Large-scale volumetric in vivo two photon calcium imaging

Abstract

Neural activity is widely distributed and highly integrated, yet individual neurons are generally thought of as the computational unit of the brain. Decoding this activity and behavior is thought to require simultaneous recording from large populations of neurons at fast timescales. Two-photon microscopy (TPM) has been combined with calcium imaging to record from hundreds of neurons in vivo at cellular resolution, yet the mammalian brain is composed of several orders of magnitude more neurons. Improving these techniques to record from larger populations has been a major goal in technology development for neuroscience. Here, we used a variety of approaches to explore the limits and techniques for maximizing the number of simultaneously recorded neurons using TPM. First, we performed a series of experiments and modeling to estimate laser power limits to large-scale TPM in vivo. We find that most applications of large-scale TPM recordings will be limited by heating, with a maximum of 200 mW average power allowable. Second, we developed a technique to perform volumetric TPM using a stereoscopic point-spread function (PSF). Using a stereoscopic PSF exploits sparsity in neural activity to scan through whole tissue volumes at a time while retaining axial information in order to uniquely decompose and segment the recorded videos. Third, we developed a computationally efficient framework for generating realistic TPM calcium imaging videos. We use these synthetic videos to evaluate current automated demixing algorithms and compare TPM imaging modalities across a variety of sample conditions. Finally, we analyze fundamental limits to TPM for calcium imaging including power, sampling, and signal constraints. We use these constraints to estimate optimal TPM performance and design a microscope that best reaches these limits. Overall, these experiments suggest that with existing technology, we may be able to use TPM to simultaneously record the neural activity of over 100,000 neurons over mouse cortex, several orders of magnitude greater than typical experiments performed today.