Research
The neural mechanisms of perceptual decision making
Perception is the process of providing a meaningful interpretation to the incoming sensory information. The study of perception goes many years back and it is based on a large body of psychophysical experiments and sensory systems electrophysiology going several decades back. Despite the detailed understanding of some of the neural circuits connecting sensory stimuli with sensation and action, many fundamental questions remain open in the area of perceptual decision making: what is the origin of the spiking variability observed in the neural responses of most brain sensory areas? Is it related to the ubiquitous behavioral variability observed in subjects when facing ambiguous stimuli? How is perception penetrated by recent previous experience? How is the stream of time-varying sensory information integrated across time to provide a stable perception? How is perception impaired in different brain disorders? To address these questions, we apply a multidisciplinary approach combining experiments on humans and rodents with computational models in several ongoing projects in our laboratory:
The neural circuitry of expectation during decision making tasks.
[PI: J. de la Rocha]
In this project, funded by the European Research Council (ERC-2015-CoG - 683209_PRIORS), we aim to characterize the neural basis of expectation, an instrumental aspect of perception, and its impact on perceptual decisions. In particular, we investigate which areas of the brain process previous experiences to generate predictive biases, which are the algorithms used in this computation and how these algorithms could be implemented by neuronal circuits. We are interested in the extent to which the representation of the sensory environment is modulated by expectation and in particular whether stimulus-evoked neural responses in early sensory areas are affected but expectation signals or by the previous history of stimuli, choices and rewards. To this end, we train rats in two-alternative forced task (2AFC) in which the weight and direction of expectation priors together with the ambiguity of an auditory stimulus are systematically manipulated. We characterize the impact of these priors on behavior while simultaneously record the spiking activity of neurons in the auditory cortex, prefrontal areas and the striatum. We also use pharmacological and optogenetic experiments to determine the role of different brain areas in the various aspects of the task. We combine our experiments with statistical models to quantitatively describe both the behavior and the activity of neural populations. The project is done in collaboration with Alex Hyafil (CRM, Barcelona) and David Robbe (INMED, Marseille).
The dynamics of sensory evidence integration during perceptual decision making
[PI: J. de la Rocha]
During perceptual discrimination task sensory information must be transformed into decision evidence and then temporally integrated in order to yield choices. We are interested in several aspects of this process: what causes the trial-to-trial response variability ubiquitous to this kind of tasks? Is the neural variability found in sensory areas introducing variability (i.e. noise) in the decision process? does the behavioral variability instead mainly reflects the dynamics of latent variables yet to be characterized? How are prior expectations integrated with the stimulus evidence? How can we design a biophysically realistic model that carries out the integration and categorization of evidence? To answer these questions we perform psychophysical experiments in humans and rodents and combined them with the analysis of latent variable dynamic models (e.g. diffusion models) as well as biophysically-inspired neural network models. We also perform electrophysiology recordings during the perceptual task which allow us to characterize the circuit mechanisms underlying the integration process. In particular we are interested in comparing categorization dynamics (such as those obtained using attractor models) with other standard models such as the drift diffusion model. Moreover, we aim to extend this type of models with urgency dynamics and experimentally determine the factors that modulate these urgency signals. Finally we are also interested in the interaction between the process of decision evidence integration and the planning and execution of the motor response (is there a bidirectional interaction? are the two processes carried out in separate circuits? ). The project is done in collaboration with Alex Roxin (CRM, Barcelona), Klaus Wimmer (CRM, Barcelona), Alex Hyafil (CRM, Barcelona) and Tobias Donner (Univ. of Hamburg).
Perceptual alterations in schizophrenia and anti-NMDAR encephalitis
[Led by Daniel Linares, PI: Albert Compte]
There is recently an interest in identifying possible perceptual effects in schizophrenia that could be used as a biomarker of the disease. Taking advantage of our parallel study with patients with anti-NMDAR encephalitis and with schizophrenia, we are testing perceptually these patients to establish (1) the robustness of reported effects in schizophrenic patients, (2) shared effects with anti-NMDAR encephalitis suggesting a common synaptic basis, (3) robust methods to test patients perceptually in the clinic to serve diagnostic or prognostic purposes in the future. This project is done in collaboration with Josep Dalmau (ICREA/IDIBAPS) and Gisela Sugranyes (IDIBAPS).
A, Illustration of the perceptual task. B, Psychometric function model fitted to one example participant. C, Motion sensitivity for small and large stimuli for all participants (scz: patients with schizophrenia, control: healthy controls). The dots show the results for each participant. The boxes indicate the mean and the 95% confidence intervals. D, Suppression index for all participants. E, Lapses for all participants.
The neural circuitry of Working Memory (WM): memory interference and disease-mediated deficits
Working memory is the cognitive function describing the maintenance and processing of information in the brain during brief periods of time. The brain areas involved in this function and the underlying circuit mechanisms are still not entirely understood. In the following interdependent projects we aim to characterize the neural circuit dynamics underlying interference of memories in working memory. We investigate the factors that can limit the accuracy of WM: how do previously stored memoranda or distractors interfere with the current content of WM? Is this interaction a bug or a feature? Can the accuracy of WM only decrease with the delay duration (e.g. diffusion) or there can be other temporal dependencies (e.g. when retrieval occurs at unexpected times)? To what extent is the maintenance of memories decoupled from future motor plans? What mechanisms underlie these interferences? What neural dynamics are associated with these processes? To address these questions we apply a multidisciplinary approach on human participants, clinical populations and rodent models in several ongoing projects in our laboratory:
Working Memory (WM) interference in the healthy primate brain
[PI: Albert Compte]
We are investigating the processes by which distractors and previous memories interfere with currently memorized items in working memory. We run psychophysics, fMRI, EEG, and TMS experiments on healthy human participants, and we analyze monkey electrophysiological recordings from collaborating laboratories (Christos Constantinidis, Julio Martinez-Trujillo, Jacqueline Gottlieb, Bijan Pesaran).
Working Memory (WM) interference in disease
[PI: Albert Compte]
We are investigating working memory in anti-NMDAR encephalitis, a disease mediated by the internalization of NMDARs and characterized by psychotic symptoms in the acute phase and protracted cognitive deficits after hospital discharge. We test patients longitudinally in this recovery phase, and we compare with parallel tests on patients with schizophrenia, for their convergent symptoms and possible shared NMDAR mechanistic substrate. The mechanistic basis of WM interference differences is explored with EEG, fMRI and MRS measurements, and possible links with NMDAR function are established with computational models for further investigation in animal studies. In addition, we include in the study the investigation of some aspects of long-term memory function and sleep, to address their dependency on NMDARs and their possible interaction with WM function. Recently, we have further included one additional clinical population with anti-LGI1 encephalitis, characterized by the immunodepletion of the synaptic protein LGI1 in limbic structures. The project is done in collaboration with Josep Dalmau (ICREA/IDIBAPS) and Pablo Jercog (IDIBAPS).
Detailed neural circuitry: exploring working memory interference with rodent models
[PI: Albert Compte and Jaime de la Rocha]
a, Sketch of the 2AFC delayed response task in head-fixed mice. Subjects are cued by a lateral speaker about the rewarded port. After a variable delay period (from 0 to 10 seconds) the two water spouts move forward instructing the animal to make a response by licking one of the two. Correct responses are rewarded with water and incorrect responses are punished with a time-out. b, Top: Proportion of correct responses responses as a function of delay length (individual mice shown on the left and average shown on the Right; n = 40 mice). Errors can be classified into memory errors caused by forgetting during the delay period and non-memory errors (or lapses) caused by a different mechanisms than memory maintenance as they can be observed at zero delay (Right). Bottom: probability to repeat the previous response as a function of delay length shows no interaction between the repeating tendency and delay, implying that repetitions are not associated with memory errors. c, Example recording using a 64 channel sillicon probe (one six channels shown) in the Anterior Lateral Motor cortex (ALM) Inset shows a schematic of the position of the recorded area. d, Spike raster-gram and PSTHs from an example neuron in correct Right stimulus trials (orange) and Left stimulus trials (blue) with delay D=10 seconds. e, Schematic of the Hidden Markov Model used to explain the behavioral data. The model consists of two modules: a WM module with dynamics describe by a double well attractor model (Left) and a Reinforcement Learning model with choice dynamics governed by a simple RL model that reinforces previous actions.
We have designed a 2AFC auditory delayed-response task and a parametric visuospatial WM task in mice. Animals perform the task in high throughput computer-controlled set-ups that generate very large data sets. We then combine recordings during the task using wide-field Calcium imaging or electrophysiology with a fine statistical characterization of the behavior and its relation to neural activity. In addition, we investigate how WM is altered in a mouse model of anti-NMDAr encephalopathy in order to understand the role of NMDAr during WM and to mechanistically characterize the executive deficits observed in patients during the period of disease recovery. These projects are done in collaboration with Josep Dalmau (ICREA/IDIBAPS) and Pablo Jercog (IDIBAPS). Finally, we study the mechanisms of interference between simultaneous tasks by modeling and analyzing electrophysiology data from a dual-task experiment in mice (collaboration with Chengyu Li, ION Shanghai).