Automated quantification of neuronal networks and single-cell calcium dynamics using calcium imaging

Abstract

Background: Recent advances in genetically engineered calcium and membrane potential indicators provide the potential to estimate the activation dynamics of individual neurons within larger, mesoscale networks (100s-1000+neurons). However, a fully integrated automated workflow for the analysis and visualization of neural microcircuits from high speed fluorescence imaging data is lacking.

New method: Here we introduce FluoroSNNAP, Fluorescence Single Neuron and Network Analysis Package. FluoroSNNAP is an open-source, interactive software developed in MATLAB for automated quantification of numerous biologically relevant features of both the calcium dynamics of single-cells and network activity patterns. FluoroSNNAP integrates and improves upon existing tools for spike detection, synchronization analysis, and inference of functional connectivity, making it most useful to experimentalists with little or no programming knowledge.

Results: We apply FluoroSNNAP to characterize the activity patterns of neuronal microcircuits undergoing developmental maturation in vitro. Separately, we highlight the utility of single-cell analysis for phenotyping a mixed population of neurons expressing a human mutant variant of the microtubule associated protein tau and wild-type tau.

Comparison with existing method(s): We show the performance of semi-automated cell segmentation using spatiotemporal independent component analysis and significant improvement in detecting calcium transients using a template-based algorithm in comparison to peak-based or wavelet-based detection methods. Our software further enables automated analysis of microcircuits, which is an improvement over existing methods.

Conclusions: We expect the dissemination of this software will facilitate a comprehensive analysis of neuronal networks, promoting the rapid interrogation of circuits in health and disease.

Keywords: Calcium imaging; Event detection; FluoroSNNAP; Functional connectivity; Neuronal phenotype; Synchrony.

Fig. 1

Overview of image analysis. A: Workflow for the analysis of time-lapse calcium imaging. After acquiring images, pseudo-automated segmentation is performed to define regions of interest (ROIs), delineating neuronal somata from background. Fluorescence vs. time is computed for each ROI and the onsets of calcium transients are determined automatically. Single-cell features of calcium transients and network-level measures of synchrony and functional connectivity are computed. A graphical user interface was created to facilitate batch processing (B), segmentation (C), inspecting the fluorescence trace and spikes of individual neurons (D) and properties of the network (E).

Fig. 2

Automated spike detection methodology. Percentile-based thresholding based on average fluorescence intensity for each region of interest (ROI) has positive predictive value of 86% for event detection if transient amplitude is nearly constant (A, example of an individual neuron’s fluorescence trace) but fails to detect low amplitude calcium transients (PPV = 63%) when there are amplitude variations from spike-to-spike (B). Event detection using continuous-wavelet transform is an alternative, more sophisticated method. However, it suffers from high false positive rate when the signal to noise ratio is low (C, 34% false positives) and the wavelet transform can introduce small phase-shifts in the event times (D). A third method for event detection used a supervised algorithm by first building a database of unique calcium transient waveforms (E, F expanded view, note the different y-axes). Each waveform was used to generate a time-varying correlation trace that indicated the level of similarity between a true calcium transient and epochs of fluorescence trace (G; correlation value computed over 5 s windows, 100-ms displacement), generating a similarity matrix (H). Local maxima in the correlation matrix corresponded to the onset of calcium transients, providing a robust method for automated event detection for a noisy trace (I, heat map shows the maximum correlation coefficient across all spike waveforms) and a trace with varying amplitude (J). Across these three methodologies, the average fraction of true calcium transients that were correctly identified (PPV) was significantly higher for the supervised detection method when compared to either the percentile or wavelet transform based methods (K). The accuracy of supervised detection method was relatively insensitive to the number of true spike waveforms beyond n = 10 (L).

Fig. 3

Classifiers for single cell calcium dynamics of neurons in the network. A: Fluorescence vs. time trace depicts automated detection of calcium transients (pink dashed lines). For each transient, the peak amplitude (a), rise time (d) and fall time (e) are determined. The resting fluorescence of each cell (c) and mean inter-event-interval (b) are also automatically computed. The resulting normalized peak intensity for an individual neuron shows a distribution of values across a network (B), while the rise and fall times show a more uniform range of values (C, red bars – rise time; gray bars – fall time). (For interpretation of the references to color in figure legend, the reader is referred to the web version of the article.)

Fig. 4

Estimate of functional connectivity. The functional connectivity of pair-wise neurons (red lines, A) was determined by testing for statistical dependencies in their calcium activity. Visual inspection (B) and correlogram (C) of the fluorescence trace of 4 neurons indicated in A shows temporal similarity between neuronal pairs (1,2) and (3,4) but not between the pair (2,4). The distribution of instantaneous phase difference between pair-wise neurons was compared to the distribution generated by resampling the fluorescence trace of one of neuron pairs (D, E; blue bars – phase difference of actual; red bars – phase difference of surrogate). An overlap in the distributions of actual and surrogate samples is indicative of a lack of functional dependency. Kolmogorov–Smirnov test was used to test statistical similarity of the two distributions. A p-value <0.01 was used as a threshold to infer functional connection; stricter statistical criteria did not change significantly the total number of functional connections beyond this point (F). (For interpretation of the references to color in figure legend, the reader is referred to the web version of the article.)

Fig. 5

Developmental changes in network activity patterns. A: Spontaneous activity patterns of neurons at different stages in development, colored-coded by age (days in vitro, DIV) and time-aligned for visual comparison, shows increasing frequency and greater complexity of activity as neurons mature. B: Network-bursts, recorded as simultaneous calcium transients in ›90% of neurons in the imaging field of view, represent the transitions between synchronized and much less synchronized states of the network. The number of synchronized states per minutes gradually increased over development and reached its peak at DIV 14, consistent with increased synaptic growth and strength. This was followed by a significant decrease in the frequency of network bursts over DIV 15–16, and a plateau DIV 18+. C: The pattern of activity became increasingly synchronized as neurons matured in culture and reached its peak at DIV 14 as revealed by increasing “red” in the synchronization matrix. D: Beyond DIV 14, spontaneous activity was moderately synchronized, characterized by a mix of synchronized network-wide bursts and intermittent flickering events. Antagonizing the GABA_A receptor with bicuculline forced synchronized calcium oscillations in mature (DIV 14+) cultures because of recurrent excitation and network disinhibition. In comparison, bicuculline greatly reduced the frequency and coordination of activity in immature cultures (DIV 8–12). (For interpretation of the references to color in figure legend, the reader is referred to the web version of the article.)

Fig. 6

Single-cell phenotypes revealed by calcium imaging. A: The spontaneous activity of mouse hippocampal neurons was recorded with the fluorescent calcium indicator, Fluo4-AM. Cultures were prepared with a mixture or neurons expressing either wildtype tau protein or a mutant version of human tau (P301S) B: Following calcium imaging, cells were fixed and probed for the expression of P301S tau protein with 12025 antibody (red – PS19, blue – DAPI). The same field of view used for calcium imaging was used during immunofluorescence microscopy to identify individual neurons that expressed the transgene (example of 4 co-registered neurons is indicated). C: Comparison of the average fluorescence trace of a representative neuron expressing the human mutant variant of tau (red) and wildtype neuron (blue) showed clear differences in the baseline fluorescence and amplitudes of calcium transients. D: Shape-based features of each calcium transient, including the fluorescence at the onset of a transient, peak amplitude, and characteristic rise and fall times, were computed and projected onto a 2-dimensional principal component space. Two clusters readily emerged in the principal component space, corresponding to neurons expressing mutant tau (red) or wild-type tau (blue). E: Resting [Ca²⁺]_i and peak amplitude were significantly larger in PS19 neurons and compared to wild-type. However, the duration of calcium transients was significantly shorter, resulting in similar calcium load between PS19 and WT neurons. PS19 n = 54, WT, n = 53. (**p <0.01, *p <0.05). (For interpretation of the references to color in figure legend, the reader is referred to the web version of the article.)