Population geometries in frontal cortex coordinate value-based decisions

Decisions often involve weighing the benefits and costs of different options, such as a small reward offered immediately versus a larger one delayed. While decision-making processes involve multiple regions of the brain, the specific neural computations carried out by these areas are not well understood. By combining large-scale electrophysiology with targeted manipulations in mice making choices between varying reward amounts and waiting periods, researchers were able to uncover differences in the way diverse frontal cortex regions process competing decision variables.

Dorsal frontal brain regions, including motor cortex, transformed competing attributes into a single decision axis shared among scenarios involving rewards and delays. The strength of projections from these dorsal areas to other brain areas predicted how likely a particular choice would be made. Blocking activity in these dorsal regions disrupted action selection regardless of the waiting period, indicating their importance for making decisions.

In contrast, ventral prefrontal cortex (vPFC) represented choices in higher-dimensional spaces that were organized according to the length of the delay. The geometry of this vPFC representation showed how strongly the waiting period influenced the decision. Disrupting vPFC activity maintained dorsal frontal circuits' representation of a shared decision axis but selectively removed the modulation by delay.

As a result, mice behaved as if the delay no longer impacted their choices. Importantly, vPFC manipulations did not affect decisions when mice only had to choose between reward amounts.

Together, these findings suggest a "generator-modulator" architecture in frontal cortex for making complex decisions. Dorsal frontal circuits act as a generator, maintaining a shared decision axis that guides choices across different contexts. Meanwhile, ventral prefrontal cortex functions as a modulator, representing competing options in distinct subspaces that adjust the decision based on the specific factors at play.