New Mechanism of Auditory Categorization in the Cortical Neuronal Populations


  A recent study published in Neuron demonstrates that neuronal ensembles in mouse auditory cortex exhibit dynamic changes to facilitate stimulus categorization when the animals are engaged in a perceptual categorization task. This work was performed by researchers in Dr. XU Ninglong’s Lab at the Center for Excellence in Brain Science and Intelligence Technology, Institute of Neuroscience, State Key Laboratory of Neuroscience, Chinese Academy of Sciences. To understand how neural circuits encode auditory perception, the researchers designed an auditory-based categorical decision-making task in head-fixed mice, and performed in vivo two-photon imaging to record populations of neurons in the layer 2/3 auditory cortex. They found that at single-cell level, neurons showed task-related activity, and the representations for auditory information were dynamically modulated during task performance. Frequency selectivity of single neurons changed toward better representations for behaviorally relevant categories. Using population decoding, they showed that a local population of neurons within auditory cortex reliably predicted animals' behavior performance.

  Why categorization is so important? Our brain receives high-dimensional and continuous sensory information from the external world, but at cognitive level the brain could only generate limited number of perceptions and corresponding actions. In order to generate meaningful perceptions to guide behavior, the brain needs to efficiently organize these sensory inputs to form concepts and produce discrete actions. Categorization is exactly the process to achieve this, in that behaviorally relevant information is extracted and sorted to produce perceptual judgments. For example, we could easily tell whether the voice in a phone call belongs to a friend or a stranger, even though the voice can often be noisy and distorted. When we think about the rainbow, although the wavelength of the light in the rainbow is changing continuously, we always have in mind the seven discreate colors. Thus categorization allows us to form discrete concepts and more easily communicate these concepts. Although categorization has been considered fundamental for cognition and widely studied in psychology and cognitive sciences, the neural mechanism of categorization begun to be investigated only recently. Such investigations would bring crucial understandings to the neural mechanisms of cognition. Some of previous studies have identified neuronal activities correlated with stimulus categorization in the prefrontal cortex (Freedman et al, Science 2001) and posterior parietal cortex (Freedman and Assad, Nature 2006). In these studies, neurons showed correlation with category membership of the stimuli, which more likely reflects the results of the computation transforming sensory information to categories. So the critical question yet to be answered is what is the neuronal computation that transforms continuous sensory information into discrete categories.

  To address this question, researchers in Dr. XU Ninglong’s lab used an auditory categorization task in mice, and combined in vivo two-photon imaging to examine how neuronal populations in the auditory cortex implement the computation of sensory to category transformation. They trained head-fixed mice to discriminate a series of pure tones (6 or 8 different frequencies) into two categories: the high-frequency category or the low-frequency category. Normally mice can reach above 80% correct rate after one-week training (Panel A and B). After mice have learned this task, the researchers performed in vivo two-photon calcium imaging using genetically encoded calcium indicator (GCaMP6s) through a chronic cranial window (panel C) to monitor large populations of neurons in the auditory cortex while the animals were performing the task.

  They found that single neuron activity showed two types of categorization-related responses, in addition to conventionally described frequency selective responses. One type of neurons showed category-identity selective response under task condition, but not under passive listening condition. As shown in Panel D of the figure below, the neuron shows strong response to the low-frequency category, but not to the high-frequency category. Another type of neurons showed selective responses to frequencies near the category boundary, and again, such selectivity only exists during task performance, but not during passive condition (Panel E). At populational level, they found that population response patterns showed dynamic changes during behavioral task, with more neurons showing selectivity to near-boundary stimuli (Panel F). Using population decoding methods, they found that local populations in the auditory cortex accurately and reliably predict animals’ choice behavior (Panel G-H).

  This work provides new insights to the understanding of how cortical neurons compute to transform continuous sensory information into discrete categories.

  This work was published online in Neuron on July 8th, entitled “Sensory-to-category transformation via dynamic reorganization of ensemble structures in mouse auditory cortex”. XIN Yu is the first author and Dr. XU Ninglong is the corresponding author. This work was supported by National Natural Science Foundation of China (grant No. 31571081); National Key R&D Program of China (grant No. 2017YFA0103900/2017YFA0103901); the Strategic Priority Research Program of the Chinese Academy of Sciences (grant No. XDB02040008); Key Research Program of Frontier Sciences, CAS (grant No. QYZDB-SSW-SMC045 ).


  Figure legend: A, Illustration of auditory based stimulus categorization task in head-fixed mouse. Under passive condition the lick ports were removed, and the animal passively listened to the auditory stimuli. B, Psychometric function of averaged across 30 sessions. C, Schematic showing two-photon imaging experiments in behaving mice. D, One example neuron showing category-selective response under task condition. E, One example neuron showing category-boundary selective responses. F, Changes in population selectivity for tone frequencies in an example imaging field between passive and task conditions. Colored patches represent frequency selectivity of individual neurons. G, Correlation between predicted choices and behavioral choices. H, The neurometric functions generated from population responses match the psychometric functions.