learning and flexibility

The overarching mission of our lab is to understand the neural mechanisms underlying learning and flexibility in the healthy brain so as to inform efforts to reverse engineer mechanistic disruptions in neurological disorders. We use behavior, molecular tools for cell-type and projection-specific access, two-photon imaging, fiber photometry, optogenetics and computational modeling.

learning versus performance

The distinction between learning and performance, first raised by Edward Tolman, posits that meaningful learning can occur even in the absence of any apparent performance improvements. We have explored this phenomenon in rodent models of learning (mice and rats, Kuchibhotla et al., 2019, Nat. Comms.), showing that animals rapidly acquire (within tens of trials) the core contingencies of a task (i.e. what actions to take in response to which sensory cues) and then engage in structured forms of exploration that leads to gradual performance improvement over hundreds to thousands of trials (Zhu and Kuchibhotla, 2024 Current Biology, Moore and Kuchibhotla, 2022, IBRO Reports). We are interested in how the sensory cortex and ascending neuromodulatory systems (including cholinergic and noradrenergic) subserve learning, performance, or both. Our recent work using longitudinal two-photon imaging, optogenetics and computational modeling demonstrates that the auditory cortex drives rapid learning via non-canonical higher-order mechanisms (Drieu et al., Biorxiv, under review) and the role of cholinergic (Zhu et al., 2023, Nature Neuroscience) and noradrenergic forms of neuromodulation. We aim to build a near ‘complete’ microcircuit view in the auditory cortex by monitoring and manipulating neuromodulatory inputs (cholinergic and noradrenergic) and local circuit elements (excitatory/inhibitory neurons and astrocytes). Moreover, have demonstrated that the auditory cortex is the default pathway for auditory discriminative learning but that the auditory cortex appears to ‘teach’ itself out of the task after extended training. We are following this teaching signal as it leaves the auditory cortex and travels to the auditory thalamus and striatum (using two-color two-photon imaging with GRIN lenses to monitor subcortical neurons and cortical axons), to define the precise circuit logic that allows sub-cortical structures to take over. We combine this with optogenetics and theoretical modeling to better understand the distributed nature of learning.

People: Ziyi Zhu, Dr. Jennifer Lawlor, Dr. Sharlen Moore, Rashi Monga, Su Jin Kim

multi-task learning

Humans and other animals can learn many different tasks throughout their lifespan, a process known as continual (or multi-task) learning. This seemingly natural ability, however, challenges most artificial neural networks requiring specified architectures that are computationally expensive. Biological brains and animal minds, however, use cognitive processes such as generalization and inference to enable continual learning. We are developing novel behaviors in combination with cutting-edge mesoscopic two-photon imaging and probabilistic optogenetics to understand how the brain ‘represents’ and ‘controls’ the learning of multiple tasks without forgetting. We are addressing three major questions in this focus area: (1) How do neural representations in the auditory and posterior parietal cortices of two tasks emerge during sequential versus simultaneous learning? (2) How are flexible rules represented in the medial prefrontal cortex and then applied to entirely novel situations? (3) How is compositionality (learning complex tasks by re-combining simpler task primitives) implemented in cortical circuits?

People: Ziyi Zhu, Fangchen Zhu, Aneesh Bal, Cecilia Shuai

Alzheimer's disease

An astonishing, yet underappreciated, aspect of AD are ‘positive’ symptoms whereby patients experience rare, transient, and contextually-triggered improvements in cognition—including lucid intervals. Memories may thus exist in the AD brain but remain largely inaccessible. We have established a new behavioral paradigm in an AD mouse model that allows us to behaviorally and neurally dissociate memory storage from retrieval (Santi et al., under review) and are now testing the extent to which neuromodulation and inhibition control retrieval. In parallel, we are building a technology platform to track dementia in human patient in order. To date, no study has systematically assessed the scientific validity of paradoxical lucidity in AD. These episodes suggest that even a highly pathological AD brain can tap into intrinsic mechanisms that transiently improve cognition. We aim to systematically explore the features, incidence, and predictability of lucid intervals in human patients using a novel, app-based, measurement approach that can be used within the patient’s home or local environment. Our long-term goal is to identify targets for therapeutic intervention.

People: Dr. Andrea Santi, Rashi Monga

habit formation

We have recently shown that cue-driven goal-directed behavior spontaneously and abruptly transitions to habit (‘volitional engagement’, Moore, et al., Biorxiv, under review). Multiple neural circuits contribute to goal-directed and habitual behavior, including the striatum. A fundamental division of labor is hypothesized between the DMS (goal-directed) and DLS (habit). Both the DMS and DLS are highly integrative structures, receiving prominent glutamatergic (cortical) and dopaminergic (midbrain) inputs. Our overarching hypothesis is that rather than sequential recruitment of DMS and then DLS, both controllers are engaged in parallel throughout learning and that projection-specific pre-motor and pre-frontal cortical signals serve as a higher-level ‘arbiter,’ determining whether the DMS or DLS controls behavior at a given moment. To test this, we are (1) validating our volitional engagement approach, (2) using optical and computational tools to monitor and manipulate inputs and outputs of the DLS/DMS across learning, and (3) extending our approach to a tasks in rats to test for generalizability.

People: Dr. Sharlen Moore, Zyan Wang

echolocating bats

echolocating bats.jpg

Bats provide an extraordinary model species to elucidate the neural circuits that enable natural behaviors. We are working in collaboration with Melville Wohlgemuth (Arizona ), Phillip Gutruf (Arizona), and Cindy Moss (JHU) to adapt optical tools, including awake two-photon imaging and wireless optogenetics, for use in the echolocating bat. Recent work in the lab has focused on how the auditory midbrain processes ethologically-relevant vocalizations (Lawlor et al., Biorxiv, under review). A future question of particular interest is whether and how the midbrain (inferior and superior colliculus) and the auditory cortex operate in concert to integrate bottom-up signals about the sensory environment with top-down control of behavioral strategies in a naturalistic context.

People: Dr. Jennifer Lawlor