8, 2193-2204. doi:10.1534/g3.118.300578

Shapiro, D. A., Renock, S., Arrington, E., Chiodo, L. A., Liu, L.-X., Sibley, D. R.,Roth, B. L. and Mailman, R. (2003). Aripiprazole, a novel atypical antipsychoticdrug with a unique and robust pharmacology. Neuropsychopharmacology 28,1400-1411. doi:10.1038/sj.npp.1300203

Smirnova, N. A., Haskew-Layton, R. E., Basso, M., Hushpulian, D. M.,Payappilly, J. B., Speer, R. E., Ahn, Y.-H., Rakhman, I., Cole, P. A., Pinto,J. T. et al. (2011). Development of Neh2-luciferase reporter and its application forhigh throughput screening and real-time monitoring of Nrf2 activators.Chem. Biol.18, 752-765. doi:10.1016/j.chembiol.2011.03.013

11

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms

Suzuki, T., Huang, C. and Fujihira, H. (2016). The cytoplasmic peptide:N-glycanase (NGLY1) – structure, expression and cellular functions.Gene 577, 1-7.doi:10.1016/j.gene.2015.11.021

Tomlin, F. M., Gerling-Driessen, U. I. M., Liu, Y.-C., Flynn, R. A., Vangala, J. R.,Lentz, C. S., Clauder-Muenster, S., Jakob, P., Mueller, W. F., Ordon ez-Rueda,D. et al. (2017). Inhibition of NGLY1 inactivates the transcription factor NRF1 andpotentiates proteasome inhibitor cytotoxicity. ACS Cent. Sci. 3, 1143-1155.doi:10.1021/acscentsci.7b00224

Tsakiri, E. N., Sykiotis, G. P., Papassideri, I. S., Terpos, E., Dimopoulos, M. A.,Gorgoulis, V. G., Bohmann, D. and Trougakos, I. P. (2013). Proteasomedysfunction in Drosophila signals to an Nrf2-dependent regulatory circuit aiming to

restore proteostasis and prevent premature aging. Aging Cell 12, 802-813. doi:10.1111/acel.12111

van Keulen, B. J., Rotteveel, J. and Finken, M. J. J. (2019). Unexplained death inpatients with NGLY1 mutations may be explained by adrenal insufficiency.Physiol. Rep. 7, e13979. doi:10.14814/phy2.13979

Wu, X., Wu, F.-H., Wang, X., Wang, L., Siedow, J. N., Zhang, W. and Pei, Z.-M.(2014). Molecular evolutionary and structural analysis of the cytosolic DNA sensorcGAS and STING. Nucleic Acids Res. 42, 8243-8257. doi:10.1093/nar/gku569

Yang, K., Huang, R., Fujihira, H., Suzuki, T. and Yan, N. (2018). N-glycanaseNGLY1 regulates mitochondrial homeostasis and inflammation through NRF1.J. Exp. Med. 215, 2600-2616. doi:10.1084/jem.20180783

12

RESEARCH ARTICLE Disease Models & Mechanisms (2019) 12, dmm040576. doi:10.1242/dmm.040576

Disea

seModels&Mechan

isms