Fast, Accurate, and Versatile Data Analysis Platform for the Quantification of Molecular Spatiotemporal Signals

Abstract

Optical recording of intricate molecular dynamics is becoming an indispensable technique for biological studies, accelerated by the development of new or improved biosensors and microscopy technology. This creates major computational challenges to extract and quantify biologically meaningful spatiotemporal patterns embedded within complex and rich data sources, many of which cannot be captured with existing methods. Here, we introduce Activity Quantification and Analysis (AQuA2), a fast, accurate, and versatile data analysis platform built upon advanced machine learning techniques. It decomposes complex live imaging-based datasets into elementary signaling events, allowing accurate and unbiased quantification of molecular activities and identification of consensus functional units. We demonstrate applications across a wide range of biosensors, cell types, organs, animal models, and imaging modalities. As exemplar findings, we show how AQuA2 identified drug-dependent interactions between neurons and astroglia, and distinct sensorimotor signal propagation patterns in the mouse spinal cord.

Keywords: AQuA2; astrocytes; cell interaction analysis; functional units; glial cells; image analysis; machine learning; molecular spatiotemporal signals; neurons; time-lapse imaging.

Figure 1.

Principles of AQuA2. (A) Typical input data with two spatial dimensions plus a time dimension is shown, with three exemplary cells illustrated in green. The example crop, centered around the middle cell, is utilized in (B), with intensity variations of five positions (labeled by different colors) illustrated by corresponding curves. (B) Event-based signal detection pipeline of AQuA2, including preprocessing, statistical test, temporal segmentation, and spatial segmentation. In the outcome of each step, different colors indicate the corresponding pixels/regions belong to the different detection results. (C) Potential outputs based on event detection pipeline, including extracted event features (size, duration, etc.), movie with event overlay, and propagation map. (D) Diagram for functional region analysis. It consists of Consensus Functional Unit (CFU) detection (cluster events with similar spatial patterns) and interaction analysis between CFUs (take event occurrence sequence as input, output significance p-value of dependency). The outcomes include CFU relation network, CFU-stimulus relation, and CFU groups. (E) Example applications of AQuA2, including signals of different cell types, different biosensors, dual-color data, and 3D data. See also Figure S1.

Figure 2.

AQuA2 improves the accuracy of AQuA. (A) Performance comparison between AQuA and AQuA2 on a two-photon ex vivo astrocyte calcium recording from a mouse slice (recorded at 1.1 Hz), which was used as a test example in the AQuA paper. Similar detection results were obtained, suggesting AQuA2 can be considered a viable alternative for applications currently employing AQuA. (B) Performance comparison on a synthetic dataset comprising large and small signals under a significantly low SNR. Among the 110 ground-truth signals, AQuA2 detects 102 and AQuA detects 105. AQuA2 generates 0 false positives, while AQuA produces 734 false positives. White triangles label the ground-truth events in the present time points. Erroneous detections are marked by white dashed circles. (C) Performance comparison on a one-photon astrocyte calcium recording from the mouse spinal cord (recorded at 45 Hz). During detection, a mask is applied to mitigate the influence of blood vessels. In the AQuA and AQuA2 detection results, each colored region presents one detected signal event. The big events were falsely split by AQuA into numerous fragments. See also supplementary videos.

Figure 3.

AQuA2 outperforms peer methods (suite2p, CaImAn, AQuA, and Begonia) on simulated data. (A)-(C) Performance (F1 score and wIoU measure, see STAR Methods) comparison between AQuA2 and peer methods under scenarios of unfixed size, unfixed location, and propagation. Top row: Illustration of signal variations. Middle row: Performance comparison under different levels of signal variation under 10dB. Bottom row: Performance comparison under different SNRs with a moderate signal variation. For all results, we used mean ± 2 x standard deviation, derived from 12 independent replications of evaluation. See also Figure S4.

Figure 4.

AQuA2 identifies the signal pattern changes in zebrafish astroglia and neurons under the addition of caffeine and reveals the correlation between these two cell types. (A) The experiment setting and average projection of the data. Zebrafish were engaged in fictive swimming in a virtual-reality (VR) environment with realistic visual feedback during swimming, recorded by a light sheet microscope. Experiments of two distinct states were conducted: normal state and under the influence of caffeine (drug state). Astroglia calcium (left) and neuronal calcium (right) were expressed using Tg(ELAVL3: GCaMP7f; GFAP: jRGECO1B). (B) Visualization of dF and AQuA2 detection results under different states. Each colored region represents one detected event. Two distinct signal patterns are found in the drug state. (C) Left: Detected CFUs across different planes under two states. Each colored region represents one extracted CFU. Right: Comparison of number and size among identified CFUs in the two states. (D) The most correlated pair of astroglial CFU and neuronal CFU on one plane, with their curves drawn on the right. Zoom-in traces of three time windows (yellow, purple, green) are also given. For the curves of drug state, neuronal spikes (gray) are additionally extracted (by suppressing small fluctuations on dF) and compared with the average dF of astroglia (red). (E) Propagation pattern of the astroglial calcium signal on different planes. The blue color shows the earliest rising time, while the red color shows the latest rising time.

Figure 5.

AQuA2 unveils differences in signal propagation of sensory and motor-evoked astrocyte calcium signals in the mouse spinal cord. (A) Experimental setting: Translaminar imaging was performed in the lumbar dorsal horn of an awake behaving Tg(GFAP:GCaMP6f) mouse on a spherical treadmill. A pressure stimulus was applied to the mouse’s proximal tail. (B) Temporal relationship between the applied sensory stimulus (tail pinch; black), mouse motor behavior (sideward and forward foot movements; purple and pink, respectively), and the average calcium activity across the field of view (blue). (C) Spatial (left) and temporal (right) activity propagation maps of the three signals indicated in B. The color spectrum depicts earlier activation in blue and later rising times in red. (D) Detected localized functional units (orange and green; top) and their corresponding calcium transients (bottom).

Figure 6.

AQuA2 quantifies signals across biosensors, cell types, organs, animal models, and imaging modalities. (A) Application of AQuA2 to quantifying the signals from mouse spinal microglia (expressed by Tg(CX3CR1:GCaMP5g)) imaging data. The comparison between dF and detection for both circular signals and the spatially complex signal (shown in a zoomed-in view of the selected orange box) is provided. On the right, the quantified propagation pattern of the spatially complex signal is given, with blue denoting early rising time and red denoting late rising time. (B) Application of AQuA2 for accurate quantification of distinct oligodendrocyte calcium dynamics within the CNS myelin sheath of zebrafish. The calcium was expressed through Tg(mbp:memGCaMP7s). The average curve of the region of interest (labeled by yellow) is depicted in the bottom-left. In the bottom middle, the propagation pattern of events is showcased, with earlier activation represented in blue and later rising times in red. On the right, a comparison between dF and AQuA2 results is provided, with each colored region representing a distinct detection. (C) Application of AQuA2 for the quantification of extracellular ATP dynamics, captured by the GRAB-ATP sensor and two-photon microscope, on the surface of astrocytes in acute cortical slices. Example representative events (marked by white triangles) are presented with their corresponding average curves (event duration is indicated by the event color). Statistical analysis of event temporal features (Rising duration and decay duration) is provided on the bottom-right. (D) Application of AQuA2 for identifying the swim-related regions on a light sheet norepinephrine (NE)-astroglia dual-color recording of zebrafish. NE was expressed by Tg(ELAVL3: GRABNE) and astroglial calcium was expressed by Tg(GFAP: jRGECO). On the left, each colored region represents a swim-related CFU. On the right, swim strength (blue) and average curves of two channels (red and yellow) are depicted. Three zoom-in figures are provided for three signals. See STAR Methods for experimental details.