Cerebral mechanisms underlying social decision making in humans: Combining computational neuroscience with intracranial recordings and model-based fMRI – BRAINCHOICE

Bayesian models of social decisions will be tested by parallel experiments in humans exploiting maximally the experimental advantages of two complementary methodological approaches:
1) model-based fMRI using Bayesian models of social interactions in healthy adults, which should specify the relationships between the neural mechanisms, behavior and the patterns of brain activation;
2) iEEG in patients with epilepsy, which is key for a comprehensive description of the temporal dynamics of populations of neurons engaged in social computations. This original approach will allow us to record spikes from single cells and Local Field Potentials (LFPs) and to perform simultaneously fMRI and iEEG in humans.
The locations of neural circuits active when people make decisions in a social setting have been identified. Key components include the ventromedial prefrontal cortex (vmPFC), the dorsomedial (dmPFC) and dorsolateral prefrontal cortex (DLPFC), parts of superior temporal sulcus (STS) including the temporo-parietal junction (TPJ) and the anterior cingulate gyrus (ACCg). Recording from this social brain network, the specific goals of this proposal are to:
(1) Unveil the computational mechanisms underlying the neural processes involved in social decision making and integrate them within the more general framework offered by POMDP models that explain how actions are selected in different contexts;
(2) Characterize the neural mechanisms for inferring other's intended action;
(3) Characterize the neural dynamics of neuronal populations from components of the social brain network when learning social hierarchy and when viewing faces varying in dominant features.
(4) Link computational processes for social cognition (aim 1) with neuronal activity, LFPs and BOLD signal (aims 2&3).

We have recorded directly LFPs from the intact orbitofrontal cortex of patients suffering from drug-refractory partial epilepsy with implanted depth electrodes as they performed a probabilistic reward learning task that required them to associate visual cues with distinct reward probabilities. Our results, published in Brain (2016), showed three successive signals: (i) around 400 ms after cue presentation, the amplitudes of the local field potentials increased with reward probability; (ii) a risk signal emerged during the late phase of reward anticipation and during the outcome phase; and (iii) an experienced value signal appeared at the time of reward delivery. This study provides the first evidence from intracranial recordings that the human orbitofrontal cortex codes reward risk both during late reward anticipation and during the outcome phase at a time scale of milliseconds.
In addition, we have investigated how the brain learns social dominance relationships. We have performed one fMRI study and one tDCS study (in collaboration with Zurich) showing that the dorsomedial prefrontal cortex (dmPFC) is necessary for such social learning.
Moreover, we have characterized the neural dynamics of human amygdala in detecting the dominance status from faces. We recorded local field potentials (LFPs) from intracranial electrodes in human amygdala while subjects viewed faces varying dominance levels. The results indicated that dominance features from a face proceeds early after stimulus onset (110 ms).
Finally, we performed 4 fMRI studies concerning the neural bases of (1) the motives underlying defection or collaboration in a public good game ; (2) beliefs updating when confronted to others’ opinions ; (3) inequity aversion for oneself or on’s group when facing a single individual or another group ; (4) interactions between confidence and a group’s opinion in a social decision making task.
These studies led to new collaborations and to public press release.

The fMRI and iEEG studies have opened new perspectives to better understand how the brain makes decisions in groups. Several factors, such as confidence in one’s own choices, the size of the group and other’s judgement can be modeled and the corresponding brain regions engaged in different social decision making processes can be characterized.
Using tDCS, we demonstrated that the dorsomedial prefrontal cortex (dmPFC) is necessary for learning of social hierarchies. Such causal role could not have been demonstraed if we only used a neuroimaging approach. At the societal level, our findings have crucial implications because they indicate that learning of social hierarchies, a fundamental mechanism needed for social organizations, can be manipulated and eventually improved by non-invasive transcranial direct current stimulation. At the clinical level, our study points to neural mechanisms that may underlie vulnerabilities to neuropsychiatric disorders mediated by the experience of iterated social defeats, which can trigger maladaptive social avoidance, behavioral inhibition, elevated glucocorticoids levels, higher vulnerability to addiction, anxiety- and depression-like symptoms.
At a more technical level, recordings of single cells in patients with epilepsy should lead to a better understanding of the neural mechanismes engaged in social decisions.
Finally, simultaneous PET-iEEG recordings in humans should link the BOLD signal and LFPs.

3 articles have been published in prestigious journals (Brain, Journal of Neuroscience et Cortex), 1 book chapter on social hiearchies (Academic Press) and several abstracts communicated in international conferences (see below).

Y Li, G Vanni-Mercier, F Mauguière, J Isnard and Dreher J-C, Reward risk coding in the orbitofrontal cortex. An intracranial recording study in humans, Brain, 2016 Jan 25. pii: awv409.
- Y. Li, G. Sescousse, C. Amiez, JC Dreher, Local morphology predicts functional organization of experienced value signals in the human orbitofrontal cortex, Journal of Neuroscience, 35(4):1648-1658, 2015
- E Metereau and JC Dreher, The medial orbitofrontal cortex encodes a general expected value signal during anticipation of both appetitive and aversive events, Cortex, 63:42-54, 2015

8 oral communications and posters at HBM and Fifth Symposium on Biology of Decision Making, 2015:
- Seongmin A. Park, J-C Dreher. Diffused responsibility modulates social brain regions in repeated cooperative decision within a group, Human Brain Mapping 2015, June 2015
- S. Park and Dreher J-C, Integration of confidence in own judgment and other’s opinion
- Bottemane L, Dreher J-C. Vicarious rewards modulate the decision threshold of the drift diffusion model
- S.A. Park and Dreher J-C. Diffused responsibility modulates social brain regions in repeated cooperative decision within a group
– Y. Li, .. and Dreher J-C. L Fifth Symposium on Biology of Decision Making
– P Wydoodt, G Sescousse, M Kehmassi and Dreher J-C. Do pathological gamblers build a composite fallacious representation of randomness?
– R Ligneul and Dreher J-C. Fifth Symposium on Biology of Decision Making
– T.S. Bortolini, P. Bado, S. Hoefle, A. Engel, J-C Dreher and J. Moll

Although social decision making is ubiquitous and central to human society, its underlying neural mechanisms remain poorly understood. There is a need for understanding social decision processes at different levels, bridging the gap between fundamental computational principles and the brain system level. In particular, the fact that complex social decision making relies on probabilistic knowledge about the possible outcomes of choices and on the intentions and cooperativeness of other individuals has been underappreciated.
The current project seeks a better understanding of the psychological and neurobiological basis of social decision-making in humans. We propose to develop a new theoretical computational neuroscience framework of social decision making. We will adopt a combination of interdisciplinary perspective, combining Bayesian models and multimodal neuroimaging, i.e. intracranial recordings (iEEG) in humans and fMRI, to investigate the neural mechanisms of decision making processes in social context.
The general goal is to characterize the computational principles and the neural mechanisms underlying social decision making. Our main hypothesis is that when we are in an interactive social setting, our brain performs Bayesian inferences using probabilistic representations of other individuals’ intentions and cooperativeness. We will use such probabilistic approaches as Bayesian inference and partially observable Markov decision processes (POMDP) to model ways in which we might predict hypothetical action outcomes, the intentions of others and whether the other is cooperative or competitive.
This theoretical framework will be tested by parallel experiments in humans exploiting maximally the experimental advantages of two complementary methodological approaches:
1) model-based fMRI using Bayesian models of social interactions in healthy adults, which should specify the relationships between the neural mechanisms, behavior and the patterns of brain activation;
2) iEEG in patients with epilepsy, which is key for a comprehensive description of the temporal dynamics of populations of neurons engaged in social computations. This original approach will allow us to record spikes from single cells and Local Field Potentials (LFPs) and to perform simultaneously fMRI and iEEG in humans.

The locations of neural circuits active when people make decisions in a social setting have been identified. Key components include the ventromedial prefrontal cortex (vmPFC), the dorsomedial (dmPFC) and dorsolateral prefrontal cortex (DLPFC), parts of superior temporal sulcus (STS) including the temporo-parietal junction (TPJ) and the anterior cingulate gyrus (ACCg). Recording from this social brain network, the specific goals of this proposal are to:
(1) Unveil the computational mechanisms underlying the neural processes involved in social decision making and integrate them within the more general framework offered by POMDP models that explain how actions are selected in different contexts;
(2) Characterize the neural mechanisms for inferring other's intended action;
(3) Characterize the neural dynamics of neuronal populations from components of the social brain network when learning social hierarchy and when viewing faces varying in dominant features.
(4) Link computational processes for social cognition (aim 1) with neuronal activity, LFPs and BOLD signal (aims 2&3).
In addition to discover whether Bayesian inferences can provide the insights regarding the computational algorithms adopted by the brain to infer the intentions of others, the ground-breaking nature of this research is to: (a) establish a mechanistic foundation for understanding the neurocomputational mechanisms underlying social choice behaviour; (b) provide a multilevel understanding of social decisions, from the system-level to the level of neuronal populations; (c) characterize the spatio-temporal neural dynamics of the social brain structures engaged in social decisions.