Fig 1.
Overview of the toolbox workflow.
Colored boxes indicate the processing modules (color code in top left corner). Dashed boxes correspond to optional procedures. Tutorial step numbers related to each procedure are indicated in the bottom-right corner of each box. Single-headed thin and thick arrows respectively depict the processing-pipeline flow, and the option to import pre-analyzed data from other methods into the stand-alone modules of the pipeline.
Fig 2.
Detection of ROIs corresponding to single neurons.
Four examples from different animal models, brain regions, imaging techniques and calcium indicators. Left: imaged optical planes. Perimeters of correct and incorrect automatically detected ROIs are shown in green and red, respectively. Yellow perimeters, detected ROIs that had to be manually curated, or undetected but manually drawn. Right: representative ΔF/F0 traces (black) and significant fluorescent transients detected (red). (A) Mouse primary visual cortex bolus injected with OGB-1 AM (two-photon imaging, 256x256 pixels, 30 Hz sampling rate). Data from Scholl et al[37]. Segmentation was performed with the “labeled nuclei” and “bigger ROI” options, and we set parameters “local contrast” to ~20 for spatial normalization of the image, thrneuropil and thrsoma to ~0.45 and ~0.095, respectively, and “minimal ROI area” to 18 pixels. (B) Mouse somatosensory cortex, where nuclei of excitatory neurons are transgenically labeled with mCherry (two-photon imaging, 256x256 pixels, 7 Hz sampling rate). Calcium dynamics monitored with GCaMP6s. Data from Peron et al[11,38]. Parameters used: “labeled nuclei” and “bigger ROI” options, “local contrast” ~43, thrneuropil ~0.15, thrsoma ~0.08, and “minimal ROI area” set to 15 pixels. (C) Transgenic zebrafish larva pan-neuronally expressing GCaMP3 (two-photon imaging, 512x256 pixels, 1 Hz sampling rate). Parameters used: “unlabeled nuclei” and “smaller ROI options”, “local contrast” ~10, thrneuropil ~0.15, thrsoma ~0.01, “minimal” and “maximal ROI areas” set to 7 and 40 pixels, and “minimal” and “maximal circularity” to ~0.48 and ~1.7. (D) Right hemisphere of the optic tectum of a transgenic zebrafish larva pan-neuronally expressing GCaMP5 (single-photon light-sheet imaging, 232x242 pixels, 100 Hz sampling rate). Parameters used: “unlabeled nuclei” and “smaller ROI options”, “local contrast” ~20, thrneuropil ~0.15, thrsoma ~0.01, “minimal ROI areas” set to 8 pixels.
Fig 3.
Correction of neuropil fluorescence contamination.
(A) Left: optical plane imaged with two-photon microscopy of the mouse somatosensory cortex (same data as Fig 2B). Right: examples of the detected ROIs (red) and their circular perisomatic masks used to calculate the local neuropil signals (white). Perisomatic masks are overlapping, but each ROI is associated with a single circular mask. Black holes inside the perisomatic masks are other detected ROIs not included in the local neuropil signal calculation. (B) Left: raw fluorescence traces obtained with the perisomatic masks shown in A (neuropil signal). Right: pair-wise correlation matrix for the signals shown in the left. Note the high temporal correlation across the traces. (C) Same as B, for the raw fluorescence traces of the ROIs shown in A (somata). (D) Same as C, for the corrected ROI fluorescence traces, obtained by subtracting the traces shown in B from the corresponding traces shown in C, with α = 0.9. Note the reduction in the temporal correlations, compared to those found in C, despite the small changes of the individual fluorescence traces. (E) Relationship between the pair-wise correlations shown in C and D.
Fig 4.
Toolbox performance in inferring neuronal activity from calcium imaging data.
(A) Two examples with different signal-to-noise ratios (SNRs) from different neurons in the cai-1 ground-truth dataset[28,39]. For each example, we show: top, GCaMP6f ΔF/F0 traces (black) and the significant fluorescent transients detected by the module (red); bottom, ticks representing the simultaneously recorded spikes (those associated with a significant calcium event are highlighted in red; asterisks mark single spikes); middle, spiking rate of the neuron calculated by temporally convolving spikes with a Gaussian filter of σ = 20 ms. Imaging was performed at 60 Hz. A spike was considered as associated with a significant calcium event if it was followed by an event of significant fluorescence within 40 ms (GCaMP6f rise time τpeak = 45± 4 ms, for 1 spike[28]). Significant transients were calculated with default toolbox parameters in Dynamic threshold mode, with a GCaMP6f τdecay = 250 ms[28]. (B) Boxplot summary of performance for all recordings (n = 37). Top: coefficients obtained when correlating the spiking rates with the raw fluorescence traces or with the ΔF/F0 traces of significant transients (where non-significant fluctuations were set to 0). Bottom: Percentage of “detected” spikes (i.e., those associated with a significant calcium event), for all recorded spikes or for single spikes only.
Fig 5.
User interface screenshots of the calculated neuronal responses associated with experimental events.
(A-E) Responses to visual stimulations with light spots at different azimuth angles of the larva’s visual field (two-photon imaging of a GCaMP3-expressing zebrafish larva). (A) ROIs are colored with an HSV color code representing their preferred azimuth angle (Peak mapping parameter; hue), azimuth selectivity (Tuning width; saturation) and average response at preferred azimuth (Response strength; value). Due to the skewed distribution of the responses, only a few responsive ROIs can be visualized. (B) Offsetting and clipping of the saturation and value channels to improve visualization. Black, original values used in A; red, rescaled values used in C. (C) Same data shown in A, but with the rescaled channel ranges. After this step, the retinotopic organization of the optic tectum becomes evident. (D and E) Screenshots of the responses of two ROIs selected by clicking on Select ROI in c. Average (black) and single-trial (gray) ΔF/F0 responses are organized according to the stimulus values, where significant trial responses are shown in red. Bottom right, tuning curves of the ROIs. Black, mean response; gray patch, standard error. Note how the responsive but less selective ROI in e is shown with a more whitish color code (low saturation). (F) Responses to visual stimulations with gratings moving in angular directions, monitored by volumetric light-sheet single-photon imaging of a GCaMP5-expressing zebrafish larva. Responses are displayed in HSV color code over the maximal intensity projection of all imaged optical sections. Note that, while the tectal neuropil indiscriminately responds to all directions (whitish ROIs), a pair of bilaterally symmetric group of ROIs in the hindbrain responds selectively to either 60° or -60° (see S1 Video for volumetric distribution of responses).
Fig 6.
Detection of neuronal assemblies for the case study.
(A) Screenshot for the selection of the zMax threshold to determine the neuronal composition of the assemblies. After setting the smooth parameter with the slider (top-right), the threshold is chosen with a mouse click on the graph of the density distribution (red arrow). (B) Screenshot showing the topography of 3 representative assemblies (out of 42). ROIs that belong to each assembly are labeled in yellow. (C) Screenshot of two user-defined anatomical axes (see tutorial) over which assemblies will be spatially organized. Each curve is automatically colored so that the combined curves reproduce the hue gradient used in Fig 5A and 5C. The chosen curves span the rostro-caudal retinotopic axis of each tectal hemisphere. (D) Screenshot of the figure obtained displaying the spatial organization of the assemblies along the selected axes. Assemblies' ROIs are colored according to the defined axis (i.e., the position of the assemblies’ spatial centroid with respect to the defined axis). The comparison with Fig 5C confirms that assemblies reproduce the tectal retinotopic functional map.
Fig 7.
Detection of neuronal assemblies in the mouse barrel cortex.
Top: raster plot of the z-scored ΔF/F0 of 277 neurons distributed in 37 assemblies, from the total imaged population of 1025 neurons. Neurons are sorted and color-coded according to the assembly to which they belong to (color bar on the right). Black trace on top, fluctuations of the number of active neurons in the total imaged population; Black trace on the left, average neuronal responses to a whisker-object contact. Bottom: activation dynamics of the detected assemblies, color-coded as in the raster plot. Vertical dotted lines indicate moments of whisker-object contact. Note that activations of assembly #11 are associated with contacts, and episodes of activity sequentially progressing through assemblies (marked by gray boxes).
Fig 8.
Neuronal assemblies obtained using whole-brain light-sheet imaging in zebrafish reveals the spatial structure of brain activity.
A GCaMP5-expressing larva was volumetrically scanned with fast multi-plane single-photon light-sheet microscopy. Due to the lack of single-neuron resolution in this imaging dataset, a grid of 9 μm-diameter hexagons was imposed over each imaged optical plane. (A-F) Some of the ROI assemblies found with the PCA-promax method, displayed over the maximal intensity projection of all imaged optical sections (for individual optical sections see S2 Video). (A-D) Pairs of symmetric unilateral assemblies. (E-F) Single bilaterally symmetric assemblies. (G) All ROI assemblies found with the PCA-promax approach, displayed over the maximal intensity projection of all imaged optical sections (for individual optical sections see S3 Video). Assemblies are colored according to the similarity of their activity dynamics (the more temporally correlated, the more similar the color of the assemblies). This color code was also used in the previous panels (except for the assembly pair shown in c, which was colored differently to facilitate visualization). (H) Same as G, but for assemblies found using k-means, color-coded according to the similarity of their activity dynamics. To allow for a comparison between G and H, the dimensionality of the dataset was reduced through PCA before clustering (otherwise, clustering did not converge). Note how this clustering reveals assemblies whose spatial organization is roughly consistent with those shown in G, but exposes a much less biologically relevant fine structure. (I) Single-trial activation dynamics of the assemblies shown in D during one of the blocks of visual stimulation. Periods of moving-grating stimulation are labeled in red, angles indicate the grating's moving directions. Note that both assemblies are activated by this OMR-inducing stimulation, with the bluish and greenish assemblies preferring 60° and -60°, respectively.
Fig 9.
Comparison of specific features of the assemblies with those of surrogate controls.
(A) Spatial layouts of a given assembly detected in the optic tectum case study, and of one example of a Random surrogate assembly (RSA, where an equal number of ROIs are randomly placed) and a Topographical surrogate assembly (TSA, where ROIs are placed preserving the inter-ROI distances of the original assembly). (B) The normalized frequency histogram of average ROI activity levels (mean significant ΔF/F0 per imaging frame) obtained for ROIs included in assemblies (top) and those included in TSAs. (C) Same as B, but for the ROI activity correlations (pair-wise Pearson correlation coefficient).