Opposing functional influence of cortical L5 neurons on functionally distinct subpopulations of L2/3

Inferring the mappings between actions and sensory feedback is a fundamental aspect of brain function. These mappings are described by two types of internal models – one that transforms motor signals into predictions of sensory feedback (forward model), and one that transforms sensory goals onto the motor commands necessary to achieve those goals (inverse model). One promising theoretical framework that provides computational ground for learning and maintaining these mappings, is that of predictive processing. The functional implementation of this algorithm in cortex implies two computational modules that interact with each other – a comparator circuit and an internal representation circuit. The key functional component of the comparator circuit – prediction-error neurons – has been previously identified in layer 2/3 across multiple cortical areas. The internal representation circuit remains enigmatic and has been hypothesized to reside in layer 5. Elucidating interactions between functional subpopulations of layer 5 and layer 2/3 becomes key for identifying prospective populations of internal representation neurons and testing the framework.
To test the circuit model’s predictions on functional influence of internal representation neurons on prediction-error neurons, we used two-photon calcium imaging in combination with cell-type specific optogenetic manipulations while mice were engaged in visuomotor behavior in a virtual reality environment. We found that optogenetic activation of Tlx3-positive layer 5 intratelencephalically projecting neurons in V1 resulted in opposing functional influence on layer 2/3 positive and negative prediction-error neurons: positive prediction-error neurons exhibited an increase in activity upon stimulation, while negative prediction-error neurons, exhibited a decrease. The observed functional connectivity in combination with other known properties of Tlx3-positive layer 5 neurons suggests that these neurons could either carry a role of inverse internal representation neurons in V1 (thus, being involved in maintaining an inverse internal model), or function as a type of internal representation neuron under a hierarchical predictive processing model.