Until recently, most research on cognitive phenomena, such as perception and decision-making was done mainly in human and non-human primates. Together with a handful of colleagues, Mainen, head of the System Neuroscience Lab, has helped to show that rodents in fact share many of primates’ cognitive abilities. Indeed, in 2008, his lab was the first to discover neural activity that reflected decision confidence in any species, a feat that was done in rats. Though research projects with human subjects have more recently started in the lab, this approach still dominates the Systems Neuroscience Lab today, where rodents are the stars, allowing the use of advanced genetic and molecular tools not available in humans and non-human primates. Using these tools, the team is able to combine multiple techniques, which allow them to record and manipulate the neural circuits that control confidence and decision-making in relevant brain regions, such as the cortex and the midbrain. The team places a major focus on the midbrain serotonin system, which they believe to play a key role, along with other neuromodulators, in regulating beliefs. Theory and modeling are also a vital component of the work done in the lab because of the inherent complexity involved.
How brains use perceptual information to create and act on models of the world, the role of confidence, uncertainty and neuromodulators in these processes
Theory, Behaviour, Electrophysiology, Optogenetics
Models and Regions
Mice, Humans / Cortex and Raphe Nuclei
We are interested in understanding the principles underlying the complex adaptive behavior of organisms.
We are interested in understanding the principles underlying the complex adaptive behavior of organisms. Starting with quantitative observations of animal behavior, we aim to integrate quantitative cellular and systems level experimental analysis of underlying neural mechanisms with theoretical, ecological and evolutionary contexts. Rats and mice provide flexible animal models that allow us to monitor and manipulate neural circuits using electrophysiological, optical and molecular techniques. We have made progress using highly-controlled studies of a simple learned odor-cued decision task and are extending our focus toward more complex behaviors. Projects in the lab are wide-ranging and continually evolving. Current topics include (i) the function of the serotonin system, (ii) sensory decision-making, (iii) the role of uncertainty in brain function and behavior.
Serotonin signaling in adaptive behavior
Comparison of neuromodulator signals in a visual decision-making task
Kcénia Bougrova, Joana Catarino
Neuromodulators have long been theorized to play special roles in regulating learning and decision-making. A particularly important question underlying these theories is what sort of signal does the release of a neuromodulator report? Dopamine neurons have been widely regarded as reporting a reward prediction error. We hypothesize that serotonin neurons report an unsigned reward prediction error or a more general state prediction error. Norepinephrine and acetylcholine neurons also hypothesized to report error-related signals. However, several of these signals are closely related and no experiments to date have actually compared the activity of all four modulators in the same task. The International Brain Laboratory (IBL) is a collaboration in which we participate with 20 other experimental and theoretical labs studying decision-making in the mouse. The IBL has developed a standardized visually-guided decision task and robust data processing architecture, as well as a growing complement of computational models. This provides an outstanding opportunity to disentangle the differential roles of these neuromodulators. To do so, we are using fiber-photometry to record bulk Ca2+ signals using the genetically encoded sensory GCaMP6 expressed in targeted populations of neuromodulatory neurons in key nuclei. These data will allow us to test theoretical predictions and to relate neuromodulatory signals to task learning and performance as well as to brain-wide electrophysiological recordings.
Collaborators: International Brain Laboratory (Eric Dewit, Peter Dayan, Charline Tessereau, Olivier Winter)
Funding: FCT, CF, Simons Foundation, Wellcome Trust
Examining the role of serotonin in predictive processing
Error signals are central to how the brain updates its internal model of the world with experience. Neuromodulators including dopamine and serotonin send axonal projections widely through the brain and are implicated in responses to salient and unpredictable events, but their precise computational functions remain incompletely understood. Work from our lab showed that serotonin neurons respond to both positive and negative errors in reward outcomes following reversal of cue-outcome contingencies in a classical conditioning task (Matias et al., 2017). This is consistent with the idea that serotonin may report an unsigned prediction error, but is also consistent with a number of diverse roles for serotonin in encoding prediction errors and surprise. To further elucidate the nature of serotonin signaling, we designed a virtual reality task for mice which allowed us to manipulate the structure of the environment including visuo-spatial associations independent of reward. We are using this task together with in vivo photometry of serotonin neurons in order to better understand the nature of serotonin signals.
Collaborators: Leopoldo Petreanu
Publications: Mattias et al., 2017
Funding: CF, FCT PhD scholarship (S. Sautory)
Serotonin and the modulation of neural population activity states
Serotonergic modulation of neural dynamics across the brain
Guido Meijer, Joana Catarino, Laura Silva
Serotonin (5-HT) is released from neurons in the dorsal raphe nucleus, projecting throughout the entire brain. How serotonergic input modulates neural dynamics in downstream target regions remains largely unknown. Therefore we are investigating how serotonin modulates neural activity at the single neuron level across the brain. To this end we target Neuropixels probes to eight insertion targets, recorded in pairs, while optogenetically stimulating serotonin release in the dorsal raphe nucleus. This approach allows us to record from a large number of brain regions and investigate how their local computations are modulated by serotonergic input.
Collaborators: International Brain Laboratory
Funding: Simons Foundation, Wellcome Trust
Atlas of Neuropixels recording sites in the mouse brain. Each line represents the region targeted by a single Neuropixel probe (colored by mouse). During a recording session, two Neuropixels are inserted in the brain simultaneously. The cylinder at the right represents an optic fiber which targets the serotonergic dorsal raphe nucleus.
The function of neuromodulators in hippocampal contextual representations
Jaime Arlandis, José Teixeira
Hippocampal cells represent environmental contexts by integrating self-motion and sensory-derived information with cognitive information about barriers, objects and reward locations in the environment into a cognitive map representation. These representations of environmental context are thought to be fundamental for mechanisms of episodic memory formation and spatial learning. However, often animals develop competing representations for the same context. How do cognitive variables (e.g. expectation) or internal states (e.g. attention/arousal) contribute to the emergence of parallel map representations? To investigate this, we record from the hippocampus and parahippocampal regions as mice navigate in virtual environments using high-density probes. The virtual reality system allows closed-loop stimulation of serotonin neurons, providing insights of this neuromodulator’s action, which is involved in the cognitive representation of expectation, in the development of competing hippocampal contextual representations.
Collaborators: Joshua Stern, Cindy Poo (Allen Institute), Daniel McNamee (Natural Intelligence Lab)
Funding: CF, ERC
Serotonergic and hippocampal representations of changes in context
While the hippocampus is classically viewed as a region involved in memory and navigation, research has found that the effect of a key class of antidepressants is dependent on structural changes in the hippocampus. Place cells are a key feature of the hippocampus. In altered contexts, place-cells remap, changing their firing patterns to form a new internal map. Recent work has thus posited that remapping might be a physiological substrate of flexible contextual inference. Given serotonin’s putative role in cognitive flexibility (Matias et al., 2017), we are exploring how serotonergic input to the hippocampus might mediate these remapping events. To do this, we perform simultaneous fiber photometry recording of serotonergic activity in the dorsal raphe and high density electrophysiology in the hippocampus while mice navigate immersive virtual worlds with sudden changes in scene. These physiological studies will guide future work to further dissect the relationship between neuromodulatory and hippocampal representations of context, potentially towards using causal perturbations to mediate the degree of remapping across ambiguous contextual changes.
Collaborators: Jaime Arlandis, José Teixeira
Publications: Matias et al., 2017
Funding: Fulbright – Luso-American Development Foundation Open Study/Research Award
Map gradually evolving.
Neural ensemble activity in prefrontal cortex during a foraging task and its modulation by serotonin
Elisabete Augusto, Fanny Cazzettes
In any given situation, the environment can be parsed in different ways to define useful decision variables (DVs) for any task, but the way in which this manifold of potential decision strategies is processed to shape behavioral policies is not known. We recorded neural ensembles in the frontal cortex of mice performing a foraging task admitting multiple DVs. Methods developed to uncover the currently employed DV revealed the use of multiple strategies and latent changes in strategy within sessions. Optogenetic manipulations showed that the secondary motor cortex (M2) is needed for mice to use the different DVs in the task. Surprisingly, we found that, regardless of the DV best explaining the behavior of each mouse, M2 activity reflected a full basis set of computations spanning a repertoire of DVs extending beyond those useful for the present task. This form of multiplexing may confer considerable advantages for learning and adaptive behavior. We are currently investigating the role of serotonin in modulation of these representations.
Publications: Cazettes et al., 2022
Collaborators: Guido Meijer, Luca Mazzucato (U. Oregon)
Funding: CF, Simons Foundation
Dynamics of biological and machine learning
The contribution of representation learning to learning variability
Learning to accurately represent the relevant state and action space of the environment is likely a challenging initial step in learning a task. Two-alternative-forced-choice tasks are popular in systems neuroscience. These are usually designed to be simple and allow for high experimental control. However, rodents usually take thousands of trials to reach asymptotic behavior and there is considerable inter-animal variability in learning profiles. In contrast, technological advances increasingly allow for richer paradigms with higher ecological validity, while maintaining high levels of experimental control. These are more naturalistic and immersive but have sensory landscapes which are potentially complex and hard to extract meaningful information from. We hypothesize that representation learning significantly contributes to explaining learning variability, both in simple tasks with reduced ecological validity and in more complex and immersive tasks. To assess this, we use a combination of data analysis and behavioral modeling in (1) mice learning to perform a simple yet unnaturalistic task, and (2) humans learning to navigate a rich and immersive paradigm, respectively.
Collaborators: Daniel McNamee (Natural Intelligence Lab), International Brain Laboratory
Funding: CF (I. Laranjeira), Simons Foundation, Wellcome Trust
Dissecting complex decision making processes
While neural networks approach human levels of performance in many complex tasks, they require much more training than humans. This may be because only humans can infer and apply generalizable principles from prior experience. However, the statistics that underlie the human learning process are poorly understood and hard to investigate in the large state spaces found in most complex tasks. We thus designed a cognitive task whose potential solutions are few enough for subjects to densely sample policy space, but complex enough to compel intelligent search. We launched the game as a smartphone-based app (hexxed.io) and collected data from ~10k human participants. Comparison of human data with artificial agents reveal features of behavior specific to humans.
Publications: Quendera et al., 2022
Collaborators: Gautam Agarwal (Claremont McKenna College)
Funding: BIAL, CF
Human latent state spaces, their control and characterization
Quantitative access to latent states through observable expressions
Eric Lacosse, Scott Rennie
The recognition of mental or “latent” states, such as emotional and cognitive states, from observable expressions is widely practiced by humans and of great importance. However, it is a process that remains very poorly understood scientifically. Latent states are believed to provide actionable information about a person’s future actions and mental health, yet traditional methods for detecting latent states rely on inherently subjective assessments by self-report or human observers. Direct neural recordings promise a direct window into latent states, but remain impractical or infeasible in natural settings. Technology to infer latent states from expressions in the face or body is potentially game changing, yet existing approaches still are hampered by reliance on human labeled samples as training data. In our previous ERC-funded project, we developed a quantitative task for mice and humans that allowed us to quantitatively probe an interesting class of latent states. We discovered that these latent states could be accessed in mice not only through neural recordings but by monitoring mice’s facial expressions. We now aim to test whether these latent states can also be accessed in humans using video monitoring of observables including facial expression and eye movements. We will also test the ability of this approach to generalize to more naturalistic settings. If successful, it would establish a new, more objective method for inferring actionable latent states in humans using non-invasive and cost-effective methods. We will study the implications of this research by exploring both the ethical and privacy implications of latent-state monitoring technology and its potential applications in the areas of drug development, psychotherapy and human-machine interactions.
Collaborators: Giorgio Gristina, Razvan Sandru
The control and characterization of altered states of consciousness
Razvan Sandru, Scott Rennie, Giorgio Gristina, Eric Lacosse
A growing body of work suggests that the serotonin system and psychedelics are key regulators of brain activity and plasticity. Psychedelics may be seen as non-specific amplifiers of subjective states in which sensory, psychological, or environmental elements may be experienced as more salient than in everyday life. This is consistent with the observation that psychedelic states are highly sensitive to the expectations and setting in which the experience occurs. Psychedelics have historically often been used in culturally significant social contexts. In contrast, modern clinical trials have focused on introspective experiences with subjects mostly isolated from the environment. Thus far, there is scarce scientific data explaining how psychedelic experiences and their outcomes are shaped by and in turn shape interaction with the environment. Despite an abundance of clinical trials, these have not attempted to distinguish the causal role of psychedelics themselves from the context in which they are given upon acute and long-term impacts. This is needed to address the lack of empirical data on whether and how the content of altered states of consciousness, or the direction of transformational changes, can be guided toward specific positive effects. In these studies we plan to manipulate specific elements of the expectations and setting of psychedelic experiences in combination with phenomenological and behavioral characterization of psychedelic experiences.
Serotonin, psychedelics and the self
Psychedelics represent a peculiar case in the history of psychopharmacology: (1) their effects seem to depend dramatically on a series of extra-pharmacological factors, including the subjects mindset and their social and physical environment; (2) studies have repeatedly pointed out to these substances’ alleged ability to alter people’s beliefs, by acting on the attribution of meaning and eliciting transformative experiences; (3) they have a long history of non-medical use which has produced a considerable amount of knowledge, alternative to, yet interacting with, the scientific one. Research on these substances has thus often crossed the border of the so-called “hard” sciences by including considerations on topics such as religion or politics. This project, based primarily at the Institute of Social Sciences (U. Lisboa), aims to develop an interdisciplinary dialogue with the scientific discourse on psychedelics by directly and critically engaging with scientists and practitioners working in the field. Topics we are considering with include are: the alteration in the attribution of meaning caused by psychedelics; the lack of diversity in psychedelic trials; the problem with cultural appropriation in psychedelic research; the history of the concept of “mystical experience” and its questionnaires; the role of social interaction in psychedelic states. We are also involved in a systematic review of psychedelic studies, and the organization of an event series on the significance of subjective experience in psychedelic research and clinical practice.
Collaborators: Chiara Pussetti (ICS-U.Lisboa), Maria Carvalho (UCP), Razvan Sandru
Funding: FCT PhD scholarship (G. Gristina)
Phenomenology and ethics of novel and uncertain experiences
Extreme or highly novel experiences can challenge ordinary beliefs concerning reality, actuality, agency and/or self. To understand how one experiences meaning in such situations and mitigates the initial uncertainty presented by novel environments, it is essential to assess as best as possible the character of individual lived experiences. Phenomenology provides the probing tools and explanatory frameworks to study lived experience with a focus on the complexity and multidimensional character of experienced meaning. We aim to integrate these tools in experimental frameworks that promote increased ecological validity and explore possible alternatives to mechanistic and representationalist approaches to experience. From this point of view we also consider bridges between neuroscience, philosophy and ethics to promote a holistic approach to knowledge centered on the betterment of societal and individual well-being.
Collaborators: Scott Rennie