DYNAMIC AND PARTIALLY SEGREGATED NEURAL REPRESENTATIONS IN THE ANTERIOR CINGULATE CORTEX SUPPORT COMPLEX TASK LEARNING

The ability to rapidly adapt to new behaviours and tasks is crucial for humans and animals living in complex, dynamic environments. Although this is a defining feature of mammalian intelligence, the mechanisms by which the brain uses existing knowledge to learn new tasks and integrate novel information remain unknown.
The Anterior Cingulate Cortex (ACC) has been shown to play a crucial role in representing task-relevant features, potentially facilitating rapid learning of novel and increasingly complex tasks. We hypothesise that the ACC enables rapid learning by updating task-specific representations through modular neural processes.
To evaluate our hypothesis, we trained mice on a progressively complex task across multiple sessions while recording in vivo neural activity in the ACC using calcium imaging. Results show that behaviourally learned task features are encoded by distinct neuronal subpopulations, some exhibiting pure selectivity, while others display mixed selectivity to task features. As task complexity increases, neurons are recruited to represent the new task-relevant features while maintaining prior encodings. Notably, the recruitment of mixed selective cells throughout learning is associated with improved task performance. Using linear classifiers, we show that task-relevant features are represented at a population level. However, at both single cell and population levels, these encodings are relatively unstable and do not generalise well over sessions.
These findings suggest that the ACC supports flexible learning through the dynamic recruitment of feature-specific and mixed-selectivity neuronal populations that integrate new task features into existing representations, providing a potential mechanism for rapid adaptation.