Removing direct photocurrent artifacts in optogenetic connectivity mapping data via constrained matrix factorization

Monosynaptic connectivity mapping is crucial for building circuit-level models of neural computation. Two-photon optogenetic stimulation, when combined with whole-cell recording, enables large-scale mapping of physiological circuit parameters. In this experimental setup, recorded postsynaptic currents are used to infer the presence and strength of connections. For many cell types, nearby connections are those we expect to be strongest. However, when the postsynaptic cell expresses opsin, optical excitation of nearby cells can induce direct photocurrents in the postsynaptic cell. These photocurrent artifacts contaminate synaptic currents, making it difficult or impossible to probe connectivity for nearby cells. To overcome this problem, we developed a computational tool, Photocurrent Removal with Constraints (PhoRC). Our method is based on a constrained matrix factorization model which leverages the fact that photocurrent kinetics are less variable than those of synaptic currents. We demonstrate on real and simulated data that PhoRC consistently removes photocurrents while preserving synaptic currents, despite variations in photocurrent kinetics across datasets. Our method allows the discovery of synaptic connections which would have been otherwise obscured by photocurrent artifacts, and may thus reveal a more complete picture of synaptic connectivity. PhoRC runs faster than real time and is available as open source software.Author summary: Mapping the presence and strength of connections between neurons is necessary for understanding how the structure of neural circuits gives rise to their function. Historically, directly measuring these connections has required difficult and time-consuming paired-patch experiments. However, recent developments in the field of two-photon optogenetics enable mapping experiments with much higher throughput. In these new experiments, a single postsynaptic neuron is recorded using an electrode, and a two-photon laser is used to optogenetically activate nearby neurons. When the laser activates a connected presynaptic neuron, the experimenter observes an electrical current on the electrode. The size of the current can then be used to determine the strength of the connection. However, this technique has a drawback: when stimulating near the postsynaptic cell, the laser can directly evoke an electrical current—called a photocurrent—even if no connection is present. This is problematic because it confounds estimates of connectivity. In this work, we developed a computational tool which removes the photocurrent artifact while preserving signals corresponding to true neuronal connections. We expect that this tool will serve as a resource for experimentalists, enabling a more complete picture of synaptic connectivity.