Periodic Reporting for period 4 - See-ACC (Cracking the Anterior Cingulate Code: Toward a Unified Theory of ACC Function)
Période du rapport: 2024-01-01 au 2025-06-30
Anterior cingulate cortex is one of the largest riddles in cognitive neuroscience and presents a major challenge to mental health research. ACC dysfunction contributes to a wide spectrum of psychiatric and neurological disorders but no one knows what it actually does. Although more than a thousand papers are published about it each year, attempts to identify its function have been confounded by the fact that a multiplicity of tasks and events activate ACC, as if it were involved in everything.
To address this conundrum, this project developed a biologically-inspired computational model of ACC function and systematically tested the model in a series of experiments involving functional magnetic resonance imaging (fMRI) and electroencephalography in both healthy human subjects and patients. The project further developed a formal theory of ACC function based on mathematical principles of non-linear dynamical systems analysis. Overall the work links high level, abstract processes associated with hierarchical reinforcement learning with lower level representations as encoded in real brains.
It is expected that this formal account of ACC will fill an important gap in the cognitive neuroscience of cognitive control and decision making, impact clinical practice, and be important for artificial intelligence and robotics research, which draws inspiration from brain-based mechanisms for cognitive control.
To address this conundrum, this project developed a biologically-inspired computational model of ACC function and systematically tested the model in a series of experiments involving functional magnetic resonance imaging (fMRI) and electroencephalography in both healthy human subjects and patients. The project further developed a formal theory of ACC function based on mathematical principles of non-linear dynamical systems analysis. Overall the work links high level, abstract processes associated with hierarchical reinforcement learning with lower level representations as encoded in real brains.
It is expected that this formal account of ACC will fill an important gap in the cognitive neuroscience of cognitive control and decision making, impact clinical practice, and be important for artificial intelligence and robotics research, which draws inspiration from brain-based mechanisms for cognitive control.
The project consists of 4 work modules; nearly all of the objectives have been achieved. A few objectives are still in progress as students complete and defend their PhD dissertations while their research is undergoing review at various scientific journals. All published references have been published in high-impact, peer-reviewed, scientific journals.
Module 1: We created a hierarchically-organized recurrent neural network model of ACC that can implement complex, goal-directed multi-step action sequences. Crucially, the model predicts how hierarchical action sequences are represented at different levels of abstraction along the expanse of ACC. The model was subsequently tested in an fMRI experiment and the results published (Colin et al., 2025). Furthermore, we developed and published a mathematical, formal theory of ACC based on dynamical systems analysis (Holroyd, 2025).
Module 2 and Module3: These modules consist of several fMRI experiments that test the computational framework developed in Module 1. For example, consistent with the model predictions, we showed that ACC monitors reward contingencies to motivate control over hierarchically-organized, goal-directed behaviors (Foinikianaki et al., 2025a). Related studies from these modules have either been published (Shahnazian et al., 2022) or are under review (Alejandro et al., 2025a; Foinikianaki et al., 2025b; Ikink et al., 2025a, 2025b). As just one example, Alejandro et al. (2025b) demonstrated that ACC builds and sustains an abstract representational bridge that links planning to action.
Module 4: We successfully completed two experiments that tested the ACC theory in patient populations. First, by recording brainwave activity from intracranial electrodes placed directly in the brains of 19 patients with epilepsy, we provided the first direct evidence that the ACC is the neural generator of brain-wave component called the reward positivity (Oerlemans et al., 2025). Second, by recording brainwaves from the scalps of 72 stroke patients with lesions to the frontal lobes including ACC, we have shown that damage to ACC disrupts the reward positivity, providing independent, causal confirmation of the results of the intracranial study (Oerlemans et al., under review).
Related: We have published two review articles (Holroyd & Verguts, 2021; Alejandro et al., 2024) and a behavioral study (Wientjes & Holroyd, 2024) related to ACC function.
Module 1: We created a hierarchically-organized recurrent neural network model of ACC that can implement complex, goal-directed multi-step action sequences. Crucially, the model predicts how hierarchical action sequences are represented at different levels of abstraction along the expanse of ACC. The model was subsequently tested in an fMRI experiment and the results published (Colin et al., 2025). Furthermore, we developed and published a mathematical, formal theory of ACC based on dynamical systems analysis (Holroyd, 2025).
Module 2 and Module3: These modules consist of several fMRI experiments that test the computational framework developed in Module 1. For example, consistent with the model predictions, we showed that ACC monitors reward contingencies to motivate control over hierarchically-organized, goal-directed behaviors (Foinikianaki et al., 2025a). Related studies from these modules have either been published (Shahnazian et al., 2022) or are under review (Alejandro et al., 2025a; Foinikianaki et al., 2025b; Ikink et al., 2025a, 2025b). As just one example, Alejandro et al. (2025b) demonstrated that ACC builds and sustains an abstract representational bridge that links planning to action.
Module 4: We successfully completed two experiments that tested the ACC theory in patient populations. First, by recording brainwave activity from intracranial electrodes placed directly in the brains of 19 patients with epilepsy, we provided the first direct evidence that the ACC is the neural generator of brain-wave component called the reward positivity (Oerlemans et al., 2025). Second, by recording brainwaves from the scalps of 72 stroke patients with lesions to the frontal lobes including ACC, we have shown that damage to ACC disrupts the reward positivity, providing independent, causal confirmation of the results of the intracranial study (Oerlemans et al., under review).
Related: We have published two review articles (Holroyd & Verguts, 2021; Alejandro et al., 2024) and a behavioral study (Wientjes & Holroyd, 2024) related to ACC function.
This project has three notable achievements that progress beyond state-of-the art:
• We are the first group to utilize artificial neural networks in conjunction with an analysis technique called representational similarity analysis to search for representations related to cognitive control in frontal cortex (Colin et al., 2025). The novelty of this advance was recognized in a review article on the subject, published in the high impact journal Trends in Cognitive Sciences, which highlighted our approach (Freund et al., 2021).
• We have proposed a new theory that integrates ACC function into a formal mathematical framework for cognitive control based on dynamical systems analysis (Holroyd, 2025). As evidence of the potential impact of this theory, this paper has formed the basis for a subsequent ERC grant proposal (“Mapping the Controllosphere”), that was recently been awarded to the P.I. (“Mapping the Controllosphere”, European Research Council Advanced Grant, €2,490,213 total award).
• Our group is the first to localize the source of scalp-related EEG activity by aggregating iEEG data across participants. Although other studies have pooled electrodes across participants to examine source activity in local brain volumes, examining brain-wide activity by pooling electrodes across the entire brain (as we did) allows for testing the hypothesis that the source is located in a predicted brain area against the alternative hypothesis that the source is located somewhere else. This enabled us to examine source activity across large volumes of the brain in a theory-neutral manner, leveraging information related to both the source’s near- and far-field potentials.
• We are the first group to utilize artificial neural networks in conjunction with an analysis technique called representational similarity analysis to search for representations related to cognitive control in frontal cortex (Colin et al., 2025). The novelty of this advance was recognized in a review article on the subject, published in the high impact journal Trends in Cognitive Sciences, which highlighted our approach (Freund et al., 2021).
• We have proposed a new theory that integrates ACC function into a formal mathematical framework for cognitive control based on dynamical systems analysis (Holroyd, 2025). As evidence of the potential impact of this theory, this paper has formed the basis for a subsequent ERC grant proposal (“Mapping the Controllosphere”), that was recently been awarded to the P.I. (“Mapping the Controllosphere”, European Research Council Advanced Grant, €2,490,213 total award).
• Our group is the first to localize the source of scalp-related EEG activity by aggregating iEEG data across participants. Although other studies have pooled electrodes across participants to examine source activity in local brain volumes, examining brain-wide activity by pooling electrodes across the entire brain (as we did) allows for testing the hypothesis that the source is located in a predicted brain area against the alternative hypothesis that the source is located somewhere else. This enabled us to examine source activity across large volumes of the brain in a theory-neutral manner, leveraging information related to both the source’s near- and far-field potentials.