SICT: automated detection and supervised inspection of fast Ca2+ transients

Figures

• 

  Figure 1. Spatiotemporal effects of filtering. As an example, the Ca2+ indicator GCaMP6f was expressed in hippocampal neurons and recorded by live epifluorescence microscopy. Three stages of signal processing are illustrated from top to bottom. For each stage, a single 2D image is depicted as grayscale image (left), as surface graph (middle) or as quantitated signal over time (right). In the right graph, the red area indicates the maximal and minimal signal of a 2D ROI consisting of 372 pixels in a region where a Ca2+ transient occurred. The grey area indicates the maximal and minimal signal in a background region (defined as all other 261772 pixels). A Ca2+ transient peak occurred at (x, y, t) = (166 µm, 157 µm, 3.48 s). (a) The raw signal demonstrates the difficulty of identifying Ca2+ events in raw data. (b) A 3D Gaussian filter with a standard deviation of 3 and 2 pixels respectively in the spatial and temporal dimension was applied to remove high-frequency spatial and temporal noise. (c) The ‘ΔF/F0’ transformation highlights pixels that have increased intensity values compared to the previous frames (left, middle). The ΔF/F0 signal exceeds the background, allowing peak detection in…

• 

  Figure 2. Program performance. True positive detection rate as a function of SNR in simulated noisy datasets. (a) Example traces of simulated raw data for a range of SNR. A pixel was chosen randomly and simulated with a SNR from 0 to 4 in steps of 0.5, using the same simulated background noise and example event. (b) The performance of automated detection was measured as the fraction of accurately detected events in simulated datasets with different SNRs. Each blue dot represents a separate test. The red line indicates a sigmoid fit of the aggregate data. Half-maximal performance was observed at SNR 1.91 with a true positive rate of 0.43. Optimal performance (99% of the maximum rate of 0.88) was reached at SNR 3.64 and higher.

• 

  Figure 3. User-dependence of Ca2+ event inspection. Four human observers inspected ROIs from the same dataset and selected ‘FastSingle’ Ca2+ events (Supplementary Fig. 3). (a) Average ΔF/F0 trace ± SD of 612 unique events selected by one or more observers. (b–i) Each colour (orange, yellow, blue and purple) marks a different observer. (b) Average ΔF/F0 trace of all events selected by each observer. (c) Number of events selected by one or multiple (1–4) observers. (d–i) Comparison of event parameters after scoring by different observers is shown in box plots. Open circles show mean value per cell; closed circle represent the mean ± SEM of n = 4 cells, N = 1 independent experiment. Statistical analysis was performed with (1)one-way repeated measures ANOVA; (2)Friedman test, (3)one-way repeated measures ANOVA with Greenhouse-Geisser correction (dashed line); pairwise comparison (line). (d) Number of events. (e) Frequency. (f) Amplitude. (g) FWHM. (h) Rise time. (i) Decay time.

• 

  Figure 4. Comparison of manual and assisted image processing. (a) Average ΔF/F0 trace ± SEM of all detected Ca2+ peaks after manual (red) or assisted image processing (black). Traces were first averaged to calculate the mean trace per cell, and then averaged to calculate the mean ± SEM from n = 14 cells, N = 1 independent experiment. (b–e) Box plots show Ca2+ transient parameters resulting from assisted and manual image processing. Open circles represent averaged value per cell; closed circles show mean ± SEM from 14 cells, N = 1 independent experiment. Statistical significance was tested with (1)paired samples t-test; (2)Wilcoxon signed-rank test. (b) Amplitude. (c) FWHM. (d) Frequency.

• 

  Figure 5. Effect of ROI definition on Ca2+ event parameter measurement. (a) Automatically identified ROIs (black) had a bigger 2D area than manually drawn ROIs (dark blue). (b–e) Each parameter was calculated using the fully assisted (black) and the manual method (dark blue), or a combination where the ROIs were manually defined and further processed by the assisted method (light blue). (b) Average ΔF/F0 trace ± SEM of 8 Ca2+ transients of 1 cell. (c–e) Box plots show Ca2+ event parameters. (c) Amplitude. (d) FWHM. (e) Integrated amplitude. Statistical significance was assessed with (1)Wilcoxon signed-rank test, (2)one-way repeated measures ANOVA or (3)one-way repeated measures ANOVA with Greenhouse-Geisser correction (dashed line); pairwise comparison (line; n = 8 Ca2+ events, N = 1 cell).

• 

  Figure 6. Caffeine affects the frequency and duration of SCTs. (a) Representative image of a neuron labelled with GCaMP6f. Red squares indicate five ROIs for which the ΔF/F0 intensity over time is shown in b. (b) Each neuron was perfused for three minutes in 1 μM TTX followed by three minutes in 1 μM TTX and 10 mM caffeine. Only the second minute of each perfusion was used to analyse the properties of SCTs (red circles). (c–k) SCT event parameters in absence (black) and presence (green) of 10 mM caffeine. (c) Average ΔF/F0 trace ± SEM of all averaged SCTs per cell. (d–m) Box plots show averaged SCTs parameter values per cell (open circles) and their overall mean (closed circles). Statistical significance was tested using (1)paired samples t-test; (2)Wilcoxon signed-rank test (n = 22 cells, N = 3 independent experiments). (d) Number of SCTs. (e) Frequency. (f) Integrated amplitude. (g) Amplitude. (h) ROIs area 2D. (i) FWHM. (j) Rise time. (k) Decay time.