Abstract
Recent advances in live Ca2+ imaging with increasing spatial and temporal resolution offer unprecedented opportunities, but also generate an unmet need for data processing. Here we developed SICT, a MATLAB program that automatically identifies rapid Ca2+ rises in time-lapse movies with low signal-to-noise ratios, using fluorescent indicators. A graphical user interface allows visual inspection of automatically detected events, reducing manual labour to less than 10% while maintaining quality control. The detection performance was tested using synthetic data with various signal-to-noise ratios. The event inspection phase was evaluated by four human observers. Reliability of the method was demonstrated in a direct comparison between manual and SICT-aided analysis. As a test case in cultured neurons, SICT detected an increase in frequency and duration of spontaneous Ca2+ transients in the presence of caffeine. This new method speeds up the analysis of elementary Ca2+ transients.
Similar content being viewed by others
Introduction
Ca2+ is a ubiquitous, conserved second messenger that regulates a range of cellular processes in all kingdoms of life1,2,3. In excitable cells, such as muscle cells and neurons, Ca2+ triggers contraction and neurotransmission, respectively4,5. Fluorescence microscopy allows the spatial and temporal measurement of intracellular Ca2+ dynamics in living cells6,7. Research questions linked to Ca2+ pathways are widespread in science and require different tissues or cell types, Ca2+ reporters and imaging hardware.
Neuronal Ca2+ transients
While classical neurotransmission is long known to be governed by action potential (AP)-evoked Ca2+ transients, there is growing interest in various other types of Ca2+ signals and their putative role in synaptic plasticity and spontaneous neurotransmission8,9,10,11. In contrast to AP-evoked transients, smaller amplitude signals - referred to as Ca2+ sparks, Ca2+ puffs, Ca2+ syntillas or spontaneous Ca2+ transients (SCTs) - occur locally in restricted subcellular compartments and are relatively difficult to detect. Ca2+ sparks/sparklets are local Ca2+ transients that are commonly observed in skeletal and cardiac muscle cells. By definition, they originate from single Ca2+ release units in the sarco- or endoplasmic reticulum where each unit represents a cluster of Ca2+ release channels such as ryanodine receptors (RyRs), inositol 1,4,5-trisphosphate receptors (IP3Rs), or a combination of both12,13. These receptors are also expressed in nervous tissues and sparks have been observed in hippocampal and dorsal root neurons14,15. So-called Ca2+ puffs are thought to represent cytosolic Ca2+ rises released from intracellular Ca2+ stores, again with a contribution of IP3Rs16. Ca2+ syntillas are defined as Ca2+ sparks in presynaptic terminals, for instance in hypothalamic neurons, where they occur spontaneously in the absence of APs and extracellular Ca2+ by a RyR-dependent mechanism17. Finally, spontaneous Ca2+ transients (SCTs) occur in the absence of AP and involve RyRs. They were observed in cerebellar Purkinje cells18 and hippocampal neurons8,11. The recurrent role of intracellular Ca2+ channels in Ca2+ sparks, puffs, syntillas and SCTs suggests a shared mechanism for these types of events. Alternatively, spontaneous Ca2+ transients can also originate from the spontaneous opening of voltage-dependent Ca2+ channels at resting membrane potential, as demonstrated in cultured hippocampal neurons10. For simplicity, we will here collectively refer to all types of events as SCTs. The local and short-lived nature of SCTs (duration in the range of 40–260 ms depending on the reporter used), together with their low amplitude (ΔF/F0 typically in the 0.3–1 range) and frequency (0.005–0.7 Hz) make the detection of these events challenging8,14,15,18.
Cytosolic Ca2+ concentrations can be monitored with both chemical and genetically-encoded Ca2+ indicators (GECI), each with their own advantages and limitations19,20,21,22. In the latter class, the GCaMP family23 has been improved by altering its Ca2+ affinity, brightness, and subcellular localization24,25,26 to yield variants with high Ca2+ affinity and fast kinetics, including GCaMP6f27. Another key factor for the detection of intracellular Ca2+ fluctuations is related to imaging hardware. For optimal temporal resolution and light sensitivity, high-quality oil-immersion objectives offer high numerical apertures. Vacuum-cooled electron multiplying CCD (EM-CCD) cameras detect photons with quantum efficiencies above 90% while maintaining reasonable noise levels. Inevitably, there is a trade-off between the image capture time (accumulating enough photons to retrieve a good signal-to-noise ratio) and time resolution (the framerate needed to capture fast biological events)19.
Image processing
High quality real-time Ca2+ imaging yields valuable datasets, but the large file sizes and the short duration of transient signals in noisy image data pose a challenge in data processing and storage. Considering the need to compare conditions in many cells per groups, manual processing of Ca2+ imaging data is laborious and sensitive to human bias by gradual changes in decision-making. Many good programs have been developed to detect changes in Ca2+ imaging data. Most are open source and available for widely used platforms as ImageJ28,29, MATLAB30,31,32,33,34,35,36,37,38,39, Python38,40 and R29 or can be run as a desktop application41. Many excellent tools, such as SeNeCA30,NeuroCa31 and FluoroSNNAP32, were designed for the optical analysis of neuronal circuits or high-throughput analysis of cell cultures, typically reporting Ca2+-based activity patterns of single or multiple neuronal cell bodies detected in vivo, ex vivo or in vitro30,31,32,36,37,39,41. Other tools like PeakCaller35 and SamuROI40 analyse Ca2+ signals in predefined regions (ROIs), offering accurate detection of biological signals in noisy data on the macro- to micro-scale, as small as dendritic spines. Automated detection of micro-scale Ca2+ signals in spatiotemporal (x,y,t) datasets has been achieved with xySpark in cardiomyocytes loaded with Fluo-4-AM (short duration with halftimes in the 10–50 ms range; spatial halfwidth of 1–5 µm)28 and with CaSCaDe in astroglia expressing GCaMP3 (mean duration of 9 s, size range 6–22 µm)33. For our purpose of studying spontaneous synaptic Ca2+ signals however, we were prompted to develop an algorithm combining all of the following features: (a) the ability to detect low-amplitude signals in noisy data with a small type II error (i.e. with a good detection rate of biological events); (b) analysing full images without user-defined ROIs; (c) the ability to process large movies (typically 10 GB in size); (d) analysing events in micro-scale subcellular compartments rather than somatic oscillations or patterns in neuronal ensembles; (e) the ability to visualize spatiotemporal signals and categorize the resulting events; and (f) tools to report descriptive parameters of the detected events including amplitude, frequency and kinetics.
In general, automated detection methods employ four steps. First, data are filtered to reduce noise. The standard deviation of a Gaussian filter is typically implemented as a user controlled setting and can be applied to the spatial or temporal dimension of a signal, or both. Second, a threshold is set based on data characteristics, such as the mean and the standard deviation of the spatial and/or temporal dimension of the data. Data are then divided into super threshold and sub threshold pixels, where super threshold pixels are considered to contain relevant biological signal whereas sub-threshold pixels are excluded from further analysis. Third, super-threshold pixels that are directly adjacent in the spatial or temporal dimension are grouped into individual segments, which can be further analysed as distinct units of the data (‘regions of interest’). A widely used method for grouping is connected-component analysis. Other strategies such as spatio-temporal independent component analysis (ICA)32,36, non-negative matrix factorization (NMF)37 and constrained non-negative matrix factorization and deconvolution38 exist but were not evaluated in our study.
Here we report SICT, an open source program that automatically detects low amplitude SCTs in noisy time-lapse imaging data. SICT allows to visually inspect each Ca2+ event and classify events for further analysis. The validity and sensitivity of the methods were confirmed using both simulated and experimental data.
Results
Automated detection of SCTs
Cultured neurons expressing the fluorescent Ca2+ indicator GCaMP6f were studied by time-lapse imaging at ≈29 Hz. Exploiting repeated measurements in the time dimension, a 3D Gaussian filter effectively reduced noise as illustrated in Fig. 1a,b. To optimize the detection of sudden increases of Ca2+, occurring within less than 5 frames (0.17 s in our experiments), ΔF/F0 values were calculated using a moving average for F0 (see methods for details). Note that this definition of F0 is unsuitable to detect slow fluctuations. A threshold was set at the median plus 3 times the interquartile range for each frame (Fig. 1c, right panel). All pixels exceeding the threshold were grouped together with neighbours in space or time, resulting in 3D ROIs containing various numbers of pixels. Each ROI thus corresponds to a putative Ca2+ event.
To reduce false positive detections, the ROIs were subsequently filtered based on two criteria. First, ROIs consisting of only a single pixel were excluded because these events typically originate from ‘hot pixel noise’ generated in the EM-CCD camera, and are inconsistent with the instrument’s point spread function. Second, ROIs occurring outside the neuronal cell structure can optionally be excluded. For this purpose, an inclusion mask was defined as all image regions with above-average pixel intensities (Supplementary Fig. 1). The latter criterion reduced the number of ROIs twofold. In 14 cells analysed, 98,4% of all SCTs (interpreted as true events after visual inspection) occurred within the neuronal inclusion mask, suggesting that this filter has a minor effect on detection sensitivity (Supplementary Fig. 1). All ROIs meeting the inclusion criteria were saved for later visual inspection and downstream analysis together with their properties (see Supplementary Table 1).
Performance of automated detection in noisy data
To test the performance of SICT in the presence of increasing noise levels, simulated datasets were produced at different SNRs by mixing synthetic Ca2+ signals with realistic background noise. Figure 2a illustrates simulated events with identical background noise in a range of SNRs. At maximum performance, 89% of the generated events were correctly detected. Of this maximum performance, 99% was reached at a SNR of 3.64. Half/maximal performance was reached at 1.91 SNR (Fig. 2b).
Assisted inspection
To allow visual inspection of the ROIs generated by automated detection, we developed a graphical user interface which is explained in detail in the Supplemental Manual. During the manual selection, ROIs can be sorted for descending quality scores which helps to identify true events (see Supplementary Fig. 6). Manual selection further allows to categorize different classes of Ca2+ signalling events. The graphical interface shows each event, both after signal processing and as raw data. Relevant sections of the time-lapse movie can be played using shortcut keys, allowing to distinguish putative biological events from various potential artefacts like moving structures, focus drift, hot pixel noise or any other possible sources of false event detection. We defined an event as true if the identified peak is consistent with the appearance of Ca2+ fluctuations in living cells.
After inspection, ROIs scored as true events were analysed for their properties such as the event frequency, amplitude, rise time, decay time, and FWHM (full width at half maximum; Supplemental Manual, step 3). Common properties are shown in Supplementary Fig. 2. Optionally, the user can label events in 9 classes based on their kinetics. Different types of Ca2+ kinetics were observed in our experiments, summarized in Supplementary Fig. 3 and Supplementary Table 2. In 14 cells analysed, fast isolated Ca2+ transients (‘FastSingle’) represented the most abundant class of events (59.69%). Other events appeared as a combination of fast and slow Ca2+ rises (‘SlowFastComplex’; 18.86%), included multiple fast Ca2+ transients (‘FastComplex’; 9.59%) or were classified as slow isolated Ca2+ transients (‘SlowSingle’; 5.95%). Other event types had minor contributions (Supplementary Fig. 3).
Human observer comparison
To evaluate the effect of differences between individual users during event inspection, four human observers were instructed to select all fast-isolated Ca2+ transients based on experience and a list of guidelines: the Ca2+ event should be isolated (and therefore not surrounded by other Ca2+ events), spatially localized in the neuronal structure and short-lived (in the order of ms). The same four recordings were processed by each observer. Before inspection, ROIs were sorted from high to low amplitude. Observers were instructed to continue searching for Ca2+ transients until no single transients were detected in the last examined 320 events.
Between the four human observers, of all 612 unique events scored by at least one observer (average trace ± SD shown in Fig. 3a), 158 events were scored by each of the four observers. One observer selected over 250 unique events that were not scored by other users (Fig. 3c). The average trace per observer did not show meaningful differences (Fig. 3b). We also compared the number of events, frequency, amplitude, FWHM, rise time and decay time (Fig. 3d–i). Collectively, the results show that the frequency of Ca2+ transients scored by different observers varies significantly. A similar difference was found for the amplitude, although with a small effect size. It is thus advisable that different experimental groups are analysed by the same person.
Manual versus assisted image processing
To validate the assisted image processing method, we compared the results from manual and assisted processing obtained from the same 14 time-lapse movies (Fig. 4). A key difference is that in manual processing, the ROIs are defined a priori by the user and do not change over time. In contrast, the assisted method includes all pixels with detectable fluorescence signals, creating a ROI that can change over time. High-intensity Ca2+ events produce more pixels with detectable signals, yielding larger ROI sizes from which the ΔF/F0 signal is averaged. Thus, the amplitude of a Ca2+ event is not directly proportional to the fluorescence intensity of the raw signal.
In line with this idea, the assisted procedure reported significantly lower peak ΔF/F0 amplitudes of 0.064 compared to 0.112 after manual detection (p < 0.001; Fig. 4a,b and Supplementary Fig. 4). The number of detected events was higher after assisted detection (frequency of 0.174 Hz compared to 0.091 Hz; Fig. 4d,e) but this difference was not significant. Furthermore, the peak duration tended to be longer after assisted detection (FWHM of 0.278 s compared to 0.175 s; Fig. 4c and Supplementary Fig. 4).
To verify if the ROI area was an important determinant of the reported SCT amplitudes, the ROI areas in the x, y dimension were compared for the same SCTs detected by the manual vs. the assisted method. The mean area of automatically identified ROIs was more than three-fold bigger (35.5 ± 1.6 µm2 compared to 10.8 ± 0.3 µm2; n = 8 events from 1 cell, Fig. 5a). To further test the effect of the different ROI definition on event amplitude and duration (Fig. 5b–d), we processed the same 8 Ca2+ transients in three different ways: (1) with ROIs generated by assisted processing as before; (2) with ROIs that were manually drawn as before; and (3) with ROIs identical to the manual method, but further processed by the assisted method. As expected, the lower amplitude of automatically processed ROIs (0.089 ± 0.012 ΔF/F0 for group 1 and 0.188 ± 0.035 for group 2; p = 0.018) was restored to a similar amplitude when the ROIs were made identical (group 3, 0.162 ± 0.036 ΔF/F0; p = 0.068 compared to group 1). Differences in the reported peak duration were not statistically significant between the three groups (Fig. 5d). To confirm the effect of the ROI area in the amplitude output, we calculated the integrated amplitude (area times amplitude), which shows a significantly higher value for the assisted processing (3.21 ± 0.51 ΔF/F0 × µm2 for group 1 and 2.08 ± 0.41 for group 2; p = 0.012) and even higher differences when the ROIs were made identical (group 3; 1.81 ± 0.42 ΔF/F0 × µm2; p < 0.001; Fig. 5e). By including all pixels with above-threshold signal, the automated ROI definition was more suitable for Ca2+ event detection in noisy data than the manual ROI definition. Thus, while the larger ROI size precludes a direct comparison of ΔF/F0 amplitudes, an optimal measure of peak intensity was obtained by integrating ΔF/F0 signal over ROI size.
Caffeine effect on Ca2+ transients
To address if the assisted program is able to detect biological differences, we tested the effect of caffeine on SCT properties. Caffeine sensitizes RyRs, which in turn triggers Ca2+ release from intracellular stores42. The effect of caffeine on intracellular Ca2+ stores has been studied in many cell types, such as smooth and skeletal muscle cells, chromaffin cells, PC12 cells and different types of neurons. Generally, caffeine application increases intracellular Ca2+ concentration in a cell type- and concentration-dependent manner43,44,45,46. The effect of caffeine on SCTs is well studied in excitable cells such as smooth muscle cells, in which caffeine increases the frequency of SCTs12. Caffeine also increases the frequency of SCTs in magnocellular hypothalamic neurons17 and superior cervical ganglion neurons47.
To investigate the effect of caffeine on SCTs in our hippocampal neurons, time-lapse images were recorded first for 3 min in the absence and then for 3 min in the presence of 10 mM caffeine. AP propagation in the neurons was blocked by TTX throughout the experiment. Ca2+ transients were analysed for the second minute of each condition to exclude image drifting artefacts and long-lasting events that were mostly noticed in the first minute after the application of caffeine (Fig. 6b). The number of SCTs was significantly increased by caffeine (from 13.8 ± 3.9 to 22 ± 3.6 events, corresponding to frequencies of 0.23 and 0.36 Hz respectively; p = 0.006, n = 22 cells, N = 3 independent experiments). The peak duration also increased from 185 ± 14 to 240 ± 36 ms (p = 0.003). A trend towards an increase was observed for the integrated peak intensity, from 1.89 ± 0.21 to 3.47 ± 0.5 ΔF/F0 × µm2 (p = 0.012; accepted α significance <0.01 for multiple parameters testing). The ROI area in presence of caffeine was 1.53-fold bigger than in its absence (Fig. 6h). In summary, these results show that SICT successfully identifies biological differences. Furthermore, we confirm that caffeine evokes complex, long-lasting Ca2+ events especially immediately after the drug application, and show that the frequency and FWHM of SCTs are increased one minute after caffeine application.
Discussion
In the last years, several programs have been developed for (semi)-automatic Ca2+ signalling analysis in different software environments. For the purpose of our study, we aimed to develop a program that speeds up the detection of small-amplitude Ca2+ elevations in noisy data.
Events detected by the SICT algorithm still require manual curation. This was a deliberate decision and a key feature which distinguishes SICT from other excellent tools28,29,30,31,32,33,34,35,40,41, such as FluoroSNNAP32, xySpark28 and NeuroCa31. All are able to detect SCTs. When adjusting parameters to strictly filter out true biological signals, we found that such filtering inevitably reduces the detection sensitivity, leading to false negatives. In addition, we did not want to make strict assumptions regarding the signal waveforms to be detected. Thus, we aimed to make the manual curation very fast by building a convenient graphical interface. Importantly, the automatically detected events are sorted for their likeliness to represent a biological signal. Various sorting methods can be chosen as explained in the section ‘ROI sorting methods’ of the SICT user manual (see Supplemental information). Events with the highest scores are shown first, typically representing convincing biological events, whereas later in the list, true events will be accompanied by more and more false positive hits. At some point, when biological signals are scarce and noise becomes predominant, the user can decide to stop the manual curation process. This decision can also be seen as a ‘filtering threshold’, but it is not a blind choice because the user is aware of the signal and noise patterns specific for the experiment.
The assisted method was validated by testing its detection sensitivity using experimental data and simulated experiments with various SNRs. The program performance reached a maximum true positive rate (TPR) of 0.89, thus leaving 11% of true events undetected even at high SNRs. As a possible cause, local pixel noise may contribute to mislocation of peak signals in the x, y or time dimension. In another study, localization errors between 0 and 1 μm were reported48 whereas our criterion for effective detection was set to 0.4 μm. Half-maximal performance with our program required SNRs of 1.91. Other published programs have claimed high performance (>98%) with SNRs as low as 0.2 albeit calculated differently49. However, the method in that study used predefined ROIs as input, and furthermore used manually selected example events to train a classifier. We avoided such conditions to be able to analyse large numbers of time-lapse movies. Furthermore, as illustrated by the example events in Fig. 2a, it may be acceptable to exclude events at SNRs below 1.5. A balance between true and false positive detection rates is an inherent dilemma in all Ca2+ imaging experiments and should be reconsidered for each individual experiment. This dilemma underlines the value of visual event inspection to aid data interpretation.
Compared to manual processing, the SICT-assisted method readily detected Ca2+ elevations with at least a similar sensitivity. SICT uses dynamic ROIs (i.e. changing over time) which are automatically generated instead of being user-defined. Higher Ca2+ signals yield larger ROIs. Therefore, the best measure of peak intensity is obtained by integrating the ΔF/F0 signal over ROI size. In contrast, ΔF/F0 amplitudes without integration should not be used as a direct measure of Ca2+ signal amplitude. For the same reason, the SICT method typically reports lower ΔF/F0 amplitudes (0.05–0.1) than manual methods (0.3–1)8,14,15,18. These lower ΔF/F0 amplitudes do not reflect lower Ca2+ peak concentrations but are a consequence of calculating the ΔF/F0 from a larger ROI, which helps to average out noise and improves the sensitivity of detection.
The SICT code is open source, allowing versatile downstream data processing in MATLAB. Both raw and analysed data are stored in a structure which allows easy data retrieval and event inspection. Overall, using the assisted processing and visual event inspection we could analyse 10 times more recordings compared to the manual method in the same amount of time. As may be expected, visual inspection by different users was a significant source of variation in the analysis, mostly affecting the event frequency. To avoid systematic differences, event inspection should be performed by the same user for all experimental groups.
As a test case, our program readily detected the expected increase in the frequency and duration of SCTs in paired recordings from hippocampal neurons after caffeine application. At the same caffeine concentration, Ca2+ syntillas in chromaffin cells previously showed a 1.73-fold frequency increase, without effect on the amplitude50. Nerve terminals stimulated with 20 mM caffeine exhibited a 2-fold frequency increase while the amplitude increased 1.18 fold17. Other studies on hippocampal neurons, which focused on SCTs, generally altered the SCTs using 30 μM ryanodine, which reduced the frequency of SCTs by blocking RyRs11. Finally, Llano and colleagues assessed SCTs in Purkinje cells after application of an agonistic concentration of ryanodine (5–10 μM), which induced a 3.8-fold frequency increase18. Our results thus agree with previous reports and demonstrate that SICT is well applicable to detect changes in SCT frequencies.
In conclusion, SICT-assisted image processing can be widely applied to investigate Ca2+ transients or other fast local fluorescence increases in live cell imaging data, providing a new tool to accelerate the discovery of fast local signalling pathways in living systems.
Methods
Primary neuron culture
Animals were housed, bred and experimentally used according to institutional guidelines and Dutch and U.S. governmental laws. Prior approval was obtained from the animal ethical committee of the VU University and VU Medical Centre, named ‘Dier ethische commissie (DEC)’, approval license FGA 11-03. Hippocampal neurons were obtained from C57BL/6 mouse brain at embryonic day E18. Unless specified otherwise, all chemicals were from Invitrogen (The Netherlands). After removal of the meninges, the hippocampi were dissected in Hank’s balanced salt solution (HBSS; Sigma), which was buffered with 10 mM HEPES. Cells were dissociated with 0.25% trypsin for 20 min at 37 °C. After washing 5 times in HBSS, the cells were suspended in Neurobasal medium, supplemented with 2% B-27, 1.8% HEPES, 0.25% glutamax and 0.1% Pen-Strep. Cells were triturated by three passes through fire-polished Pasteur pipettes and counted using a Fuchs-Rosenthal chamber. High density continental cultures were created by seeding 25k neurons on a layer of 25k rat glia per well on 18 mm coverslips in 12-well plates or 50k neurons on 50k glia on 25 mm coverslips in 6-well plates. Glass coverslips were pretreated by washing in ethanol and sprayed with 0.1 mg/ml Poly-D-lysine, 0.2 mg/ml rat tail collagen (BD Biosciences) in 10.2 mM acetic acid. Cultures were kept in a humidified incubator at 37 °C and 5% CO2. After 8 days, half of the medium was replaced. Neurons were infected with lentiviral particles encoding GCaMP6f27 at 7 days in vitro (DIV) and analysed between DIV14-18.
Ca2+ imaging
Ca2+ imaging was performed using a custom-built epifluorescence setup called FAINT (for Flash activation of Action potential-Independent NeuroTransmission). It uses an inverted microscope (IX73, Olympus) illuminated by a polychromeV monochromator (TILL Photonics) with a 150 W Xenon high stability lamp in combination with a suitable GFP filter set (GFP filterset 49002, Chroma). Coverslips were placed in the imaging chamber with artificial cerebrospinal fluid (ACSF) as extracellular solution containing (in mM): 140 NaCl, 2.4 KCl, 10 HEPES, 10 glucose, 4 CaCl2, 4 MgCl2 (pH 7.3, 300 mOsm). 1 μM of tetrodotoxin (TTX, Ascent) and 10 mM caffeine were added when specified. GCaMP6f was excited at a wavelength of 480 nm. Imaging was performed using a 40x-oil immersion objective (Olympus UAPON40xO340-2, NA 1.35) at RT (22–24 °C). Images were acquired by Live Acquisition software (LA; FEI) at a frequency of 28.77 Hz for 3–5 minutes using an Andor Ixon Ultra 897 electron-multiplying CCD (EM-CCD) camera with the following settings: vacuum cooling to −70 °C, EM gain of 200, pre-amplification of 2, 40% lamp intensity, 30 ms exposure. Images had a spatial resolution of 512 × 512 pixels each calibrated to cover 0.16 μm2 of the specimen. Intensity values were recorded in 16-bit unsigned integer format, producing roughly 10 gigabytes per time-lapse movie. Data were exported in ‘.raw’ format using the ‘Offline Analysis’ data streaming option in Live Acquisition software, together with its associated ‘.mtd’ file which contains all metadata of the experiment. To investigate the effect of caffeine on SCTs, the same cell was first perfused in presence of 1 µM TTX for 3 min, followed by 3 min in 1 μM TTX and 10 mM caffeine.
Automatic event detection
Source code is accessible in the Supplemental Software. All computations were performed in MATLAB version 2016b51 unless specified otherwise. The automatic detection of ROIs is run by a graphical user interface (see Supplemental Manual, step 1). Data was imported as a 3D matrix (x, y, time) of unsigned 16-bit integer numbers indicating GCaMP6f fluorescence intensity. Instead of the often used 2D filter48 we used a 3D Gaussian filter (imgaussfilt3) to reduce high-frequency noise by weighted averaging each pixel with its neighbours, using a standard deviation (sigma) of 3 pixels for spatial, and 2 frames for temporal filtering. To calculate ΔF/F0 for each pixel, the baseline intensity (F0) was determined as a moving average of 15 to 5 time points prior to the current data point. The ΔF/F0 was then calculated as (F − F0)/F0. For an example of typical raw and calculated data values, see Supplementary Fig. 8. The threshold to select Ca2+ transients was calculated for each frame by taking the median plus three times the interquartile range. The interquartile range is an effective measure of noise levels because it is not affected by the occurrence of Ca2+ fluctuations or baseline drift.
The above-threshold data points were clustered into 3-dimensional ROIs using the bwconncomp function. The resulting ROIs (each a collection of voxels in the x,y,t dimension) represent putative Ca2+ elevations. Information for each ROI was saved to an output file for further analysis, including raw intensity values, the peak x, y and t coordinates and the weighted centroid (for a full list of properties see Supplementary Table 1).
Assisted event inspection
Kinetic parameters for single events are calculated as detailed in the Supplementary Table 3. In short, these included the baseline and amplitude of the ΔF/F0 trace; the 10–90% rise time, 90-10% decay time and the full width at half maximum (FWHM). For kinetic parameters, time values were estimated beyond framerate resolution by linear interpolation. To report descriptive parameters of selected SCTs, the parameters were first calculated per event, averaged per cell and then averaged to get the overall mean.
Manual image processing
For comparison, we also performed manual processing by opening the image stack in ImageJ (NIH). The first 500 frames (17s) were removed to allow removal of a bleaching component by subtraction of a mono-exponential curve fitted to the average frame intensity over time. ROIs were manually drawn, together fully covering visible neuronal structures in the image. Average pixel intensities of each ROI were calculated for each frame, graphed and visually inspected. The ΔF/F0 was calculated using a F0 value calculated as the average from the last 25 data points of the trace. Peak amplitude was calculated as maximum ΔF/F0 during the peak minus a baseline value (averaged value from the 10 data points of the same ΔF/F0 trace, 50 frames before the peak). The FWHM was calculated as the time of the closest data point between the half-maximal signal in the rising and decay phase of the peak. The frequency of events was calculated as the total number of events divided by the total imaging time in each experiment. The parameters were calculated per event, averaged per cell and then averaged to get the overall mean.
Simulations of algorithm performance in noisy data
To assess the detection performance, a range of signal-to-noise ratios (SNRs) was tested using simulated data. Background noise was simulated by taking raw data frames and calculating mean and standard deviation. These values were used to generate a matrix with the same spatial dimensions. To reduce processing time, a decreased time dimension was used. Per simulation, 100 Ca2+ events at 10 Hz (on average 0.35 per imaging frame) were generated by adding values of 1 at random locations within a cell mask in a matrix of all zeros. The mask corresponded to all pixels that had, on average in the time dimension, a higher-than-average intensity compared to the whole field of view (likely corresponding to a fluorescently labelled cell). The events were then spread out in the spatial dimension using a 2D Gaussian filter with a standard deviation of one pixel in both the x and y direction. Subsequently, the signal was convoluted in the time dimension with a normalized example event, for which we used the Ca2+ signal from a neuron firing a single action potential. This example event was multiplied by the standard deviation of the corresponding pixel and multiplied with a SNR factor. Finally, the matrix that contained background noise was summated with the matrix that contained simulated events and saved in the same format as experimental data files.
This procedure was repeated with SNRs varying from 0.01 to 7 in 0.01 increments. The files were automatically processed by SICT and the output automatically analysed to determine how many events had been detected, using the criterium that SICT detected a ROI with a spatial accuracy of 1 pixel (0.4 µm) and a temporal accuracy of 10 frames (±348 ms).
Data analysis
Statistical analysis was performed using SPSS v.23.0 (IBM Corp., Armonk, NY, USA). Data are reported as mean ± SEM, except when specified otherwise. Data were checked for normality using the Shapiro-Wilk or Kolmogorov-Smirnov test. When required, other assumptions were tested: homogeneity of the variance was assessed with Levene’s test; the sphericity assumption was tested with Mauchly’s test. For normally distributed data, a parametric test was run; otherwise a non-parametric test was performed. If the sphericity assumption was not met, the Greenhouse-Geisser correction was used to adjust the degrees of freedom. For two groups, a paired samples t-test or Wilcoxon signed-rank test was performed. In case of more than 2 groups, one-way repeated measures ANOVA or Friedman test was used.
Where necessary, the alpha significance threshold was adjusted for multiple testing. In case of more than 2 groups, a Bonferroni correction was used to adjust the p-value for pairwise comparison, keeping the alpha significance threshold at 0.05. The effect size was calculated for the paired samples t-test as \({\rm{r}}=\sqrt{{{\rm{t}}}^{2}/({{\rm{t}}}^{2}+{\rm{df}})}\), for Wilcoxon signed-rank test as \({\rm{r}}=|{\rm{Z}}/\surd {\rm{N}}|\), for one-way repeated measures ANOVA as \({\rm{r}}=\sqrt{{{\rm{\eta }}}^{2}}\) and, for Friedman test as \({\rm{r}}=\sqrt{{{\rm{\chi }}}^{2}/({{\rm{\chi }}}^{2}+{\rm{N}})}\). The effect size is reported for all p-values lower than 0.05. Statistical details for each figure are reported in Supplementary Table 4. In box plots, open circles represent individual measurements and filled circles depict mean ± SEM. The central line shows the median and statistical outliers are marked by red ‘plus’ symbols.
Code availability statement
MATLAB code can be retrieved without restrictions from the VU Institutional Research Data Management system (Groffen, AJA, 2018, “Replication data for: SICT: automated detection and supervised inspection of fast Ca2+ transients”, https://hdl.handle.net/10411/UA6UZG, DataverseNL). An instruction is included in the Supplementary Information.
Data Availability Statement
Replication data are deposited in the VU Institutional Research Data Management system (Groffen, AJA, 2018, “Replication data for: SICT: automated detection and supervised inspection of fast Ca2+ transients”, https://hdl.handle.net/10411/UA6UZG, DataverseNL).
References
Plattner, H. & Verkhratsky, A. The ancient roots of calcium signalling evolutionary tree. Cell Calcium 57, 123–132 (2015).
Clapham, D. E. Calcium Signaling. Cell 131, 1047–1058 (2007).
Berridge, M. J., Lipp, P. & Bootman, M. D. The versatility and universality of calcium signalling. Nat. Rev. Mol. Cell Biol. 1, 11–21 (2000).
Neher, E. & Sakaba, T. Multiple roles of calcium ions in the regulation of neurotransmitter release. Neuron 59, 861–72 (2008).
Dulhunty, A. F. Excitation-contraction coupling from the 1950s into the new millennium. Clin. Exp. Pharmacol. Physiol. 33, 763–772 (2006).
Palmer, A. E. & Tsien, R. Y. Measuring calcium signaling using genetically targetable fluorescent indicators. Nat. Protoc. 1, 1057–65 (2006).
Schneggenburger, R. & Neher, E. Intracellular calcium dependence of transmitter release rates at a fast central synapse. Nature 406, 889–93 (2000).
Emptage, N. J., Reid, C. A. & Fine, A. Calcium Stores in Hippocampal Synaptic Boutons Mediate Short-Term Plasticity, Store-Operated Ca 2 Entry, and Spontaneous Transmitter Release 29, 197–208 (2001).
Groffen, A. J. et al. Doc2b is a high-affinity Ca2+ sensor for spontaneous neurotransmitter release. Science 327, 1614–8 (2010).
Ermolyuk, Y. S. et al. Differential triggering of spontaneous glutamate release by P/Q−, N− and R-type Ca2+ channels. Nat. Neurosci. 16, 1754–63 (2013).
Reese, A. L. & Kavalali, E. T. Spontaneous neurotransmission signals through store-driven Ca2+ transients to maintain synaptic homeostasis. Elife 4, 1–15 (2015).
Cheng, H. & Lederer, W. J. Calcium sparks. Physiol. Rev. 88, 1491–545 (2008).
Cheng, H., Lederer, W. J. & Cannell, M. B. Calcium sparks: elementary events underlying excitation-contraction coupling in heart muscle. Science 262, 740–4 (1993).
Koizumi, S. et al. Characterization of elementary Ca2+ release signals in NGF-differentiated PC12 cells and hippocampal neurons. Neuron 22, 125–37 (1999).
Ouyang, K. et al. Ca2+ sparks and secretion in dorsal root ganglion neurons. Proc. Natl. Acad. Sci. 102, 2–7 (2005).
Wiltgen, S. M., Dickinson, G. D., Swaminathan, D. & Parker, I. Termination of calcium puffs and coupled closings of inositol trisphosphate receptor channels. Cell Calcium 56, 157–168 (2014).
De Crescenzo, V. et al. Ca2+ syntillas, miniature Ca2+ release events in terminals of hypothalamic neurons, are increased in frequency by depolarization in the absence of Ca2+ influx. J. Neurosci. 24, 1226–35 (2004).
Llano, I. et al. Presynaptic calcium stores underlie large-amplitude miniature IPSCs and spontaneous calcium transients. Nat. Neurosci. 3, 1256–65 (2000).
Takahashi, A., Camacho, P., Lechleiter, J. D. & Herman, B. Measurement of intracellular calcium. Physiol. Rev. 79, 1089–1125 (1999).
Grienberger, C. & Konnerth, A. Imaging Calcium in Neurons. Neuron 73, 862–885 (2012).
Whitaker, M. Genetically-encoded probes for measurement of intracellular calcium. Methods Cell Biol. 99, 153–182 (2010).
Paredes, R. M., Etzler, J. C., Watts, L. T. & Lechleiter, J. D. Chemical Calcium Indicators. Methods 46, 143–151 (2009).
Nagai, T., Sawano, A., Park, E. S. & Miyawaki, A. Circularly permuted green fluorescent proteins engineered to sense Ca2+. Proc. Natl. Acad. Sci. USA 98, 3197–3202 (2001).
Akerboom, J. et al. Optimization of a GCaMP Calcium Indicator for Neural Activity Imaging. J. Neurosci. 32, 13819–13840 (2012).
Horikawa, K. Recent progress in the development of genetically encoded Ca2+ indicators. J. Med. Investig. 62, 24–8 (2015).
Mao, T., O’Connor, D. H., Scheuss, V., Nakai, J. & Svoboda, K. Characterization and subcellular targeting of GCaMP-type genetically-encoded calcium indicators. PLoS One 3, 1–10 (2008).
Chen, T. W. et al. Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300 (2013).
Steele, E. M. & Steele, D. S. Automated Detection and Analysis of Ca2+ Sparks in x-y Image Stacks Using a Thresholding Algorithm Implemented within the Open-Source Image Analysis Platform ImageJ. Biophys. J. 106, 566–576 (2014).
Prada, J. et al. An open source tool for automatic spatiotemporal assessment of calcium transients and local ‘signal-close-to-noise’ activity in calcium imaging data. PLoS Computational Biology 14 (2018).
Tomek, J., Novak, O. & Syka, J. Two-Photon Processor and SeNeCA: a freely available software package to process data from two-photon calcium imaging at speeds down to several milliseconds per frame. J. Neurophysiol. 110, 243–56 (2013).
Jang, M. J. & Nam, Y. NeuroCa: integrated framework for systematic analysis of spatiotemporal neuronal activity patterns from large-scale optical recording data. Neurophotonics 2, 035003 (2015).
Patel, T. P., Man, K., Firestein, B. L. & Meaney, D. F. Automated quantification of neuronal networks and single-cell calcium dynamics using calcium imaging. J. Neurosci. Methods 243, 26–38 (2015).
Agarwal, A. et al. Transient Opening of the Mitochondrial Permeability Transition Pore Induces Microdomain Calcium Transients in Astrocyte Processes. Neuron 93, 587–605.e7 (2017).
Bindocci, E. et al. Neuroscience: Three-dimensional Ca2+ imaging advances understanding of astrocyte biology. Science (80-.). 356 (2017).
Artimovich, E., Jackson, R. K., Kilander, M. B. C., Lin, Y. C. & Nestor, M. W. PeakCaller: An automated graphical interface for the quantification of intracellular calcium obtained by high-content screening. BMC Neurosci. 18, 1–15 (2017).
Mukamel, E. A., Nimmerjahn, A. & Schnitzer, M. J. Automated analysis of cellular signals from large-scale calcium imaging data. Neuron 63, 747–760 (2009).
Maruyama, R. et al. Detecting cells using non-negative matrix factorization on calcium imaging data. Neural Networks 55, 11–19 (2014).
Pnevmatikakis, E. A. et al. Simultaneous Denoising, Deconvolution, and Demixing of Calcium Imaging Data Eftychios. Neuron 89, 285–299 (2016).
Romano, S. A. et al. An integrated calcium imaging processing toolbox for the analysis of neuronal population dynamics. PLoS Comput. Biol. 13, e1005526 (2017).
Rueckl, M. et al. SamuROI, a Python-Based Software Tool for Visualization and Analysis of Dynamic Time Series Imaging at Multiple Spatial Scales. Front. Neuroinform. 11, 44 (2017).
Hamed, M., Odul, J., Miller, A. S., Kawakami, K. & Asanobu, K. NAS: Neuron analyzer suite for automatic analysis of neuronal activities from calcium imaging data. Int. J. Pharma Med. Biol. Sci. 4, 167–170 (2015).
Kong, H. et al. Caffeine induces Ca2+ release by reducing the threshold for luminal Ca2+ activation of the ryanodine receptor. Biochem. J. 414, 441–452 (2008).
Shmigol, A., Kirischuk, S., Kostyuk, P. & Verkhratsky, A. Different properties of caffeine-sensitive Ca2+ stores in peripheral and central mammalian neurones. Pflügers Arch. - Eur. J. Physiol. 426, 174–176 (1994).
Rizzuto, R. & Pozzan, T. Microdomains of intracellular Ca2+: molecular determinants and functional consequences. Physiol. Rev. 86, 369–408 (2006).
Verkhratsky, A. & Shmigol, A. Calcium-induced calcium release in neurones. Cell Calcium 19, 1–14 (1996).
Brini, M. et al. Subcellular analysis of Ca2+ homeostasis in primary cultures of skeletal muscle myotubes. Mol. Biol. Cell 8, 129–43 (1997).
Yao, L. et al. Ca2+ sparks and Ca2+ glows in superior cervical ganglion neurons. Acta Pharmacol. Sin. 27, 848–52 (2006).
Ellefsen, K. L., Settle, B., Parker, I. & Smith, I. F. An algorithm for automated detection, localization and measurement of local calcium signals from camera-based imaging. Cell Calcium 56, 147–156 (2014).
Szymanska, A. F. et al. Accurate detection of low signal-to-noise ratio neuronal calcium transient waves using a matched filter. J. Neurosci. Methods 259, 1–12 (2015).
Zhuge, R. et al. Syntillas release Ca2+ at a site different from the microdomain where exocytosis occurs in mouse chromaffin cells. Biophys. J. 90, 2027–37 (2006).
MathWorks, T. MATLAB Release 2016b, The MathWorks, Inc., Natick, Massachusetts, United States (2016).
Acknowledgements
We thank Robbert Zalm, Jurjen Broeke, Desiree Schut, Frank den Oudsten and Joost Hoetjes for excellent technical assistance and Matthijs Verhage for critically reading the manuscript. This research was funded by the Netherlands Organization for Health Research and Development (ZonMW project 91113022).
Author information
Authors and Affiliations
Contributions
T.B. developed the SICT program with input from R.M., A.J.G. Ca2+ imaging and manual analysis was performed by R.M. Simulations were designed and run by T.B. Supervised inspection of SICT-detected events was performed by R.M., Q.B., T.B. and R.L. Caffeine application experiments and data analysis were done by R.M. R.M., A.J.G. and T.B. designed the study, interpreted data and wrote the manuscript.
Corresponding author
Ethics declarations
Competing Interests
The authors declare no competing interests.
Additional information
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Mancini, R., van der Bijl, T., Bourgeois-Jaarsma, Q. et al. SICT: automated detection and supervised inspection of fast Ca2+ transients. Sci Rep 8, 15523 (2018). https://doi.org/10.1038/s41598-018-33847-4
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-018-33847-4
Keywords
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.