CalciumZero: a toolbox for fluorescence calcium imaging on iPSC derived brain organoids

Abstract

Calcium plays an important role in regulating various neuronal activities in human brains. Investigating the dynamics of the calcium level in neurons is essential not just for understanding the pathophysiology of neuropsychiatric disorders but also as a quantitative gauge to evaluate the influence of drugs on neuron activities. Accessing human brain tissue to study neuron activities has historically been challenging due to ethical concerns. However, a significant breakthrough in the field has emerged with the advent of utilizing patient-derived human induced pluripotent stem cells (iPSCs) to culture neurons and develop brain organoids. This innovative approach provides a promising modeling system to overcome these critical obstacles. Many robust calcium imaging analysis tools have been developed for calcium activity analysis. However, most of the tools are designed for calcium signal detection only. There are limited choices for in-depth downstream applications, particularly in discerning differences between patient and normal calcium dynamics and their responses to drug treatment obtained from human iPSC-based models. Moreover, end-user researchers usually face a considerable challenge in mastering the entire analysis procedure and obtaining critical outputs due to the steep learning curve associated with these available tools. Therefore, we developed CalciumZero, a user-friendly toolbox to satisfy the unmet needs in calcium activity studies in human iPSC-based 3D-organoid/neurosphere models. CalciumZero includes a graphical user interface (GUI), which provides end-user iconic visualization and smooth adjustments on parameter tuning. It streamlines the entire analysis process, offering full automation with just one click after parameter optimization. In addition, it includes supplementary features to statistically evaluate the impact on disease etiology and the detection of drug candidate effects on calcium activities. These evaluations will enhance the analysis of imaging data obtained from patient iPSC-derived brain organoid/neurosphere models, providing a more comprehensive understanding of the results.

1 Introduction

Abnormal neuronal activity patterns, including hyperactivity and hypoactivity, have been observed in various neuropsychiatric disorders in various brain regions affecting information processing, sensory integration, and social interactions [4]. Understanding the precise mechanisms that drive altered neuronal activity in psychiatric disorders such as schizophrenia and autism is crucial. It will help identify strategies to normalize these patterns of neuronal activity in patients with neuropsychiatric disorders.

However, it was difficult to directly study individual neuronal and neural network activities in humans due to ethical reasons. One of the major breakthroughs in modeling human neuronal activity is to use of iPSC-derived brain organoid models. [23]. These models facilitate the dissection of disease mechanisms and provide a platform for drug effect evaluation in the human brain context. In addition, various chemical or genetically encoded sensors have been invented to indicate the dynamics of neuronal activities [11, 14]. For example, GCaMP [19] is a synthetic fusion protein of green fluorescent protein (GFP), calmodulin (CaM), and M13, a peptide sequence from myosin light-chain kinase, which is a genetically encoded calcium sensor. GCaMP calcium imaging [28] is a technique that utilizes genetically encoded calcium indicators to visualize and monitor neuronal activity in real time. This imaging method enables researchers to study the dynamics of individual neurons or neuronal populations, helping to unravel the complexities of neural circuits and their involvement in various cognitive processes and disease conditions.

Calcium imaging analysis, however, typically requires preprocessing to filter noise. Cropping and stabilization are the two important preprocessing components. Recent advancements in calcium imaging analysis tools, such as EZcalcium [2], CaImAn [3], and SmaRT2P [12], have greatly improved the handling of complex and voluminous imaging data. These tools provide automated, scalable solutions that incorporate advanced image processing techniques for motion correction, neuron segmentation, and functional connectivity analysis. Although they enhance the precision and efficiency of neuronal activity studies, integrating sophisticated computational methods like machine learning to deepen our understanding of neural functions and interactions , they impose high memory requirement due to the nature of machine learning. Thus, it is necessary to perform cropping to reduce the runtime memory.

The objectives of this project are two-fold: First, to create a user-friendly open-source toolbox designed specifically for the analysis of calcium (and other sensors for neuronal activity) imaging data obtained from patient iPSC-derived brain organoid models. Second, to expedite the process for end-users to obtain crucial outputs from well-established calcium imaging analysis pipelines and provide supplementary options for identifying disease mechanisms and assessing the effects of drugs on neuronal activities. Overall, we make the following contributions in this paper:

• We build CalciumZero, an open-source toolbox with a user-friendly GUI to support fully automated calcium imaging analysis pipeline.

• We make CalciumZero to be cross-platform compatible with little or no programming experience required. We provide standard application installers for direct GUI interaction.

• We develop a new calcium imaging analysis framework, featuring robust clustering, phase detection, statistical testing, peak calling, and phase analysis. This comprehensive approach enhances our ability to identify and quantify key dynamics in the data, ensuring efficient analyses.

2 Materials and methods

We developed CalciumZero a toolbox to improve end-user experience on calcium (and potentially other neuronal activity biosensor) imaging analysis. CalciumZero can be installed on common operating systems, such as Windows, Linux, and Mac, without programming experience. End-users do not need to take care of system packages/software such as specific libraries, while the toolbox provides options for the end-user to obtain sufficient details on the quality of the results and biological interpretation. In this section, we provide a general overview of our toolbox development.

2.1 Data acquisition

2.1.1 Generation of dorsal forebrain cerebral organoids/neurospheres

Dorsal forebrain patterned organoids were generated using an established protocol that was described in previous studies [5, 22]. Brain organoids were generated from 4 patients (22q11.2DS) and 4 controls as described previously [22]. At 40 Days In Vitro (DIV 40), organoids were transferred to maintenance media containing 1% N2 and 2% FBS. After DIV 80, they were maintained in media containing 1% N2, 2% FBS, and 1% B27 supplement until experiments. For drug effect testing experiments, neural progenitors were selected from DIV 12 brain organoids and differentiated into neurospheres as described previously [24].

2.1.2 Calcium imaging

A novel lentivirus vector has been developed and designed to drive the expression of GCaMP6s which is a genetically encoded calcium indicator known for its high sensitivity and fast kinetics. This vector, referred to as Lenti-Ubi-mRuby2-GSG-p2A-GCaMP6s, is engineered to utilize a ubiquitous promoter (Ubi), which ensures robust expression. Additionally, the vector includes a fusion of mRuby2, a red fluorescent protein, linked via a flexible GSG linker to the self-cleaving peptide p2A, and facilitating the independent expression of GCaMP6s. This innovative construct provides a powerful tool for monitoring intracellular calcium dynamics with a high temporal resolution, making it highly valuable for a wide range of neurobiological and cellular research applications. Brain organoids were transduced with the lentivirus vector 14 days before recording. $\hbox {Ca}^{2+}$ activity as indicated by GCaMP6 fluorescence intensity was assessed visually before image acquisition. GCaMP6s signal was recorded for 25 min using an Olympus IX81 epi-fluorescence microscope at 4$\times$ magnification, with a frame rate of 1 frame per second (fps), and recorded for 1500 frames. At the 500-second mark, a glutamate solution was added to the culture medium at a final concentration of 100 $\mu$M for brain organoids or 6.25 $\mu$M for neurospheres. Calcium activity was measured in a set of brain organoids and a set of neurospheres derived from brain organoids.

CalciumZero takes advantage of CaImAn [3], an advanced machine learning based calcium signal detection algorithm, and integrates it with both preprocessing and postprocessing tasks within its calcium imaging analysis tools to ensure more consistent and reliable results. The flowchart depicting the four primary tasks within this pipeline is presented in Fig. 1. These tasks include segmentation and stabilization for preprocessing the input data, peak detection, clustering, feature extraction, and the statistical analysis.

2.2 Analysis pipeline overview

Fig. 1

CalciumZero pipeline overview

2.2.1 Segmentation and cropping

Compared to the 2D culture, human brain organoids have a more well-defined round shape. However, the size and morphology of human brain organoids can vary significantly due to various factors, including culture conditions, developmental stages, and the use of different iPSC lines. The preprocessing steps, such as object segmentation and cropping, are crucial not only for optimizing memory usage in identifying the Regions of Interest (ROIs) but also for eliminating potential artifact-related background noises. Signals outside the boundary regions are discarded, and a standardized cropping process is applied to each input image, thereby reducing the overall image size. Moreover, in batch-mode processing, all input image sequences must be resized uniformly to meet the requirements of the CaImAn library for efficient batch processing.

To achieve this, we implement a histogram-based grayscale image segmentation followed by object detection, enabling the cropping of image sequences without omitting any significant objects. This approach effectively reduces both processing time and memory consumption. For enhanced accuracy, we further eliminate all background pixels surrounding the organoid, thus minimizing the likelihood of false object identification within the background.

To obtain the desired segmented image, we utilize an algorithm with an adaptive threshold [17]. This algorithm adjusts to local variations in image intensity, making it particularly suitable for images with varying levels of illumination or contrast from frame to frame. Considering that the luminosity of the image sequence exhibits minimal variation, we have relaxed the strictness of threshold determination. In addition to determining the threshold on a frame-by-frame basis, we also provide the option to find a coarse threshold with a stride greater than one. By applying the threshold, grayscale segmented image sequences with only black and white pixels are generated.

2.2.2 Stabilization

After reducing the input image size, it is essential to perform stabilization to account for affine transformations within the image sequences. Due to signal fluctuations caused by variations in the activity status of elements, conventional image registration algorithms often prove inadequate. Among the widely used algorithms, the Lucas-Kanade algorithm [9] and the Non-Rigid Motion Correction (NoRMcorre) algorithm [16] are notable. Although NoRMcorre offers faster computation times, our findings indicate that the Lucas-Kanade algorithm delivers better stabilization results.

To address the issue of prolonged execution times, we implement parallel processing for multiple input files, effectively reducing latency. We chose to use ImageJ2 [20], a TIFF-friendly tool, along with the Image Stabilizer plugin [8]. This plugin leverages the Lucas-Kanade algorithm to align neighboring slices to a reference slice while continuously updating the reference. The stabilizer is capable of addressing affine transformations and provides adjustable parameters to achieve optimal stabilization results.

2.2.3 Peak detection, clustering, and feature extraction

We directly utilize CaImAn [3] to perform CNMF-E algorithm to identify the ROIs. Within the identified ROIs, the peaks of these ROIs, which reflect the dynamics of intracellular calcium level, are extracted with a Python peak calling pipeline. We used and enhanced the peak detection and feature extraction algorithm in [1]. The features of individual peaks include rise, decay, amplitude, and frequency. In addition, the network features such as synchronized neuronal activity score will be calculated and the synchronized neuronal activity pattern will be visualized as a heatmap.

2.2.4 Statistical analysis

After peak detection and feature extraction, statistical analyses, including the Kolmogorov-Smirnov test [10] and Welch’s t-test [21], are conducted to explore variations between patient and control groups. To further analyze the time series data, a Self-Organizing Map (SOM) clustering algorithm [15] is applied. This process involves scaling the data and training the SOM to organize the series into clusters based on their characteristics. Following clustering, Dynamic Time Warping Barycenter Averaging (DBA) [15] is used to compute the average time series for each cluster, providing insights into the typical patterns within each group. This clustering approach categorizes signals into distinct groups, facilitating statistical analyses and exploration of variations between patient and control groups.

Subsequently, phase detection is conducted on each clustered time series, identifying distinct phases and capturing crucial dynamics such as timing and duration, which are visualized to aid in interpretation. To determine the significance of observed treatments, a Wilcoxon signed-rank test [25] is performed, assessing the statistical significance of differences and calculating metrics such as fold changes to visualize treatment effects. The final analytical steps involve peak calling on clustered data and phase analysis; peak calling isolates significant events in the time series, while phase analysis measures aspects such as frequency, baseline, rise, and decay of each phase, providing a comprehensive view of the time series dynamics.

3 Results

3.1 CalciumZero toolbox

CalciumZero is an open-source toolbox that is available for downloading from its GitHub repository. The installation process requires minimal to no programming knowledge, making it accessible to a wide range of users. For optimal utilization of CalciumZero, users can follow the detailed installation instructions provided. The toolbox includes several application installers, with versions available for Windows, Linux, and MacOS.

After downloading the installer, users can simply launch it to begin the installation process. Depending on the operating system, the installation typically takes about two minutes. Upon completion, the toolbox can be launched, at which point a GUI which will appear within seconds. Although CalciumZero functions as a fully automated calcium imaging toolbox, users have options to execute individual tasks separately for manual optimization. For instance, one might choose to pre-process input files initially, and then utilize the pre-processed output as input for subsequent tasks at a later time. Once the parameters are fixed, they could be used in a batch process mode. Compared with other tools (Table 1), CalciumZero is the first toolbox that incorporates both cropping and stabilization. It also provides enriched analytic feature extraction which usually requires an additional toolbox and manual operations.

Table 1 Features supported by different toolboxes

The GUI of CalciumZero is depicted in Fig. 2. During the CaImAn analysis step in the batch mode, CalciumZero performs preprocessing on the images collectively to generate uniform results. It is important to note that CalciumZero adjusts these images to conform to the required shape specified by the CaImAn library, as will be discussed in Sect. 3.2. CalciumZero offers two preprocessing options: a batch mode, which processes images collectively to generate unified input images, and a single-instance mode, which allows for the individual processing of images. In single-instance mode, each process operates independently, enabling users to apply different configurations to each instance. This flexibility allows users to process multiple data samples, fine-tuning the pipeline to achieve optimal analysis results. Additionally, the processes can be executed in parallel, thereby reducing the overall execution time.

Fig. 2

CalciumZero Quality Control GUI

3.2 Segmentation and cropping

An example of segmentation and object retrieval is illustrated in Fig. 3. This procedure allows for a reduction in preprocessing time for large input datasets, with the coarse-grind method [17] to yield a threshold close to the optimal value. The object segmentation and cropping processes are designed to minimize memory usage and eliminate potential artificial background noise. Representative experimental results, as presented in Fig. 4, demonstrate that these techniques significantly reduce computation time for subsequent pipeline tasks. The observed speedup, ranging from 4$\times$ to 10$\times$ with minimal margin, is anticipated due to the substantial reduction in image size of post-cropping. This efficiency gain is achieved by preventing unnecessary computations on background pixels and noise, thereby streamlining the processing pipeline.

Fig. 3

Segmentation and cropping results. A. Original input image; B. Segmented image; C. Background-removed cropped image

Fig. 4

Average computation time improvement by segmentation and cropping. Y axis is the average computation time of 6 samples. Significance levels: ns ($P> 0.05$), *($P\le 0.05$), **($P\le 0.01$), ***($P\le 0.001$)

3.3 Stabilization

Our image stabilization process effectively enhances the stability of captured images, resulting in significantly reduced motion blur and distortion. These methods perform robustly across challenging conditions, ensuring consistent image integrity and clarity. The stabilization process effectively corrects affine transformations in the input images. A stabilization example of frames before and after stabilization is shown in Fig. 5. The red dashed line shows the reference direction, while the green dashed line indicates the directions of the object before and after stabilization. It can be seen that the post-stabilization image shows a lower distortion angle.

Fig. 5

Stabilization result. The red dash lines represent the reference direction of the object. The green dash lines represent the directions of the object before and after stabilization, respectively

To visualize the CaImAn output while gaining the seamless transition from Jupyter Notebook to CalciumZero, we integrate multiple analytical diagrams and output images from CaImAn into a quality control tab in the application. Specifically, CalciumZero provides a ROI visualization window that helps users determine how the identified ROIs are accepted or rejected, to gain a better sense of the performance of the calcium imaging task, as shown in Fig. 2.

3.4 Peak detection, clustering, and feature extraction

In our results, Fig. 6 exemplifies the refined peak detection and feature extraction capabilities of our Python-based peak calling pipeline (PyPeakCaller), which is adapted from the PeakCaller software originally developed in MATLAB [1]. The representative time course in Fig. 6A illustrates the dynamics of intracellular calcium levels within the ROIs identified by CaImAn. The extracted peaks indicate significant events, showcasing the tool’s efficiency in detecting variations in calcium signaling.

Fig. 6

Representative PyPeakCaller results. A. Representative time course; B. Raster plot; C. Histogram of individual peaks height; D. Histogram of individual peak fall and rise time; E. Features of synchronization

The raster plot in Fig. 6B provides a detailed visual summary of the activity across multiple neurons, offering insights into the temporal patterns of calcium spikes. Figure 6C and D present histograms of individual peak heights, along with their rise and fall times respectively, emphasizing the variability and characteristic features of the calcium peaks detected across the study.

Additionally, the synchronization features of neuronal activity are highlighted in Fig. 6E, which presents various synchronization metrics and visualizes these patterns through a heatmap. This visualization procedure not only demonstrates the synchronized neuronal activity but also provides quantitative measurements to enhance our understanding of neuronal communications and interactions. Collectively, these results underscore the robustness of our analytical approach in capturing and quantifying complex biological dynamics.

3.5 Statistical analysis

We present two common applications of statistical analysis using our toolbox.

3.5.1 Statistical analysis of differences between patient and control groups

One common question in modeling psychiatric disorders is whether there is any significant difference between patient and control neurons in their neuronal activity pattern [27]. If so, what aspects are altered in patients? CalciumZero pipeline provides an optional pipeline to address these questions. The end-user provides an annotation file for their experimental conditions. The pipeline will group the data according to these experimental conditions and apply statistical tests to determine if there is any statistically significant difference neuronal peak features between conditions and visualize the results. We used a calcium imaging dataset that we previously obtained from our 22q11.2 microdeletion syndrome (22q11.2DS) brain organoid model [18] as an example to show the procedure. Deletions at the chromosome 22q11.2 locus, also called DiGeorge syndrome (OMIM#188400) are the most frequent deletion in humans. Adults and children with 22q11.2DS demonstrate cognitive, social, and emotional impairments [13]. 22q11.2 deletions are also one of the strongest genetic risk factors for schizophrenia [26]. Recently, our and other studies have repeatedly demonstrated that the neuronal activity is altered in the patients with 22q11.2DS [18, 6]. We use our dataset to demonstrate the pattern detection procedure with the CalciumZero pipeline.

Peak data from multiple patients and controls were combined in R, and key features such as rise time, fall time, peak height, and the number of peaks per trace were analyzed. Kolmogorov-Smirnov and Welch’s t-tests were conducted to identify significant differences between the patient and control groups. The results (p-value<2.2e$-$16) showed significant differences in rise time, fall time, and number of peaks between the groups, indicating distinct neuronal network synchronization patterns. This comprehensive pipeline enabled the identification of specific peak feature disparities, which are visually represented in Figs. 7 and 8, respectively.

Fig. 7

Difference between patients and controls in peak features

Fig. 8

Difference between patients and controls in neuronal network features. Significance levels: ns ($P> 0.05$), *($P\le 0.05$), **($P\le 0.01$), ***($P\le 0.001$)

Fig. 9

Drug effects on peak features: representative clusters

3.5.2 Statistical analysis of drug effects

Another critical application of the CalciumZero calcium imaging pipeline is to determine the effects of various drugs on neuronal activity patterns. To this end, we recorded the calcium activity in organoids at resting conditions and then stimulated the organoids with glutamate, a neuronal transmitter. We then applied the CalciumZero pipeline to analyze the details of the drug effects on calcium activity patterns. We first conducted a pattern clustering analysis to classify the neurons with similar calcium activity patterns (Fig. 9). Second, we determined if each cluster has a common phase transition before and after drug treatment. Finally, we extracted the features (individual neurons as well as neuronal networks) of each phase (Fig. 10). A Wilcoxon signed-rank test was conducted to determine if there is any significant change for each feature before and after treatment. We demonstrated that the peak height is altered after glutamate stimulation in all the clusters (Fig. 11).

Fig. 10

.Drug effects on peak features: peak change detection for cluster 4

Fig. 11

Drug effects on peak features: statistical analysis of peak height across all clusters under glutamate stimulation. Significance levels: ns ($P> 0.05$), *($P\le 0.05$), **($P\le 0.01$), ***($P\le 0.001$)

4 Discussion

The goal of this project is to develop a user-friendly open-source toolbox specifically tailored to analyze calcium imaging data for detecting disease mechanisms and drug effects in patient iPSC-derived brain organoids/neurospheres. Our main priority was to provide pipelines to get key outcomes that are critical for the end users to understand the pattern changes of cellular calcium dynamics between patients and controls. Compared with existing general-purpose software packages, our toolbox optimizes the pre- and post-processes for the investigators who work on patient iPSC-derived brain organoids to determine disease mechanisms and drug effects. In addition, we further integrated novel peak detection techniques aiming at identifying sudden changes in calcium levels from time series signals obtained.

The CaImAn toolbox provides methods to do motion correction, signal extraction, and component evaluation. In the context of fluorescence calcium imaging, there are approximately 50 parameters that may affect the outcome. In CalciumZero toolbox, we have a set of default values for those parameters. If users want to tune them, CalciumZero has a user-friendly window to adjust those parameters, as shown in Fig. 12. To better compare and visualize the differences across different parameter settings, we also provide a Jupyter Notebook version of CalciumZero to ease the process of parameter optimization if needed.

Fig. 12

CalciumZero parameter adjustment interface

To ensure that CalciumZero does not monopolize all CPU resources, the toolbox incorporates a dynamically adjusted threshold that limits the maximum number of pipeline instances that can run concurrently. On most systems, a maximum of four instances can be processed in parallel. However, users have the option to manually adjust this setting based on their system’s computing capabilities. To achieve the desired performance, the primary system requirement is to have sufficient RAM capacity. Calcium imaging analysis is a memory-intensive task, for which we recommend a minimum of 32 GB of RAM, particularly when parallel processing is intended. In general, 24 GB of RAM is expected to be adequate for standard operations.

Our analysis in CalciumZero demonstrates treatment efficacy by clustering, phase detection, peak calling, and statistical testing. Clustering the time series enables focused comparisons between patient and control groups, revealing distinct patterns. Phase detection within these clusters highlights differences in timing and duration, suggesting treatment impacts on synchronized neuronal activity. Peak calling further identifies specific features-such as rise, decay, and frequency-attributable to treatment effects. Finally, statistical testing confirms significant differences, underscoring that the treatment induces measurable changes in neuronal dynamics.

Many aspects still need to be improved in the future. For example, we plan to better visualize the results where multiple pipeline processors are running at the same time. It will display different settings and provide a clearer boundary among different pipeline processors. For pipeline processors with different settings on the same data, it will also provide direct comparisons for users to find an optimal setting.

5 Conclusion

Overall, we developed CalciumZero, an open-source tool with a user-friendly GUI that provides a self-containing pipeline to conduct fluorescence calcium imaging analysis for brain organoid models. We make CalciumZero to be cross-platform compatible with little or no programming experience required. The tool is designed to bridge the gaps between pre- and post-processes, streamlining the statistical analysis with an application on the psychiatric disorders.