Neuronal Coding of Choice and Action-Selection during Decision-Making in Behaving Mice

Project Information

NEUROGOAL

Grant agreement ID: 677878

DOI

10.3030/677878

Project closed

EC signature date 24 March 2016

Start date 1 July 2016

End date 30 June 2022

Funded under

• EXCELLENT SCIENCE - European Research Council (ERC)

Total cost

€ 1 499 933,00

EU contribution

€ 1 499 933,00

Coordinated by

CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE CNRS
France

Periodic Reporting for period 4 - NEUROGOAL (Neuronal Coding of Choice and Action-Selection during Decision-Making in Behaving Mice)

Reporting period: 2021-01-01 to 2022-06-30

Summary of the context and overall objectives of the project

Our daily life is a complex chain of decisions and actions that shapes our behaviors. Individuals tend to choose the best action possible (‘action-selection’) among different alternatives through “goal-directed” decision-making. To learn and achieve an optimal behavior, individuals must: (i) Predict the potential cost (e.g. risk) and benefit (e.g. reward) that might occur as a consequence of an action (‘outcome’). This ‘action-value’ function is learned from the causal consequences of an action (‘action-outcome’ association), and the subjective value of different outcomes (‘outcome-value’ associations); (ii) Compare ‘action-value’ functions and select the action with the greatest value. The probability of selecting one of two choices is called ‘action-selection’ and determined by the difference in their ‘action-value’ functions; (iii) Update the ‘action-value’ function according to the difference between the predicted and the obtained reward. The development of neuroeconomics and artificial intelligence, as well as maladaptive decision-making found in many neuropsychiatric disorders highlight the crucial importance of this process. However, despite the growing interest of action-outcome’ and ‘outcome-value’ associations over the past few years, the neuronal correlates of choices that drive goal-directed ‘action-selection’ as well as the synaptic underpinnings have been largely neglected. Here, we aimed to resolve several outstanding questions: 1) Given that the comparison between choice alternatives should occur in the cortex as a precursor of choice and action, are multiple choices represented in the cortex by specific patterns of cell activation? Are they pre-existing or encoded through learning? 2) Given that the reward values are supposed to be encoded in several cortical structures through the help of subcortical structures, how are these different systems interconnected, and how do they cooperate to implement action values in the cortex and further influence choice? 3) Given that maladaptive decision-making is found in many psychiatric disorders including autism, are the synaptic and cellular underpinnings of decision-making altered in our mouse model of autism? Overall, our work revealed that higher-order motor areas, which include the secondary motor cortex (M2) in rodents, appear to optimally orchestrate the selection of action, during which M2 superficial neurons form with the prefrontal cortex a network that maintains and updates deterministic choice bias.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

To study cognitive/behavioral flexibility in the face of changing contingencies, we first developed a new self-paced decision-task that gave access to both the learning and the execution phase of a decision-making task, for which we implemented computational models (Aime et al., eLife2020). Importantly, our computational model of this task went far beyond simple measures of behavior (e.g. lever pressing), which was thus crucial to tie neural activities to hidden cognitive variables on every trial (e.g. action and decision values). We indeed used two-photon laser scanning microscopy (2PLSM) in vivo and implemented sophisticated experimental strategies (fast volumetric and dual-color two-photon calcium imaging of somas, dendrites and long-range axons, silicon probes, GRIN lens-mediated imaging, and miniscope…) to monitor in foraging mice the structure and activity of M2 superficial neurons on a trial-by-trial basis in expert mice as well as, and for the first time, during learning. As compared to previous studies, this was achieved by tracking the same population of neurons, at very high spatial and temporal resolution and over long-time frames through the implantation of a non-invasive cranial window (Neuron 2019; Cell Reports 2020; PNAS 2021; eLife 2020; Cell Reports 2022). We collected a tremendous amount of data, and our capacity to handle and analyze such large data sets was clearly a limiting factor that was underestimated. To address this issue, we developed new algorithms to extract automatically the activity sources from in vivo two-photon microscopy images, which were released as a free and open-source library to help the community in the field and beyond (biorxiv, 2021, 2022). Our data indicate for the first time that M2 superficial neurons form with the mPFC a network that maintains and updates deterministic choice bias. This decision-value is remarkably persistent in M2 neurons during decision epoch and beyond, notably when high decision values are high, suggesting that network states ultimately settle into stereotyped network activity pattern.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

By using these new methods and strategies, we made several fundamental discoveries (published or submitted to high-profile journals) that have significantly advanced the field of decision-making beyond the state of the art. Notably, we demonstrated the key and intricate function of disinhibition, nonlinear dendritic integration of coincident cortico-cortical and subcortico-cortical input streams in cortical superficial layer, and the trafficking of AMPA receptors in the induction of synaptic plasticity in vivo (Neuron 2019), the initiation of cortical remapping and adaptive behaviors (Cell Reports 2020; PNAS 2021), learning (eLife 2020) and development (Cell Reports 2022). Showing that dendrites are able to generate local computations that influence how animals perceive and adapt to the world, and by which mechanisms, is a break-through in neuroscience, since in the past they were mostly seen as passive elements of the neurons, just funneling information to the soma. Importantly, we found that the firing rates of superficial M2 neurons is gradually modified over long-time scale during reinforcement learning of a foraging task, in a way that quantitatively reflects decision-value, which we found to be dynamically gated by superficial inputs from the basolateral amygdala (eLife 2020). Taken together, our data indicate that M2 pyramidal neurons are indeed critical for action-selection and decision-value signaling.
Key aspects of several psychiatric and neurological disorders involve disturbances of action-selection and decision-value signaling: Parkinson's disease, Tourette's syndrome, attention/hyperactivity disorder, drug addiction, Schizophrenia. Therefore, understanding the neural substrates of decision will tackle medical, societal and economical challenges by potent breakthrough discoveries. Nevertheless, the considerable potential of in vivo neuronal imaging at the micrometer-scale is far from being fully realized, in part because the exploitation of the wealth of experimental data is hampered by the lack of robust, fast automated analysis tools. We designed novel computational methods for the analysis of large-scale in vivo microscopy experiments and provided them to the community to promote this cutting-edge experimental field and advance our understanding of neuronal and cognitive processes. It will thus positively support the dissemination of knowledge and technologies in Europe.

summaryforpublication.png