Abstract
Dyrk1a triplication in Down's syndrome and its overexpression in Alzheimer's disease suggest a role for increased DYRK1A activity in the abnormal metabolism of APP. Transport defects are early phenotypes in the progression of Alzheimer's disease, which lead to APP processing impairments. However, whether DYRK1A regulates the intracellular transport and delivery of APP in human neurons remains unknown. From a proteomic dataset of human cerebral organoids treated with harmine, a DYRK1A inhibitor, we found expression changes in protein clusters associated with the control of microtubule-based transport and in close interaction with the APP vesicle. Live imaging of APP axonal transport in human-derived neurons treated with harmine or overexpressing a dominant negative DYRK1A revealed a reduction in APP vesicle density and enhanced the stochastic behavior of retrograde vesicle transport. Moreover, harmine increased the fraction of slow segmental velocities and changed speed transitions supporting a DYRK1A-mediated effect in the exchange of active motor configuration. Contrarily, the overexpression of DYRK1A in human polarized neurons increased the axonal density of APP vesicles and enhanced the processivity of retrograde APP. In addition, increased DYRK1A activity induced faster retrograde segmental velocities together with significant changes in slow to fast anterograde and retrograde speed transitions, suggesting the facilitation of the active motor configuration. Our results highlight DYRK1A as a modulator of the axonal transport machinery driving APP intracellular distribution in neurons, and stress DYRK1A inhibition as a putative therapeutic intervention to restore APP axonal transport in Down's syndrome and Alzheimer's disease.
SIGNIFICANCE STATEMENT Axonal transport defects are early events in the progression of neurodegenerative diseases, such as Alzheimer's disease. However, the molecular mechanisms underlying transport defects remain elusive. Dyrk1a kinase is triplicated in Down's syndrome and overexpressed in Alzheimer's disease, suggesting that DYRK1A dysfunction affects molecular pathways leading to early-onset neurodegeneration. Here, we show by live imaging of human-derived neurons that DYRK1A activity differentially regulates the intracellular trafficking of APP. Further, single-particle analysis revealed DYRK1A as a modulator of axonal transport and the configuration of active motors within the APP vesicle. Our work highlights DYRK1A as a regulator of APP axonal transport and metabolism, supporting DYRK1A inhibition as a therapeutic strategy to restore intracellular dynamics in Alzheimer's disease.
Introduction
Axonal transport defects in Alzheimer's disease (AD) affect the intracellular distribution of APP, and the extent to which it is exposed to amyloid β-peptide (Aβ) producing enzymes (Stokin et al., 2005; Thinakaran and Koo, 2008; Choy et al., 2012). The phosphorylation of APP regulates its transport and its intracellular localization (Lee et al., 2003; Gibbs et al., 2015). A synergistic effect between APP and the dual-specificity tyrosine phosphorylation regulated kinase 1A (DYRK1A) has been proposed to lead to early amyloid pathology onset (Wegiel et al., 2011). Both genes are located on chromosome 21 within the critical region triplicated in Down syndrome (DS) (Kimura et al., 2007; Wegiel et al., 2008). Moreover, increased dyrk1a expression was found in postmortem brains of sporadic AD (Kimura et al., 2007). Harmine, a β-carboline with high potency to inhibit DYRK1A, reduces the phosphorylation of APP (Frost et al., 2011; H. Kim et al., 2016; Kawakubo et al., 2017). However, whether APP neuronal distribution is regulated by DYRK1A activity remains elusive.
Dyrk1a overexpression in mice enhanced the phosphorylation of APP at Thr668 and increased the production of Aβ (Martí et al., 2003; Ryoo et al., 2008; Wegiel et al., 2011). Moreover, DYRK1A acts as a priming kinase of tau protein enhancing subsequent phosphorylation by GSK3β (Woods et al., 2001; Liu et al., 2007). Interestingly, microduplications of chromosome 21, including dyrk1a but not app, are implicated in tau-dependent neurodegeneration in DS (Schnabel et al., 2018). Therefore, DYRK1A activity has emerged as a relevant target for the treatment of neurodegenerative diseases (Kimura et al., 2007). Its inhibition or genetic normalization in AD and DS models reduces APP levels and Aβ-related pathologies (Branca et al., 2017; García-Cerro et al., 2017). Harmine and other DYRK1A inhibitors also reduced the phosphorylation of tau at distinct disease-related sites and diminished tau pathology, confirming its specificity (Frost et al., 2011; Melchior et al., 2019). Moreover, its long-term use is associated with a delay in the onset of both amyloid plaques and NFTs (Velazquez et al., 2019). DYRK1A can be considered a transport regulator because of its ability to control the phosphorylation of microtubule-associated proteins. The growing list of DYRK1A cytoskeletal targets, including tau, tubulin, α-synuclein, SEPT4, MAP1A, MAP1B, and N-WASP (Woods et al., 2001; E. J. Kim et al., 2006; Sitz et al., 2008; Scales et al., 2009; Park et al., 2012; Ori-McKenney et al., 2016), supports a relevant role in the regulation of the cytoskeleton structure. Moreover, other functions of DYRK1A have been associated with the regulation of endocytosis and vesicle recycling because of the differential phosphorylation of dynamin, amphiphysin, and synaptojanin (Chen-Hwang et al., 2002; Adayev et al., 2006; Murakami et al., 2009). However, the role of DYRK1A in the regulation of APP intracellular distribution remains unknown.
Axonal transport depends on kinesin and dynein motors to ensure anterograde and retrograde distribution of vesicular cargoes in polarized microtubules. This complex multistep process is highly regulated by the action of different kinases (Hirokawa et al., 2010; Gibbs et al., 2015). A hybrid tug-of-war/coordination model was proposed to explain bidirectional cargo transport, driven by teams of active and opposing kinesin and dynein motors (Hendricks et al., 2012; Lacovich et al., 2017). The exchange between active conformations in a cooperative multimotor arrangement paradigm determines the processive behavior and speed transitions of cargoes (Leidel et al., 2012; Ferro et al., 2019). APP is packed in Golgi- or endosome-derived vesicles sorted from the soma to the presynapse and back to the cell body (Koo et al., 1990; Kaether et al., 2000; Falzone and Stokin, 2012). The axonal transport of APP is necessary to maintain synaptic boutons, promote axonal growth, and react to injury (Torroja et al., 1999; Leyssen et al., 2005; Soldano et al., 2013; Wang et al., 2017). Changes in APP transport are linked to its processing and vice versa; increased β-secretase-mediated cleavage reduces the anterograde delivery of APP, and proteasome inhibition or selective UV irradiation decreases APP anterograde transport by enhancing its processing (Rodrigues et al., 2012; Almenar-Queralt et al., 2014; Niederst et al., 2015; Otero et al., 2018). APP phosphorylation leads to enhanced processing (Lee et al., 2003) because of a transport-dependent mechanism (Muresan and Muresan, 2005; Acevedo et al., 2014). DYRK1A has been shown to phosphorylate tau, APP, microtubule-associated proteins, and proteins involved in membrane internalization (Wegiel et al., 2011). However, it is still unknown whether DYRK1A can regulate APP intracellular distribution.
Here we hypothesized that DYRK1A activity regulates axonal transport. Our results support DYRK1A as a novel player in the intracellular dynamics of APP, which is relevant for its metabolism and for the progression of neurodegeneration in AD.
Materials and Methods
Cell culture
Human induced pluripotent stem cells (hiPSCs) were obtained from UCSD (Craig Venter, male hiPSC, CVhiPSC, MTA UCSD). hiPSCs were maintained on an inactivated mouse embryonic fibroblast feeder layer at 37°C 5% CO2, in KO DMEM (Invitrogen) supplemented with 20% Knockout Serum Replacement, 2 mm nonessential amino acids, 2 mm L-glutamine, 100 U/ml penicillin, 50 µg/ml streptomycin, 0.1 mm β-mercaptoethanol, and 4 ng/ml of bFGF. All the reagents were obtained from Invitrogen. Mouse neuroblastoma N2a cells from passages 10-30 (American Type Culture Collection) were propagated using 0.25% trypsin and grown at 37°C in 5% CO2 in DMEM supplemented with 10% FBS, 1% penicillin/streptomycin, and 1% glutamax.
Antibodies
Primary antibodies are as follows: DYRK1A (Abcam, ab65220, 1:300), Beta-Tubulin 3 (TUJ) (Aves, TUJ8847981, 1:300), NESTIN (Sigma-Aldrich, N5413, 1:500), MAP2 (Sigma-Aldrich, M3696, 1:200), phospho-tau (Ser202) (Peter Davies, CP13, 1:1000), ANK-G (N106/36) (Neuromab, 75-146, 1: 200), and phospho-APP (Thr668) (Cell Signaling, 3823S, 1:1000). Secondary antibodies against mouse and rabbit IgG conjugated to AlexaFluor-564 (1:400) or AlexaFluor-488 (1:400).
Bioinformatic analysis
The proteomic dataset of harmine-modified proteins available at Karmirian et al. (2021) was uploaded to pathway enrichment analysis using Metascape database at a p value cutoff = 0.05, considering Gene Ontology Biological Process annotation (Zhou et al., 2019; Karmirian et al., 2021). Protein interaction networks were identified using the STRING database, considering medium and high confidence interactions. This software uses curated connectivity information from literature to determine interaction networks among the differentially expressed proteins and proteins of interest known to participate in the enriched processes.
Gene expression analysis by qRT-PCR
Total RNA was isolated from organoids using PureLink RNA Mini Kit (Invitrogen, 12183018A). At least five organoids per batch at 45 d were collected after 24 h of exposure to harmine 7.5 µm or DMSO (vehicle). The amount of RNA was evaluated using NanoDrop 2000 (Thermo Fisher Scientific, ND2000USCAN). Each sample was digested with 4 U of DNase I Amplification Grade (Invitrogen, 18068-01). Reverse transcription was performed with 1000 ng of digested RNA with the M-MLV kit (Invitrogen, 28 025-013) in a total volume of 20 µl. RT conditions were as follows: 10 min at 25°C, 50 min at 37°C, and 15 min at 70°C. To perform the relative quantification of APP, KLC1, KIF1B, MAP7D1, MAP6, and SOD1 by real-time PCR, specific pairs of primers were used for each gene. Primer sequences were as follows: APP forward: 5′-TCTCGTTCCTGACAAGTGCAA-3′; APP reverse: 5′-GCAAGTTGGTACTCTTCTCACTG-3′; KLC1 forward: 5′-CAGCTAACCTACTGAATGATGCC-3′; KLC1 reverse: 5′-GCTCTTTTACACAACGGCTCT-3′; MAP7D1 forward: 5′-ATGAAGCAGCCATCCAACGG-3′; MAP7D1 reverse: 5′-AGGTTAACTGCCGACACGG-3′; KIF1B forward: 5′-TGGCAGTTACTTCCTACACAGA-3′; KIF1B reverse: 5′-GGGAACGGCTACTTGTTTCAT-3′; MAP6 forward: 5′-CAGCCTCTACAGCGAACCC-3′; MAP6 reverse: 5′-TTCCTCGTGGGCTTATGGC-3′; SOD1 forward: 5′-ACAAAGATGGTGTGGCCGAT-3′; SOD1 reverse: 5′-AACGACTTCCAGCGTTTCCT-3′. Real-time PCRs were performed in triplicate with 10 ng of cDNA per reaction at a final reaction volume of 10 µl in MicroAmp Optical 96-Well Reaction Plate (Applied Biosystems, N8010560) containing 1× GoTaq qPCR Master Mix (Promega), 300 nM CXR Reference Dye, and 250 nM final concentration of APP, KLC1, MAP7D1, and SOD1 designed primers or 500 nM final concentration of KIF1B and MAP6 designed primers. The reaction and analyses were performed in StepOnePlus Real-Time PCR System (Applied Biosystems, 4376600) under the standard cycling conditions: initial GoTaq Hot Start Polymerase activation at 95°C for 2 min, followed by 40 cycles of denaturation at 95°C for 15 s and annealing/extension at 60°C for 1 min. Data were analyzed as fold change (2-ΔΔCt) with gene expression normalized by HPRT1 (housekeeping gene). Data from four independent experiments were pooled and analyzed using GraphPad Prism software, and graphs were plotted in R.
Human neuron differentiation
Neurons were derived from iPS cells as previously described (Pozo Devoto et al., 2017). Briefly, optimal size colonies were enzymatically detached and grown in suspension for 48 h to allow embryoid body formation. Embryoid bodies were neuronally induced in suspension for 3 d: NIM: DMEM/F12 (Invitrogen, 12634-010), 1% N2 supplement (Invitrogen, 17502-048), 1% MEM-NEAA (Invitrogen, 11140-050), 280 UI/ml heparin (Sigma-Aldrich, H3393), and 1% Pen-Strep (Invitrogen, 15140-122). Embryoid bodies were attached to laminin-coated 6-well plates and grown for 7-14 d to allow the formation and maturation of neural tube-like rosettes enriched in neural progenitors. Neural precursors were manually picked and transferred to 25 cm2 flasks in NIM medium supplemented with B27 and ascorbic acid (NIM + 2% B27 supplement, Invitrogen, 17504-044), 0.1% ascorbic acid (Sigma-Aldrich, A1300000) for up to 1 month, changing media every 2 d. Neural rosettes were dissociated in 1% trypsin (Invitrogen, 15090-046) and accutase (1:1) (Invitrogen, A11105) for 8 min at 37°C and later blocked using 0.5 mg/ml trypsin inhibitor (Invitrogen, R007100). The suspension was centrifuged for 5 min at 1000 rpm. The pellet was washed with DMEM/F12, disaggregated to single-cell, and resuspended in neural differentiation media: neurobasal medium supplemented with 1% N2 supplement (Invitrogen, 17502-048), 2% B27 supplement (Invitrogen, 17504-044). Cells were plated over poly-ornithine (Sigma-Aldrich, P4957) and laminin-coated (Invitrogen, 23017-015) coverslips (0.1 mg/ml and 20 µg/ml, respectively) into 24-well plates and maintained in 500 µl/well complete neural differentiation media + 1 µg/ml laminin (Invitrogen, 23017-015), 1 µm cAMP (Sigma-Aldrich, A6885), 200 µg/ml ascorbic acid (Sigma-Aldrich, A1300000), 10 ng/ml BDNF (Invitrogen, 10908-010), and 10 ng/ml GDNF (Invitrogen, PHC7041). Half of the media was replaced every 3 d.
Expression vectors
To register the transport of APP, the pcDNA3-APP-yellow fluorescent protein (YFP) vector was used to express a fusion between APP695 and YFP (Kaether et al., 2000). pCAG-DYRK1A-eGFP and pCAG-DYRK1AK188R-eGFP were kindly gifted by D'Arcangelo from Rutgers University (Yabut et al., 2010). To generate pcDNA3-DYRK1A-mCherry, DYRK1A sequence was amplified by PCR without stop codon from pCAG-DYRK1A-eGFP with primer 5′-GCGCCCATCACACTGGGCCACCATGCATACAGGAGGAGAGACTTCAG-3′ (including EcoRI and Kozak sites) and 5′-CAACAGAGTCCTGTAGCTAGCTCGGCGGCCGCGCGC-3′ (containing the NotI site). DYRK1A insert was subcloned into TOPO 2.1 (Invitrogen, K450002) and sequenced to validate integrity. DYRK1A was inserted in frame with mCherry fluorescent marker from pcDNA3-mCherry using a subcloning strategy that included EcoRI and NotI restriction sites.
Human neurons and N2a transfection and harmine treatment
Neurons were transfected on DIV 14 after plating. pcDNA3-APP-YFP transfection was performed with 1 µg of plasmidic DNA in a transfection mixture of OptiMEM (Invitrogen, 31985-062) and Lipofectamine 2000 (Invitrogen, 11668-019). Two hours after transfection, transfection medium was replaced and neurons were treated with vehicle or harmine 7.5 µm (Cayman, 10010324). Cotransfection was performed using 1 µg of each pcDNA3-APP-YFP and pcDNA3-DYRK1A-mCherry. Forty-eight hours after transfection/cotransfection (D16), neurons were analyzed by live-cell imaging for transport analysis, fixed for immunocytochemistry, or processed to obtain protein for Western blot analyses. N2a cells were transfected using Lipofectamine 2000 (Invitrogen, 11668-019) following standard protocols.
Live-cell imaging
Imaging of live cells and kymograph generation were performed as previously described (Falzone and Stokin, 2012). Briefly, 30 s movies of APP-YFP moving particles in neurons were recorded using an inverted epifluorescence microscope (Olympus IX81) connected to a CCD camera (Olympus DP71/12.5 megapixels). Cultures were observed under a 100× lens (numerical aperture, 1.25) and maintained at 37°C, 5% CO2, and 10% humidity using a CO2 humid chamber and heated stage (Olympus). Directionality was determined by tracking fluorescent axons. To avoid introducing biases, imaging was performed in axons at their middle part separated by at least 2 FOV distance (∼200 µm) from cell bodies and from axonal tips. Kymographs were generated from the recordings with ImageJ using the multiple kymograph plug-in (Otero et al., 2014).
Development of semiautomatic tracking toolbox and axonal transport dynamics analysis
Using custom-made MATLAB algorithms, a complete toolbox was developed to semiautomatically track vesicle trajectories from kymographs. Briefly, a median filter obtained for impulsive noise removal with kernels 3-5 pixels wide was applied to the original kymograph. The image was convolved with an averaging filter (convolution with dimension kernel ∼20-25 pixels in diameter) for background estimation. Canny filter on grayscale image was used to obtain a binary image of edge location. From this grayscale image with two relevant edges accented, a hysteresis binarization step was performed, obtaining 1-pixel-thick edge elements allowing an adequate evaluation in the subsequent step of obtaining critical points in the image: starting points of tracks, end of tracks, and intersections. Segmental velocities, run lengths, pauses, and reversions were computed from trajectories using custom-made MATLAB application. For the calculation of segmental velocities, processive trajectories were divided into 20 frames of duration, producing a linear approximation with the least-squares method, and filtering trajectories with outliers and significant differences between the dataset and the fitted curve. The slope of the regression was considered as the velocity of the segment (see Fig. 3A). The point at which the movement is equal to 0, filtering the points produced by the noise of the trajectory, was used for the calculation of run lengths, pauses, and reversions. Trajectories were defined as anterograde or retrograde depending on their average movement (see Fig. 2I). A trajectory was defined as stationary if it met the stationary criterion, that is, the vesicle moved <0.05 pixel per frame (<0.16 µm/s). Similarly, pauses were considered when segments follow a stationary criterion for >5 frames (625 ms) moving <0.05 pixel per frame (<0.16 µm/s). The script merges continuous segments of the same type and segments that are too small. The difference in micrometers between the final point and the initial point of a determined continuous segment of the trajectory was considered as the “run length.” The points separating anterograde and retrograde segments were defined as reversions. For the analysis of the segmental velocity distributions, the parameters of the Gaussian mixture model were determined by an expectation maximization algorithm using a function (gmdistribution) in MATLAB. A bootstrapping with resampling procedure with n = 1000 was implemented to compute the intervals of confidence of the Gaussian mixture parameter estimates. Velocity transitions were extracted using an algorithm to detect optimal cuts between segments obtained from trajectories (see Fig. 3B). Briefly, two successive steps are defined to calculate particle fluctuation threshold (th) and changes because of velocity transitions defined as the relative angle (α) formed between two consecutive segments. Th describes vesicle wiggling movement apart from its traveling direction because of the medium viscosity, and it is calculated averaging all perpendicular fluctuations along the trajectory in 5 pixel segments. Alpha represents relative changes in direction between segments. Linear regressions are computed to adjust segments and to obtain the angle they form (α i). Each time, angles calculated dynamically were compared with the predefined value (Th). When threshold between angles is exceeded (sin(α i) > Th), then a change in direction is detected and accounted as a velocity transition. An example illustrating this procedure is shown in Figure 3B.
Control and harmine segmental velocities modeled as a combination of normal distributionsa
Experimental design
cviPSCs (male, XY karyotype) were differentiated into functional and polarized human neurons using previously described protocols (Pozo Devoto et al., 2017). Fourteen days after terminal plating, neurons were transfected with fluorescently fused APP-YFP vector and treated with harmine for 48 h. DMSO treatment was used as control condition. DYRK1A and DYRK1AK188R overexpression was achieved by cotransfecting the respective vectors with APP-YFP. mCherry vector was used as cotransfection control. Protein collection, immunofluorescence, and live imaging were performed as shown in Figures 2 and 4. At least three independent differentiation procedures were performed to obtain neurons and collect data for each experimental design. Movies collected from live imaging were transformed to kymographs as described, and single-particle tracking and analysis were performed using developed scripts. Runs data obtained from analysis were plotted and compared in Figures 3C–F and 5A–D. Anterograde and retrograde segmental velocities were differentially extracted to plot a three modal distribution Gaussian mixture model in Figures 3G, H and 5E, F. Velocity transitions were extracted using custom-made scripts described above and data plotted as 2D kernel density for analysis in Figures 3I–K and 5G–I. Statistical analysis and n of each quantification are described in the figure legends.
Control and overexpressed DYRK1A segmental velocities modeled as a combination of normal distributionsa
Statistical analysis
Data processing and statistical analyses were performed using R software. All statistical details of performed experiments, such as statistical tests, and values of n, are described in the figure legends. Briefly, normality of the data was assessed using the Shapiro–Wilk test. p value of <0.05 in the Shapiro–Wilk test did not reject the null hypothesis that the data are normally distributed. Student's t test was performed if samples were normally distributed. If the p value was <0.05 in Shapiro–Wilk test, the null hypothesis that the data are normally distributed was rejected and a nonparametric Mann–Whitney U test was performed. The error in segmental velocity distributions was obtained using a bootstrapping with resampling procedure with n = 1000 to compute the intervals of confidence of the Gaussian mixture parameter estimates adjusted to three modes. Significant differences were determined by differences of nonoverlapping errors in the mode center or fraction between the control and the experimental condition. Statistical differences on velocity transitions distributions were analyzed using 10 simulated random samples of each condition obtained by a bootstrapping procedure with resampling. Transition patterns were plotted in normalized hexbins using R (package “hexbin”). Control, experimental, and simulated hexbin plots were compared using the Pearson correlation coefficient. Pearson correlation coefficient obtained for control against simulated and for control against experimental was obtained, and statistical differences determined when p value was <0.05.
Software accessibility
The code generated for data analysis is available on request from the corresponding author.
Western blotting
Total protein from neurons or cell lines was collected in 100 µl of lysis buffer [50 mm Tri-HCl (pH 7.5), 150 mm NaCl, 1% Igepal (Sigma-Aldrich, I8896) and 1× protease inhibitor cocktail (Sigma-Aldrich, P2714)] and centrifuged for 10 min at 10,000 rpm (8000 × g) at 4°C. Protein concentration of the supernatant was measured using BCA Protein Assay Kit (Pierce, 23225). Equal amounts of protein (30 µg) in 10 µl of 4× Laemmli sample buffer (250 mm Tri-HCl, pH 6.8, 0.04% bromophenol blue, 40% glycerol, 8% SDS, and 2% β-mercaptoethanol) were loaded onto 12% SDS-polyacrylamide gel, and See-Plus 2 (Invitrogen, LC5925) was used as a molecular-weight marker. Proteins were transferred onto nitrocellulose membranes using a wet system in 25 mm Tris base, 190 mm glycine, and 20% methanol. Membranes were blocked in 5% BSA in 0.1% Tween-20 in TBS (TBS-T) for 1 h and incubated in 1% BSA in TBS-T with primary antibody overnight at 4°C. After washing, membranes were incubated with HRP-conjugated secondary antibody for 4 h at room temperature. Westerns were developed using an ECL kit (Pierce, 32106). Scanned images were analyzed using ImageJ software.
Immunofluorescence and image collection
PBS washed cells were fixed (4% PFA, 4% sucrose in PBS) for 30 min at 37°C and permeabilized with 0.1% Triton X-100 for 10 min at room temperature (RT). Cells were blocked for 1 h at RT (3% BSA, 0.1% Triton X-100, and 10% goat serum in PBS) and incubated with primary antibodies in blocking solution overnight at 4°C. Samples were incubated with secondary antibodies for 2 h at RT, and DAPI (300 nm) was included in washes for 30 min. Cells were mounted on slides with DPX (Sigma-Aldrich, 06522). Image analysis was performed using Image J. Background subtraction was performed using a rolling ball radius of 50 pixels. The same ROIs were applied to images for all data channels (488/555), and the fluorescent intensities were obtained. The intensity data generated by ImageJ were then passed to R for further processing and analysis.
Results
Harmine modulates microtubule-based transport processes in human cerebral organoids
DYRK1A has a myriad of phosphorylation targets; however, whether its inhibition impacts on cellular processes that regulate the metabolism of APP remains unknown. To search for novel intracellular pathways modulated by DYRK1A inhibition, we analyzed global changes in protein expression from a proteomic dataset obtained from harmine-treated human brain organoids at 7.5 µm for 24 h (Dakic et al., 2016) compared with control condition (DMSO) (Karmirian et al., 2021). Harmine, known as a potent DYRK1A inhibitor, was used to identify protein expression changes in human brain organoids following a pathway enrichment analysis of biological processes (Fig. 1). Metascape analysis revealed that cytoskeleton-dependent intracellular transport (LogP = −5.71) and microtubule cytoskeleton organization (LogP = −3.78) were among the most enriched processes modified after harmine treatment (Fig. 1A). Within these processes, microtubule-based transport, mitochondria localization, microtubule-mediated transport, and axonal transport reflected changes in neuronal intracellular trafficking (Fig. 1A). From the total amount of proteins differentially expressed in enriched processes, we identified three groups corresponding to microtubule cytoskeleton organization (MAP7D1, MAP6, TUBB2A, TUBA4A, TBCE, TBCB, TUBB1, MAST4), molecular motor proteins (KIF27, KLC1, KIF1B, DNAH14, MYO19, MYO1C, MYO5C), and adaptors and regulators of vesicle trafficking and sorting (RAB2A, RAB18, GAS8, SOD1, HSPA8, TRAK2, CCDC8, PDCD6IP) (Fig. 1B). Since DYRK1A has also been described as a transcriptional regulator (Di Vona et al., 2015), we tested whether the specific changes in protein levels observed by proteomics were because of modifications in mRNA expression. qRT-PCR was performed to measure candidate mARNs of upregulated proteins KLC1, MAP6, and SOD1; and downregulated proteins KIF1B and MAP7D1 revealing no significant changes in mRNA amount in harmine-treated organoids compared with control (Fig. 1C). In addition, no changes in APP mRNA expression were observed in cerebral organoids treated with harmine (Fig. 1C). Since DYRK1A exerts a role in APP processing (Kimura et al., 2007), and we found changes in microtubule-based transport after DYRK1A inhibition, we performed a protein–protein cluster interaction analysis between the APP vesicle and the processes modified by harmine treatment (Fig. 1D) considered CLSTN1, PSEN1, GAP43, SYT1, STX7, and RAB3 as resident proteins in the APP vesicle (Vagnoni et al., 2012; Almenar-Queralt et al., 2014) (Fig. 1D). We found high and medium confidence interactions between harmine-modified proteins and the APP vesicle, plus relevant protein associations among the harmine-modified clusters (Fig. 1D). These results suggest a putative role for DYRK1A inhibition in the expression (transcription, translation, and/or accumulation) of proteins that are involved in the intracellular trafficking of APP loaded vesicles.
Harmine modulates microtubule-based transport processes in human cerebral organoids. A, Enriched Gene Ontology processes corresponding to the number of upregulated or downregulated proteins obtained from a dataset of organoids after control or harmine treatment plotted as –10Log(P)FDR p < 0.05. B, Fold change of upregulated (green) or downregulated (red) proteins found in enriched processes after harmine treatment (7.5 μm for 24 h). C, qRT-PCR to measure mRNA expression in control and harmine-treated organoids. mRNA from KLC1, MAP6, SOD1, MAP7D1, and APP was amplified from four independent experiments (N = 4). Values are expressed as fold-change relative to the untreated condition. Box and whiskers represent median; 25th and 75th percentiles; plus 1.5 interquartile range. Dots represent independent measures. D, Interactome network from processes enriched after harmine treatment (axonal cytoskeletal, microtubule-associated proteins, molecular motors, and transport regulator proteins) was plotted to show the interactions with the APP vesicle. Protein interactions were obtained using high (≥0.7) and medium (≥0.4) confidence in STRING version 10.0 software. The global protein repository network is composed of 38 nodes (proteins) and 155 edges (interactions).
Axonal APP vesicle density is reduced after harmine treatment
Since proteome analysis revealed that harmine modulates microtubule-based transport related proteins, we further investigated the effect of harmine on the axonal transport of APP vesicles (Fig. 2). Highly polarized human-derived neurons were obtained from the differentiation of CV-hiPSC following a previously validated protocol (Fig. 2A) (Pozo Devoto et al., 2017). Fourteen-days-old human neuronal cultures showed enriched expression of neuronal markers (βIII-tubulin) and few neural precursor cells expressing nestin (Fig. 2B). Phosphorylated tau and ankyrin-G staining were observed in neuronal projections and at the axon initial segment, respectively, revealing the presence of polarized axonal structures and specialized regions associated with the generation of action potentials (Fig. 2C). Human neurons treated with harmine showed no morphologic or structural changes and maintained similar axon initial segment cytoskeletal structure compared with control (DMSO) (Fig. 2E). Immunofluorescence staining against phosphorylated APP (Thr668-pAPP) confirmed harmine effect in treated neurons showing a significant reduction of Thr668-pAPP when normalized to total APP both at the neuronal soma and axonal projections (Fig. 2F–H) (Ryoo et al., 2008). To test the effect of harmine on APP axonal transport, we performed high-resolution/speed live-cell imaging in control and harmine-treated neurons transfected with pcDNA-APP-YFP (Fig. 2A,D). Thirty second movies were generated to extract the dynamics of fluorescent APP vesicles (APP-YFP) (Fig. 2D; Movies 1, 2). APP-YFP transport within axons was registered and transformed to kymographs, from which trajectories were tracked semiautomatically using a custom-made MATLAB application (Fig. 2K). APP vesicles moving anterogradely correspond to net right descending trajectories, while left descending trajectories were accounted as retrograde moving APP vesicles (Fig. 2K). Stationary APP corresponds to nonmoving vesicles represented by vertical trajectories within the 30 s time-frames (Fig. 2K). Interestingly, harmine induced a significant decrease in the density of axonal APP vesicles (Fig. 2L). These changes were observed because of a selective reduction in the density of retrograde APP vesicles, without affecting the anterograde or stationary vesicle density in harmine-treated neurons compared with control (Fig. 2M). Although harmine is 10 times more potent at inhibiting DYRK1A than to other kinases of the DYRK family (Göckler et al., 2009), we tested DYRK1A specificity in axonal APP vesicle reduction by overexpressing a DYRK1A dead kinase mutant (DYRK1AK188R) with dominant negative function (Yin et al., 2012). Cotransfection of APP-YFP and DYRK1AK188R led to a reduction of overall APP vesicle density in axons (Fig. 2L). The DYRK1AK188R-mediated reduction was observed in anterograde, stationary and retrograde densities (Fig. 2M). This reduction in APP vesicle density after the inhibition of DYRK1A activity was not because of changes in full-length APP expression as observed in Western blots from control and harmine-treated cells (Fig. 2I,J). Together, these results suggest that impaired DYRK1A activity reduces the overall density of APP vesicles in axons.
Retrograde APP vesicle density is reduced after DYRK1A inhibition. A, Schematic representation of the experimental design used to analyze APP axonal transport dynamics in control (DMSO) or harmine (7.5 μm for 48 h) treated human neurons. B, C, High-resolution spinning disk confocal microscopy images of human neuronal cultures at D14 immunostained for βIII tubulin and nestin (B), ankyrin-G and phosphorylated tau (C). D, Epifluorescence images of human neuron culture at D16 transfected with pcDNA-CMV-APP-YFP (APP-YFP). E, Epifluorescence images of human neuron at D16 immunostained for ankyrin-G and MAP2 after 48 h of DMSO or harmine (7.5 μm) treatment. F, High-resolution spinning disk confocal microscopy images of human neuronal cultures at D16 immunostained for phosphorylated (Thr668-pAPP) and total APP after 48 h of control or harmine treatment. High magnifications of axonal projections highlighted in white boxes. G, H, Quantification of fluorescence integrated density of Thr668-pAPP normalized to total APP in soma (G) and projections (H). Soma: n = 130 control, 124 harmine; projections: n = 202 control, 176 harmine, from 3 independent experiments. I, Western blots from control (DMSO) or harmine (7.5 μm for 48 h) treated N2A homogenates showing APP (≈100 kDa) and tubulin (55 kDa) expression. Tubulin was used as loading control. J. Quantification of APP optical density normalized to tubulin (N = 3). Data are mean ± SEM. Student's t test. K, Kymographs obtained from a 30 s movie (8 frames/seconds) recorded in axons from neurons transfected with APP-YFP and treated with vehicle or harmine (7.5 μm for 48 h). Colored lines indicate trajectories recovered from a semiautomatic tracking tool box system. Average of total (L) or anterograde, stationary, and retrograde (M) APP vesicle density in 100 µm axon length from control (DMSO), harmine-treated, and DYRK1AK188R transfected neurons. Kymographs n = 98 DMSO, 116 harmine, 38 DYRK1AK188R from 3 independent experiments. Data are mean ± SEM. *p < 0.05; ***p < 0.001; Student's t test. Scale bars: B, 50 µm; C, D, E, L, 10 µm; F, 20 µm.
APP axonal transport in a control DMSO-treated human-derived neuron at day 16. This is a representative movie showing the axonal transport of APP vesicles in real time in a neuron that has been treated with DMSO and transfected with APP-YFP. Movie showing time in seconds (8 frames/s) is oriented so that right- to left-moving vesicles correspond to anterograde transport and left to right to retrograde movement. Scale bar, 10 µm.
APP axonal transport in a harmine-treated human-derived neuron at day 16. This is a representative movie showing the axonal transport of APP vesicles in real time in a neuron that has been treated with harmine (7.5 µm) for 48 h and transfected with APP-YFP. Movie showing time in seconds (8 frames/s) is oriented so that right- to left-moving vesicles correspond to anterograde transport and left to right to retrograde movement. Scale bar, 10 µm.
Harmine regulates the processive movement of APP vesicles
Vesicular processive movement depends on the coordinated stepping of kinesin and dynein opposing motors that determine run directionality, length, and duration (Reis et al., 2012; Lacovich et al., 2017). Therefore, we tested whether harmine also regulates APP single-particle dynamics. Single-particle trajectories of moving APP-YFP vesicles were extracted using validated custom-made MATLAB application (Lacovich et al., 2017; Alloatti et al., 2018) (Fig. 3A; see Materials and Methods). Continuous segments corresponding to moving trajectories under defined properties were considered runs (Fig. 3A; see Materials and Methods). In-depth analysis of transport dynamics showed that APP vesicle run length and duration were not affected after harmine treatment (Fig. 3C,D). However, net retrograde APP moving vesicles revealed a pronounced stochastic behavior with significant increases in the number of retrograde and anterograde runs associated with its movement (Fig. 3A,E,F). Therefore, harmine enhanced the directional changes of retrograde-biased moving APP vesicles.
Harmine regulates the processive movement of APP vesicles. A, Representative kymographs showing trajectories of APP vesicles in axons from human-derived neurons considered in single-particle dynamics analysis. Representation of runs (blue), reversions (white), pauses (red), and segmental velocities calculated from semiautomated tracking and MATLAB processing. B, Kymograph (left) and representation of a velocity transition (Velocity n to n + 1) between two calculated runs showing the linear regressions obtained from segments n and n + 1 to find the optimal cutting point (right). C-K, APP axonal transport properties obtained from analyzing moving vesicles after DYRK1A inhibition. Average run length distance (C) and duration (D) of net anterograde or retrograde APP vesicles. Anterograde runs n = 1287 DMSO, 1475 harmine; retrograde runs n = 1247 DMSO, 1431 harmine. Average number of homologous (E) and heterologous (F) runs in anterograde and retrograde moving APP vesicles. Anterograde tracks n = 570 DMSO, 651 harmine; retrograde tracks n = 548 DMSO, 553 harmine. *p < 0.05; **p < 0.01; Student's t test. Anterograde (G) and retrograde (H) probability distribution of segmental velocities of moving APP-YFP vesicles represented as a Gaussian mixture model with three modes (A-C) in control (DMSO, dark blue) and harmine-treated (light blue) human neurons. Mode center (dotted line) and mode fraction are displayed in Table 1. Segmental velocities n = 7694 antero, 6412 retro DMSO; 9225 antero, 7177 retro harmine. Significant differences in mode center (*) and mode fraction (#). I, J, 2D Kernel density plots showing the probability of transitions from velocity n to n + 1 in control (I) and harmine (J) treated neurons represented from red to green as probability increases for each transition in the four quadrants: Q1, anterograde-anterograde; Q2, anterograde-retrograde; Q3, retrograde-retrograde; Q4, retrograde-anterograde. K, Box and whiskers plot represents the Pearson correlation coefficient when comparing DMSO versus DMSO simulations (white box) and DMSO versus harmine (green box). Data are median, IQR, minimum and maximum. ***p < 0.001 (Student's t test).
The transport of APP in living cells depends on a concerted action of several motors in a cooperative multimotor arrangement (Schneider et al., 2015; Ferro et al., 2019). Kinesin and dynein forces interact in a tug-of-war/coordination model to exert directional transport in axonal polarized microtubules (Hendricks et al., 2012; Lacovich et al., 2017). Therefore, we studied whether harmine modifies the distribution of segmental velocities of APP. Segmental velocities extracted from kymographs (Fig. 3A,G,H) showed a multimodal distribution that was analyzed in terms of a Gaussian-mixture-model with three components (A, B, and C) attributed to the contribution of different configurations of motors transporting the cargo, as previously described (see Materials and Methods) (Lacovich et al., 2017). The distribution of segmental velocities obtained after harmine treatment showed similar anterograde velocity modes (A, B, C) but mild fraction exchanges between anterograde B and C modes compared with control neurons (Fig. 3G; Table 1). Contrarily, harmine treatment significantly increased the retrograde fraction of slow Mode A moving vesicles and decreased the fraction of fast Mode C (Fig. 3H; Table 1). However, only mild shifts toward faster retrograde velocities in Modes B and C were observed when plotting retrograde velocity distribution (Fig. 3H; Table 1), suggesting an overall reduction in retrograde movement.
The statistical properties of APP trajectories depend on the underlying configuration of the motors driving the cargo. Particularly, the transition rate between consecutive runs with two given velocities in a multimotor system can increase or decrease depending on the probability of molecular motor complexes switching their active configuration (Leidel et al., 2012; Reis et al., 2012). Thus, we studied in detail the transitions between consecutive runs to understand whether harmine modulates the exchange of active motors in APP vesicles. The probability of a given velocity (n) to change to a next velocity (n + 1) in the anterograde or retrograde direction within a run were represented as 2D Kernel density probability plots (Fig. 3I,J). The transport of APP under control conditions revealed higher probability of transitions between anterograde to anterograde velocities (Q1) and retrograde to retrograde velocities (Q3) than those observed for reversions (Q2, Q4) (Fig. 3I), similar to previous results obtained in other living systems (Leidel et al., 2012). Pearson correlation analysis between the distribution of control and harmine speed transitions revealed significant differences after treatment (Fig. 3K), but not when compared against their simulations generated by a bootstrapping procedure, (see Materials and Methods). Interestingly, 2D Kernel density analysis within anterograde to anterograde quadrant (Q1) revealed an increase in the intermediate to fast velocity transitions after harmine treatment (Fig. 3J, black arrow). Similar increases in intermediate to fast velocity transitions within the retrograde to retrograde quadrant (Q3) were observed after harmine treatment (Fig. 3J, black arrow). These changes in the distribution of velocity transitions are in agreement with the changes observed in the proportion of segmental velocities suggesting that harmine modulates the transport of APP by affecting the exchange of active motor configurations.
DYRK1A overexpression increases the axonal APP vesicle density
Genomic DYRK1A triplication in DS and its overexpression in sporadic AD suggest that increased DYRK1A activity induces detrimental effects in neuronal function that lead to neurodegeneration (Kimura et al., 2007; Ryoo et al., 2008). Since we found that harmine-mediated DYRK1A inhibition modifies the density and dynamics of APP vesicles within axons, we studied whether this can be specifically attributed to DYRK1A activity. Therefore, we analyzed the impact of DYRK1A overexpression in APP axonal transport. To avoid disturbing the early role DYRK1A plays in neural differentiation and polarization (Yabut et al., 2010), we induced its overexpression for 48 h in already polarized human neuronal cultures at D14 (Fig. 4A,C). A vector driving DYRK1A expression fused to the mCherry fluorescent marker (DYRK1A-mCherry) was subcloned and transfected in N2A neuroblastoma cell lines to confirm DYRK1A overexpression in Western blots (Fig. 4B). At D16, human neurons cotransfected with APP-YFP and DYRK1A-mCherry showed a significant (100%) increase in DYRK1A fluorescence intensity compared with nontransfected neurons (Fig. 4C,D). Again, we observed no changes in full-length APP expression in Western blots from control and DYRK1A-overexpressing cells (Fig. 2E,F). Using this setting, we analyzed APP axonal transport in control (APP-YFP + mCherry) and DYRK1A-overexpressing neurons (APP-YFP + DYRK1A-mCherry). Thirty second movies were registered and transformed to kymographs to obtain single trajectories for analysis (Fig. 4G; Movies 3, 4). DYRK1A overexpression induced a significant increase (25%) in axonal APP vesicle density, opposite to the effect observed with harmine (Fig. 4H). Moreover, this change was because of a selective increase in retrograde and stationary APP vesicle density, without anterograde impairments (Fig. 4I); hence, these results support the proposition that DYRK1A activity might be involved in the regulation of APP vesicle distribution within axons.
DYRK1A overexpression increases the retrograde loading of APP vesicles into axonal transport. A, Schematic representation of experimental design used to analyze the effect of DYRK1A overexpression on APP-YFP axonal transport dynamics in D16 human neurons. B, Western blots from control (pcDNA3-mCherry) or DYRK1A (pcDNA-CMV-DYRK1A-mChrerry) transfected N2A homogenates showing endogenous DYRK1A (86 kDa) and DYRK1A-mCherry (116 kDa) expression. Tubulin was used as loading control. C, Epifluorescence images of human neuron culture at D16 cotransfected with DYRK1A-mChrerry and APP-YFP for 48 h. Immunostaining for DYRK1A and GFP showing cotransfected (white arrowhead) and nontransfected (yellow) neurons. D, Quantification of normalized fluorescent integrated density in DYRK1A-overexpressing (oe) neurons compared with nontransfected (endo) neurons. n = 14 nontransfected, 19 DYRK1A transfected neurons. E, Western blots from control (mCherry), DYRK1A-overexpressing N2A homogenates showing APP (≈100 kDa) and tubulin (55 kDa) expression. Tubulin was used as loading control. F, Quantification of APP optical density normalized to tubulin. n = 3. Data are mean ± SEM. One-way ANOVA. G, Kymographs obtained from a 30 s movie (8 frames/seconds) recorded in axons from neurons cotransfected with APP-YFP + mCherry (top) or APP-YFP + DYRK1A-mCherry (bottom). Colored lines indicate trajectories recovered from a semiautomatic tracking tool box system. Average of total (H) or anterograde, stationary, and retrograde (I) APP vesicle density in 100 µm axon length from control (mCherry) and DYRK1A-overexpressing (DYRK1A-mCherry) neurons. Kymographs n = 62 APP-YFP + mCherry, 54 APP-YFP + DYRK1A-mCherry from three independent experiments. Data are mean ± SEM. *p < 0.05; **p < 0.01; ***p < 0.001; Student's t test. Scale bars: C, E, G, 10 µm.
APP axonal transport in a control mCherry cotransfected human-derived neuron at day 16. This is a representative movie showing the axonal transport of APP vesicles in real time in a neuron that has been cotransfected with mCherry and APP-YFP vectors. Movie showing time in seconds (8 frames/s) is oriented so that right- to left-moving vesicles correspond to anterograde transport and left to right to retrograde movement. Scale bar, 10 µm.
APP axonal transport in a DYRK1A-overexpressing cotransfected human-derived neuron at day 16. This is a representative movie showing the axonal transport of APP vesicles in real time in a neuron that has been cotransfected with DYRK1A and APP-YFP vectors. Movie showing time in seconds (8 frames/s) is oriented so that right- to left-moving vesicles correspond to anterograde transport and left to right to retrograde movement. Scale bar, 10 µm.
DYRK1A overexpression enhances the processivity, velocity, and speed transitions of APP vesicles
To unravel whether DYRK1A activity is regulating molecular motor processivity, we extracted similar transport parameters after overexpressing DYRK1A. In-depth transport analysis revealed that DYRK1A overexpression induced a significant increase (22%) in the length of the retrograde runs (Fig. 5A) without changing retrograde run duration (Fig. 5B). This result suggests an enhancement in the processivity of retrograde APP movement. In addition, no changes were observed in the number of homologous or heterologous runs both in anterograde and retrograde net moving vesicles for APP-YFP + DYRK1A-mCherry compared with control (Fig. 5C,D). Therefore, DYRK1A overexpression exerts opposing effects on retrograde transport compared with harmine treatment, while no significant changes in anterograde run lengths or duration were observed in both experimental conditions. To assess whether the overexpression of DYRK1A modifies the distribution of segmental velocities of APP, we plotted the Gaussian mixture model with three components (Fig. 5E,F). DYRK1A overexpression reduced the fraction of anterograde velocities corresponding to slow Mode A and increased the proportion of Modes B and C (Fig. 5E; Table 2). No changes were observed for the average anterograde segmental velocities for each DYRK1A-overexpressing mode compared with mCherry control (Fig. 5E; Table 2). However, DYRK1A overexpression induced significant increases toward faster velocities in the retrograde direction (Modes A, B, and C) compared with control (Fig. 5F; Table 2). Moreover, significant reductions in the fraction of slower velocities (Mode A) were observed after DYRK1A overexpression (Fig. 5F; Table 2).
DYRK1A overexpression enhances the processivity and velocity of retrograde APP vesicles. A-D, APP axonal transport properties obtained from analyzing moving vesicles after DYRK1A overexpression. Average run length distance (A) and duration (B) of net anterograde or retrograde APP vesicles. Anterograde runs: n = 1069 mCherry, 1035 DYRK1A-mCherry; retrograde runs: n = 808 mCherry, 889 DYRK1A-mCherry. Average number of homologous (C) and heterologous (D) runs in anterograde and retrograde moving APP vesicles. Anterograde tracks: n = 461 mCherry, 434 DYRK1A-mCherry; retrograde tracks: n = 275 mCherry, 317 DYRK1A-mCherry. ***p < 0.001 (Student's t test). Anterograde (E) and retrograde (F) probability distribution of segmental velocities of moving APP-YFP vesicles represented as a Gaussian mixture model with three modes (A-C) in control (mCherry, gray) and DYRK1A-mCherry (red) human neurons. Mode center (dotted line) and mode fraction are displayed in Table 2. Segmental velocities n = 5954 antero, 3177 retro mCherry; 5788 antero, 3762 retro DYRK1A-mCherry. Significant differences in mode center (*) and mode fraction (#). G, H, 2D kernel density plots showing the probability of transitions from velocity n to n + 1 in control (G) and DYRK1A-overexpressing (H) neurons represented from red to green as probability increases in the four quadrants: Q1, anterograde-anterograde; Q2, anterograde-retrograde; Q3, retrograde-retrograde; Q4, retrograde-anterograde. I, Box and whiskers plot represents the Pearson correlation coefficient when comparing mCherry versus mCherry simulations (white box) and mCherry versus DYRK1A-mCherry (green box). Data are median, IQR, minimum and maximum. ***p < 0.001 (Student's t test).
To understand whether DYRK1A overexpression modulates the exchange in active molecular motors within the moving APP vesicle, we performed 2D Kernel density analysis of the velocity transitions as previously described (Fig. 5G,H). Similar configurations were observed with higher probability of velocity transitions in Q1 and Q3 compared with those observed for reversions (Q2, Q4) (Fig. 5G,H). The analyses of the probability distribution between DYRK1A overexpression (DYRK1A-mCherry) and control (mCherry) showed pronounced changes of slow to fast velocity transitions and vice versa in the anterograde to anterograde quadrant (Q1) (Fig. 5H, black arrows). Similarly, enhanced effects on slow to fast velocity transitions were observed in the retrograde to retrograde quadrant (Q3) (Fig. 5H, black arrow). These results suggest that increased DYRK1A expression enhances the probability for active motor exchange within the APP vesicle toward configuration favoring faster velocities.
Discussion
The protein kinase DYRK1A, encoded in DS critical region of chromosome 21, has been described as a modifier of APP, tau, and cytoskeletal proteins, thus gaining relevant attention in the induction of the pathology and neurodegeneration associated wth DS and AD. While distinct mechanisms have been described for cytoplasmic DYRK1A concerning the modification of microtubule-associated proteins and adaptors, little is known about the global intracellular processes in which DYRK1A may be involved. In addition, whether DYRK1A can modify neuronal intracellular trafficking, and specifically be involved in APP distribution remained unknown. Our experiments reveal new roles for DYRK1A in the control of intracellular dynamics through the regulation of the axonal transport of the APP vesicle. Together, our findings highlight that harmine or the overexpression of DYRK1A exert differential effects on APP transport. Therefore, the modulation of DYRK1A activity may emerge as a therapeutic target to restore transport defects in neurodegenerative diseases, such as AD and DS.
Brain organoids derived from iPSCs are being used to propel our understanding of drug effect, dissect neuronal intracellular processes, and overcome interspecies differences that should also reduce the demand of animal facilities (Lancaster and Knoblich, 2014; Chiaradia and Lancaster, 2020). We used a proteome analysis performed in brain organoids treated with harmine, a powerful inhibitor of DYRK1A, to identify relevant information about the molecular pathways affected by this specific pharmacological treatment (Frost et al., 2011; H. Kim et al., 2016). Harmine treatment induced changes in different protein profiles (Karmirian et al., 2021); however, the main enriched biological processes we highlight were associated with cytoskeleton organization and microtubule-dependent transport. This is consistent with previous works that revealed the impact that DYRK1A has on neuronal morphogenesis through the regulation of cytoskeletal dynamics (Martinez de Lagran et al., 2012; Ori-McKenney et al., 2016). Moreover, we found that harmine modulates the protein profiles of subunits of kinesin, dynein and myosin molecular motors, as well as different adaptor proteins associated with the control of vesicle transport. DYRK1A nuclear function mediates the transcriptional regulation of many RNA polymerase II genes raising the possibility that changes in protein levels associated with the transport process might be because of the regulation of gene expression (Di Vona et al., 2015). However, mARN candidates for upregulated and downregulated proteins did not show significant changes in mRNA amount after harmine treatment. These results pointed to the question of whether harmine-mediated DYRK1A inhibition may play a role in axonal transport regulation.
DYRK1A inhibition has been proposed as a putative therapeutic strategy in AD because of its ability to reduce APP phosphorylation and to restore abnormal phenotypes in AD mouse models (H. Kim et al., 2016; Melchior et al., 2019; Velazquez et al., 2019). Moreover, harmine treatment showed improved spatial learning and memory in mouse models of AD (He et al., 2015). Since harmine effects can be also attributed to the inhibition of other DYRK family members expressed in the brain (DYRK1B, DYRK2, and CLK1) (Tarpley et al., 2021), we validated our results by overexpressing active WT DYRK1A, and a kinase dead mutant DYRK1AK188R. Noteworthy, we do not expect in our experimental setting that the effect of harmine could be attributed to the inhibition of the monoamine oxidase MAO (Tarpley et al., 2021) since, in the brain, it is secreted by astrocytes and enriched in the extracellular space, but missing in our neuronal 2D cultures. Interestingly, we showed in iPSC-derived neurons that both harmine and the dominant mutant DYRK1AK188R expression reduced the overall density of APP vesicles within axons. Contrarily, WT DYRK1A overexpression led to an increase in APP axonal density through the enhancement of stationary and retrograde vesicles. The phosphorylation of APP has been associated with its increased transport since it localizes to the tips and neurites of N2A neuroblastoma cells (Muresan and Muresan, 2005). The phosphorylation of APP at Thr668 by DYRK1A can also be accounted as a regulatory step for APP internalization and colocalization within early endosomes (Ryoo et al., 2008; Muresan et al., 2009). APP cleavage into the pathologic amyloidogenic form occurs mainly in the intracellular endo-lysosomal compartment (Thinakaran and Koo, 2008; Otero et al., 2018). Therefore, consistent with our results, the increased vesicle density observed after DYRK1A overexpression may provide a critical retrograde pool of APP to the endo-lysosomal system for β-secretase processing (Almenar-Queralt et al., 2014; Otero et al., 2018). Interestingly, since harmine reduced the retrograde density of APP vesicles, DYRK1A inhibition may delay its amyloidogenic processing, supporting a therapeutic approach directed to sustain APP functions and reduce amyloid production (Sosa et al., 2017; Velazquez et al., 2019). In addition, since APP is found within newly formed endocytic CCV (Nordstedt et al., 1993; Marquez-Sterling et al., 1997), and DYRK1A regulates the assembly of the endocytic apparatus through the phosphorylation of multiple endocytic proteins required for CCV formation (Murakami et al., 2009, 2012), it is possible that DYRK1A might play a role in the internalization of APP. Interestingly, dynamin phosphorylation by DYRK1A appears to have a dual role in the interaction of dynamin with amphiphysin and endophilin, a key step in the internalization of synaptic vesicles (Chen-Hwang et al., 2002). Experimental evidence suggests that DYRK1A overexpression delays this initial process by reducing the assembly of CCV (Y. Kim et al., 2010; Murakami et al., 2012). However, this dual role of DYRK1A activity in mediating protein binding for the formation of CCV and in the uncoating of endocytosed CCV may result in a complex mechanism necessary for the internalization and loading of APP into the retrograde transport (Murakami et al., 2012). Our results are consistent with a role for DYRK1A in the control of APP distribution within axons, although the exact players mediating this process need to be further investigated.
Axonal transport is highly regulated by the phosphorylation of molecular motors, cargoes, and cytoskeletal proteins that control dynamic features of movement (Hirokawa et al., 2010; Gibbs et al., 2015; Yogev et al., 2016). Among the many targets of DYRK1A, different tubulin subunits and microtubule-associated proteins affecting cytoskeletal architecture are regulated by phosphorylation (Scales et al., 2009; Ori-McKenney et al., 2016). Nevertheless, the effect of DYRK1A on microtubule-mediated transport has not been directly studied. Therefore, we analyzed single-particle dynamics of axonal APP after DYRK1A inhibition or overexpression to unravel its effect on axonal transport. Harmine treatment increased the stochastic behavior of APP transport by increasing both the retrograde and anterograde runs in retrograde moving vesicles. Contrarily, DYRK1A overexpression facilitated the retrograde processivity by increasing the distance of the retrograde runs. Experimental evidence suggests that the composition of microtubule-associated protein is essential for recruiting or repelling molecular motors to the axonal microtubule lattice (Lacovich et al., 2017; Chaudhary et al., 2019), while other microtubule binding proteins act as molecular brakes for retrograde transport (Schwenk et al., 2014). Our results are consistent with the suggestion that DYRK1A activity can modify the microtubule-associated code (Ori-McKenney et al., 2016; Monroy et al., 2018), leading to a specific configuration that may facilitate retrograde axonal transport. Therefore, DYRK1A may control the switches in motor configurations while vesicles are moving, since we found changes in speed transitions after DYRK1A inhibition and more profound effects after overexpression, suggesting that DYRK1A is regulating the exchange between on/off states of molecular motors (Keller and Bustamante, 2000). Noteworthy, DYRK1A overexpression revealed significant increases in retrograde segmental velocities which lead to a facilitation of dynein motor activity, a process that can be mediated by DYRK1A-dependent modification of microtubule-associated proteins (Monroy et al., 2020). Interestingly, on an obstacle encounter, cargoes transported by dynein-dynactin often remain attached to the microtubule and undergo microtubule sliding (Dixit et al., 2008). Moreover, differential effects are exerted on kinesin and dynein motors by microtubule protein decoration (Dixit et al., 2008; Lacovich et al., 2017; Monroy et al., 2018). While our experiments did not rule out whether DYRK1A kinase activity exert a direct or indirect effect over the axonal transport of APP, we support a model in which local DYRK1A may modify many different steps in the regulation of axonal transport. Therefore, future efforts should be directed to dissect the effect of DYRK1A on individual proteins that either regulate vesicle loading, microtubule protein decoration, and/or kinesin/dynein motor activity in APP transport.
Our work highlights DYRK1A as a modulator of microtubule-associated proteins, motors and adaptors that in combination, play a cooperative role in the regulation of the axonal transport of APP. Moreover, we validated the effect of harmine in human neurons as a modulator of relevant neuronal processes that regulate intracellular dynamics. These results stress the relevance of DYRK1A in the regulation of APP metabolism in neurodegenerative diseases and shed light on a novel molecular pathway that can be targeted for therapeutic interventions in DS and AD.
Footnotes
T.L.F. was supported by Consejo Nacional de Investigaciones Científicas y Técnicas PUE Institucional IBCN 2018, Universidad de Buenos Aires Grant UBACyT 2017/2019, and Agencia Nacional de Promoción Científica y Tecnológica Grants PICT 2017-1648 and PICT 2019-0217. I.F.B. was supported by Facultad de Medicina, UBA fellowship, IBRO, and Boehringer Ingelheim.
The authors declare no competing financial interests.
- Correspondence should be addressed to Tomás L. Falzone at tfalzone{at}fmed.uba.ar
