AI-guided screen identifies probucol-mediated mitophagy enhancement through modulation of lipid droplets

  1. Natalia Moskal
  2. Naomi P. Visanji
  3. Olena Gorbenko
  4. Vijay Narasimhan
  5. Hannah Tyrrell
  6. Jess Nash
  7. Peter N. Lewis
  8. G. Angus McQuibban

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

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 and mitochondria. Conversely, lipid droplet expansion, which occurs following mitochondrial damage was suppressed by probucol and probucol-mediated mitophagy enhancement required lipid droplets. Probucol-mediated lipid droplet dynamics changes may prime the cell for a more efficient mitophagic response to mitochondrial damage.

Article activity feed

  1. Review Commons

    Note: This rebuttal was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    The authors do not wish to provide a response at this time.

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    This study finds that probucol enhances mitophagy and mitochondrial quality control, suggesting its target ABCA1 may be a good target for the treatment of Parkinson's disease. Using an AI approach based on reported enhancers of mitophagy, they identify a list of 79 compounds to test in an assay based on Parkin- and CCCP-mediated mitophagy. One of the hits is probucol, an inhibitor of ABCA1, an ATP-dependent lipid transporter. Probucol causes a slight increase in mitochondrial turnover in HeLa cells, and increases Paraquat-driven mitophagy in flies. To explore the effects of lipids on the probucol-induced effects on mitophagy, it is observed that overexpression of human ABCA1 in flies suppresses the effect of Paraquat, and that inhibitors of DGAT reduces mitophagy as well, which may result from altered lipid droplets metabolism. Probucol alleviates motor phenotypes in fly and zebrafish PD models induced with mitochondrial dysfunction. The rough-eye fly phenotype observed with overexpression of Rho-7 (PARL ortholog) is modulated by ABCA expression, in addition to modulation of Parkin/PINK1. Finally, probucol appears to work at late stage mitophagy by modify LC3 lipidation and increasing lysosomal areas.

    Overall, this is an interesting paper, which makes a good case for the consideration of probucol as a potential treatment for PD. This is an improved version of a manuscript from the same group I had reviewed for another journal, and many of the issues raised have been resolved since. Yet, I remain to be convinced about some points:

    Major points

    1. The data provided does not support the claim that the AI selection of compounds had anything to do with the "high-rate" of success. To be able to claim this, there should be a comparison with a randomly selected set of ~80 compounds from the 3231 candidates, and show that for those, there is a lower chance of finding hits, if any at al. This section needs to be improved and a more compelling case presented. Just to be clear, it doesn't take anything away from the rest of the paper, but the authors make claim about this AI method that I don't believe are justified. Either provide more data or change the text and tone down the claims about the AI method.
    2. The effects of probucol on LD area and mitophagy in cells are interesting, but the data presented are not sufficient to claim that the effects are ABCA1-dependent. To do this, they must measure LD area and mitophagy in cells with shRNA against ABCA1, followed by CCCP and probucol (or DMSO). Probucol should have no effect when ABCA1 is knocked down. The experiments shown in Appendix Fig. S4 don't exactly show that; S4C lacks probucol treatment, and S4D/E measure mitochondrial volume, which is not mitophagy. Furthermore, it would important to show at least one dose-response curve for probucol in mitophagy, in cells. What is the IC50? Does it match that of the known Kd for ABCA1? This would be further evidence that the effects of probucol are "on-target".
    3. In Figure 5A, what is the baseline climbing % WITHOUT paraquat? This is important to show to provide a comparison point for the level of rescue induced by probucol.
    4. The epistatic relationships for the rough eye phenotype (REP) are confusing and need to be better explained/presented. A diagram showing the effect of each gene on the phenotype would be useful. Also, given that the effect of probucol in cells are Parkin/PINK1-independent, it is somewhat confusing to find that hABC1 effects are Parkin-dependent for the REP. Please clarify.
    5. In Figure 7E, again, data for the baseline climbing % is lacking. Furthermore, it is not clear why the % climbing of ABCA RNAi+DMSO+PQ (bottom graph) is higher than the control condition mCherry RNAi+DMSO (top graph). Are these % for the bottom and top graphs comparable? If not, then the bottom graph should also include a baseline condition without paraquat. Finally, why is paraquat not reducing climbing in ABCA RNAi or Atg7 RNAi? Controls are lacking in these experiments (as in Figure 5A above).

    Minor point

    1. In Figure 5C, experiment with SR3677 is shown, but mentioned nowhere in the text. I understand this is a ROCK inhibitor, but this should be mentioned.

    Significance

    Nature and significance: This is an interesting report, but there is no major conceptual advance in our understanding of mitophagy and neurodegeneration. Nonetheless, the discovery of a new potential target (ABCA1) for treating mitochondrial dysfunction of worth reporting.

    Comparison with existing knowledge: the report is original, but there have been a few reports of the effect of perbucol in Parkinson's model (e.g. Ray et al 2014, Cell Death Discov, PMID 24407237), or ABCA1 variants (Ya & Lu, 2017, Med Sci Monitor). Those should be acknowledged.

    Audience: researchers in the field of Parkinson's research, but also mitophagy and lipid metabolism.

    My expertise: Parkin/PINK1 pathway, mitophagy, pharmacology, structural biology.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    In their manuscript, Moskal et al. performed an AI-based screen to find potential mitophagy inducers and subsequently test them in model organisms of human neurodegenerative disorders, whereby mitophagy induction might be beneficial. The authors found that probucol, a retired lipid-lowering drug, is a candidate mitophagy inducer at least in cell culture and flies. Furthermore, they reported physiological benefits in association with probucol treatment to flies fed paraquat, a toxin that causes loss of dopaminergic neurons. This manuscript has some potential especially as it implicates an unknown target for mitophagy, ABCA. However, the paper can't be published in its current form for the several reasons:

    Major comments:

    The sequence in which the experiments were presented in this paper is illogical and sometimes confusing: (i) in page 5, when the mitophagy-inducing effect of probucol was shown in paraquat-treated flies, the corresponding physiological relevance of this observation, ie effects on climbing ability and survival, should have been reported immediately and not later on in another section of the manuscript. Since paraquat induces mitophagy, It is not very intuitive to imagine that further increase in response to probucol is beneficial. This leaves the reader wondering whether further increase in mitophagy is beneficial or detrimental. In fact, excessive mitochondrial clearance has been shown to render caloric restriction - the most robust antiaging and autophagy-inducing intervention, detrimental (PMID: 30929899). (ii) Similarly, causality testing of the role of ABCA-autophagy axis in mediating the salutary effects of probucol in flies, which were reported in the very end of the manuscript should have been mentioned early on upon introducing such health-promoting effects. (iii) Another example is also LC3 lipidation and lysosome abundance, which support the pro-autophagic effect of probucol. Instead of being reported with other evidence supporting mitophagy induction at the beginning of the manuscript (page 5) were kept for no obvious reason to be mentioned in the very end of the results (page 9) after all other mechanistic and in vivo physiological testing results were reported!

    As such, the manuscript needs extensive re-writing, thereby also removing unnecessary data that only distracts from the main message. By that, I mainly refer to the rough eye model induced by expression of rhomboid-7 in the fly eye. The sudden shift towards using the rho-7 fly eye model along with extensive characterization of the role of pink1 and parkin is distracting from the main objective of the study: finding a novel mitophagy inducer in probucol and describe its mechanism of action. Instead of reporting causality testing of autophagy in paraquat-treated flies, where probucol induced mitophagy, the authors decided to discuss the fly eye model and go on to show the role of mitophagy, pink1 and parkin in this model. I suggest removing this model from the current manuscript and dedicate a future manuscript to the experiments performed there because they do not serve the purpose of the study. In fact, mitophagy induction in this model is harmful. As such, this model only dilutes the message and lowers confidence in the robustness of the reported cascade of events involving probucol>ABCA>mitophagy>physiological benefits.

    Regardless, the experiments testing causality of autophagy and ABCA in Fig. 7e are misleading. There are no corresponding negative controls showing the benefits of probucol in absence of ABCA and Atg7 RNAi. The authors should not rely on a different set of experiments (Fig. 5a-b) done at a different time and another cohort of flies.

    The authors applied an interesting screening approach using semantic textual similarity to a pre-defined list of positive controls. These positive controls comprised of 7 mitophagy inducers that primarily act on the NAD+-sirtuin pathway. NAD+, as also sirtuins, induce a myriad of effects, not only mitophagy. In fact, some might even argue that autophagy/mitophagy are only partly involved in the effects of these compounds (PMID: 34843394). Why did not the authors expand their positive control list to include other classes of mitophagy inducers? the reference (ref. 10) used has many other potential positive controls that do not necessarily increase apoptosis and mitochondrial damage. Furthermore, the sematic screen has been limited to papers published until 2014. Why was that the case? In fact, a crude Pubmed (not Medline) search using the term "mitophagy" returns 935 hits (till 2014), while from 2015 onwards it returns more than 5800 papers! This clearly shows that much more has been done after 2014. The search should be expanded, otherwise the authors are missing out on all major development that happened in the field.

    The authors also reported that they filtered out the candidate molecules with any association to either the term "apoptosis" or "mitochondrial damage". How could you differentiate between causation or protection from mitochondrial damage when a compound is mention in the context of apoptosis/mito damage using semantic fingerprint?

    How does ABCA1 KO and overexpression affect the pro-mitophagic action of probucol? This has been shown for DGAT, but not for ABCA which is a central finding in this manuscript.

    • In page 1, the following statement is problematic: "we focused on the ultimate endpoint of mitophagy-clearance of damaged mitochondria from cells. Ultimately, if this step is improved, then the negative consequences of mitochondrial damage in the dopaminergic neurons may be mitigated". Increasing the clearance of damaged mitochondria is not necessarily beneficial. In fact, excessive clearance of damaged mitochondria renders autophagy inducing interventions harmful as mentioned above. It is also not clear whether and how the authors could tell that only damaged mitochondria were sequestrated. Even in absence of mitophagy induction at baseline (in absence of CCCP), this does not imply that only damaged mitochondria are cleared when damage is induced. Lack of evidence is not an evidence for lack of an effect.
    • In page 4, the authors report that "the degradation of VDAC1 was increased at several time points by probucol treatment, but not in the absence of mitochondrial damage (Figure 2D, E)" the graph shows DSMO and probucol not CCCP vs DSMO, is this a typo?
    • Fig 7A-B is very confusing. Why did not the authors use protease inhibitors to properly evaluate autophagy flux if that was the purpose? also how come starvation does not induce autophagy in these cells? what was the time point tested? Were the cells starved as a positive (or worse: negative) control?
    • The authors reported that their screen could efficiently predict olaparbi upon leaving it out (top 3.9% of the 3231 compounds screened). How did the other 6 positive controls fair in this cross validation using the leave-one-out approach? please report this for the other 6 positive compounds as well.

    Minor

    • rather small sample size in most of the experiments with n=3-4 (eg, Fig 5 D, E F and G).
    • Some figures do not show the data points and thus it is not possible to tell how big was the exact sample size (eg, Fig. 5A)
    • how do the authors explain the discrepancy between Fig. 4G and Fig. 3D where PQ increased % of red-only/total mito area in one but not in the other?
    • How can the results in Fig 6 quantified? If qualitative then many more representative images should have been presented.
    • There appears to be a deleted lane from the western blot in Fig S2A. This needs to be declared in the legend, along with a justification.
    • Western blots in general do not show the molecular weight. Please add this throughout the manuscript
    • A lot of abbreviations are mentioned without spelling out what they stand for, eg, CCCP, ABCA, DGAT.
    • Typo in page 1: the loss of which is (not are) responsible for the classical motor..
    • Line numbers were not included in the submitted manuscript, thus I could not provide the authors with the exact position of any of the issues mentioned above.

    Significance

    Despite reporting novel findings, the manuscript has major flaws in terms of experimental approach and presentation of the results. The authors need to revise their screen, as also restructure the manuscript to be better reflect the findings, thereby improving the significance of both the physiological and mechanistic value of the study.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    Summary:

    In the manuscript by Moskal et al, the authors utilize an AI-based approach to identify mitophagy modulating compounds based on parameters from already FDA-approved drugs. From this screen, the authors find that the compound Probucol enhances stress-induced mitophagy. It appears to do this via its canonical target, ABCA1, via a PINK1/Parkin-independent mechanism. Importantly, the authors show that probucol improves phenotypes in fly and animal models with mitochondrial perturbations, suggesting that this pathway may hold some therapeutic value.

    Major comments:

    • I will mention that I have already reviewed a version of this manuscript at another location and here the authors have addressed my previous concerns.
    • The key conclusions are robust and shown in multiple cell types and models.
    • I do not have any requests for additional experiments, given that those present already sufficiently support the claims of the paper.

    Minor comments:

    • I am a little confused over the descriptions of the rough eye phenotype in Fig. 6. In the text on p7, the authors state that rho-7 overexpression causes the eyes to appear rough. Yet in the figure the eyes appear smoother - indeed the legend says they have "glossy appearing eyes". This seems the opposite to rough - unless I am missing something? Perhaps rewording the text would help with this.

    Significance

    • Here the authors identify a small molecule, probucol, that enhances mitophagy and in doing so implicate the ABCA1 pathway of lipid transport in this process. A great strength of the work is that in vivo data (in flies and fish) are obtained, which has clear disease-relevance. The manuscript is timely, in that lipid signaling and lipid droplets are being shown to have a strong regulatory role in autophagy.
    • The authors use a novel AI-based approach to identify mitophagy-inducing compounds, and while this seems a good approach, I do not feel I have the expertise here to critically review this aspect. It is also noteworthy that the initial assay was designed to look at modulators of Parkin-dependent mitophagy, but the compound discovered appears to act independently of PINK1/Parkin.
    • My main expertise is in autophagy and mitophagy cell biology and I do think the findings in this manuscript will be of interest to this field and those focussed on therapeutic approaches for diseases where impaired mitochondrial function has been implicated.
    • The manuscript does lack a mechanistic understanding of the mode of action of probucol - how it is enhancing mitophagy is not clear. However, it is up to the authors and editors of the destination journal as to whether more work is done here, or in a follow-up study.
  5. Version published to 10.1101/2022.05.18.492560v2 on bioRxiv
  6. Version published to 10.1101/2022.05.18.492560v1 on bioRxiv