Sample preparation
hiPSC-derived neurons
Fibroblast donors were male and of European ancestry—these research subjects were enrolled in the Sibling Study of Schizophrenia at the National Institute of Mental Health in the Clinical Brain Disorders Branch (NIMH, protocol 95M0150, NCT00001486, Annual Report number: ZIA MH002942053, DRW PI) as previously described [13]. Early passage fibroblasts (< 5 passages) were reprogrammed into hiPSCs as previously described [14], and subsequently differentiated through neural progenitor stages into cortical neurons. Neurons were co-cultured in 24-well ibidi plates (Cat. No. 82406, ibidi GmbH, Munich, Germany) with astrocytes prepared from the cortices of neonatal rats to promote neuronal maturity as previously described [13, 15]; and were maintained with partial media changes twice a week for up to 10 weeks (Day in Vitro (DIV70)).
Animals
Timed-pregnant Wistar rats for astrocyte cultures were obtained from Charles River Laboratories (Wilmington, MA, USA; stock Crl:WI003). To obtain fetal tissue, pregnant dams were euthanized by carbon dioxide asphyxiation followed by cervical dislocation. Mice were purchased from Jackson laboratories (Bar Harbor, ME, C57BL6/J; stock #000,664), and bred for the generation of postnatal day 0 mice primary neuronal cultures. To obtain neonatal tissue, pups were anesthetized by being placed on ice, followed by rapid decapitation, and dams were returned to the breeding colony. All rodents were housed in a temperature-controlled environment with a 12:12 light/dark cycle and ad libitum access to standard laboratory chow and water.
Mouse primary cortical cultures
Mouse cortical neurons were cultured as previously described with modifications [16]. Briefly, on the day of birth, mice were anesthetized by being placed on ice, then rapidly decapitated and their cortices removed. Cortical tissue was dissociated using papain, and plated at a density of 2.5 × 10^5 per well on a 24-well ibidi plate (Cat. No. 82406, ibidi GmbH, Munich, Germany) coated with poly-D-lysine and laminin. Neurons were maintained in culture with partial media changes every 2 days, and imaged between DIV14 and DIV15.
Viral transduction
hiPSC-derived neurons were transduced at DIV23 with adeno-associated virus expressing mRuby2 and GCaMP6s under the control of a synapsin promoter (MOI ~ 6 × 10^4, Addgene viral prep # 50,942-AAV1 [17]. Following a full media exchange on DIV25, neurons were cultured for at least 21 days and imaged on DIV 42 or 63. Mouse primary cultures were transduced with 1:10 viral concentration used in human experiments of the same virus (human synapsin 1 promoter was ubiquitously expressed in mouse neurons). Mouse primary cultures were infected at DIV5–DIV8 prior to DIV14–DIV15 recordings.
Image acquisition
LSM780 confocal microscope
Primary mouse cortical cultures and hiPSC-derived neurons were imaged in culture media on a Zeiss LSM780 equipped with a 10X/0.45NA objective, a temperature- and atmospheric-controlled enclosure to maintain neurons at 37° and 5% CO2. A reference image was acquired for each field of mRuby fluorescence followed by a time-series was acquired at 4 Hz for 8 min. In some cases, tetrodotoxin (TTX, 1uM) was then added to block synaptic transmission and incubated for at least 5 min prior to imaging to equilibrate.
Spinning disk confocal microscope
Neurons were removed from culture media and were continuously perfused with artificial cerebro-spinal fluid (ACSF) containing (in mM): 128 NaCl, 30 glucose, 25 HEPES, 5 KCl, 2 CaCl2, and 1 MgCl2 (pH 7.3) [15]. Imaging was performed at DIV56 or DIV70 on a custom-built Zeiss AxioExaminer Z.1 equipped with a live-slice Yokogawa spinning disk module, Flash4.0 V3 sCMOS camera, and a 20X/1.0NA water immersion objective. A reference image was acquired using mRuby fluorescence, then a time-series was acquired at 10 Hz for 5 min. For experiments in which pharmacological blockers were added, TTX (1 uM) was included in the perfusate for at least 5 min prior to imaging.
Acquisition parameters
From all microscopes, two image types are collected: a time-series of GCaMP6s fluorescence and a reference image of mRuby to demarcate infected neurons. The reference image of the LSM780 scope is downsampled using the MATLAB function imresize to match the time-series image in X and Y dimensions.
Scope
|
Reference image
X Y
|
Pixel to micron
|
Time-series image X Y
|
Pixel to micron
|
---|
LSM 780
|
1024 × 1024 pixel
|
0.83 × 0.83 μm per pixel
|
256 × 256 pixel
|
3.32 × 3.32 μm per pixel
|
Spinning disk
|
1024 × 640 pixel
|
0.645 × 0.645 μm per pixel
|
1024 × 640 pixel
|
0.645 × 0.645 μm per pixel
|
Toolbox installation and software requirements
All data processing for CaPTure is conducted in MATLAB (Version 2017a or later). The processing pipeline is divided into several steps as described below, the execution of which are explained in the following repository https://github.com/LieberInstitute/CaImg_cellcultures. The repository consists of a `toolbox` directory whose path needs to be added to the MATLAB working directory to run any of the processing steps. The directions to download and install the toolbox are described in the ‘installation’ step of the repository.
Statistics
To calculate the effect of pharmacological manipulations, the lmerTest R package [18] was used for performing linear mixed effects modeling as a function of treatment main effect (Baseline versus TTX) and used cell line and the cell culture experimenter as the random intercepts to control for variability in the cell culturing process.
Using CaPTure
In this section we describe the analysis workflow: first we identify ROIs by segmenting neurons in the cell-fill channel, and then extract fluorescence intensity. Then we identify "peaks,” which are used to calculate per-image and per-cell summary and aggregate metrics to assess network and cellular activity.
Step1: Convert .czi time series files to.mat files
A time-series of images was collected for each imaging field, and saved using the Zeiss proprietary.czi file format to maintain image metadata. Since all of the image data processing is performed in MATLAB, we recommend that users convert the raw data to MATLAB format for fast and easy access. We use the Bio-Formats package called `bfmatlab` [19] to load the. czi data into MATLAB and the MATLAB save function to save it to .mat format. The `bfmatlab` package supports the conversion of multiple proprietary file formats obtained from different microscope systems, thus enabling the use of CaPTure on calcium imaging data obtained from various systems.
Step2: Identify ROIs
CaPTure allows the user to automate detection of ROIs, and then to select ROIs based on their shape or size. The strategy allows us to detect cells that express the cell-type specific GECI, but are inactive. From each reference image, we identify infected neurons from which to measure calcium dynamics (Fig. 1A). Neurons have a complex morphology, and we aimed to identify signal from the soma, and not from surrounding neuropil. Thus, we used the MATLAB function ‘imhmin’ to suppress the background signal coming from the neurites (Fig. 1B). We then used the region growing technique [20] for segmenting ROIs from the red image, where the pixel with the minimum fluorescence intensity of the image is chosen as the initial seed location, and the region is iteratively grown by comparing all unallocated neighboring pixels to the seed region. The difference between the intensity value of each pixel and the mean of the region is used as a measure of similarity. The pixel with the smallest difference measured this way is allocated to the respective region. This process stops when the intensity difference between the region mean and that of the new pixel becomes larger than a user specified threshold, in this case, the standard deviation of the image (Fig. 1C). The fully grown region is termed the background, thus leaving out the regions with high intensity which become the final segmented ROIs (Fig. 1D). To select for neurons and to remove noise, debris and neuropil from further inclusion in the data, we used eccentricity (a measure of the roundness of the ROI calculated by the MATLAB function ‘regionprops3’) and a minimum size threshold to filter out ROIs from neuropil and noise (Fig. 1E, F). The output of Step 2 provides the identification of all ROIs. The output of alternative segmentation algorithms [21] can be integrated and used for extraction of downstream activity traces.
Step3: Extract traces from each ROI
Calcium imaging allows for measurement of calcium levels in each individual cell by measuring dynamic fluorescence intensity. From each identified neuron, i.e., ROI, we extract calcium signals by measuring the fluorescence intensity over time. Traces (signal) are extracted from the green video using the ROI segmentations from Step 2. Each point on the trace is the average intensity of all the pixels of the segmented ROI at that Z frame in the green video. The output of Step 3 (Fig. 2) is raw traces for each ROI. For ease of illustration, in subsequent figures we focus on three ROIs: ROI 16-low activity (light teal), ROI 19-moderate activity (medium blue) and ROI 23-high activity (royal blue).
Step4: Extract delta fluorescence/fluorescence (dff) from step3
Due to fluctuations in viral transfection efficiency, baseline activity, expression of the virus and the position of the cell within the sample, there can be differences in the baseline fluorescence intensity fluctuations between ROIs. We thus normalized dynamic fluorescent intensity to baseline by calculating the change in fluorescence using a rolling average [22] to obtain the DF/F following standard methodology. The output of Step 4 provides normalized traces with smoothing (Fig. 3).
Step5: Construction of correlation maps
To identify calcium events, a correlation map is constructed to compare the pattern of fluorescence intensity changes with known motifs representing calcium events. Prior to the calculation of the correlation map, the dff traces needed to be interpolated because the motif library, created by FluoroSNNAP [12], utilized a frame rate of 10 frames/sec (Fig. 4A). We utilized the FluoroSNNAP motifs (Additional file 1: Figure S1: 1–16) and constructed seven motifs (Additional file 1: Figure S1: 17–23) based on observations from our data. A matrix (‘Ca’, rows = motifs, columns = x axis of the trace) of correlation coefficients of all motifs across the trace is computed (Fig. 4B). The correlation coefficients are set to a value of zero at locations across the trace where the intensity/height of the trace are below a certain threshold that represents the background, to avoid noise (Fig. 4C). The output of Step 5 aligns normalized traces to motifs (Fig. 4).
Step6: Extract event location and duration
We next extract the event location and duration for each event in each ROI (Fig. 5). A final row matrix is computed by picking the maximum correlation coefficient from each column of ‘Ca’. The points that exceed the user given correlation threshold (0–1) on the row matrix represent the events of that trace. A high correlation threshold might result in missing some events, while a low correlation threshold will potentially pick noise as events, so an optimal threshold of ~ 0.7 was used for our datasets (Fig. 5B). The total number of all the consecutive points/frames that cross the threshold is taken as the event duration in frames. The output of Step 6 counts and classifies motifs (Fig. 5). We illustrate the occurrence of each motif in our example data set (Fig. 6A), and the occurrence of each motif within each ROI (Fig. 6B).
Step 6A (optional): Synchronicity
Because neurons in in vitro networks are highly interconnected, we aimed to estimate the degree to which calcium events were synchronous across a given field. To do this we quantified how synchronous the calcium activity is between the ROIs of a given field using the functions (‘SCA’) provided by the FluoroSNNAP package (Fig. 7). The package provides different methods to quantify synchrony including phase correlation, entropy, and Fourier Transforms of the calcium traces and events. We used the correlation method applied on calcium activity and corresponding surrogate traces of pairwise neurons in a field to quantify the network synchronicity [12]. Eigenvalue decomposition is used on the pairwise correlation matrix of the ROIs, which decomposes the matrix into clusters of ROIs with similar activity and quantifies the synchronization of each ROI cluster. The output of Step 6 shows the degree to which events in each ROI are correlated with events in other ROIs.
Step7: Extract final data
A custom MATLAB script was written to extract two types of metrics: individual ROI metrics in the file long_dat and image metrics in the file man. This allows us to make comparisons across individual cells and across fields. The final man.csv file represents the image level summary statistics (in columns) for each image (in rows) in the dataset.
Name
|
Image name
|
---|
Metadata
|
Biological and metadata associated with the image
|
num_ROI
|
Number of cells identified in the red image
|
num_active_ROI
|
Number of cells that fire at least one calcium event
|
prop_active_ROI
|
Proportion of active cells in the image
|
corrSYN
|
Synchronicity index describes how synchronous are the cells in the image in firing events
|
motif(1–23)
|
Frequency of occurrence of each motif in the time series of the all the ROIs in the image
|
The final long_data.csv file represents the ROI level summary statistics (in columns) for each ROI (in row) in the dataset.
Name
|
Image name which the ROI belongs to
|
---|
Metadata
|
Biological and metadata associated with the image
|
events_ROI
|
Number of calcium events that a cell produced
|
avg_width
|
Average duration (frames) of events for that ROI
|
Volume
|
Number of pixels in red image that corresponds to the ROI
|
Eccentricity
|
Describes if the ROI is more elongated or more circular in shape. An ROI whose eccentricity is 0 is actually a circle, while an ROI whose eccentricity is 1 is a line segment
|
motif(1–23)
|
Frequency of occurrence of each motif in the time series of the ROI
|