Advances in research on the neural loop mechanism of flexible classification decision

On May 5th, Neuron published an online research paper entitled "Neural Circuit Mechanism of Flexible Perceptual Decision Based on Task Structure Information".

The research was completed by the research group Xu Ninglong, a researcher at the Chinese Academy of Sciences’ Brain Science and Intelligent Technology Innovation Center (Institute of Neuroscience), Shanghai Brain Science and Brain-like Research Center, and State Key Laboratory of Neuroscience.
The task structure knowledge realizes the brain computing mechanism for flexible decision-making of perceptual categories.

Researchers used neural loop optical recording and optical manipulation technology, combined with animal behavior paradigm, and found that the feedback information from the orbital frontal lobe supports the auditory cortex to encode auditory classification standards.
This neural loop operation conforms to the reinforcement based on task state inference Learning calculation model can realize flexible information classification based on task rules.

Researchers have obtained causal evidence through precise intervention in this feedback loop, supporting that the loop calculation mechanism does play a key role in the flexible classification of animals.

The research results are conducive to the exploration of the neural loop operation principle of knowledge-based flexible cognitive behavior.

In a changing environment, organisms can make specific behavioral responses based on different sensory information.

In mammals, thanks to the evolution of the neocortex of the brain, this function has been highly developed.
Animals can not only make simple stimuli-responses, but also can extract common structures in a changeable environment through implicit learning.
Form an internal model or knowledge.

This kind of knowledge about environmental structure can effectively guide animals to form inferences about changes in environmental conditions, resulting in highly adaptive intelligent behaviors.

Studying the neural mechanism of this flexible decision-making intelligent behavior is the key to uncovering the mystery of brain cognition.

Although the neuroscience community has conducted a lot of research on the neural activity of the prefrontal brain area, including the orbital frontal lobe, in animal behavior, a key question that has not been answered is how to pass complex circuits between different brain areas.
Link to work together to achieve flexible decision-making based on inference? In order to study the above problems, Xu Ninglong’s research group established a behavioral paradigm of flexible perception and decision-making on mice with fixed heads, so that under experimental conditions, the process of flexible decision-making can be controlled and analyzed.
In the behavioral paradigm, the neural circuit operation mechanism behind it is studied.

This behavioral paradigm is based on the auditory classification task developed in the laboratory.
In each round, the mouse will receive a sound stimulus of a specific frequency, and classify the stimulus according to the frequency, and pass the classification result through The behavior of licking left or right is reported, and a certain amount of water is given as a reward.

In order to introduce the component of behavioral flexibility into the task, the researchers added two different category dividing lines to the task to determine the classification rules.

The dividing line will change in different rounds to realize the switching of task rules.

In order to obtain rewards, mice need to infer the current classification rules based on the behavior results, and make different classifications of the same sound stimulus according to the classification rules.

Through the design of such a behavioral paradigm, researchers use simple changes in voice frequency and reward rules to guide mice to perform flexible decision-making behaviors based on task rules.

After training, mice can not only make flexible selection of sound stimuli according to task requirements, but also gradually establish a strategy of inferring task rule changes by using the estimation of category boundaries.

Under this behavior strategy, when the category boundary changes, mice can not only quickly change their choices, but also use incomplete feedback information in the behavior results to make inferences about the categories of sound stimuli that have not been experienced.

This task simulates the intelligent behavior of animals using their understanding of task structure information, inferring environmental changes based on incomplete information, and making flexible and adaptive choices for the same sensory stimuli.

Because the task state estimation based on speculation is an implicit cognitive variable, in order to study the neural coding and loop calculation of such cognitive variables, the researchers established a reinforcement learning model that includes task structure and state speculation.

This model can not only accurately simulate the behavior of mice, but also estimate implicit cognitive variables.

On this basis, the researchers used two-photon in vivo calcium imaging technology to record single-cell resolution populations of neurons in the auditory cortex when the mice performed the above behavioral tasks.

The recording results show that the auditory cortex neurons not only represent the sound stimulus itself, but some neurons also directly encode the implicit cognitive variable of inferring the category boundary.

The discovery reveals how neurons in the brain encode subjective cognitive processes in real time.

In order to further study how the cross-brain neural circuit realizes the calculation of category boundary changes based on behavioral results, the researchers speculated based on the results of previous anatomical and functional studies that the top-down orbitofrontal lobe-auditory cortex feedback loop may be flexible The auditory classification task regulates the encoding of implicit classification boundaries in the auditory cortex.

In order to explore this possibility, the researchers used all-optical recording and manipulation technology to perform two-photon calcium imaging of neurons in the auditory cortex and at the same time inhibit the ipsilateral orbitofrontal lobe-acoustic cortex projection through optogenetic technology.

The results show that the auditory cortex's coding of category boundaries does depend on this cross-brain projection loop.

In order to obtain further causal evidence to verify whether the orbitofrontal lobe-auditory cortex loop is really involved in the flexible auditory classification behavior, the researchers used chemical genetics to inhibit the bilateral orbital frontal lobe-auditory cortex projection and found that the mice were classified flexibly at this time The ability of the auditory cortex is significantly impaired, which supports the direct involvement of the orbitofrontal lobe-auditory cortex projection in the intelligent behavior of auditory flexible classification.
This result is also related to the orbitofrontal lobe’s feedback information to update the auditory cortex’s loop calculation function to encode the category boundary.
Consistent.

Finally, in order to further verify whether the orbitofrontal lobe-auditory cortex feedback loop actually performs the calculation function of inferring the category boundary line based on the task structure, the researchers used two-photon in-vivo calcium imaging to project the neuron axis from the orbital frontal lobe to the auditory cortex.
The activity was recorded, and it was found that these axons did indeed encode the feedback information needed to update the category boundary inference.

In summary, this research reveals how a well-defined neural circuit encodes and calculates task state speculation based on structural knowledge, thereby mediating intelligent behaviors for flexible decision-making.

This research has played an important role in promoting the understanding of intelligent behavior and cognitive function from the level of neural circuit mechanisms.

The research was completed by Liu Yanhe, a PhD student in the Perceptual Neural Basic Research Group of the Center of Excellence for Brain Intelligence, and others under the guidance of Xu Ninglong.
Laboratory members Xin Yu, Cui Lele, Pan Jingwei and others made important contributions.

The research work was supported by Yang Tianming, a researcher at the Brain Intelligence Excellence Center, and was funded by the National Natural Science Foundation of China, the Ministry of Science and Technology, the Chinese Academy of Sciences and the Shanghai Science and Technology Commission.

Figure 1.
Brain Intelligence Center of Excellence has made progress in the research of flexible classification decision neural circuit mechanism.
Figure 2.
(A) Task design.

Mice are trained to lick left or right, and report their classification of sound frequencies in order to obtain a certain amount of water as a reward.

The sound stimulus used in the task is seven pure tones evenly distributed between 7 kHz and 28 kHz according to the interval.

In different passages, the sound boundary is located at 10 kHz (low frequency boundary) and 20 kHz (high frequency boundary).

(B) Psychophysical curves of mice in different types of passages in an experiment.

The dotted line represents the dividing line of the mouse's subjective sound stimulation frequency.

The error represents the 95% confidence interval.

Green is the psychophysical curve in the low-frequency dividing passage, and orange is the psycho-physical curve in the high-frequency dividing passage.

(C) Inhibition of orbitofrontal lobe-acoustic cortex projection affects the flexibility of auditory classification.

(D) The use of two-photon in-body calcium imaging to record neuronal activity in the auditory cortex while inhibiting the orbital frontal lobe-acoustic cortex projection activity.

(E) The coding of paragraph types by neurons in the auditory cortex depends on the orbitofrontal lobe-auditory cortex projection activity.

(F) The activity of auditory cortex neurons with selectivity for paragraph type characterizes the mouse's estimation of the category boundary.
This information characterization depends on the orbitofrontal lobe-acoustic cortex projection activity.

(G) Two-photon in vivo calcium imaging recording of orbital frontal lobe-acoustic cortex axons.

(H) Orbitofrontal lobe-acoustic cortex axons show selectivity for mouse choices and rewards.

(I) Orbitofrontal lobe-auditory cortex axon activity provides feedback information to the auditory cortex.

Figure 3.
Source of experimental image: Center for Excellence in Brain Science and Intelligent Technology, Chinese Academy of Sciences