Identifying and targeting abnormal mitochondrial localization associated with psychoses

  1. Marzieh Haghighi
  2. Donna McPhie
  3. Mohammad Rohban
  4. David J Logan
  5. Johan Fredin Haslum
  6. Jessica D Ewald
  7. Kyle W Karhohs
  8. Erin Weisbart
  9. Beth A Cimini
  10. Shantanu Singh
  11. Bruce M Cohen
  12. Anne E Carpenter

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Therapeutics working by novel mechanisms are needed for patients with psychiatric conditions. Cell-based assays to identify candidates that reverse observed abnormalities could accelerate the process. Here, we imaged peripheral cells (skin fibroblasts) of 168 patients, stained for DNA, actin, and mitochondria. We found mitochondria tend to be farther from the cell border for patients who experience psychosis (including subsets of individuals with bipolar disorder, schizophrenia, and schizoaffective disorder). We observed a reverse trend, albeit not statistically significant, for patients diagnosed with major depression. Because the mitochondrial dispersion phenotype could be identified by a single metric, we were able to readily query existing databases of cells stained for their mitochondria and treated with various chemical or genetic perturbations. We identified compounds and genes both negatively and positively affecting the psychosis-associated phenotype, including some known to impact psychiatric conditions. Developing therapeutics with novel mechanisms is a complex multi-step challenge. This cell-based assay holds promise for virtual and physical screening to identify candidates for treating psychiatric conditions.

Article activity feed

  1. Review Commons

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

    Learn more at Review Commons

    Reply to the reviewers

    The authors provide a full response to review alongside the revised manuscript.

  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 builds upon previous work in schizophrenia and other disorders using fibroblasts derived from patients, assessing mitochondrial phenotypes and then using these to identify compounds which reverse these phenotypes. The study is one of the largest of its kind performed to date with 168 patients included. The authors undertake mitochondrial phenotyping and machine learning of the outputted images to be segregate the patients based on clinical features and the associated cellular phenotype. The authors then go on to screening virtually publicly available datasets of cancer cells treated with compounds and also genetic modulations. In doing so, they can identify compounds which modulate the phenotypes and therefore might be of value to test in the patient derived lines. The study has strengths in the large number of samples, the advanced machine learning and the virtual screening. Furthermore, the authors highlight and discuss the limitations of the study well. There are some weaknesses which the authors can address. Firstly in the introduction, although it is comprehensive in some areas, in other areas for example outlining the fibroblast mitochondrial phenotype and indeed the use of patient fibroblasts to identify compounds, there is significant literature missing, particularly in Parkinson's Disease where screening in fibroblasts has resulted in compounds entering Phase 3 clinical trials. In addition to the studies using 100 or more PD patient fibroblast lines for phenotyping and patient stratification have not been included. It would be useful if the authors could comment on the robustness of the phenotypes identified in the fibroblasts over multiple passages. This is important when considering the biological and disease relevance of the phenotypes and it is not something the authors show or comment on. In discussing the genetic manipulations it would be useful to comment on the genes identified in more detail particularly those which are not known to be associated with changes in mitochondrial phenotypes.

    Significance

    This study builds on work from multiple labs investigating the utility of fibroblasts to identify phenotypes and find potential novel therapeutics. The size of the cohort and the advanced machine learning methods are a particular strength and this advances the field in this area. The availability of the data and code is a strength to allow others to replicate the findings. The lack of experimental validation of any of the compounds or genes identified by the virtual screening is a weakness which could be addressed.

  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 study, Haghighi et al. seek to build upon prior literature linking alterations in mitochondrial network distribution with various kinds of psychosis. Correlations between subcellular mitochondrial localization and different psychological states is an interesting and potentially fruitful frontier and should be explored; however, despite their ambitious strategy to screen 168 skin fibroblasts from patients experiencing psychosis, and examine various online image databases, there is a concerning number of issues related to the image-analysis approach. The foremost of these is a lack of direct measures of mitochondrial distribution, which might serve to validate their proposed MITO-SLOPE protocol. There is also a worrisome lack of robust controls, which are critical in light of how admittedly subtle some of the distribution phenotypes may be. Overall, the aim to screen differences in mitochondrial distribution is a laudable goal and, in the context of psychological disorders, could be helpful in identifying new therapeutic targets; but the methodology employed in this study does not seem to be sufficiently rigorous to be able to leverage this approach for screening purposes.

    I have extensive experience investigating mitochondria with advanced imaging technologies, including super-resolution microscopy as well as high-throughput and 4D imaging modalities. I am also familiar with standard as well as machine-learning approaches for quantifying mitochondrial morphology as well as distribution or trafficking. In my opinion, this study requires substantial revision, both in terms of the indirect and often opaque image-analysis pipeline as well as the inclusion of orthogonal experiments, which could serve to lessen concerns regarding purported differences in mitochondrial distribution, which are so difficult to discern as to be imperceptible. It is worth noting, too, that this study appears to be predicated, in many ways, upon a 2010 study (Cataldo et al.) of mitochondria in patients with bipolar disorder, which appears to reflect its own lack of critical controls for cell size.

    Major comments:

    The authors state, in the first paragraph of the results section: "By eye, we observed that samples from patients in the control and MDD categories show a more fine-grained, dispersed mitochondrial network extending to the edges of the cell, whereas patients in the categories experiencing psychosis tend to show an agglomerated, thicker network more concentrated around the nucleus. The pattern is subtle and heterogeneous across a cell population." The pattern is indeed subtle. I am concerned that it is so subtle as to be imperceptible. Firstly, it is important to note that the mitochondrial reticulum in BP, SZ, and SZA is more difficult to differentiate, by eye, because the signal appears to be saturated in places, such that the boundaries of individual mitochondria are indistinguishable due to differences in contrast or possibly from the fluorescence intensity itself. Although the authors indicate in the legend that the intensity of the mitochondrial fluorescence was adjusted "for visual clarity," it appears that the contrast needs to be decreased in the BP, SZ, and SZA conditions. It is also important to note that MitoTrackers load into mitochondria in a membrane-potential-dependent fashion. Did the authors detect differences in membrane potential between these groups? While imaging, was the same laser power and gain utilized from condition to condition? With this being said, it is not clear that mitochondria in control and MDD categories have different morphologies from the other conditions. It is also not clear what "fine-grained" means in this context. Is this a comment on aspect ratio? If so, it would be better to use standard terminology. (Why are there large red circular structures in the nucleus? These are likely not mitochondria, so why are they showing up in the channel with MitoTracker?) It is also not evident that one condition has more dispersed mitochondria than another. Given that the authors appear to be making this a central claim of their manuscript, it would seem appropriate to highlight specifically the regions of the different cells that they believe exhibit meaningful differences. If I attempt to look at the merged image, which is important because it is really the only way that one can gauge the relative distance of the mitochondrial network from the edge of the cell, there would seem to be no obvious differences between the conditions. Another key point that I think important to mention, given that it is frequently referenced in this manuscript, Cataldo et al., 2010 indicate that mitochondria in patient fibroblasts with bipolar disorder (BD) are more perinuclear than those in control. However, a cursory inspection of the images from this study (e.g., Figure 2A-B; Figure 4A-D; and Figure 6A-H) unambiguously demonstrate that the BD cells are smaller than the control cells. Of course, if the cells are smaller, the distance from the nucleus will tend to be shorter. In Cataldo et al., 2010, the authors state, "We also measured cell area, cell length, cell width, and cell perimeter of the fibroblasts used in this analysis to verify that the observed mitochondrial distributional differences were not simply a result of BD cells being smaller, shorter, or fatter. No significant differences in any of these measurements were seen based on diagnosis after two sample t tests." Notably, the data is not shown, so it is difficult to appreciate what the variance of the population of cells from control and BD would look like, but it must be said, nevertheless, that the representative images in this paper all point to the BD cells being smaller. In light of this, it would be helpful if Haghighi et al. could add scale bars to all the images (e.g., in Figure 2), so readers can ascertain whether all the cells are portrayed at the same scale and are of similar areas.

    As the authors indicate, interpretable measures of mitochondrial morphology include values like size and shape. It is concerning, therefore, that Figure 3 purports to identify a number of significantly different mitochondrial "features" in the patient groups experiencing psychosis, but they do not appear to make an effort to clarify how any of these features might reflect ground truths of mitochondrial architecture, which can be understood directly by values such as aspect ratio, circularity, area, number organelles, number of nodes or branching points in a network, etc. Unless the authors can specifically tie their machine-learning classifications to standard mitochondrial shape descriptors, their classifications will remain opaque and therefore of limited credibility or value. One way to improve the validation of their machine-learning classification methods would be to use empirically sound methods for manipulating a mitochondrial morphology and distribution, which could serve as positive or negative controls. For example, treatment of cells with the uncoupler FCCP would induce mitochondrial fragmentation, treatment with cycloheximide results in stress-induced mitochondrial hyperfusion (SIMH), or treatment with Nocodazole would block mitochondrial trafficking. Treating control cells with these chemicals would help to establish baseline measurements for how far the patient cells are deviating from untreated controls, in one direction or another. Such considerations, I think, are especially important when the mitochondrial phenotypes are so subtle. I agree with the authors' argument that, for the purposes of screening, it is best to focus on a single metric. Based on their apparent discernment of the subtle differences in mitochondrial distribution in patients experiencing psychosis, they opted to examine possible differences in network density. To this end, they developed "MITO-SLOPE." Out of multiple categories of features, they highlight the following as the most powerful for establishing differences in mitochondrial network density:

    "(a) A subset of texture measures in the nuclei and cytoplasm area of the mito channel. (b) A subset of features measuring the intensity of the mitochondria area across the cell."

    Within the concentric bins around the cell nuclei, they measure:

    • FracAtD: Fraction of total stain in an object at a given radius.
    • MeanFrac: Mean fractional intensity at a given radius, calculated as the fraction of total intensity normalized by the fraction of pixels at a given radius.
    • RadialCV: Coefficient of variation of intensity within a ring, calculated across 8 slices."

    While the authors have recommended the use of a single metric for purposes of screening, MITO-SLOPE appears to represent a bundle of metrics, which, in the end, do not amount to a clear readout of what is being measured. From my point of view, if one were interested in measuring mitochondrial distribution, then, in an ideal situation, one would measure the average distance of all the mitochondria from the center of the nucleus. And, since the size of the cell is critical for establishing relative distances to the boundaries or periphery of the cell, one would normalize this metric by cellular area. Thus, the readout would be: [average mitochondrial distance from the nuclear center (µm)]/[cellular area (µm2)]. An even simpler metric could be: [average mitochondrial distance from nuclear center (µm)]/[average cytoplasmic radius (µm)]. When talking about mitochondrial distribution, we typically think in terms of where is the mitochondrial network, on average, in relation to the nucleus (perinuclear) or to the edge of the cell (peripheral). By quantifying the actual mean distance of the mitochondrial network in relation to both the nucleus and the bona fide cell extremities, via the metrics I described above, one can obtain direct measurements of the truly meaningful values related to mitochondrial distribution. It seems deviating from these approaches introduces more and more opportunities for confounding variables.

    However, the MITO-SLOPE analysis does not seem to consider this metric. Is this, or a similar variation, not the most direct way to establish differences in the mitochondrial network distribution? I would, of course, at least want to see a discussion of why the authors have not chosen to use the most direct form of quantification for this purely spatial value. Why opt for a multifaceted measurement of a relatively straightforward quantity, when a simpler form of quantification would not only suffice but arguably be more likely to capture the ground truth? With this being said, it is not clear to me why, within MITO-SLOPE there seems to be a reliance on measuring the "intensity" of the mitochondria. (And what intensity is it? Mean intensity per ROI?) Of course, particularly if MitoTrackers were used for staining mitochondria, there will be heterogeneity in fluorescence intensity from organelle to organelle, which introduces potential confounders into the workflow. Furthermore, as indicated above, to know if the subcellular distribution of mitochondria is truly altered, it is essential to know if the cell size has likewise changed. Therefore, any unbiased measure of mitochondrial distribution must take into consideration the size of the cell; however, based on the information provided about MITO-SLOPE, it does not appear that the authors are accounting for possible variations in cell size that might account for alterations in mitochondrial network distribution - i.e., a smaller cell will have a more constrained area in which mitochondria will be able to disperse - thus, not accounting for cell size (area) will yield ambiguous results. For example, how can we know if mitochondrial motility is impaired or if the cell is simply smaller and there is less space in which to move? Another complexity, here, is if the cell boundaries were not accounted for via staining of actin, etc., then establishing a true cell boundary will be very challenging. How many bins are sufficient to capture the whole cell? Just 12? Furthermore, human fibroblasts have a tendency to be quite large (sometimes several hundred microns from end to end); how can the authors account for the whole cell, particularly in cases where part of the cell is beyond the field of view or cells are growing on top of each other, as is often the case?

    In Figure 6, there is no control image that could be used as a frame of reference. I have extensive experience imaging A549 cells. The mitochondria in these images appear to be highly fragmented. The staining patterns, particularly of the cells treated with divalproex-sodium, are quite dim, indicating mitochondrial depolarization. Of course, depolarization affects the fluorescence intensity of mitochondria stained with vital dyes, such as MitoTrackers, which will, in turn, presumably affect the values obtained from MITO-SLOPE, which appear to rely on intensity gradients, rather than more concrete spatial coordinates. Also, as indicated above, it is unclear how the authors are establishing the edges of cells without a marker of the plasma membrane or cytoskeleton.

    The authors note that "Divalproex-sodium is a benzodiazepine receptor agonist and HDAC inhibitor (Rahman et al. 2025) used to manage a variety of seizure disorders (Willmore 2003) and bipolar disorder(Bond et al. 2010; Cipriani et al. 2013); it shows a positive MITO-SLOPE which is the direction expected to normalize the centralized mitochondrial localization associated with psychosis." Insofar as this recommends the drug for use in "normalizing" perinuclear mitochondria within neurons, it would seem only prudent to mention that this drug also appears to induce mitochondrial depolarization and fragmentation, which are both associated with a range of severe human pathologies. I would caution the authors to not highlight one potential benefit while omitting an obvious side effect involving what appears to be significant perturbation of mitochondrial structure and function. What is the point of normalizing mitochondrial distribution if the mitochondria being redistributed are dysfunctional?

    The authors note, in Figure 7, that their MITO-SLOPE analysis was unable to discern a statistically significant difference in cells with specific knockouts of genes associated with mitochondrial trafficking. If the MITO-SLOPE cannot discern a difference in the context of a substantial abrogation of mitochondrial transport capacity, how is it that it could detect meaningful differences where there is only a "subtle" change in distribution? This result would seem to militate strongly against the efficacy of this analysis pipeline and would not recommend its use for unbiased screening and discovery.

    Minor comments:

    For Figure 6 b and c, "µm" should be "µM."

    The introduction and discussion could be more concise.

    Significance

    This study attempts to fill an important gap in knowledge relating to mitochondrial distribution and psychological disorders. It aims to perform an initial screen to try to validate a novel analysis pipeline called MITO-SLOPE, however, the study appears to lack analytical rigor, both in terms of the underlying cell biology together with the approach for quantification, itself. Conceptually, this study has great promise, but the authors will need to improve their pipeline prior to publication, which will likely require fundamental revisions, including an array of orthogonal measures (largely lacking here) as well as detailed demonstrations of how the segmentation actually works and ultimately yields data reflecting demonstrable mitochondrial trafficking/distribution defects.

  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:

    The manuscript by Haghighi and McPhie et al. builds upon their previous findings by exploring the mitochondrial localization as a disease-associated phenotype in mental disorders, particularly in psychotic disorders. They recruited a cohort of patients diagnosed with schizophrenia, schizoaffective disorder, bipolar disorder and MDD. By taking advantage of skin biopsies, they screened patient-derived fibroblasts for aberrant mitochondrial localization and morphology using common staining techniques. Then, they use a machine learning approach to classify patients into their respective groups, which was effective for BP, SZA and pooled psychotic patients. Authors then develop a single feature for phenotyping, Mito-SLOPE, a metric of mitochondria density distribution across a cell by radial areas. With this metric, psychotic patients tend to have more nuclear-localized than edge-localized mitochondria; whereas MDD patients show a trend for higher edge-to-nucleus distribution. To find candidate drugs, authors screen publicly available datasets of cells treated with small compounds using mito-SLOPE. Furthermore, authors then apply mitoSLOPE on a CRISPR screen dataset, showcasing the role of mitochondrial dynamics genes and three genes of interest because of their association with psychosis. Finally, they identified the top genes whose KO or overexpression may explain (or reverse) the mitoSLOPE phenotype.

    Overall, the manuscript is well-written, the conclusions are supported within their limitations and this work represents an advancement in the field. I recommend it for publication provided these concerns are addressed:

    Major comments:

    1. The mitoSLOPE measure is very interesting and most likely reflects a subtle changes in mitochondrial transport. How does the microtubule network look like in the patient fibroblasts, are there obvious alterations in e.g. their posttranslational modifications? Is there a difference in mito transport speed or pausing frequency?
    2. I concur with the exclusion of compounds that obviously alter cell shape, as the authors mention for the cancer therapeutics. Some cancer therapeutics actually affect microtubule dynamics (see 1st point), which may underlie their effect on both cell shape and mitoSLOPE. To undertand the mechanism of action, the top hits should also be tested for the integrity of the microtubular network and mitochondrial transport parameters.
    3. While I agree with the authors' reasoning that the observed phenotype could be a result of the disease or the result of a compensatory mechanism, their hypothesis could be experimentally tested by addition of any of the top hits in order to reverse mitoSLOPE in their patient cell lines. It may not have worked for Lithium in their last manuscript, but the mechanism of action of the novel compounds could be cell intrinsic.
    4. Does recreation of the CRISPR cell line in their hands produce the same phenotype?
    5. Additionally, the observed phenotypes could also be a product of the medication taken by the patients. Deeper patient data from the cohort may be relevant to put the findings in context. How were patients diagnosed? Which medications were the patients taking? Was substance abuse present? In Mertens et al, Lithium responders and Lithium non-responders showed a differential mitochondrial response, how does this affect their dataset?
    6. While MDD itself is not a psychotic disorder, it can still present with psychotic features. Was this evaluated during the recruitment? Also important, were they on antipsychotic medication in addition to antidepressant therapy?
    7. The fact that CACNA1C is excluded from the "unbiased" hit discovery (Fig 8) undermines the power of the filtering criteria selected by the authors. Authors should include some discussion around this.

    Minor comments:

    1. Colored images should be made colorblind-accessible. This applies to microscopy images and graphs.
    2. Fig 3: Exact p-values should be reported in the graphs
    3. Fig. 5 and Fig 7a-b: It is not immediately clear what the lines in these graphs represent. Is it the individual drug/gene hits in a pre-ranked manner?
    4. Fig 6 b-c: should the "m" be capitalized for Molarity?
    5. The annotation of divalproex/valproic acid as a "benzodiazepine receptor agonist" is incorrect. While it is known to enhance GABAergic neurotransmission, the mechanism is supported to be through GABA synthesis rather than being a GABA-A receptor agonist (see eg. PMID: 23407051).
    6. Supplementary Fig 3 and 4 could be swapped to match the main text order.
    7. One reference was inaccessible: Anon, Phenomics-Enabled Discovery and Optimization of Small Molecule RBM39 Degraders as Alternative to CDK12 Targeting in High-Grade Serious Ovarian Cancer (HGSOC).

    Significance

    Recently, mitochondria have emerged as mediators of anxious behavior and are increasingly studied in the context of neuropsychiatric disorders. However, the molecular mechanisms that connect altered mitochondrial performance to specific neuropathological conditions are unknown. This study extends our knowledge in this realm. While it is in principle an extension of earlier work from the authors (Cataldo, A.M. et al. Am. J. Pathol. 2010), it has added value due to the application of their automated analysis to publicly available datasets, providing a clear technical advance. This identified known as well as novel compounds that could revert the mitochondrial phenotype and makes this study specifically interesting to an audience interested in translational research. The strength of the manuscript certainly lies in the large number of examples studied and their well-rounded discussion of their findings. It is limited by the fact that the phenotype of neuropsychiatric conditions is studied in peripheral cells, and thus may not be a simple cell-autonomous response but a compensatory, systemic response that is not easy to replicate in a fibroblast in isolation. No mechanistic insight is gained on the underlying cell biology in the current format.

  5. Version published to 10.1101/2025.10.08.676630 on bioRxiv