Research

Here are some ongoing and completed research projects.

Neural circuits for vision, cognition, and their interaction

Sifting of visual features in the superior colliculus

Sensory information is critical for guiding behavior. A general principle of sensory processing in the brain is hierarchical feature extraction, in which more complex features are extracted as the signal goes higher up the processing chain. While this process of "sifting" for the behaviorally relevant bits of information has been established in the visual cortex, most mammals have another, more evolutionarily ancient visual pathway that goes to the superior colliculus (SC). The SC is known to be important for innate visually guided behaviors, such as detecting an approaching predator. In my PhD work, I investigated visual processing of behaviorally relevant stimuli in the mouse SC. I focused on the looming stimulus: dark, expanding discs presented in the upper visual field of the mouse, which simulate an approaching aerial predator and drive a robust defensive reaction such as flight or freezing. By recording throughout the layers of the SC in awake, head-fixed mice, I showed that in the superficial SC, many neurons behave like the retinal ganglion cells from which they receive input. But as one moves down into the intermediate layer, one encounters cells that are highly specific to the looming stimulus, invariant to the stimulus location over a wide swath of the visual field, and emphasize novel appearances by habituating to repeated presentations. This demonstrates that the principle of sifting for behaviorally relevant information applies to the subcortical visual pathway and serves as an example of complex visual computation in the SC.

Vision gain-modulates abstract nonvisual representations in the primary visual cortex

The "outside-in" approach (in which we track sensory feature representations at successive stages of processing) is successful at identifying fundamental computations of the sensory brain. But representation of features by itself is not enough to drive behavior, as sensory information often has to be combined with internal signals about the animal's state, behavioral context, and goals. In my postdoctoral work, I investigated how these two streams of information are integrated in the rat primary visual cortex (V1). V1 is well-known for retinotopic organization and low-level feature detection, but is also modulated by nonvisual signals, such as eye, head, and body movements, stimulus history, attention, and space. But it is unknown whether V1 maintains explicit representations of cognitive variables and how they interact with visual input. We recorded in freely moving rat V1 and hippocampal CA1 while the animal repeatedly alternated between executing four paths on a W-track. In complete darkness, the majority of V1 neurons represented the animal's generalized progress on paths that share the same action sequence ("directional path progression"). In the presence of visual stimuli, these tunings were not replaced with visual responses, but instead their amplitudes were multiplicatively modulated by the stimuli. As a result, the generalized path progression code became specific to each path. Finally, these neurons were preferentially coordinated with CA1 during hippocampal ripples. Together, these results demonstrate that V1 combines internally generated cognitive signals with visual stimuli via gain-modulation, allowing sensory input to anchor an internal representation of task state to specific locations during active behavior.

Neural recording technology

Electrode pooling

Understanding the mechanism behind brain function requires recording the activity of many neurons simultaneously. A standard method to do this is extracellular electrophysiology, in which electrodes implanted into the brain pick up action potentials of nearby neurons. While recent technological advances have scaled up the number of electrodes on neural probes, each electrode still needs to be connected to the backend electronics via a wire. This poses a significant limitation to the number of electrodes that can be used, as naively increasing it requires more wires and a thicker device that displaces and kills more neurons than it records. To solve this problem, we developed a new method called electrode pooling that uses a single wire to serve many electrodes through a set of controllable switches. This leads to pooled neural signals, which can be separated through spike sorting. Using a popular neural probe (Neuropixels 1.0), we implemented electrode pooling and demonstrated that this can lead to a 2–3-fold increase in the number of electrodes that can be simultaneously used, significantly boosting the yield of electrical recordings.

Open and reproducible neuroscience

Spyglass

Advances in neuroscience are made by analyzing large, complex datasets generated from multiple sources of data. But deriving consistent and reproducible insights from data remains a complex and time-consuming task, as neuroscientists lack the tools to track the entire provenance of analyses and to easily share them with the community. To address this issue, we created Spyglass, an open-source software framework designed to promote the shareability and reproducibility of data analysis in neuroscience. Spyglass integrates standardized formats with reliable open-source tools, offering a comprehensive solution for managing neurophysiological and behavioral data. It provides well-defined and reproducible pipelines for analyzing electrophysiology data, including core functions like spike sorting. In addition, Spyglass simplifies collaboration by enabling the sharing of final and intermediate results across custom, complex, multi-step pipelines as well as web-based visualizations.