Automatic deep learning-driven label-free image-guided patch clamp system

Abstract

Patch clamp recording of neurons is a labor-intensive and time-consuming procedure. Here, we demonstrate a tool that fully automatically performs electrophysiological recordings in label-free tissue slices. The automation covers the detection of cells in label-free images, calibration of the micropipette movement, approach to the cell with the pipette, formation of the whole-cell configuration, and recording. The cell detection is based on deep learning. The model is trained on a new image database of neurons in unlabeled brain tissue slices. The pipette tip detection and approaching phase use image analysis techniques for precise movements. High-quality measurements are performed on hundreds of human and rodent neurons. We also demonstrate that further molecular and anatomical analysis can be performed on the recorded cells. The software has a diary module that automatically logs patch clamp events. Our tool can multiply the number of daily measurements to help brain research.

Fig. 1. Steps of DIGAP procedures.

1: Pipette calibration by the user, 2: pipette replacement after recording, 3: image-based automatic pipette tip detection, 4: automatic cell detection, 5: pipette navigation to the target cell, 6: 3D cell tracking, 7: pressure regulation, 8: gigaseal formation, 9: break-in, 10: electrophysiological recording, 11: nucleus and cytoplasm harvesting, 12: anatomical reconstruction of the recorded cell.

Fig. 2. Hardware setup of the DIGAP system.

a Microscope with a motorized stage. b Micromanipulator. c Controller electronics for manipulators. d Patch clamp amplifier. e Pressure controller module. f Computer with the controller software.

Fig. 3. The developed algorithms for the DIGAP system.

a Result of the Pipette Hunter detection model shown in three different projections of the image stack. Initial state (blue contour) and the result (green contour) of our pipette localization algorithm are shown. b Training dataset generation: 265 image stacks (60–100 images per stack with 1 μm frame distance along the Z-axis) captured from human and rodent neocortical slices with DIC videomicroscopy (left). 31,720 objects as healthy cells (green boxes) labeled on every slice of the image stack by four experts. c After the training session, the DIGAP system detects cells in unstained living neocortical tissues. d Accuracy of the automated cell detection pipeline. e Lateral tracking of the cell movement (n = 174). DIC images of the targeted (in blue box) and patched cell (in green box). The cell drifted from its initial location (arrows in the right panel) during the pipette maneuver. f, g Z-tracking of the cell movement (n = 174). The template image was captured at the optimal focal depth (in red boxes) before starting the tracking. During the pipette movement, image stacks were captured from the targeted cell (upper panels) such that the middle slice was taken of the most recent focus position. The bottom row shows the differences between the template and the image of the corresponding Z position. The lowest standard deviation value of the difference images (plots) shows the direction of the cell drift in the Z-axis. Source Data is available as a Source Data file.

Fig. 4. A representative example of a visual patch clamping procedure.

a Trajectory of the pipette tip (red line) with obstacle avoidance (numbered) in the tissue and the spatial location of the detected cells (green boxes). The steps of the avoidance algorithm are the following. 1: The pipette is moved forward in the initial trajectory until an obstacle is hit. 2: The pipette is pulled back. 3: The pipette is moved laterally in a spiral pattern until the resistance is back to normal. 4: The obstacle is passed. 5: The pipette is readjusted to the trajectory. 6: The approaching is continued. b Plots of the depth of the pipette tip in the tissue, the applied air pressure, and the measured pipette tip resistance during the approach. c Image of a cell before and after performing patch clamp recording on it. Source Data is available as a Source Data file.

Fig. 5. GUI of the software.

a Main window with an image stack loaded and the built-in labeling tool started. b Monitoring window to check the pressure and resistance values. Pressure values can be set here when operating manually, or the measurement can be restarted from different subphases here. c Main window when browsing the detected cells, initiated with the Find and Patch button. The measurement can be started by selecting a cell. d The Patch Clamp Diary module showing a plot with annotations of a sample and measurements in it.

Fig. 6. Electrophysiological properties of the cells patched by DIGAP.

a Main electrophysiological parameters from the successful automatic patch clamp recordings. The box plots show the series resistance (R_s, left panel), the membrane resistance (R_m, middle panel), and the resting membrane potential (right panel) of all successful measurements (n = 47 for rat and n = 41 for human samples). The boxes show the median, 25 and 75 percentiles, and min/max values, and the whiskers are 1.5 interquartile ranges. b Different cell types are identified according to firing features: pyr pyramidal cell, bAD burst adapting, cNAC continuous non-accommodating, cSTUT continuous stuttering, bSTUT burst stuttering, dSTUT delayed stuttering, cAD continuous adapting, dNAC delayed non-accomodating. c Individual neurons’ action potential half-widths are presented as a function of the same neuron’s R_m. Note the segregation of excitatory and inhibitory neuronal classes. Dataset is recorded from rodent samples (Panel c and d colors correspond to panel b). d The proportion of recorded cell types. e–g Same plots as b–d, representing the dataset recorded in human neocortical slices. Source Data is available as a Source Data file.

Fig. 7. Anatomical and molecular biological investigation of neurons patched by DIGAP.

a Two anatomically reconstructed human autopatched neurons. The darker colors represent somata and dendrites of the pyramidal (green) and the interneuron (red) cells. The brighter color shows the axonal arborization. The firing patterns of the cells are the same color as their reconstructions. b mRNA copy numbers of a housekeeping (RPS18, black bars) and the aquaporin 1 (AQP1, red bars) gene from four representative human pyramidal cells. Source Data is available as a Source Data file.