Neural dynamics of an extended frontal lobe network in goal-subgoal problem solving

Valentina Mione*, Jascha Achterberg*, Makoto Kusunoki, Mark J. Buckley, and John Duncan
*equal contribution

Below is a summary of the project. The full preprint can be found on: (Link to be added soon)

Abstract

Complex behavior calls for hierarchical representation of current state, goal, and component moves. In the human brain, a network of “multiple-demand” (MD) regions underpins cognitive control. We recorded from four putative homologs to human MD regions in the frontal lobe ventrolateral (vlPFC), dorsomedial (dmPFC), dorsal premotor (dPM) and insula/orbitofrontal (I/O) cortex - as monkeys solved an on-screen spatial maze. Across regions there was wide variation in strength of encoding task features. Sensory input and current state were strongly coded in vlPFC, goal most stably in dmPFC, and move most rapidly in vlPFC and dPM. I/O responded during revision of a prepared route. Across regions, an abstract, hierarchical code of problem structure marked progress from problem start to end. We suggest that, across an extended frontal network, partially separated but widely reproduced codes build the structured control program of organized behavior.

Task & recordings

Monkeys solve a multi-step maze task which requires them to navigate a 2D grid from the start location to 1 of 4 goal locations, using saccades. Monkeys start at the center, need to remember the current goal location (presented at the start of the trial), and navigate using presented choice options:

Depending on available choice options, goal locations can be reached via 2-step or 4-step routes:

We record 1374 neurons from 4 regions of frontal cortex (FC), using semi-chronic microelectrode arrays.

On the figure above PS refers to the principal sulcus and AS to the arcuate sulcus. The number of neuron recorded by region are:

vlPFC: 315
dPM: 341
dmPFC: 360
I/O (intermediate agranular insula and immediately adjacent orbitofrontal cortex): 358

Extracting neural subspaces

To understand the computations within FC, we want to identify the neural subspaces in which key variables are represented (goal, next move):

Combine data from 2nd step & 4th step to create PCA space.
Project data from across the trial into space to observe the dynamics of the geometry.

For each region we ask: Which variables determine the shape & dynamics of projections?

Codes for goal and location

We project the population activity for trials into the ‘Goal space’ and then measure the average distance of projections when grouping them by ‘current goal’ or ‘current position’.

Illustration: Grouped by shape and color. Distances lower for color.

Then we compare this distance to cases with different goals & positions (‘Baseline’). Data from 2nd & 3rd choice. P-values via bootstrap. We see that the shape of projections of different regions are determined by different variables. vlPFC is mostly driven by the current location, dmPFC is more strongly driven by the current goal, and dPM and I/O capture a mixture of both variables.
(Not depicted: dmPFC has stable goal code throughout trial.)

Codes for next move

Same as (4) but using the ‘Move space’ and calculating distances with regards to the first move or second move. ‘Baseline’ are moves which are neither first nor second move. We see that the move code develops in vlPFC before it reaches dPM. The other two regions to not show a strong move code.

Generally, the ‘Move space’ is orthogonal to the ‘Goal space’, except for in dmPFC.

Hierarchical choice code

Correlation analysis of neuronal population activity reveals hierarchical coding of abstract problem structure across brain regions. All regions demonstrated strongest correlations within the same step and time window, indicating neural encoding of progress through problem-solving. Secondary correlations showed region-specific differences: most areas maintained stronger within-step than between-step correlations (temporal coding), with this pattern less pronounced in vlPFC. These relationships persisted even when controlling for temporal proximity, suggesting genuine representation of abstract problem structure rather than mere temporal adjacency.

Conclusions

Our findings reveal a distributed frontal network with both specialized and overlapping functions that coordinates multi-step, goal-directed behavior. While regions show partial specializations—vlPFC rapidly processes sensory input and current state, dmPFC maintains stable goal representation, dPM codes movement parameters, and I/O signals route complexity—substantial information sharing occurs across all regions. The network is characterized by orthogonal representational spaces that minimize interference between different task aspects, and a shared hierarchical coding of abstract problem structure. This organization suggests how the primate brain constructs flexible control programs for complex, temporally extended behaviors through regional specialization embedded within a broadly interconnected system.

Page updated

Google Sites

Report abuse