Akihiro Funamizu

Learning-stage-dependent, field-specific, map plasticity in the rat auditory cortex during appetitive operant conditioning

Cortical reorganizations during acquisition of motor skills and experience-dependent recovery after deafferentation consist of several distinct phases, in which expansion of receptive fields is followed by the shrinkage and use-dependent refinement. In perceptual learning, however, such non-monotonic, stage-dependent plasticity remains elusive in the sensory cortex. In the present study, microelectrode mapping characterized plasticity in the rat auditory cortex, including primary, anterior, and ventral/suprarhinal auditory fields (A1, AAF, and VAF/SRAF), at the early and late stages of appetitive operant conditioning. We first demonstrate that most plasticity at the early stage was tentative, and that long-lasting plasticity after extended training was able to be categorized into either early- or late-stage-dominant plasticity. Second, training-induced plasticity occurred both locally and globally with a specific temporal order. Conditioned-stimulus (CS) frequency used in the task tended to be locally over-represented in AAF at the early stage and in VAF/SRAF at the late stage. The behavioral relevance of neural responses suggests that the local plasticity also occurred in A1 at the early stage. In parallel, the tone-responsive area globally shrank at the late stage independently of CS frequency, and this shrinkage was also correlated with the behavioral improvements. Thus, the stage-dependent plasticity may commonly underlie cortical reorganization in the perceptual learning, yet the interactions of local and global plasticity have led to more complicated reorganization than previously thought. Field-specific plasticity has important implications for how each field subserves in the learning; for example, consistent with recent notions, A1 should construct filters to better identify auditory objects at the early stage, while VAF/SRAF contribute to hierarchical computation and storage at the late stage.

Progressive plasticity of auditory cortex during appetitive operant conditioning

In stimulus-response-outcome learning, different regions in the cortico-basal ganglia network are progressively involved according to the stage of learning. However, the involvement of sensory cortex remains ellusive even though massive cortical projections to the striatum imply its significant role in this learning. Here we show that the global tonotopic representation in the auditory cortex changed progressively depending on the stage of training in auditory operant conditioning. At the early stage, tone-responsive areas mainly in the core cortex expanded, while both the core and belt cortices shrank at the late stage as behavior became conditioned. Taken together with previous findings, this progressive global plasticity from the core to belt cortices suggests differentiated roles in these areas: the core cortex serves as a filter to better identify auditory objects for hierarchical computation within the belt cortex, while the belt stores auditory objects and affects decision making through direct projections to limbic system and higher association cortex. Thus, the progressive plasticity in the present study reflects a shift from identification to storage of a behaviorally relevant auditory object, which is potentially associated with a habitual behavior.

Pre-Attentive, Context-Specific Representation of Fear Memory in the Auditory Cortex of Rat

Neural representation in the auditory cortex is rapidly modulated by both top-down attention and bottom-up stimulus properties, in order to improve perception in a given context. Learning-induced, pre-attentive, map plasticity has been also studied in the anesthetized cortex; however, little attention has been paid to rapid, context-dependent modulation. We hypothesize that context-specific learning leads to pre-attentively modulated, multiplex representation in the auditory cortex. Here, we investigate map plasticity in the auditory cortices of anesthetized rats conditioned in a context-dependent manner, such that a conditioned stimulus (CS) of a 20-kHz tone and an unconditioned stimulus (US) of a mild electrical shock were associated only under a noisy auditory context, but not in silence. After the conditioning, although no distinct plasticity was found in the tonotopic map, tone-evoked responses were more noise-resistive than pre-conditioning. Yet, the conditioned group showed a reduced spread of activation to each tone with noise, but not with silence, associated with a sharpening of frequency tuning. The encoding accuracy index of neurons showed that conditioning deteriorated the accuracy of tone-frequency representations in noisy condition at off-CS regions, but not at CS regions, suggesting that arbitrary tones around the frequency of the CS were more likely perceived as the CS in a specific context, where CS was associated with US. These results together demonstrate that learning-induced plasticity in the auditory cortex occurs in a context-dependent manner.

Neural substrate of dynamic Bayesian inference in the cerebral cortex

Dynamic Bayesian inference allows a system to infer the environmental state under conditions of limited sensory observation. Using a goal-reaching task, we found that posterior parietal cortex (PPC) and adjacent posteromedial cortex (PM) implemented the two fundamental features of dynamic Bayesian inference: prediction of hidden states using an internal state transition model and updating the prediction with new sensory evidence. We optically imaged the activity of neurons in mouse PPC and PM layers 2, 3 and 5 in an acoustic virtual-reality system. As mice approached a reward site, anticipatory licking increased even when sound cues were intermittently presented; this was disturbed by PPC silencing. Probabilistic population decoding revealed that neurons in PPC and PM represented goal distances during sound omission (prediction), particularly in PPC layers 3 and 5, and prediction improved with the observation of cue sounds (updating). Our results illustrate how cerebral cortex realizes mental simulation using an action-dependent dynamic model.

Distributed representation of tone frequency in highly decodable spatio-temporal activity in the auditory cortex

Although the place code of tone frequency, or tonotopic map, has been widely accepted in the auditory cortex, tone-evoked activation becomes less frequency-specific at moderate or high sound pressure levels. This implies that sound frequency is not represented by a simple place code but that the information is distributed spatio-temporally irrespective of the focal activation. In this study, using a decoding-based analysis, we investigated multi-unit activities in the auditory cortices of anesthetized rats to elucidate how a tone frequency is represented in the spatio-temporal neural pattern. We attempted sequential dimensionality reduction (SDR), a specific implementation of recursive feature elimination (RFE) with support vector machine (SVM), to identify the optimal spatio-temporal window patterns for decoding test frequency. SDR selected approximately a quarter of the windows, and SDR-identified window patterns led to significantly better decoding than spatial patterns, in which temporal structures were eliminated, or high-spike-rate patterns, in which windows with high spike rates were selectively extracted. Thus, the test frequency is also encoded in temporal as well as spatial structures of neural activities and low-spike-rate windows. Yet, SDR recruited more high-spike-rate windows than low-spike-rate windows, resulting in a highly dispersive pattern that probably offers an advantage of discrimination ability. Further investigation of SVM weights suggested that low-spike-rate windows play significant roles in fine frequency differentiation. These findings support the hypothesis that the auditory cortex adopts a distributed code in tone frequency representation, in which high- and low-spike-rate activities play mutually complementary roles.

Uncertainty in action-value estimation affects both action choice and learning rate of the choice behaviors of rats.

The estimation of reward outcomes for action candidates is essential for decision making. In this study, we examined whether and how the uncertainty in reward outcome estimation affects the action choice and learning rate. We designed a choice task in which rats selected either the left‐poking or right‐poking hole and received a reward of a food pellet stochastically. The reward probabilities of the left and right holes were chosen from six settings (high, 100% vs. 66%; mid, 66% vs. 33%; low, 33% vs. 0% for the left vs. right holes, and the opposites) in every 20–549 trials. We used Bayesian Q‐learning models to estimate the time course of the probability distribution of action values and tested if they better explain the behaviors of rats than standard Q‐learning models that estimate only the mean of action values. Model comparison by cross‐validation revealed that a Bayesian Q‐learning model with an asymmetric update for reward and non‐reward outcomes fit the choice time course of the rats best. In the action‐choice equation of the Bayesian Q‐learning model, the estimated coefficient for the variance of action value was positive, meaning that rats were uncertainty seeking. Further analysis of the Bayesian Q‐learning model suggested that the uncertainty facilitated the effective learning rate. These results suggest that the rats consider uncertainty in action‐value estimation and that they have an uncertainty‐seeking action policy and uncertainty‐dependent modulation of the effective learning rate.

Condition interference in rats performing a choice task with switched variable- and fixed-reward conditions.

Because humans and animals encounter various situations, the ability to adaptively decide upon responses to any situation is essential. To date, however, decision processes and the underlying neural substrates have been investigated under specific conditions; thus, little is known about how various conditions influence one another in these processes. In this study, we designed a binary choice task with variable- and fixed-reward conditions and investigated neural activities of the prelimbic cortex and dorsomedial striatum in rats. Variable- and fixed-reward conditions induced flexible and inflexible behaviors, respectively; one of the two conditions was randomly assigned in each trial for testing the possibility of condition interference. Rats were successfully conditioned such that they could find the better reward holes of variable-reward-condition and fixed-reward-condition trials. A learning interference model, which updated expected rewards (i.e., values) used in variable-reward-condition trials on the basis of combined experiences of both conditions, better fit choice behaviors than conventional models which updated values in each condition independently. Thus, although rats distinguished the trial condition, they updated values in a condition-interference manner. Our electrophysiological study suggests that this interfering value-updating is mediated by the prelimbic cortex and dorsomedial striatum. First, some prelimbic cortical and striatal neurons represented the action-reward associations irrespective of trial conditions. Second, the striatal neurons kept tracking the values of variable-reward condition even in fixed-reward-condition trials, such that values were possibly interferingly updated even in the fixed-reward condition.

Action-dependent state prediction in mouse posterior parietal cortex

Model-based decision making requires prediction of future states by action-dependent state transition models. To investigate their neural implementation, mice were trained to do an auditory virtual navigation task and neuronal activity was recorded in the posterior parietal cortex (PPC) and the posteromedial cortex (PM), with the genetically encoded calcium indicator GCaMP6f after gene transfer by AAV2/1 and 2-photon microscopy. A mouse was head restrained and maneuvered a spherical treadmill. 12 speakers around the treadmill provided an auditory virtual environment. The direction and amplitude of sound pulses emulated the location of the sound source, which was moved according to the mouses locomotion on the treadmill. When the mouse reached the sound source and licked a spout, it got a water reward. The task consisted of two conditions: continuous condition in which the guiding sound was presented continuously and intermittent condition in which the sound was presented intermittently. In both conditions, mice increased lickings as they approached the sound source, indicating that mice recognized the sound-source position and predicted a reward. In intermittent condition, the anticipatory licking was increased even when the sound was omitted, suggesting that mice maintained the predicted sound-source position based on their own actions. We optically recorded calcium transients of up to 500 neurons simultaneously in each of layers 2, 3 and 5 in 8 mice. A subset of neurons increased the activities as mice approached the goal. This increase of activities was observed both with and without sound inputs in all the layers of PPC, while the increase was observed only during sound inputs in PM. To test how the activities in PPC and PM contributed to the prediction of goal distance, we conducted a decoding analysis. Probabilistic decoder predicted the goal distance from the recorded population activities: the decoder was trained with the data in continuous condition. In PPC, the predicted distance significantly decreased both with and without sound inputs consistently with the actual distance to the goal. These results suggest that PPC realizes action-dependent state prediction in the absence of sensory input.

Model-free and model-based strategy for rats’ action selection

Recent researches have revealed that the amygdala plays a central role in the acquisition and the expression of fear, although few studies have investigated comprehensive gene expression along with the acquisition of fear memory. To investigate gene expression following fear conditioning, we examined cued fear conditioning paradigm in C57BL/6J male mice. Two hours after fear conditioning, the amygdala was dissected out bilaterally from the coronal slices. Then total RNA was purified. Gene expression was analysed by Serial Analysis of Gene Expression (SAGE) resulting 707 up-regulated genes and 1858 down-regulated genes in the fear conditioned sample relative to the control. We found that cytoskeleton related genes (Actb, MAP1), and several vesicle exocytosis related genes (Stx12, Cplx2, and SLC17A6) were up-regulated, and neurodegenerative disease related genes (Gsk3b, Hap1) were down-regulated in the amygdala after fear conditioning. Changes in cytoskeletal and exocytotic gene expressions might be involved in the fear memory acquisition.

Different neural activities require different decoders

In this study, we attempted to identify the most influential features of input data for neural decoding across different decoders. For the example of decoders, we used support vector machine (SVM), k-nearest neighbor method (KNN) and canonical discriminate analysis (CDA) and decoded the tone-induced neural activities in a rat auditory cortex into the test tone frequencies. We proposed an algorithm of sequential dimensionality reduction (SDR) to identify the neural activity pattern which increases the prediction accuracy of each decoder. The algorithm reduced input data one by one without deteriorating the prediction accuracy as far as possible. The accuracy of SVM and KNN improved when neural activities had high spike rates and high dispersiveness. On the other hand, CDA performed better on sparse neural activities. Thus, according to spike rates and dispersiveness of neural activities, an efficient decoder can change. Moreover, considering the different algorithms between SVM - KNN and CDA, we hypothesized that disperse and sparse neural activities have an advantage in discrimination and memory, respectively.