3D-CASH RANDOM-ACCESS LEVERAGES THE <EM>IN-VIVO</EM> OPTICAL RECORDING OF ACTION POTENTIALS BY CLOSELY TRACKING THE TIME RESPONSE OF AN ULTRA-FAST MEMBRANE VOLTAGE REPORTER

Random-access sampling has long been considered as an effective strategy to improve sampling speed and signal-to-noise ratio (SNR) in neuron voltage imaging. Practical in-vivo implementations, however, struggled with target drift-induced recording artefacts and excessive photo-bleaching. As we demonstrated recently, 3D Custom-Access Serial Holography (3D-CASH), an augmented volumetric random-access scan mode, alleviates both issues. While directing every laser pulse to a different 3D target location, CASH shapes the single-pulse two-photon excitation volume to cover a larger portion of the targeted cell compartment, for better robustness against target motions and higher photon yield.
Here, we present 3D-CASH recordings (4-20 kHz) from up to 10 principal cells in layer2/3 of visual cortex in awake mice behaving on a treadmill. Target cells expressed FORCE1f, a recently developed genetically-encoded voltage indicator (GEVI), with sub-millisecond response time and unprecedented dynamic range. Single spike responses exhibited a SNR >4, a spike-averaged peak amplitude of 0.9 and a 1.2 ms FWHM. The experimentally determined single spike waveform is well predicted by a kinetic model of the GEVI voltage gate. For low SNR recordings (<6), spike detection was enhanced by a denoising convolutional neural network trained on the FORCE1f kinetics. For experimental validation, we recorded neuron responses to visual noise and moving contrast bars to confirm that the optical traces match the temporal fidelity of classical electrophysiology.