The project has generated compelling convergent evidence that maintenance (subserving WM) and integration (subserving DM) of information over time are implemented in shared neural circuits. We used recurrent neural network (RNN) modelling to provide initial proof-of-principle that a single circuit, with characteristics matching those of a specific class of neurobiologically plausible models (‘ring attractors’), can produce both robust WM and flexible information integration depending on the task context. We used non-invasive scalp electroencephalography (EEG) combined with state-of-the-art multivariate decoding methods to show that human participants performing WM and DM tasks appear to exploit this same shared-circuit solution. We also demonstrated that while the human participants successfully adapted across WM and DM contexts, their behaviour in both was contaminated by shared sources of noise and bias that likely originate within the shared memory/decision circuit. Furthermore, we succeeded in identifying an apparently prominent role for pupil-linked arousal systems in adaptation of the shared circuit across WM vs DM contexts. Altogether, these findings amount to a fundamental new insight into the neural basis and close relationship between two higher cognitive functions. They have been received very well at scientific conferences and we are currently preparing them for publication.
Over the course of the project, the researcher has also published two research articles in collaboration with international laboratories, both of which addressed questions relevant to the project. Work with collaborators at Leiden University further illuminated the role of pupil-linked arousal systems in higher cognition – in this case, in the implementation of ‘cognitive control’ on the classic Stroop task (Tromp, Nieuwenhuis & Murphy, 2022, Computational Brain & Behavior). Other work with collaborators at UKE Hamburg assessed how humans flexibly switch between distinct sensory-motor mapping rules during DM (van den Brink et al., 2022, Neuron).
Lastly, at time of writing we have also collected 60% of a behavioural/EEG/pupillometry dataset of older adults performing the same WM/DM paradigm used above, with the aim of exposing potential shared sources of age-related deficits in WM and DM. Preliminary analysis of these data indicate that degraded sensory encoding is a dominant source of age-related increase in WM and DM error. Data collection and analysis for this part of the project are ongoing; once complete, we will write up the results and submit them for publication.
