If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
1 These authors contributed equally to this work. 2 Current address: Wave Life Sciences, Cambridge, MA 02138, USA. 3 Current address: Ascidian Therapeutics, Boston, MA, 02135, USA. 4 Current address: Novartis Institute of Biomedical Research, Cambridge, MA, 02139, USA. 5 Current address: Pfizer Rare Disease Research Unit, 1 Portland Street, Cambridge, MA, 02139, USA. 6 Current address: Verge Genomics, 2 Tower Pl, South San Francisco, CA, 94,080, USA. 7 Current address: Rgenta Therapeutics, 700 Main St, Cambridge, MA 02139, USA. 8 Current address: Arbor Biotechnologies, 20 Acorn Park Drive, Cambridge, MA 02140, USA.
The development of phenotypic assays with appropriate analyses is an important step in the drug discovery process. Assays using induced pluripotent stem cell (iPSC)-derived human neurons are emerging as powerful tools for drug discovery in neurological disease. We have previously shown that longitudinal single cell tracking enabled the quantification of survival and death of neurons after overexpression of α-synuclein with a familial Parkinson's disease mutation (A53T). The reliance of this method on manual counting, however, rendered the process labor intensive, time consuming and error prone. To overcome these hurdles, we have developed automated detection algorithms for neurons using the BioStation CT live imaging system and CL-Quant software. In the current study, we use these algorithms to successfully measure the risk of neuronal death caused by overexpression of α-synuclein (A53T) with similar accuracy and improved consistency as compared to manual counting. This novel method also provides additional key readouts of neuronal fitness including total neurite length and the number of neurite nodes projecting from the cell body. Finally, the algorithm reveals the neuroprotective effects of brain-derived neurotrophic factor (BDNF) treatment in neurons overexpressing α-synuclein (A53T). These data show that an automated algorithm improves the consistency and considerably shortens the analysis time of assessing neuronal health, making this method advantageous for small molecule screening for inhibitors of synucleinopathy and other neurodegenerative diseases.
Predictive cellular models for human disease pathologies play critical roles in target validation, drug discovery and the development of new therapeutics. Transformed cell lines, primary cells, and more recently, iPSC-derived human cells have been used in assay development, each with different advantages and weaknesses. Transformed cell lines can be easily modified to model disease pathways and develop convenient assays, however, they have inherent limitations that may complicate interpretation of results, leading to subsequent failures in preclinical and clinical development [
. Primary neuronal cultures isolated from rodents provide more physiologically relevant in vitro systems for modeling neurological diseases, yet they can have limited insight into human disease pathophysiology. With the remarkable advancements in stem cell technology, human iPSC-derived neurons represent a more clinically appropriate cellular model to study disease-relevant pathologies [
Longitudinal characterization of transcriptomic, functional, and morphological features in human iPSC-derived neurons and their application to investigate translational progranulin disease biology.
α-synuclein is a key pathological driver in multiple neurodegenerative diseases. Collectively known as “synucleinopathies”, Parkinson's disease (PD), dementia with Lewy bodies (DLB) and multiple system atrophy (MSA) all exhibit pathological cellular degeneration caused by abnormal misfolding, accumulation and aggregation of α-synuclein. Genetic linkage of mutations and amplifications of the α-synuclein gene (SNCA) to human disease validates using α-synuclein overexpression in preclinical cellular models to investigate disease-relevant pathophysiology [
. In this context, we have recently established an assay that monitors survival of iPSC-derived human cortical neurons following the overexpression of α-synuclein. The overexpression of either WT or a PD-related mutant (A53T) α-synuclein in human iPSC-derived cortical neurons induces neurotoxicity, and a subsequent decrease in survival [
The survival assay is an established method to determine the cumulative risk of death, and Cox cumulative hazard analysis is widely used to describe the outcomes using hazard ratio (HR) and its confidence interval (CI) [
. Traditionally, this manual tracking has been performed by personnel trained to distinguish live from dead neurons and is highly labor-intensive and time-consuming. For example, counting neurons from one 96-well plate can take more than two weeks to complete. In addition, inconsistencies in counting criteria within or between individual counters may introduce variability, necessitating recounting of the same data set by different individuals.
To address these challenges, we took advantage of an automated live imaging system (BioStation CT) and machine learning-based image analysis software (CL-Quant) to develop an automated analysis method. The algorithm is used with the CL-Quant software and yielded results with neurons transfected with α-synuclein A53T that are similar to those obtained with manual counting, but with improved consistency and reproducibility. The final algorithm also revealed the beneficial effects of BDNF treatment on neuronal survival, neurite outgrowth and healthy time measurements, and established the capacity of the algorithm to detect a drug treatment effect. These results indicate that automated analysis of neuronal survival with our novel algorithm is suitable for the future evaluation of neuroprotective small molecules in models of synucleinopathy and potentially other neurodegenerative disorders.
Materials and methods
Plasmids
Expression construct for α-synuclein A53T was obtained and licensed from the Whitehead Institute (Massachusetts Institute of Technology, Cambridge, MA). pSF-CAG plasmid was purchased from Oxford Genetics (Oxford, UK). mApple or EGFP was cloned into pSF-CAG. MAP2 promoter was then switched using Gibson assembly to produce pSF-MAP2-mApple. (NEB, Ipswich, MA).
Neuron plating, culture, and transfection
iPSC-derived human cortical neurons were purchased from Fujifilm Cellular Dynamics (iCell GABA neurons, cat# R1013) and plated onto Corning BioCoat Poly-d-Lysine 96-well plates (cat# 3842) which were coated with Laminin (Sigma L2020; 1:300 dilution in water) one day prior to neuron plating. Approximately 55,000 neurons were seeded per well. Culture medium (200 µl/well) was changed the next day and then one-half of the medium was changed every 3–4 days. At 7 days after plating, red fluorescence tracer plasmid (pSF-MAP2-mApple) and empty vector or α-synuclein A53T mutant plasmid were introduced into neurons using Lipofectamine 2000 (Thermo Fisher) according to the manufacturer's guide, with slight modification as described previously [
]. The same procedure was used to introduce pSF-CAG-EGFP and pSF-CAG-mApple into neurons to confirm the co-transfection in Supplementary Fig. 1. For the BDNF treatment experiment in Fig. 4, BDNF was added to culture medium (50 ng/ml as final concentration) immediately after transfection.
Automated time-lapse live imaging
After transfection, the plate was inserted into BioStation CT (Nikon) which consists of a cell incubator and automated imaging system. Six technical replicates (wells) of neurons were automatically imaged with the 10x objective lens every 12 h for 14 consecutive days. In each well, 4 × 4 tiled images were captured. Individual neurons were tracked by monitoring red fluorescence with the CL-Quant software (Nikon) and Algorithm 3 developed by Nikon. A cell ID number was given to each cell body and tracked during the imaging period.
Survival analysis with Cox proportional hazard model
The risk of neuron death was determined over time, and Cox proportional hazard model was used to estimate cumulative risk of death and hazard ratio [
. The hazard ratio is a calculation of the risk ratio between the empty vector transfected group and the α-synuclein A53T transfected group. By setting the hazard ratio of the Vector 25 ng transfected group equal to 1, we can estimate the hazard ratio of other groups. Then, by comparing these hazard ratios, it can be determined how much the risk of cell death will be increased by α-synuclein A53T expression. The hazard ratio calculation was performed using the open-source ‘Survival package’ of the R program, which was also utilized to conduct statistical analysis and create plots (https://cran.r-project.org/web/packages/survival/index.html).
Algorithm development and optimization
The "segmentation part" and the "tracking part" are the two components that make up the algorithms. Characteristics of the previous method [
] are comparable to those of Algorithm 1. Although it demonstrated acceptable segmentation and tracking capabilities for "motor neurons" as defined in the research, it did not demonstrate good sensitivity for our "cortical neurons" analysis. Therefore, we modified the algorithm that produced Algorithm 3 by adding a top hat filtering technique (kernel size 20 µm) to the "segmentation part". The two algorithms share the same "tracking part".
Results
Development of algorithms for the automated detection of neuronal cell bodies, neurites, and nodes
Utilizing the neuronal survival assay with manual counting analysis, we have previously shown that overexpression of a mutant form of α-synuclein (A53T) in human iPSC-derived cortical neurons induces death over time [
]. Here, we set out to improve this method by developing and optimizing algorithms for automated analysis. iPSC-derived human cortical neurons were transiently transfected with plasmid encoding α-synuclein A53T or empty vector, together with a plasmid encoding mApple, a fluorescent protein used to trace the cell body and neurites of neurons (Fig. 1A). We chose red fluorescence protein mApple to avoid phototoxicity since it is generally known that repetitive imaging for GFP can induce it. In addition, the red channel filter on BioStation CT provides better quality than green channel.
Fig. 1Development of algorithms for automated analysis of neuronal survival assay using BioStation CT and CL-Quant. (A) Workflow diagram showing the development of automated algorithms for neuronal survival analysis. iPSC-derived human neurons were transfected at DIV8 and maintained in BioStation CT with images taken every 12 h. The collected images were quantified by manual counting to create a reference data set for the subsequent development of automated algorithms. (B) Representative fluorescence and brightfield images acquired by the automated imaging system of BioStation CT. Neurons were transfected with red fluorescence tracer plasmid (pSF-MAP2-mApple). (C) Example of automated detection shows cell bodies in red, neurites in yellow and nodes in blue. The information can be plotted to show cell survival, neurite outgrowth and neuronal health over time in culture.
Both fluorescence and brightfield images were captured every 12 h over two weeks using the BioStation CT live imaging instrument (Fig. 1B). Images were analyzed by both manual counting and automated analysis using CL-Quant software. Neuronal cell bodies, neurites and nodes were successfully detected by CL-Quant using our algorithms (Fig. 1C). After completion of imaging, manual tracking data was used for comparison to improve the algorithm.
Algorithm 3 detects α-synuclein A53T-induced neuronal death with improved sensitivity and accuracy
To assess the sensitivity and accuracy of the automated algorithms as compared to manual counting, a test set of data was generated by overexpressing various doses (25, 50, 75, 100, 125 ng) of α-synuclein A53T and control plasmids. Overexpression of α-synuclein A53T significantly increased the risk of neuron death represented as hazard ratio (HR) in a dose-dependent manner when analyzed by manual counting (Fig. 2A).
Fig. 2Comparison of manual counting, Algorithm 1, and Algorithm 3 in quantifying neuronal survival. iPSC-derived human neurons were transfected with varying doses of α-synuclein A53T or control plasmids along with mApple tracer. Plots of calculated cumulative hazard over time, with statistical parameters in table below. (A) α-synuclein A53T expression caused a dose-dependent increased risk of neuron death relative to control, as assessed by manual counting. (B) Our initial algorithm (Algorithm 1), which distinguishes cell bodies based on the fixed fluorescent threshold, was not able to detect weak fluorescence in cell bodies. This resulted in a relative inability to differentiate α-synuclein-induced toxicity according to plasmid amount. (C) Modifications of Algorithm 1 to develop Algorithm 3, which utilizes a flexible fluorescence threshold, enhanced the detection of cell bodies with lower fluorescence levels. This revealed a dose-dependent increased risk of neuron death relative to control, similar to what was observed with manual counting. Cox proportional hazard analysis was used to estimate relative risk of death or hazard ratio (HR), and the P value was determined by the logrank test. CI, confidence interval; N, number of neurons. p < *** 0.001, ** 0.01, * 0.05, compared to the lowest dose of the control vector. (Representative result of three independent experiments.).
The initial algorithm (Algorithm 1) detected the α-synuclein-induced cell death but with much lower sensitivity than the results obtained with manual counting, as indicated by lower hazard ratios and fewer cells counted (Fig. 2B). When using Algorithm 1, the hazard ratio of α-synuclein A53T transfected groups ranged from 1.144 to 1.592 depending on the plasmid amount, whereas manual counting resulted in 1.484 to 2.085. The number of neurons counted was 20–30% lower in all groups when compared to manual counting. Further comparative analysis revealed that neurons expressing weak fluorescence were included in manual counting but were excluded in the automated detection by Algorithm 1 (Supplementary Fig. 2). A significant update to the algorithm as detailed in the Materials and Methods section resulted in a new version, Algorithm 3. While Algorithm 1 detects cell bodies based on the ‘fixed’ threshold of fluorescence levels, Algorithm 3 has a ‘flexible’ cut-off to equalize background intensity level of the images (see Materials and Methods). This modification resulted in hazard ratios (1.456 - 2.066) that were much closer to those of manual counting (1.484 - 2.085) (Fig. 2C). In addition, Algorithm 3 was able to detect plasmid amount-dependent toxicity in the Vector control group as measured by transfection efficiency and hazard ratio with statistical significance (75, 100, 125 ng), indicating the algorithm's sensitivity. Therefore, Algorithm 3 was used in subsequent experiments for further validation.
Algorithm 3 sensitively measures decreases in neurite length and node number induced by α-synuclein A53T
In addition to neuronal survival, neurite length and node number are other important indicators in assessing neuronal health. As CL-Quant is able to detect neurites and nodes using our algorithm (Fig. 1C), we next evaluated whether the algorithm could correctly and efficiently capture changes in neurites and node numbers induced by α-synuclein A53T toxicity. In control cells transfected with empty vector, total neurite length increased 1.5∼2.3 fold in a time-dependent manner when images were compared to the first frame. However, in α-synuclein A53T transfected neurons, neurite length decreased over time and in a dose-dependent manner according to the amount of plasmid transfected (eg. ∼55% decrease of neurite in 125 ng group when compared to the first frame) (Fig. 3A and 3B).
Fig. 3Algorithm 3 reveals the effect of α-synuclein A53T on neurite length and node number over time in culture. (A) Neurons transfected with α-synuclein A53T showed decreased total neurite length compared to empty vector transfected neurons (y-axis) over time (x-axis), suggesting synuclein-dependent degeneration of neurites. Total neurite length in each frame of images was normalized to the first frame for comparison of the slopes between groups. (B) Area under the curve in (A) was calculated (multiplication of x-axis and y-axis) and plotted using Graph Pad Prism™ software. (Mean ± SEM). (C) The number of nodes per neuron (y-axis) was reduced over time (x-axis) in cells overexpressing α-synuclein A53T compared to empty vector transfected neurons, indicating a synuclein-dependent decrease of neuronal health. (D) Area under the curve in (C) was calculated (multiplication of x-axis and y-axis) and plotted by Graph Pad Prism™ software. (Mean ± SEM). (E) Healthy time (the duration that neurons display three or more nodes) was significantly shortened in α-synuclein A53T overexpressing neurons in a dose-dependent manner. Note that the healthy time even in empty vector transfected neurons shows an inverse correlation to the amount of plasmid transfected, demonstrating the sensitivity of our algorithm to reveal toxicity due to the transfection alone. (Mean ± SD). For (B), (D) and (E), statistical differences were assessed by one-way ANOVA with a post-hoc Dunnett's test. p < *** 0.001, ** 0.01, * 0.05, compared to the lowest dose of the control vector. (Representative result of three independent experiments.).
Similarly, the number of nodes (neurite projection sites from the cell body) was reduced in α-synuclein A53T overexpressing neurons in a time-dependent manner (Fig. 3C and 3D). For example, the node number decreased by 42% when neurons were transfected with 125 ng of plasmid.
We next compared the ‘healthy time’ of neurons, defined as the duration where neurons displayed three or more nodes during the two weeks of imaging. Analysis with Algorithm 3 indicated that the healthy time of neurons expressing α-synuclein A53T was significantly shortened as compared to that of neurons transfected with the empty control vector (Fig. 3E). The healthy time was decreased by 75% for neurons transfected with 125 ng of plasmid as compared to control neurons. Interestingly, we also observed that the healthy time was reduced by 24–29% in control group neurons upon transfection with higher amounts of the empty vector plasmid, suggesting the algorithm was sensitive enough to detect even small increases in toxicity. These results highlight the capacity of Algorithm 3 to accurately detect the neurites and nodes and demonstrates that these variables are a sensitive reflection of neuronal fitness (i.e., overall health).
Algorithm 3 reveals the neuroprotective effects of BDNF
We next sought to determine whether Algorithm 3 could detect treatment effects that reduce α-synuclein toxicity. To this end, we used BDNF as a positive control because it is known to increase neuronal survival and neurite outgrowth [
. BDNF (50 ng/ml) was added to the culture medium immediately after transfection and this treatment (Vector+BDNF and A53T+BDNF) significantly improved all the endpoints (1.9- and 1.6-fold increase of survival; 2.6- and 1.7-fold increase of total neurite length, 1.9- and 1.4-fold increase of node number; 2.3- and 1.6-fold increase of healthy time) when compared to vehicle-treated cells (Vector+Vehicle and A53T+Vehicle) (Fig. 4A-G). These results indicate that Algorithm 3 successfully captures the beneficial effects of BDNF on neuronal health, and validates the algorithm for future small molecule evaluation and novel drug discovery.
Fig. 4Algorithm 3 reveals the beneficial effects of BDNF on survival, neurite length, node number, and healthy time. (A) Representative images at Frame 3 and Frame 20 during two weeks of imaging. Each frame has a 12 hr interval. (B) Addition of BDNF (50 ng/ml) to neurons after transfection of 25 ng of A53T plasmid) decreased the risk of neuron death (i.e., hazard ratio) both in vector and α-synuclein A53T groups (Vector+BDNF and A53T+BDNF, respectively), compared to vehicle treated groups (Vector+vehicle and A53T+vehicle, respectively). Cox proportional hazard analysis was used to estimate relative risk of death, or hazard ratio (HR) and the P value was determined by the logrank test. CI confidence interval; N, number of neurons. p < *** 0.001, ** 0.01, * 0.05. (C) BDNF treatment increased neurite length (y-axis) over time (x-axis) both in vector and α-synuclein A53T groups (Vector+BDNF and A53T+BDNF, respectively), compared to vehicle treated groups (Vector+vehicle and A53T+vehicle, respectively). (D) Area under the curve in (C) was calculated (multiplication of x-axis and y-axis) and plotted using Graph Pad Prism™ software. (Mean ± SEM). (E) Node numbers per neuron (y-axis) was increased over time (x-axis) by BDNF treatment both in vector and α-synuclein-A53T groups (Vector+BDNF and A53T+BDNF) compared to vehicle treated groups (Vector+vehicle and A53T+vehicle). (F) Area under the curve in (E) was calculated (multiplication of x-axis and y-axis) and plotted by Graph Pad Prism™ software. (Mean ± SEM). (G) Healthy time in neurons was increased by BDNF treatment in accordance with neurite and node number analysis. (Mean ± SD). For (D), (F) and (G), statistical differences were assessed by unpaired t-tests. p < *** 0.001, ** 0.01, * 0.05, compared to vehicle control. (Representative result of three independent experiments.).
Over the past decade there have been significant technical advances in the development of high-content imaging systems and related software which have enabled automated image acquisition and analyses [
Automated fluorescence lifetime imaging high-content analysis of Förster resonance energy transfer between endogenously labeled kinetochore proteins in live budding yeast cells.
. This in turn, has allowed the development of sensitive and accurate phenotypic assays for target validation and compound evaluation to identify potential treatments for various diseases including neurodegenerative disorders [
. As previously reported, BioStation CT is an integrated instrument for live cell imaging and CL-Quant is a machine learning-based software for analyzing images with teach-by-example interfaces, which can provide various digital quantifications in stem cells, neurons, as well as 3-dimensional cultures of cancer cell lines [
. Here, we have developed and optimized an algorithm for use with CL-Quant software that enables the automated detection of neuronal cell bodies, neurite projections and node numbers. We demonstrate how this algorithm permits the automated analysis of neuronal survival and health in iPSC-derived human cortical neurons overexpressing the familial PD mutant form of α-synuclein (A53T).
α-synuclein pathology is widespread in the PD patient brain, most notably affecting dopaminergic neurons within the substantia nigra pars compacta, but is also prevalent in cortical regions. The accumulation of α-synuclein in neurons leads to cellular dysfunction and neurotoxicity in cell and animal models [
. While identifying measurable α-synuclein phenotypes in human iPSC-derived neurons is challenging, it is nonetheless highly desirable due to the obvious benefits of having a tractable, physiologically-relevant, and cell-based system to model synucleinopathies. To this end, longitudinal tracking of individual neuron survival to study α-synuclein A53T-induced neurotoxicity in vitro provides an enhanced sensitivity and robustness, allowing for the detection of phenotypes often too subtle to measure with other neuronal assay systems [
. One disadvantage of the original assay format is the labor-intensive manual cell counting and tracking of hundreds of individual neurons over time, which has limited the utility to a very low throughput confirmatory assay.
With the information obtained by CL-Quant software running our novel algorithm, we can obtain cumulative hazard ratios, total neurite length, number of nodes, and healthy time which each reflects the cellular fitness of neurons. By adjusting the threshold of fluorescence detection of Algorithm 1, the improved algorithm (Algorithm 3) was able to distinguish different levels of cellular fitness according to the amount of α-synuclein plasmid transfected. The neuronal survival, neurite length, number of nodes, and healthy time all showed inverse correlations with α-synuclein concentration or “dose” in neurons assessed using automated analysis. Algorithm 3 was also able to detect the beneficial effects of BDNF treatment in neurons, which supports the application of this algorithm to small molecule testing and drug discovery efforts. Indeed, we were able to successfully evaluate the effectiveness of YTX-7739 in iPSC-derived human cortical neurons [
A brain-penetrant stearoyl-CoA desaturase inhibitor reverses α-synuclein toxicity in synucleinopathy models in vitro and in Parkinson's disease-like mice.
], which is the lead compound of Yumanity Therapeutics and has finished the Phase I clinical trial for Parkinson's disease in 2021.
The method is also amenable to target identification and confirmation using siRNA approaches (data not shown). In addition, the automated algorithm shortened the analysis time to 10 h per 96-well plate, a remarkable improvement from ∼2 weeks per plate by manual counting. Finally, the algorithm eliminated any concerns of intra- or inter-counter variability (Supplementary Fig. 2).
Taken together, the automated algorithm described here overcomes the limitations of manual counting with similar precision, accuracy and better consistency. These factors will increase the throughput and allow for a more rigorous screening assay. It can serve as a useful tool to evaluate potential therapeutic compounds and it will enable us to address complex hypotheses in this highly challenging space of neurodegenerative disease drug discovery. Future applications could be expanded to different cell types (ie: glutamatergic and dopaminergic neurons) and different proteinopathies [
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Automated fluorescence lifetime imaging high-content analysis of Förster resonance energy transfer between endogenously labeled kinetochore proteins in live budding yeast cells.
A brain-penetrant stearoyl-CoA desaturase inhibitor reverses α-synuclein toxicity in synucleinopathy models in vitro and in Parkinson's disease-like mice.
Longitudinal characterization of transcriptomic, functional, and morphological features in human iPSC-derived neurons and their application to investigate translational progranulin disease biology.
05186-X&id=gr1.jpg){kind=link}
05186-X&id=gr2.jpg){kind=link}
05186-X&id=gr3.jpg){kind=link}
05186-X&id=gr4.jpg){kind=link}