Fig 1.
Overview of image analysis by NA3.
a, Motoneuron loaded with a fluorescent calcium indicator. Average intensity projection of 1000 images to show the neuronal morphology. The neuron shifts spontaneously from a low activity state to a high activity state. Rainbow pseudocolor shows low intensity values in blue and high intensity values in red. b, Workflow for the analysis of calcium imaging raw data. After threshold determination according to a rule, the user defines the window size for x,y-grid, and chooses a signal-to-noise ratio value to tune the stringency of the tool. Two computations are started: (1) the signal intensity calculation, (2) the wavelet transform. A result pdf is automatically created. c, The signal-to-noise ratio (SNR) defines the stringency of the computation. The signal average threshold (SAT) can be used to set a signal threshold. The SAT can be close to the black level without having a strong impact on the computation result. d, The documentation pdf defines activity events, marks them, and counts them. All traces showing one or more activity events are given and used to count the total activity. An overview image is created that shows the grid over the first image of the movie, red circles to visualize the activity state of a grid window, and the total activity value, which is a computed value to describe the overall activity state of the neurons. The number of counted activity events per grid window events is shown in a text fil in the results folder. e, Overview of the NA3 workflow. The tool combines functions in ImageJ with “R”. Video processing and signal extraction occurs in ImageJ, before the signals are automatically transferred to “R”. In “R”, the event computing takes place. The result is created in “R” and exported as a pdf-file. The results are also transferred back to ImageJ to allow an interactive access to the data for image processing, ROI selection, or data evaluation.
Table 1.
Feature comparison.
Fig 2.
Calcium activity assessment with the activity detector tool.
a, Principle of calcium activity event detection. Activity event identification is shown for a representative grid window (per se representing a ROI) on a single motoneuron. This motoneuron (shown in the inlet, and in Fig 1) shows global calcium transients. 2600 images (frames; x-axis) were analyzed. The grey trace shows the raw mean intensity values of a representative grid window. After extraction of the image signal in a grid window, all local maxima of the intensity signal are identified at several scales (y-axis) and signal candidates are selected and marked (blue dots). Details are explained in the methods: Strategy for calcium event (peak) identification. b, Effect of tuning parameters on calcium activity event detection. The total number of computed activity events (y-axis) in relation to changes in the user-dependent signal-to-noise ratio (SNR). Two activity stages of the motoneuron are compared. The low activity state (in a, frame 1–1300) and the high activity state (frame 1301–2600). Discrimination of the high activity state and the low activity state is very effective over a broad range of SNR values from 1.5 to 4. The signal average threshold was set to an intensity value of 6 (up-rounded mean intensity value seen in the background). A conservative SAT value was selected and modified at a SNR of 2 (blue square; SAT = 5 a.u.; purple circle; SAT = 6 a.u.). c, Data documentation 1: x,y-t summary. The image shows the distribution and number of calcium activity events raised by a spontaneously active motoneuron. Such an image is automatically generated by the program. The user-dependent tuning parameters for this analysis are given. The image field 142 x 130 pixel was automatically split in a grid of 8 x 8 pixel (WS 8 px). Magenta circles indicate areas with calcium events. The smaller the diameter, the less activity is found in the corresponding grid window. All detected calcium activity events are summed up to offer the value ‘total activity’. d, Data documentation 2: the individual traces represent changes in fluorescence in one grid window. The tool automatically generates traces (black line) representing a grid window and shows raw bit values (y-axis) over the frame number (x-axis). Calcium activity events detected by the tool are labeled with a little red square at the peak point. The upper panel describes the graph in grid 5/14 (x/y-axis) in the somatic region of the motoneuron. Here raw bit values ranged from about 65 to 110. In the lower panel a region in the growth cone of the motoneuron was analyzed (grid 13/4; x/y-axis). Here, raw mean bit values in the grid range from 6 to 10. Note the robust detection of global activity despite an almost 10-fold difference in the mean intensity values in the corresponding grid window.
Fig 3.
Computation of action potential-induced calcium events.
a, Experimental approach. Whole-cell patch clamp recording and parallel calcium imaging of single cells. Here, the calcium indicator was applied with the help of the patch clamp pipette. b,c, Analysis of true-positive responses (TPR) and false-positive responses (FPR) based on parallel calcium imaging and patch clamp recording. Action potentials were induced by current injection (12 times, 200 pA, interstimulus interval: 5 s) for different times (10–500 ms). Current injection of 200 pA for 10 ms (upper panel in c) or 100 ms (lower panel in c) induced single action potentials (indicated by the blue label and vertical line. Calcium imaging was performed at 20 Hz. Calcium event labeling by four computational approaches. Four computational strategies for calcium event definition were applied; our CWT-approach, deconvolution, template-matching, and definition of significant signals above a computed baseline (‘Romano toolbox’). d, Experimental approach. Hippocampal neurons were cultured for 24 days in vitro. At this age, neurons develop glutamatergic synapses with mature hallmarks [39] and become spontaneously active. Cells were loaded with OGB1-AM and investigated with patch clamp recording and calcium imaging at 20 Hz. e, f, Electrophysiological recording of action potentials induced by spontaneous activity. In this trace, twelve action potentials are marked and correlate with AP-induced calcium spikes. All twelve calcium events (ROI, cell soma in a) were labeled by the CWT computation under these low SNR conditions.
Fig 4.
Calcium spike-detection under stringent computing conditions.
a, Confocal imaging of synchronously spiking glia-derived neurons (re-computing of earlier data; [40]. Cells were loaded with calcium indicator OGB1 to label glial cells and neurons. Spontaneous activity was induced by inhibition of GABAergic signaling with bicuculline to induce the spiking behavior in the neural network. A magenta arrow labels all neurons. Computed loci are shown. The average signal trace representing all computed loci is shown. b, Imaging was performed under low-light conditions. The raw image represents this imaging situation. Loci of computed activity events are shown on a brightness-contrast corrected image of the neurons. Under control conditions, spontaneous spiking is observed on the somata, but also in the periphery of the neurons. Spike blockade (TTX, CNQX) correlates with a reduced number of computed activity events. Graph: Raw intensity values are plotted against the frame number. The calcium spikes are efficiently blocked by TTX and CNQX. Scale bar in a: 100μm; in b: 50 μm.
Fig 5.
Activity profile of cultured hippocampal neurons before and after activity induction.
a, Average intensity: Hippocampal neurons loaded with a fluorescent calcium indicator. Control; neurons at a low activity state. cLTP; neurons in a global activity state after treatment with a chemical LTP induction solution. b, Activity distribution under control conditions. Calcium imaging was performed for 55 seconds with a speed of 20 Hz. Virtual total activity number (WS8, SNR 2.5, SAT 11, MAC 2): 2135 activity events. Note that regions of high activity are indicated by circles with a larger diameter (Fig 5d). Under cLTP treatment, neurons increase their total activity number to 4785 events. Yellow marks point to grid windows shown in c. c, Under control conditions, activity events close to the baseline trace are found at some positions. After cLTP treatment local and global calcium signals (global: red arrow) are detected. Calcium spikes and local calcium transients are marked. d, Calcium spikes in loci of synchronous activity. (left) Spontaneous activity profile of hippocampal neurons under control conditions. Neuronal somata do not exhibit a spiking behavior. Calcium spikes are identified in the periphery in indicated grid windows. Yellow squares point to grid windows, which are shown on the right with the corresponding signal trace and the activity marks. Blue arrows point to other grid windows with this synchronous activity pattern. Out-of-synchronicity events are also detected (Red arrow on the right). Some obvious calcium spikes were overseen by the computation due to the stringency parameters used for this analysis (purple arrows). In Fig 6e we show that the spikes are detected when the same data are analyzed at SNR values of 2.0 and 1.5.
Fig 6.
Wavelet-based activity detection in noise signals and spike-detection precision in a spontaneous active hippocampal neuron.
a Activity distribution under control conditions (movie as in Fig 4b, left panel). Here, analysis was performed with the following parameters: WS8, SNR 2.5, SAT 2, MAC 1). 4092 activity events were computed. b, Noise video analysis. A homogenous fluorescence signal was imaged (identical camera settings). Analysis was performed with the following parameters: WS8, SNR 2.5, SAT 2, MAC 1). 53 noise events were computed. Some signals are camera-based (graph 31/21). c, Calcium spikes in loci of synchronous activity. Loci are marked in (a). All cell soma ROI (magenta) was computed with the ROI tool in NA3.d, Typical signal traces found in the noise video. Grid windows are indicated in (b). e, Spike-detection precision. Areas showing 23 synchronous spikes (1/13–41/23) are compared with a subthreshold SAT value, at different SNR values. The graph shows the underestimation in the number of spikes in the y-axis. f, Comparison of computed events in the noise video (in b) compared to spontaneous active neurons (in a). Settings were: WS8, SNR variable, SAT 2, MAC 1. All SNR values allow the discrimination between the noise state and the active state. The higher the SNR, the better is the stringency of the tool. Small activity events are underestimated under high SNR conditions.
Fig 7.
Activity profile of hippocampal neurons after calcium spike blockade.
a, Average intensity: Hippocampal neurons loaded with a fluorescent calcium indicator. Pseudocolor images: neurons before a spike, during a calcium spike (seen in the lower left cell), and after spike blockade. 3000 images, acquired at 20Hz are analyzed (WS8, SAT7, SNR 3). b, Under control conditions, one neuron (the lower, left) shows a high spiking activity, as indicated by the activity pattern (red circles in the grid windows). After acute treatment with an inhibitor cocktail (TTX, CNQX, APV), spiking behavior is blocked (yellow squares; traces in c), non-spike-like activity events become visible on the soma and in the periphery. Shape and number of residual activity signals are quite diverse (bright blue squares; traces in c). c; Grid window-specific signal traces with the corresponding activity marks (red) as indicated in b. d, Activity hotspot in the periphery, in a varicosity-like structure. The activity hotspot is indicated by a contrast-enhanced average intensity projection. Image 496 shows the low activity state, while image 2124 indicates the high activity state. The corresponding grid window is indicated in (b).
Fig 8.
Parallel computing of activity events and variance area.
a, Activity map of spontaneous activity of hippocampal neurons (DIV 10) before and after acute activity blockade. b, Calcium spike formation (left trace) is blocked by the inhibitor cocktail (right trace). c—e, Raw traces representing typical calcium signals in non-spiking areas (black trace). The yellow band indicates the variance area in a sliding window of 30 images. Activity marks are indicated as red dots. The activity state of the grid windows is described with two parameters, the variance area, and the number of activity events. c, Signal trace with one or two computed activity events. The variance area is almost identical before and after spike block. d, Signal fluctuations, represented by the variance area, become smaller under spike block conditions. Furthermore, the activity inhibitor cocktail reduces the number of activity events. e, This grid window over a neuritic element shows high signal fluctuation, which correlates with a higher number of activity events. Spike-blockade reduces the variance area and the number of activity events. However, a local activity event is not blocked by the inhibitor cocktail. Structural elements of local activity on the basis of this analysis are shown in Fig 9.
Fig 9.
Structures with high rates of local activity after calcium spike blockade.
a, Average intensity image of hippocampal neurons loaded with the calcium indicator. Grid windows with high rates of local activity are shown in yellow (see c). b, Activity map (see also Fig 8). c, Five regions of interest are indicated (trace 1 –trace 5). The upper three represent growth cone-like structures, the lower two traces represent hotspots on neurites. Note the variability and the diverse character of the calcium signal patterns detected by the computational approach.
Fig 10.
Computing of signals close to the noise level.
a, Typical hippocampal neuron, loaded with calcium indicator. Two ROIs are indicated. b,c, Calcium traces representing the yellow and magenta ROI in a. Removal of extracellular calcium causes a decline in the calcium indicator signal. This correlates with a reduced number of computed activity events (red dots). d-e, Number of computed activity events are shown on the x,y-grid. Under calcium-free conditions (cyan), a low number of activity events was found in the signal trace. In the presence of extracellular calcium, more activity events are computed by the algorithm. Regions of activity are found in the soma, but also on distal neurites. g,h, Summary graphs for computed activity events and variance area. The virtual activity value has a strong discriminative power to describe the experimental situation (calcium-free versus calcium present). n = 164 ROIs, mean ± SEM; one way ANOVA (Kruskal-Wallis) followed by Dunn’s multiple comparison test; p-values: ** < 0.001; **** < 0.0001. i, Representative signal analysis (yellow ROI in a) with deconvolution (blue dots), template-matching (green dots), and our CWT-approach. Phases of homeostatic activity in presence of extracellular calcium are best described with help of template-matching and CWT-computation.