(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .
An AI-guided screen identifies probucol as an enhancer of mitophagy through modulation of lipid droplets [1]
['Natalia Moskal', 'Department Of Biochemistry', 'University Of Toronto', 'Toronto', 'Naomi P. Visanji', 'Edmund J Safra Program In Parkinson S Disease', 'Morton', 'Gloria Shulman Movement Disorders Centre', 'Toronto Western Hospital', 'Olena Gorbenko']
Date: 2023-03
Failures in mitophagy, a process by which damaged mitochondria are cleared, results in neurodegeneration, while enhancing mitophagy promotes the survival of dopaminergic neurons. Using an artificial intelligence platform, we employed a natural language processing approach to evaluate the semantic similarity of candidate molecules to a set of well-established mitophagy enhancers. Top candidates were screened in a cell-based mitochondrial clearance assay. Probucol, a lipid-lowering drug, was validated across several orthogonal mitophagy assays. In vivo, probucol improved survival, locomotor function, and dopaminergic neuron loss in zebrafish and fly models of mitochondrial damage. Probucol functioned independently of PINK1/Parkin, but its effects on mitophagy and in vivo depended on ABCA1, which negatively regulated mitophagy following mitochondrial damage. Autophagosome and lysosomal markers were elevated by probucol treatment in addition to increased contact between lipid droplets (LDs) and mitochondria. Conversely, LD expansion, which occurs following mitochondrial damage, was suppressed by probucol and probucol-mediated mitophagy enhancement required LDs. Probucol-mediated LD dynamics changes may prime the cell for a more efficient mitophagic response to mitochondrial damage.
Funding: This study was supported by the Canadian Institutes of Health Research (PJT-156186 to G.A.M.). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Copyright: © 2023 Moskal et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
In a previous screen, we focused on compounds that accelerated the transition of the E3 ubiquitin ligase, Parkin, from the cytosol to the mitochondria-a step, which is integral to this mitophagy pathway [ 15 ]. While this step is highly amenable to microscopy-based phenotypic screening, the limitations of this approach include possible omission of hits that function downstream of Parkin or that target other mitophagy pathways [ 16 ]. In this present screening iteration, we evaluated the clearance of damaged mitochondria from cells. Ultimately, if this downstream step is improved, then the negative consequences of mitochondrial damage in the dopaminergic neurons may be mitigated [ 17 ]. This effort led to the identification of a compound that enhances mitophagy following mitochondrial damage and led us to elucidate its mechanism of action, resulting in the identification of a putative newfound role for the ATP binding cassette transporter A1 (ABCA1) in mitophagy, through its effects on lipid droplets (LDs) dynamics.
Here, using a similar approach, we screened a candidate list of molecules from the DrugBank database for similarity to known mitophagy enhancers. Many of the drugs have already been safely administered to humans. Repurposing drugs from other indications offers the opportunity to accelerate the clinical trials pipeline, given the presence of preexisting pharmacological and toxicological information about candidate compounds [ 14 ].
A review by Georgopoulos and colleagues describes several compounds currently known to stimulate and potentiate mitophagy [ 10 ]. However, most of these compounds also induce mitochondrial damage or apoptosis. While these are bona fide mitophagy enhancers, induction of mitochondrial damage or apoptosis would likely preclude their clinical use as this may worsen PD pathogenesis. To address the unmet need for mitophagy enhancers with therapeutic potential, we employed a computational approach using artificial intelligence (AI) to identify previously uncharacterized mitophagy enhancers. We have previously successfully used this strategy that detects patterns and associations across several large datasets to identify compounds that are similar to a user-defined training set of positive controls, to identify drugs with disease modifying potential for PD that target aggregation of alpha synuclein [ 11 – 13 ].
Evidence for the importance of mitophagy in PD pathogenesis comes from both sporadic and genetic cases. Several disease-causing, loss-of-function mutations in genes encoding proteins that mediate mitophagy have been identified, including PINK1 and PRKN [ 2 , 3 ]. Additionally, several disease-causing mutations in genes not directly associated with mitophagy have secondary negative effects on mitochondrial health or on the retrograde transport of damaged mitochondria to axons, such as SNCA, GBA, and LRRK2 [ 4 – 6 ]. Besides genetic causes of PD, mitochondrial damage and mitophagy impairment have also been widely implicated in sporadic disease, including the inactivation of proteins that mediate mitophagy [ 7 – 9 ]. The clear connection between this pathway and the health of dopaminergic neurons emboldened us to search for novel mitophagy-enhancing compounds as potential disease-modifying therapeutics for PD.
Clearance of damaged mitochondria is important for the survival of dopaminergic neurons, the loss of which is responsible for the classical motor symptoms of Parkinson disease (PD). PD is a progressive neurodegenerative disease characterized by rigidity, akinesia, and bradykinesia as well as wide ranging nonmotor symptoms such as anxiety, depression, sleep disturbances, and loss of smell [ 1 ]. Current treatment options address symptoms but fail to disrupt the progression of the disease, thus a disease-modifying treatment remains a major unmet need. Mitochondrial dysfunction is clearly implicated in PD pathogenesis, so enhancing the removal of damaged mitochondria (mitophagy) might have potential as a therapeutic intervention [ 1 ].
However, probucol treatment no longer improved climbing when ABCA RNAi was expressed in dopaminergic neurons ( Fig 4E ). These findings show that probucol treatment increased downstream autophagy steps such as autophagosome and lysosome biogenesis, seemingly priming the cells for a more rapid and efficient mitophagy response when damage strikes. ABCA, probucol’s target, likely facilitates this effect through alterations to LD dynamics ( Fig 4F ).
To determine whether probucol’s effect on mitophagy were responsible for its ability to improve paraquat-induced climbing impairment, climbing assays were performed as described in Fig 2A , but the dopaminergic neuron-specific TH-GAL4 driver was used to drive expression of RNAi targeting candidate genes. In this manner, we dissected which factors were dispensable for probucol-mediated climbing improvements. Consistently with experiments in cells that showed that probucol had no effect on Parkin subcellular distribution ( S5 Fig ), parkin RNAi did not abrogate the effects of probucol-mediated improvement of climbing impairment, which was evident in both the mCherry RNAi and parkin RNAi -expressing flies ( Fig 4E ).
An endogenously mCherry-tagged Lamp reporter fly line was employed to assess lysosomes in vivo. Lamp is a lysosomal protein, which increases in abundance following paraquat treatment [ 33 ]. Dopaminergic neurons were segmented with TH immunostaining, and the area occupied by Lamp-positive lysosomes within dopaminergic neurons was measured. As expected, paraquat addition to fly food increased abundance of Lamp-positive puncta ( Fig 4C ). Probucol treatment increased the area of lysosomes in the dopaminergic neurons, under basal conditions where no exogenous stimulus was added to trigger lysosome accumulation ( Fig 4C and 4D ).
( A ) HeLa cells were either incubated in DMEM, DMEM with CCCP, or HBSS media for 6 hours in the presence or absence of bafilomycin. Lysates were separated by SDS-PAGE and immunoblotting was performed using antibodies that recognize LC3 and tubulin, as a loading control. ( B ) Densitometry was performed to measure the levels of lipidated LC3-II, which were normalized to the tubulin loading control. ( C ) Endogenously tagged Lamp-3xmCherry flies were fed food supplemented with DMSO or probucol in the presence and absence of paraquat. ( D ) The Lamp area in each cell was measured as a percentage of total cell area, as defined by segmentation of TH-positive dopaminergic neurons. ( E ) RNAi targeting either parkin or ABCA was expressed in the dopaminergic neurons with the TH-Gal4 driver. The effect of probucol on paraquat-induced climbing impairment was assessed in flies from the various genotypes. Climbing as a percentage of flies to cross 12.5 cm height is displayed. ( F ) Hypothetical mechanism of mitophagy enhancement by probucol may involve up-regulation of Lamp-positive late endosomes/lysosomes and LC3-II-positive mature autophagosomes, possibly arising from mobilization of LDs adjacent to mitochondria. LD expansion occurs following mitochondrial damage, but in the presence of probucol, the abundance of LDs is reduced, resulting in increased abundance of mito-lysosomes. Data information: Bars represent mean values and error bars correspond to SEM from at least 3 independent biological replicates, which are indicated with data points in graphs B and D. Statistical analysis to assess differences between DMSO and probucol within the same treatment condition or genotype group was performed using one-way ANOVA tests with multiple comparison correction in B and E, while unpaired t tests were used to compare DMSO and probucol in D. *, *, and *** indicate p-value <0.05, <0.01, and <0.005, respectively. The data underlying the graphs shown in the figure can be found in S1 Data . CCCP, carbonyl cyanide m-chlorophenyl hydrazone; LD, lipid droplet; TH, tyrosine hydroxylase.
A recent report found that autophagy must be tuned to provide sufficient dynamic range to resolve differences in LC3 lipidation following manipulations such as drug treatments [ 32 ]. This can be accomplished by employing bafilomycin, an inhibitor of lysosome–autophagosome fusion at low doses so the effects of manipulations would be apparent. Experiments were repeated in HeLa cells, which lack endogenous Parkin with the addition of low-dose bafilomycin treatment. While the difference between DMSO- and probucol-treated groups under basal conditions was not as robust as in HEK293 cells, the addition of bafilomycin revealed a difference in LC3 lipidation between DMSO- and probucol-treated cells in the CCCP group ( Fig 4A and 4B ).
Since early mitophagy steps were unaltered by probucol, downstream steps were assessed next. Immunoblotting can be used to evaluate the lipidation status of LC3. The lower molecular weight, lipidated form of LC3 (LC3-II), correlates with autophagosome levels and therefore increases as autophagy proceeds. However, it is important to note that increased LC3-II levels alone, in the absence of cotreatments with lysosome inhibitors, cannot directly indicate increased autophagic flux [ 31 ]. LC3-II levels increased following probucol treatment under basal conditions ( S9 Fig ) in HEK293 cells with endogenous Parkin present at low levels. LC3-II levels were predictably higher upon mitophagy induction with CCCP treatment compared to basal conditions, as other studies show. However, the difference in LC3-II between DMSO- and probucol-treated cells did not persist under these conditions.
Since probucol no longer exerted effects on mitophagy without LDs and when ABCA1 levels are reduced, we probed whether ABCA1 affected probucol-mediated LD expansion. This critical step in probucol’s mechanism of action on mitophagy was no longer evident when ABCA1 levels were reduced with shRNA (Figs 3F and S7B ).
Finally, to evaluate the importance of LDs on the mitophagy enhancement conferred by probucol treatment, diacylglycerol acyltransferases (DGAT) 1 and DGAT2 inhibitor treatment was added to suppress the LD expansion, which occurred following prolonged mitochondrial stress (Figs 3C and S8 ). The addition of DGAT inhibitors to probucol treatment in the context of paraquat-induced mitochondrial dysfunction in flies attenuated probucol’s enhancement of mitophagy in the dopaminergic neurons of flies ( Fig 3E ). Mitophagy was no longer elevated by probucol in the context of paraquat, when DGAT inhibitors are present. Since DGAT inhibitors reduced LD area and attenuated probucol’s effect on mitophagy, this suggests probucol’s effect on LDs may be responsible for its subsequent effects on mitophagy.
LD-mitochondria contacts have been observed in several studies. Peridroplet mitochondria localized to contact sites exhibit altered bioenergetic properties and serve as a source of ATP for LD expansion [ 30 ]. However, under conditions of high autophagic flux, LDs are thought to interact with mitochondria to buffer the lipotoxic species that are produced as byproducts of autophagy [ 28 ]. Interestingly, a greater proportion of LDs colocalized with mitochondria in probucol-treated cells under basal conditions where mitochondria appeared intact and visibly elongated (Figs 3D and S8 ). Under the conditions in which this increased contact is observed, mitophagic flux is not elevated ( Fig 1C and 1D ), so it is unlikely that the contacts facilitate buffering of lipotoxic byproducts like in starvation-induced autophagy [ 28 ].
( A ) TH-positive dopaminergic neurons were segmented in BODIPY-stained brains from flies fed food supplemented with the indicated combinations of probucol and paraquat. ( B ) The percentage of LD area over the total TH-positive cell area is quantified from A. ( C ) The percentage of LD area over total cell area in HeLa cells treated with either DMSO or probucol alone or in the presence of CCCP with and without DGAT inhibitors. ( D ) The overlap between BODIPY-stained LDs and mitochondria was assessed using the Manders Correlation Coefficient in HeLa cells treated as described in C). ( E ) Probucol treatment was coadministered along with paraquat in the presence and absence of DGAT inhibitors. The percentage of red-only mitochondrial area was measured. ( F ) HeLa cells were transfected with pLKO1 and shABCA1 plasmids and treated with the indicated combination of probucol and CCCP for 24 hours. BODIPY staining was used to quantify the percentage of cell area occupied by LDs. Data information: Results are representative of at least 3 independent biological replicates, as indicated by the data points. Mean values are displayed with error bars, which represent the SEM. * and *** represent p-values <0.05 and 0.005, respectively. Statistical analysis was performed using ANOVA analysis, with multiple comparison correction for E and F. Unpaired Student t test analysis was performed to compare DMSO and probucol groups in B, C, and D. The data underlying the graphs shown in the figure can be found in S1 Data . CCCP, carbonyl cyanide m-chlorophenyl hydrazone; DGAT, diacylglycerol acyltransferase; LD, lipid droplet; TH, tyrosine hydroxylase.
The extent of crosstalk between lipid homeostasis and autophagy reported in other studies and ABCA1’s role in lipid efflux led us to test whether probucol altered LDs. LDs have been found to form in response to starvation-induced autophagy and deferiprone-induced mitophagy [ 28 , 29 ]. LDs similarly increased significantly in cells treated with CCCP for 24 hours (Figs 3C and S8 ). Surprisingly, probucol treatment reduced the LD area expansion that occurred following mitophagy induction but had no effect on LD area under basal conditions (Figs 3C and S8 ). LD area also increased in the dopaminergic neurons when flies consumed paraquat in their food. Probucol supplementation decreased the LD area expansion to levels comparable to those under basal conditions ( Fig 3A and 3B ).
We also reduced ABCA1 levels with shRNA ( S7B Fig ) in cells expressing mitoQC and Cerulean-Parkin and treated cells as in Fig 1A . While probucol increased the percentage of red-only mitochondrial area indicative of mito-lysosomes in cells transfected with the control pLKO1 vector, probucol no longer increased mitophagy in cells transfected with shABCA1 (Figs 1F and S7C ).
Human ABCA1 and mitoQC transgenes were coexpressed, or mitoQC was expressed alone in the dopaminergic neurons of flies (Figs 1E , S7A and S7D ). Once again, paraquat treatment increased the percentage of red-only mitochondrial puncta, which represent mitochondria localized to lysosomes. Basal mitophagy was unaffected by overexpression of the transgene. However, in flies fed paraquat, ABCA1 reduced the percentage of mitochondrial area localized to lysosomes ( Fig 1E ). Immunoblotting was performed to confirm expression of human ABCA1 ( S7A Fig ).
To determine whether probucol’s canonical target, ABCA1, is involved in its effects on mitophagy, we performed several genetic manipulations to change ABCA1 levels in cell culture or in the dopaminergic neurons of flies. The same mitophagy assay used to characterize probucol’s effects was also performed in this context.
To directly probe the cell type impacted in PD, Tg(dat:EGFP) zebrafish embryos were coincubated with MPP + and probucol or vehicle control. Dopaminergic neurons can easily be visualized and counted in this transgenic model 27 . Following 1 day of incubation in MPP + , the number of dopaminergic neurons in the ventral diencephalon (vDC), which is analogous to the human nigrostriatal region, was reduced. In contrast, probucol-fed zebrafish retained more of their dopaminergic vDC neurons ( Fig 2E and 2F ). This reduced loss of dopaminergic neurons likely led to the probucol-mediated improvements to PD-related phenotypes.
In addition to toxin-based models of PD, we also tested probucol’s effect in a genetic model of mitochondrial dysfunction. Specifically, in heteroplasmic flies with approximately 90% of mitochondrial DNA (mtDNA) that contained a temperature-sensitive mutation in mitochondrial cytochrome c oxidase subunit I (mt:ColI T300I ). Shifting these flies to a nonpermissive temperature causes mitochondrial dysfunction resulting in climbing defects and significant reduction of life span [ 22 , 24 , 25 ]. Climbing and life span were improved in heteroplasmic mt:ColI T300I flies fed food containing probucol rather than DMSO ( S6 Fig ). Much like paraquat, the phenotypes displayed by these flies arise from mitochondrial dysfunction, and probucol reduced their severity, possibly by promoting mitophagy in this context as well, given that mitophagy has previously been shown to remove mitochondria bearing deleterious mutations [ 26 ].
Zebrafish embryos were incubated in MPP + in combination with probucol or DMSO. Following 6 days of incubation, the movement of adult zebrafish was captured using ZebraBox. The distance travelled by the zebrafish exposed to DMSO in combination with MPP + was visibly reduced compared to zebrafish incubated in DMSO alone, but the addition of probucol to the MPP + increased the distance travelled by the zebrafish ( Fig 2C and 2D ). SR3677, a chemical inhibitor of Rho-associated protein kinase 2 (ROCK2) was employed as a positive control. We previously characterized this compound as a mitophagy enhancer and found that it improved locomotor decline in flies fed paraquat [ 22 ].
( A ) Effect of probucol administration on paraquat-induced climbing defect. The percentage of flies to climb across a height of 12.5 cm. ( B ) Effect of probucol administration on survival, in the presence and absence of paraquat cotreatments. ( C ) Single particle tracking traces to visualize distance travelled by zebrafish in wells following administration of DMSO, probucol, and MPP + , as indicated. ( D ) Distance travelled by zebrafish in each treatment group in C, in addition to following treatment with positive control compound SR3677. ( E ) Tg(dat:EGFP) zebrafish brains were imaged following treatment with DMSO, probucol, and MPP + . Dashed lines highlight the vDC region of interest. ( F ) The number of vDC neurons in each treatment group in E was quantified. Data information: At least 3 independent biological replicates were performed for each experiment, with individual replicates depicted by data points in F. Bars and error bars represent mean and SEM values, respectively. Unpaired Student t tests were performed to analyze data in A, log-rank tests were used to analyze survival data, and one-way ANOVA analysis was performed to analyze data in D and F along with Dunnett’s multiple comparison correction. *, **, ***, and **** indicate p-values <0.05, <0.01, <0.005, and <0.001, respectively. The data underlying the graphs shown in the figure can be found in S1 Data . MPP+, 1-methyl-4-phenylpyridinium; PD, Parkinson disease; vDC, ventral diencephalon.
Drosophila and Danio rerio serve as PD model systems, as they replicate much of human PD pathogenesis and display phenotypes, which reflect human disease presentation. Flies and zebrafish exhibit loss of dopaminergic neurons resulting in impaired locomotion following exposure to PD-causing toxins such as 1-methyl-4-phenylpyridinium (MPP + ) and paraquat [ 21 , 23 ]. After replicating these features in the two model organisms, we tested probucol’s disease-modifying potential. The climbing ability and survival of flies declined following paraquat administration, but cotreatment with probucol improved locomotor function and survival ( Fig 2A and 2B ).
To further substantiate the findings in cells, the mitoQC reporter was expressed in the dopaminergic neurons of flies fed food supplemented with DMSO or probucol alone or in combination with paraquat, a mitochondrial toxin that causes PD in humans and PD-related phenotypes such as loss of dopaminergic neurons and locomotor impairments in model organisms such as flies [ 21 ]. Paraquat induces a mitophagy response characterized by an increase in mito-lysosomes in dopaminergic neurons [ 22 ]. Mitophagy further increased in flies coadministered probucol along with paraquat ( Fig 1C and 1D ). Basal mitophagy did not differ in the dopaminergic neurons of flies fed either DMSO- or probucol-supplemented food.
We also probed steps further upstream in the mitophagy cascade, including PINK1-mediated phosphorylation of ubiquitin at S65 and recruitment of Parkin to damaged mitochondria ( S5 Fig ). Phosphorylation of mitochondrial ubiquitin S65 following mitochondrial damage did not further increase following probucol treatment based on immunostaining and immunoblotting experiments ( S5A, S5B and S5D Fig ). Parkin recruitment to damaged mitochondria likewise was not increased by probucol treatment ( S5C Fig ), indicating that this mitophagy enhancer likely exerts its effect on steps further downstream of Parkin recruitment or through other Parkin-independent mitophagy pathways [ 16 ].
Compounds were retested in an orthogonal mitochondrial clearance assay; probucol increased the extent to which mitochondrial substrates were degraded following induction of mitochondrial damage. Immunoblotting was performed to visualize levels of the outer mitochondrial membrane protein voltage-dependent anion-selective channel protein 1 (VDAC1) and the inner mitochondrial membrane substrate ATP synthase F1 subunit alpha (ATP5A) ( S2D, S2E and S4 Figs). Predictably, VDAC1 levels declined over the CCCP time course. While probucol treatment does not increase VDAC1 degradation under basal conditions, it promoted enhanced degradation of VDAC1 over the CCCP time course ( S2D and S2E Fig ).
( A ) The mitoQC reporter was expressed in HeLa cells along with Cerulean-Parkin. Mitophagy was stimulated by treating cells with CCCP for 6 hours along with either DMSO or candidate molecules from the screen. Mitochondria appear as red-only puncta when localized to acidic cellular compartments. ( B ) Cells with >5 red-only puncta were defined as mitophagic, and the mean percentage of mitophagic cells in each treatment condition is displayed. ( C ) Seven-day old flies expressing the mitoQC reporter in dopaminergic neurons were fed food supplemented with probucol in combination with either paraquat or water. ( D ) Immunostaining using antibody against TH segmented and defined dopaminergic neurons. The mean percentage of red-only mitochondrial area in each dopaminergic neuron is displayed. ( E ) Control flies and flies coexpressing the human ABCA1 transgene and the mitoQC reporter were administered probucol in the presence and absence of paraquat. The mean percentage of red-only mitochondrial area in TH-positive neurons is displayed. See S4D Fig for corresponding images. ( F ) Mitophagy was assessed in HeLa cells stably expressing mitoQC and Cerulean-Parkin, which were transfected with either pLKO1 vector or with shABCA1. The mean percentage of red-only mitochondrial area per cell is displayed. See S4C Fig for corresponding images. Data information: Results are representative of at least 3 biological replicates, each indicated by data points in B, D, E, and F. Bars represent mean values and error bars represent SEM. *, **, and *** indicate p-values <0.05, 0.01, and 0.005, respectively. At least 40 cells were assessed for B, D, E, and F, respectively. Dopaminergic neurons from at least 2 fly brains were analyzed for each treatment with at least 10 ROIs per brain. Statistical analysis was performed using ANOVA and Dunnett’s multiple comparison correction. The data underlying the graphs shown in the figure can be found in S1 Data . ABCA1, ATP binding cassette transporter A1; CCCP, carbonyl cyanide m-chlorophenyl hydrazone; ROI, region of interest; TH, tyrosine hydroxylase.
Upstream of the degradation of damaged mitochondria is the targeting of mitochondria to lysosomes. The mitoQC assay can be used in both cells and in vivo to probe this precise step in mitophagy. Briefly, cells were transfected with Cerulean-Parkin and RG-OMP25, a plasmid containing mCherry and GFP in tandem with the mitochondrial targeting sequence of OMP25 ( Fig 1A ). Upon localization to acidic lysosomes, GFP signal is quenched, which distinguished mitochondria localized to lysosomes because they appear as puncta with red-only signal. Less than 5 red-only puncta are present in cells under normal conditions, but the number of puncta increases following induction of mitochondrial damage [ 20 ]. Cells with greater than 5 red-only puncta corresponding to mito-lysosomes were classified as “mitophagic” ( Fig 1A ). Following 6-hour CCCP treatment, 40.25 ± 6.3% cells are mitophagic compared to 13.32 ± 4.83% under normal conditions. A higher percentage of probucol-treated cells were mitophagic (66.29 ± 5.6%; Fig 1B ), compared to treatments with the other candidate compounds.
The top 79 most similar candidates identified by IBM Watson for Drug Discovery were screened in HeLa cells stably expressing GFP Parkin and mito-DsRed and subjected to prolonged treatment with carbonyl cyanide m-chlorophenyl hydrazone (CCCP), which depolarizes the mitochondrial membrane potential. Loss of mito-DsRed signal occurs in mitophagic cells, which comprise a large percentage of the population ( S2C and S3B Figs) [ 18 ]. In cells pretreated with DMSO instead of small molecules, 78 ± 2.83% of cells have low/no mito-DsRed signal ( S3B Fig ). MAD z-scores were calculated, as previously described for each of the 79 compounds ( S2A Fig and S3 Dataset ) [ 19 ]. Three compounds were selected for retesting based on MAD z-score values ( S2A Fig ), thereby producing a hit rate of 3.8%.
Ultimately, we assessed 3,231 candidate drugs from the DrugBank database for semantic similarity to the training drugs ( S1C Fig ). IBM Watson for Drug Discovery creates text fingerprints for all the training set molecules, in addition to 3,231 candidate molecules from the DrugBank database. We computed a similarity score (to the training set) score for each candidate entity, and they ranked drugs from highest to lowest accordingly ( S1 Dataset ).
We employed a computational approach using AI (IBM Watson for Drug Discovery) to identify drugs amenable to repurposing as PD therapeutics. Specifically, an in silico screen was performed to identify potential mitophagy enhancers from the DrugBank database ( www.drugbank.ca ), based on their similarity to positive control mitophagy enhancers. Our training set was based on a review article about mitophagy modulators, and it comprised the following 7 drugs: PTEN-induced putative kinase 1 (PINK1) activator kinetin; poly ADP-ribose polymerase (PARP) inhibitor olaparib, p53 inhibitor pifithrin-alpha; nicotinamide (NAD + accumulation); and sirtuin1 activators resveratrol, fisetin, and SRT1720 [ 10 ] ( S1C Fig ). Extensive model validation was performed prior to commencing the screen in cells ( S1 Fig ; see Methods for details.)
Discussion
Using our screening approach, several previously characterized mitophagy enhancers were recovered both in silico and in cells. In silico screening identified staurosporine among the top hits (17/3,231, top 0.5%), which is a well-characterized activator of mitophagy, but it was excluded from further experiments as it induces apoptosis [34]. In our cell-based screen, dichlorocopper ranked as 1/79 in our screen. Copper binds to and increases the kinase activity of autophagy regulatory kinases ULK1 and ULK2 [35,36]. ULK1/2 mediate autophagosome formation downstream of PINK1/Parkin [37,38]. Dichlorocopper would not have been identified had our mitophagy screen focused on Parkin recruitment, but nevertheless enhanced mitophagy, so the design of this screen may be superior to our previous screening approach [22]. The recovery of compounds with established mitophagy-promoting effects gave us confidence in the predictive power and efficacy of our dual screen. By filtering out compounds that induce mitochondrial damage or apoptosis, our effort was focused on identifying new compounds and mechanisms leading to mitophagy enhancement.
Interestingly, the set of compounds used to train our in silico model to identify mitophagy enhancers largely consisted of SIRT1 agonists (S1C Fig). SIRT1 affects mitophagy by up-regulating the mitophagy receptor BNIP3 in aged mouse kidney [39]. Several compounds function by increasing the cellular NAD+ pool, which is a cofactor for SIRT1. This bias led us to speculate that our dual screen would identify several more mitophagy enhancers that function through this common mechanism shared among the training set. Despite the bias, only 1 of the 3 final hits (3-methoxybenzamide) was a SIRT1 agonist. Interestingly, a recent Phase I clinical trial has demonstrated efficacy for nicotinamide, an NAD+ precursor, in PD, so the mechanism of action encompassed in the training set is likely nevertheless a relevant disease target [40].
Ultimately, the screen identified probucol, a drug used to treat hypercholesterolemia prior to the advent of statins. The target, which probucol inhibits, is the ATP-binding cassette transporter ABCA1, and diverse assays in cells and flies, which involved genetically reducing ABCA levels, abolished probucol’s effects on mitophagy in cells and climbing in flies, while expressing human ABCA1 reduced mitophagy in dopaminergic neurons [41]. Experiments probing distinct steps in the mitophagy pathway found that probucol impacted mitophagy and in vivo phenotypes independent of PINK1/Parkin but required LDs, as pharmacological inhibition of LD biosynthesis abrogated probucol’s mitophagy enhancing effect.
Under normal conditions, probucol had several relevant effects: It (1) increased LD-mitochondria contacts; (2) increased late endosomes/lysosomes; and (3) increased autophagosome lipidation. Importantly, probucol did not increase mitolysosome abundance under normal conditions but does so following mitochondrial damage. LDs adjacent to mitochondria can supply fatty acids during nutrient stress [42] and can interact with and transfer lipids and proteins to both lysosomes and autophagosomes [43,44]. LD mobilization by lipase PNPLA5 is required to facilitate the formation of autophagic membranes, including in the context of mitochondrial damage [45]. Given that we observed increased abundance of late endosomes/lysosomes and mature autophagosomes under basal conditions, we speculate that the latter is true and may occur adjacent to mitochondria under basal conditions, but further investigation is required. Ultimately, the increased abundance of two components, which can subsequently fuse to form mito-lysosomes when mitophagy is induced, likely primes the cell for a more efficient and protective degradative response.
Cells and flies were subjected to prolonged mitochondrial damage in the form of CCCP treatment or paraquat feeding. Under these treatment conditions, we observed an increase in LD abundance. This is consistent with studies by Nguyen and colleagues and by Long and colleagues, which demonstrate increase LD levels upon starvation-induced autophagy and deferiprone-induced mitophagy [28,29]. The two studies attribute different roles for LDs in these contexts. LDs buffer lipotoxic species, which are generated as a byproduct of autophagic degradation and facilitate the subcellular transition endolysosomes undergo from the peripheries of the cell towards damaged perinuclear mitochondria, respectively. We did not investigate the reason for LD expansion occurs upon mitochondrial damage, but we did find that LDs were necessary for probucol’s effects on mitophagy.
Lysosome position is critical for effective macroautophagy and found to be disrupted by inhibition of LD biosynthesis in deferiprone-induced mitophagy [29,46]. The increased abundance of LDs at mitochondria may facilitate the positioning of lysosomes away from the peripheries where they become active [46,47]. In vivo, clear overlap between LDs and lysosomes was visible both in dopaminergic neurons and in other cells of the fly brain, supporting this possibility.
Interestingly, LD expansion following mitochondrial damage was reduced in probucol-treated cells and flies. This feature of probucol’s mechanism may be particularly relevant, given recent studies that found increased LD accumulation in the dopaminergic neurons of PD patients [48]. Likewise, the reduced accumulation supports the idea that LDs may be mobilized by probucol treatment to facilitate mitophagy [45]. Probucol mitigated both LD accumulation and mitochondrial damage—two features of PD pathogenesis. Our data points to probucol’s canonical target, ABCA1, are required for both probucol’s effects on mitophagy and on LD expansion. Targeting a point of crosstalk between these two pathogenesis pathways may be advantageous.
ABCA1 R219K gene polymorphisms impacts PD progression, as measured using the Hoehn and Yahr scale [49]. The K allele, which is associated with slower PD progression, also affects the lipid profile of those who carry the genotype. Compared to individuals carrying ABCA1 R219K RR or RK, ABCA1 R219K K genotype carriers have elevated high-density lipoprotein cholesterol and lower triglyceride levels [50]. It remains to be determined whether the differences in lipid profiles are responsible for clinical differences between these groups.
Across toxin-based and genetic models of mitochondrial damage and PD in two different species, probucol improved survival, locomotor function, and reduced the loss of dopaminergic neurons. A prior paper also found probucol-induced improvements to phenotypes caused by increased free radical production in C. elegans. In this study, how probucol facilitates these improvements was not examined but was postulated to occur via its antioxidant properties [51]. Given probucol’s promising effects in several preclinical animal models, it might be fruitful to mine human epidemiologic data for any associations between probucol treatment and reduced risk of PD, since it is still in use in Japan and China. Statins also increase mitophagy, but in a Parkin-dependent manner, unlike probucol [52]. Whether statins impact LD expansion following mitochondrial damage may represent an interesting avenue for future inquiry.
In conclusion, our study showcased a dual in silico/cell-based screening methodology, which identified known and new mechanisms leading to mitophagy enhancement. ABCA1, which localizes to endolysosomes in cells and regulates lipid homeostasis, likely acts as a mediator of crosstalk between LD dynamics and mitophagy since LDs are required for the mitophagy enhancement conferred by probucol [53].
[END]
---
[1] Url:
https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3001977
Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.
via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/
