For a primary screen, rat hippocampal neurons were transfected with a mitochondrially targeted DsRed (MitoDsRed) on Day in Vitro (DIV) 6 or 7 and on DIV 8 or 9 test compounds were applied for 1 h before imaging mitochondrial movement. The images were analyzed with our previously published algorithm called PATHS that tracks more than 10,000 mitochondria per well in a 96-well plate format (Shlevkov et al., 2019). For the analysis, two parameters were used: the mean of the integrated distance traveled by mitochondria and the Kolmogorov–Smirnov (KS) Z-score of the integrated distance traveled. The integrated distance represents how far each individual mitochondrion traveled within the imaged field, summing the movement in all directions rather than only the net displacement from its starting position (Shlevkov et al., 2019). The KS Z-score of the integrated distance represents the statistical significance of the integrated distance value in a compound-treated condition relative to the mock-treated condition. The efficacy of the screen was confirmed by comparing the vehicle (DMSO) to calcimycin, which arrests mitochondrial movement by elevating cytosolic Ca²⁺ (Supplementary Figure 1). We previously screened 1,641 compounds from the Selleck Bioactive collection and Biomol collection in this manner. Six enhancers of mitochondrial motility were found with Z-scores greater than 3 in each of two replicates: two actin depolymerizers (Latrunculin B and Cytochalasin B), three kinase inhibitors that acted through Aurora Kinase B (BMS-754807, Hesperadin, and TAK-901), and one peptidase inhibitor (AAF-CMK) (Shlevkov et al., 2019; Figures 1A,B,D). To discover additional enhancers, we returned to the output of that screen and selected 20 compounds that had a Z-score greater than 2. These compounds (Figures 1B,C) were rescreened at three concentrations: 0.3, 3, and 13 μM (Figure 1C) and from that rescreening 4 promising hits were brought forward for further investigation (indirubin, cantharidin, adefovir dipovoxil, and damnacanthal). We also used the screening platform to test in duplicate 240 compounds from a library of compounds with known mechanisms of action (Supplementary Table 1). We identified three, febuxostat, 7,8-Dihydroxyflavone hydrate (DHF) and RWJ-50271, as having Z-scores greater than 2 in both replicates (Figure 1D). Because RWJ-50271 is a frequently encountered false-positive, only febuxostat and DHF were brought forward for further testing. The six selected compounds were then rescreened after 1 h treatment with doses ranging from 1 to 10 μM (Figures 2A–F). In this rescreening, only indirubin and cantharidin had a reasonable dose/response relationship for enhancing the movement of the neuronal mitochondria (Figures 2A–F). To further examine their efficacy, we established their dose-response relationships over a broader range of concentrations (Figures 2G,H). The resulting estimates of their effective concentration (EC50) values were 1.3 μM for cantharidin and 1.5 μM for indirubin and these two compounds were selected for further investigation.

Although the primary screen quantified mitochondrial motility, it did not discriminate between anterograde and retrograde movements and, though biased toward analysis of axons, contained some dendritic mitochondria as well. Therefore, to characterize further indirubin’s effect on axonal mitochondria, we used the more conventional method of kymography on individual identified axons. Mitochondria in cortical neurons were labeled by transfection with MitoDsRed on DIV 6 or 7 and imaged 2 days post-transfection. Prior to imaging, neurons were incubated with DMSO (control) or 10 μM indirubin for 1 h. To quantify axonal mitochondrial movement, a 70–115 μm length of an individual axon was imaged for 3 min. Movement was analyzed using the Kymolyzer algorithm (Basu and Schwarz, 2020) to extract the density of mitochondria per μm of axon, the percent of mitochondria that were motile during the imaging period in each direction, and their velocities. The percent motile was 9.9% in control (DMSO) axons but increased to 20.9% with indirubin (Figures 3A,B). The enhancement of motility was evident in both directions: indirubin increased the anterograde percent motile from 5.5 to 12.5% and retrograde from 4.3 to 8.4% (Figure 3B). The average velocity of mitochondria was also increased for both anterograde from 0.10 to 0.21% and retrograde from 0.14 to 0.21% in indirubin-treated neurons (Figure 3C). Mitochondrial density along the axon was not appreciably altered (Figure 3D).

To determine whether indirubin influenced mitochondrial movement in non-neuronal cells, we applied a customized algorithm called QuoVadoPro (Basu and Schwarz, 2020) to HeLa cells transfected with MitoDsRed. This algorithm is well-suited for analysis of cells in which mitochondria can be in a reticulum rather than discrete organelles, and where microtubules are not arrayed in parallel like in an axon. The algorithm emphasizes progressive movement over jiggle and the pixel variance of the fluorescently tagged mitochondria serves as a proxy for motility. Compared with control Hela cells, the pixel variance of MitoDsRed from 3 min time-lapse movies was at most slightly increased (11%, p = 0.10) after the indirubin treatment (Figures 3E,F). We also examined the effects of indirubin on other organelle movement in Hela cells. The pixel variance of mCherry-Rab7, which labels late endosome/lysosomes, was not increased in indirubin-treated cells (Figures 3G,H). Thus, indirubin was most effective in neurons and may be selectively enhancing mitochondrial transport.

Indirubin inhibits glycogen synthase kinase 3 beta (GSK-3β) and cyclin-dependent kinases (CDKs) including CDK1, CDK2, CDK4, and CDK5 (Hoessel et al., 1999). In addition, an indirubin derivative, indirubin E804, that acts as type I CDK16 inhibitor, influences neurite outgrowth and vesicular transport within axons (Dixon-Clarke et al., 2017). To probe whether any of these kinases are the cellular target responsible for the action of indirubin on mitochondrial motility, we identified additional inhibitors of these kinases and tested them at three concentrations. GSK-3β was of particular interest because it has previously been shown to inhibit mitochondrial transport when activated (Yang et al., 2017). GSK-3β inhibitors can be categorized as either ATP-competitive or non-ATP-competitive inhibitors (Orena et al., 2000; Eldar-Finkelman and Martinez, 2011; Kalra et al., 2017; Levin et al., 2017; Dan et al., 2020). We selected four ATP-competitive inhibitors (AR-A014418, indirubin-3′-monoxime, 6-BIO, and TWS119) and two ATP-noncompetitive inhibitors (tideglusib and Li+) (Phiel and Klein, 2001; Noble et al., 2005; Tolosa et al., 2014) to test in our assay, but none of these resembled indirubin and enhanced mitochondrial movement; some caused a modest inhibition (Supplementary Figure 2A). We also tested compounds annotated as selective inhibitors of CDK1, CDK2, CDK4, and CDK5. Compared with the mock condition, mitochondrial motility was decreased after the treatment with AUZ (K03861), a selective CDK2 inhibitor (Tang et al., 2018; Supplementary Figure 2B). R547, roscovitine, and SU9516, three inhibitors reported to parallel indirubin in their inhibition of CDKs were also tested. R547 is an ATP-competitive inhibitor for CDK1, CDK 2 and CDK4 (DePinto et al., 2006), roscovitine, is an ATP-competitive inhibitor for CDK1, CDK2, and CDK5 that releases cytochrome c from mitochondria and induces apoptosis (De Azevedo et al., 1997; Edamatsu et al., 2000). SU9516 has been described as a potent selective inhibitor of CDK1, CDK2 and a weak inhibitor for CDK4 (Lane et al., 2001). None of these CDK inhibitors, however, enhanced mitochondrial motility in our assay and indeed several reduced it (Supplementary Figure 2B). This was also the case with dabrafenib and rebastinib, CDK16 inhibitors (Dixon-Clarke et al., 2017; Supplementary Figure 2B). Five compounds were re-examined over a broader concentration range, but still failed to mimic Indirubin’s ability to enhance mitochondrial transport in a dose-response manner (Supplementary Figures 2C–G). The failure of these kinase inhibitors to mimic indirubin appears rule out these CDKs and GSK-3β as the cellular targets responsible for the enhanced mitochondrial transport; we have not been able to associate this action of indirubin with any of its annotated target kinases.

Cantharidin is a potent inhibitor of the serine/threonine protein phosphatases types 1 (PP1) and 2A (PP2A) and a weak inhibitor of type 2B (PP2B) (Honkanen, 1993). The reported half-maximal inhibitory concentration (IC50) values for cantharidin on these targets are ∼0.16 μM for PP2A and ∼1.70 μM for PP1, and >500-fold higher IC50 value for PP2B than PP1 in PC12 cells (Li and Casida, 1992; Honkanen, 1993). To determine whether any of these phosphatases are responsible for the effect of cantharidin on mitochondria motility, we examined cortical neurons treated with two other protein phosphatase inhibitors, LB-100, and okadaic acid (Figure 4). LB-100 inhibits the activity of PP2A with an IC50 value of ∼0.4 μM and for PP1 at ∼80 μM in the human glioblastoma multiforme cell line U87MG (Lu et al., 2009). In contrast, okadaic acid potently inhibits PP2A at ∼0.04 nM and PP1 at 10–200-fold higher concentrations than PP2A in PC12 cells (Honkanen, 1993; Lu et al., 2009; Hu et al., 2017). IC50 values for these compounds have not been determined in rat primary neurons. To explore the effectiveness of these compounds on mitochondrial motility, rat cortical neurons on DIV 8 or 9 were treated with the compounds for 1 h prior to imaging using PATHS to establish a dose-response curve. Compared with control cells, mitochondria motility was significantly increased in neurons treated with 3 μM LB-100 (Figures 4A,B). From the sigmoidal dose-response curve, an IC50 value for LB-100 to enhance mitochondrial motility was at ∼1.28 μM (Figure 4E), implicating PP2A as the most likely target. Unlike LB-100, however, okadaic acid did not enhance mitochondrial motility and no clear dose-response relationship could be established over a broad range of concentrations (Figures 4C,D,F).

The influence of cantharidin, LB-100 and okadaic acid on mitochondria was also examined by kymography in the axons of cortical neurons transfected with MitoDsRed (Figure 5). The percent of mitochondria undergoing anterograde transport increased upon treatment with either 10 μM cantharidin (166% of control) or 3 μM LB-100 (291% of control). Retrograde transport was also increased by cantharidin (249%) and LB-100 (271%) (Figures 5A,B). Both compounds also increased the velocity in both directions and the density of mitochondria (Figures 5C,D). Consistent with the data in Figure 4, however, okadaic acid did not consistently increase mitochondrial transport in either direction (Figures 5A–C), nor did it detectably alter mitochondrial density (Figure 5D). To examine the effects of cantharidin and LB-100 in non-neuronal cells, we again used the QuoVadoPro algorithm in HeLa cells, as in Figure 3. Both compounds increased mitochondrial motility (Figures 5E,F) but not the motility of mCherry-Rab7-tagged endosomes (Figures 5G,H). Thus, cantharidin and LB-100 will enhance mitochondrial transport in both neurons and HeLa cells and this action is at least in part selective for mitochondria.

From this pharmacological analysis, PP2A emerged as a likely candidate to be the relevant cellular target of cantharidin and LB-100, although the lack of efficacy of okadaic acid was a discordant finding. To test directly whether PP2A was an inhibitor of mitochondrial transport, we used shRNA to knockdown the catalytic subunit of PP2A (PP2CA) in neurons. We established the knockdown efficiency of PP2CA shRNA by introducing them by lentiviral transduction into cortical neurons. After 48–72 h, PP2CA levels were assessed by Western blot (Figure 6A) and found to be reduced by 32% relative to neurons transduced with a scrambled shRNAs (Figures 6A,B). When mitochondrial transport was then assayed by kymography in cortical axons, the PP2CA shRNA increased the percent of motile mitochondria to 22.2% compared to 12.7% with the control shRNA (Figures 6C,D). Most prominently, it was the percent of retrograde transport that increased from 4.5% (control) to 13.1% (PP2A shRNA), with a more than twofold increase in the average velocity of retrograde transport as well (Figures 6D,E). The effects on anterograde transport were more modest and lacked compelling statistical significance (p = 0.3) (Figures 6D,E). The effects of the shRNA on retrograde motility were not due to off-target effects; expression of a PP2CA that was resistant to the shRNA restored mitochondrial movement to levels in control axons (Figures 6D,E). As observed with pharmacological inhibition, mitochondrial number per axonal length was not detectably altered by knockdown of PP2CA (Figure 6F). We examined the selectivity of PP2CA knockdown for mitochondrial transport by applying kymography to mCherry-Rab7 tagged endosomes. Compared with control axons, neither the average speed, nor the number of Rab7 organelles were appreciably altered in PP2CA-knockdown axons, though there may have been a modest increase in the percent that were motile from 10.7 to 11.7% (p = 0.12) (Figures 7A–D). Thus, the shRNA data confirm the finding implied by the pharmacology of cantharidin and LB-100 that PP2A is restraining mitochondrial motility in cortical axons (Figures 6D–F).

pDsRed2-Mito (MitoDsRed) was a gift of G. Hajnoczky (Thomas Jefferson University, Philadelphia, PA, United States). For shRNA constructs, pLKO.1-non-target shRNAs and pLKO.1-PP2Ac shRNAs (TRCN0000002484) were obtained from Sigma-Aldrich (St. Louis, MI, United States). The EGFP fragment was amplified by PCR from mEGFP.N1 and cloned into pLKO.1-non-target-shRNA (Scrambled) and pLKO.1-hPP2A-shRNA (PP2A shRNA) to generate pLKO.1-non-target-EGFP and pLKO.1-hPP2A-shRNA-EGFP. pLKO.1-non-target-EGFP and pLKO.1-hPP2A-shRNA-EGFP expressing lentivirus was produced at Boston Children’s Hospital (BCH) Viral Core. PP2A-shRNA-Resistant-BFP was generated using Q5 Hot Start High-Fidelity site-directed mutagenesis kit (New England Biolabs, #M0494, Ipswich, MA, United States). mCherry-Rab7 (#55127) and mEGFP-N1 (#54767) were obtained from Addgene.

Hela cells were maintained in DMEM, high glucose, GlutaMAX™ (ThermoFisher Scientific, Waltham, MA, United States) supplemented with 10% FBS (Atlanta Biological, Minneapolis, MN, United States), 1% penicillin/streptomycin (ThermoFisher Scientific) in 5% CO2 at 37°C. To transfect the cells, Lipofectamine2000 (ThermoFisher Scientific) and TransIT-LT1 (Mirus, Madison, WI, United States) were used. Primary cortical neurons isolated from E18 rat embryos (Charles River) were seeded on 20 μg/mL poly-D-lysine (Sigma-Aldrich), and 3.5 μg/mL laminin (ThermoFisher Scientific) coated plates. Neurons were maintained in Neurobasal Medium (Life Technologies, Carlsbad, CA, United States) supplemented with 2% B27 (GIBCO, Carlsbad, CA, United States), 100 U/mL penicillin/streptomycin (ThermoFisher Scientific), in 5% CO2 at 37°C. For spot-groove cultures, cortical neurons were plated on 20 μg/mL poly-D-lysine (Sigma-Aldrich), and 3.5 μg/mL laminin (ThermoFisher Scientific) coated μ-24-well plates with polymer coverslip bottoms (ibidi, Fitchburg, WI, United States). Each plate was scratched with a CAMP-PR pin rake (Tyler Research Corporation, Edmonton, AB, Canada) to promote separate growth of neuronal cell bodies and projecting axons in the plate. For this culture, approximately 1 × 10⁴ neuronal cells per well were seed in the center of the 24-well plates. After 1 h, more media was added and cells were maintained in those conditions until transfection.

Rat cortical neurons were washed once with ice-cold PBS and homogenized in cold NP-40 buffer (1% Nonidet P-40, 50 μM Tris–HCl, 5 mM EDTA, and protease inhibitor cocktail). The primary antibodies were anti-PP2A-Cα/β (1:1,000; Santa Cruz Biotechnology), anti-α-tubulin (1:1,000; Sigma-Aldrich). Western blots are done for three independent experiments.

MitoDsRed was introduced into Hela Cells or cortical neurons at DIV6-7 using Lipofectamine2000 transfection reagent (ThermoFisher Scientific). Both Hela cells and neurons were imaged 2 days post-transfection. Prior to imaging, neurons were incubated with the indicated compound in either regular growth media or phenol-free Hibernate E media (BrainBits). The MitoDsRed was imaged 1–3 h post incubation with compounds.

Hela cells and cortical neurons were imaged live for 3 min on a TiEclipse microscope. The analysis of the mitochondrial variance score in Hela cells and axonal mitochondrial transport in cortical axons were analyzed with the custom scripted QuoVadoPro, and Kymolyzer algorithms (Basu et al., 2020; Basu and Schwarz, 2020). For analysis of large-scale images, time-lapse images obtained from the ArrayScan XTI imaging platform were analyzed using CellProfiler pipelines (Carpenter et al., 2006; Jaqaman et al., 2008) and PATH algorithms in MATLAB (MathWorks, Natick, MA, United States). From the algorithms, stationary and motile mitochondria were distinguished, and then the movement made by each motile mitochondrion was analyzed per field to derive the mean of integrated distance. The KS Z-score was used to determine the statistical significance. For dose-response curves, each concentration was converted to log concentration (nM). The mean of integrated distance value was converted to the percent maximum activity by determining the low and high values within the individual data set and internally normalized. For plotting, the log (agonist) versus normalized response-variable slope were used in Prism (GraphPad, San Diego, CA, United States).

For high-content data analysis, MATLAB was used for Z-score measurements, using the formula Z = (x − μ)/σ for each parameter. Mean±SEM. for each parameter were obtained from samples described in Figures 2, 4 and Supplementary Figures 2, 3. For a low-content data analysis, the statistical analyses were done using Prism (GraphPad). For a two-sample comparison, the normal distribution of the mean value was tested using D’Agostino and Pearson omnibus normality test and the variance was tested using F-test. If the value is normally distributed, the Student’s t-test was used for statistical analysis. Otherwise, nonparametric Mann–Whitney test was used. For multi-sample comparison, the normal distribution of the mean value was tested using D’Agostino and Pearson omnibus normality test and the variance was tested using Bartlett’s test. For the normally distributed value, one-way ANOVA and Tukey’s post hoc test was used for statistical analysis. Otherwise, nonparametric Kruskal–Wallis test was used. The method of statistical analysis is specified in each figure legend.