Building sensory and abstract representations

Sensory representations

To make sense of the world around us, we use our sensory organs (our eyes, ears, nose, mouth and skin) to gather information. This information is sent to specialised regions in a part of the brain called the neocortex, which combines these sensory inputs and extracts important features to create internal representations. These sensory representations are what account for our perception of the external world and influence how we act upon it.

At SWC, we are working to understand the mechanisms by which the neocortex generates sensory representations. Our research primarily focuses on the visual areas of the neocortex as we can precisely control visual input in our experiments. 

We use multiple methods including two-photon imaging, whole cell recordings and anatomical approaches to uncover how single neurons and networks of neurons result in representations of external stimuli in the neocortex. We measure how neurons connect to each other and how they form subnetworks that encode sensory features of the environment.

Current research

The Margrie, Hofer and Mrsic-Flogel labs are working to comprehensively dissect the organisation of circuits in the visual cortex with respect to different classes of excitatory projection neurons and inhibitory interneurons. The aim of this research is to integrate these data into a comprehensive model of single neurons and networks during sensory processing and provide the basis for understanding how dynamical modes arise from circuit architecture. This work involves close collaboration with computational researchers at GCNU.

The Mrsic-Flogel lab is working closely with the Znamenskiy lab* at the Francis Crick Institute to determine how object depth is computed from optic flow signals and which cortical areas and projection pathways implement this computation.

The Hofer lab is investigating cortical areas that process visual motion to determine how they distinguish movement in the environment from self-generated optic flow, and is testing how the thalamus contributes to this computation. The Hofer lab is also testing the theory that high-order thalamus acts as a switchboard to prioritise and differentially route visual information to target areas depending on behavioural relevance.

The Margrie lab is collaborating with the Burgess lab* to test how visual information is integrated with head direction signals to generate egocentric and allocentric representations of the visual environment.

The Branco lab is studying how sensory stimuli are analysed and identified as threats by midbrain circuits. In collaboration with the Branco lab, the Hofer lab is investigating how, through thalamic inhibition, these midbrain circuits can be regulated by and coordinated with cortical pathways during visual processing.

*Petr Znamenskiy and Neil Burgess are both part of our affiliates programme at SWC. 

Abstract representations

In addition to sensory representations of the world around us, we also need representations for more abstract concepts such as space and knowledge. 

In 1978, John O’Keefe proposed a theory that the hippocampus and associated areas function as a cognitive map for spatial memory, forming an inner GPS in the brain. This theory of hippocampal function explained the finding of place cells and predicted the discovery of grid and head direction cells.

At SWC, we are working to expand on this theory to further understand how animals navigate. We are also exploring how such representations are used in learning and also how animals represent social information.

Current research

The O’Keefe lab is working towards identifying the mechanisms by which different sensory and internal inputs give rise to place cell formation, and determining the contribution of spatial representations in navigation. 

The team are testing the theory that the hippocampus generates a goal-directed vector to the goal from the animal’s current location, by identifying how this goal vector is represented in place cell activity during navigation on the honeycomb maze. The O’Keefe lab is also working closely with the Branco lab to look at the neuronal activity in an associated area called the retrosplenial cortex during escape-to-shelter behaviour. 

Additionally, the Margrie lab and Clopath lab are working to understand how internal motion and external visual cues are integrated in retrosplenial cortex to enable object-based navigation. 

As part of the affiliates programme at SWC, Neil Burgess is testing predictions of an important recent generalisation of the spatial theories about place cell and grid cell firing. The theory proposes that states and transitions representing arbitrary bodies of knowledge are represented by place and grid cells respectively. Grid cell firing patterns are thought to be eigenvectors of the transition matrix, making precise predictions for their firing patterns in both spatial and non-spatial tasks. 

The Burgess lab is also working with the Margrie lab and Branco lab to investigate the role of retrosplenial cortex circuits in coordinate transformation during spatial memory. They are focusing on escape behaviour and also exploiting new technologies that allow neural measurements during bodily translation or rotation.

Furthermore, the Hofer lab is testing the hypothesis that the thalamus facilitates coordinate transformations between areas and they are investigating the underlying circuit mechanisms.

The Saxe lab is exploring how neural representations change across a distributed network of interconnected brain areas during learning. The team are also investigating the neural representations underlying task switching, multitasking and abstraction.

The Duan lab is exploring how information about other individuals is represented in the brain, as for social animals individuals rarely act in isolation but instead interact with multiple others.

The Isogai and O’Keefe labs are also researching the neural mechanisms by which social information is represented in the brain, specifically the amygdala and hypothalamus, and how this representation is used to select from a repertoire of specific social behaviours. The Isogai lab is studying odour signals, called pheromones, and trying to understand how they give rise to the representation of social information including sex, age, and reproductive and social status, as well as how this representation drives the decision to engage in social interactions such as parenting, aggression and mating.

Generative models, inference and predictions

Sensory and abstract representations are brought together to form generative models of our environment, which are updated with external and internal information to allow us to predict the world around us. At SWC, researchers focus on understanding how these models are represented in neural activity and how the brain learns and stores prior knowledge (‘priors’).

A key assumption of the generative model is that it can be used to influence sensory processing to predict, disambiguate or prioritise sensory input. Previous research at SWC revealed that sensory representations not only reflect the identity of external stimuli but also contain information about stimulus context, behavioural relevance as well as current goals, expectations, actions or past experience. 

Our latest research is revealing how diverse contextual signals are integrated with sensory information, and how such signals emerge through experience and learning.

Current research

The Akrami lab is using behavioural and physiological approaches, along with computational modelling, to study the neural underpinnings of the formation, storage and utilisation of experience-dependent priors. The team focuses on understanding how probabilities are represented in the brain and is using this knowledge to develop new circuit-level models of the computations that underlie formation of priors. They are also testing how statistical learning of regularities in sensory inputs or abstract representations can be stored in neural circuits.

The Saxe lab is exploring how prior knowledge impacts learning. The team are investigating the dynamics of learning new tasks with a neural network that already has pre-existing knowledge.

The Hofer, Margrie and Mrsic-Flogel labs are working to understand the underlying circuit mechanisms in visually-guided behaviour. They use anatomical tracing to identify candidate regions and then measure and disrupt feedback from identified regions during visually-guided behaviours. Furthermore they are testing how these inputs influence the local network activity in relation to different excitatory and inhibitory cell classes in visual cortex, and are working with GCNU to build a model of how long-range predictive signals influence local computations during sensory processing.

The Murray and Mrsic-Flogel labs are investigating how animals adapt to deviations from expected sensory feedback. They work on the cerebellum, which is known to store the internal model of sensorimotor transformations required for movement and also learns to control the timing of actions in response to predictable changes in sensory input. They have shown that cerebellar output is crucial for generating activity dynamics in premotor cortex during planning and they are now investigating the contributions of the cerebellum to sensorimotor representations in the neocortex as animals adapt to deviations from expected sensory feedback.

The Erlich lab is working closely with the Duan lab to understand social cognition and how the brain predicts actions of other agents in the world. While technical limitations have meant that systems neuroscience has traditionally focused on studying individual animals, the Erlich and Duan labs are making use of recent advances in computer vision to track multiple animals in real-time and extract detailed information about their behaviour. 

Additionally, the Erlich lab is working to understand how information that comes into our nervous system in a sensory reference frame is used to update a stable world-model that enables goal-direction behaviour.

Decision-making and action selection

Reaching a decision requires interpreting information about the current environment and comparing available actions. But how does the brain transform information about the world into purposeful actions? 

At SWC, we aim to establish a theoretical framework for how decisions are made and map this framework onto computations in neural circuits.

In some cases, evolution has resulted in sensory stimuli having intrinsic values that directly select appropriate actions. For example, the smell of pups in rodents drives mothers to initiate maternal behaviours. Similarly, when a dark shadow appears overhead, rodents initiate appropriate defensive behaviours such as escaping or freezing. 

In contrast to these innate behaviours, the brain can also learn the value of stimuli and resultant actions, thereby allowing it to generate adaptive behavioural responses through experience. At SWC we are testing the hypothesis that the basal ganglia are the key structures supporting this process via reinforcement leaning, enabling the coupling of sensory context to specific actions.

We are also exploring how the brain guides behavioural choices under conditions of uncertainty, such as when only partial sensory evidence is available. We know that such decisions must rely on combining prior knowledge with existing sensory input, but where and how this is achieved in the brain is not currently well understood.

Current research

The Branco lab are working to determine the biophysical and circuit mechanisms which enable the integration of sensory evidence of threat into escape decisions. They are also looking to
describe how knowledge of the environment and the internal state modulate the escape response. 

As context and/or prior experience can shape the decision to engage in defensive behaviour, the Branco, Hofer, Stephenson-Jones and Margrie labs are also working to determine the contribution of inhibitory circuits in the thalamus and basal ganglia in adaptive regulation of escape. 

The Isogai lab is studying how the amygdala orchestrates specific innate social behaviours. Furthermore, the Duan and Erlich labs are studying the neural mechanisms of social decisions by training animals in ethologically relevant foraging tasks that can be quantified in a game-theoretic framework. Their aim is to discover where and how information about the other player is represented in the brain and how this information is integrated with reward information to guide choice behaviours in a social context.

The Duan lab is researching risky decisions and “when does it pay to gamble”. The team are using mouse models of risk decision making to study how interconnected cortico-subcortical networks allow the brain to compute expected value, to compare the values of different options, to select an option and transform that selection into an action. 

Furthermore, the Erlich lab is working to understand the neurobiology of economic decisions, such as choosing between £10 today and £20 in a year. The team are training humans in non-verbal tasks, to bridge the gap between human and animal studies.

The Stephenson-Jones and Mrsic-Flogel labs are collaborating to determine how sensory representations are used to drive specific, learned actions in the basal ganglia, and to identify circuit mechanisms for learning and implementation of this sensorimotor transformation, using chronic recordings and activity manipulations in different classes of cortico-striatal projection neurons.

Furthermore, a collaboration between the Burgess*, Stephenson-Jones and Branco labs is testing models of hippocampal-striatal interactions linking neural responses to navigation and planning and also how ‘model-based’ and ‘model-free’ reinforcement learning combine. 

The Akrami and Mrsic-Flogel labs are also working closely with the Burgess lab and DeepMind to reveal the mechanisms by which prior knowledge and sensory evidence are integrated into the process of selecting appropriate actions. 

The Akrami lab are working to uncover the circuit mechanisms underlying the joint representation of the interaction of current sensory evidence and recent sensory history by the parietal cortex. 

The Mrsic-Flogel lab are testing the hypothesis that premotor cortex integrates decision-relevant information from sensory and parietal cortical areas to initiate action selection, using novel decision-making behavioural paradigms with targeted activity measurements and perturbations of identified neuronal populations in candidate brain regions. 

The teams are also collaborating with GCNU to generate a model that explains animals’ decision process, by validating against trial-by-trial performance in the task on one hand and by testing how accurately it predicts the decision-related neural activity on the other.

*Neil Burgess is part of our affiliates programme at SWC. 
 

Working memory

The ability to store information for short periods of time, known as working memory, is vital for many aspects of cognition and behaviour, especially decision-making. Working memory involves the representation of relevant information by combining current sensory evidence and prior knowledge. Exactly how the brain generates, maintains and updates working memory is poorly understood.

Several groups at SWC are tackling different aspects of working memory and its circuit implementation using innate or tailored learned behavioural tasks.

Although the properties of persistent activity that support working memory may differ between brain areas, we are looking to determine whether the circuit mechanisms by which it is generated share common principles. We aim to generate biologically realistic models that capture experimental data and to explain persistent activity dynamics during behaviour. 

Current research

The Akrami lab is working to reveal the neural basis of sensory working memory formation, its maintenance, update and recall, in particular in conjunction with its role in cognitive control. The Hofer and Akrami labs are collaborating to understand the role of thalamocortical circuits in this process.

The Mrsic-Flogel lab is testing the idea that reciprocal interactions between cerebellum and frontal cortex, and between parietal and frontal cortex, interact to maintain working memory of motor plans and behavioural context. The Akrami and Mrsic-Flogel labs are also seeking to determine how different brain regions are recruited to support various timescales of working memory.

The Branco lab is seeking to determine how working memory of shelter location is implemented, updated and maintained as animals explore their environment. The team are also collaborating with the Margrie and Burgess* labs to identify the mechanisms by which working memory guides orienting, by studying interactions between retrosplenial cortex and superior colliculus. 

The O’Keefe lab is studying the role of persistent activity in the amygdala in maintaining the representation of conspecifics during social behaviour, and how it may be generated via reciprocal communication between ventral hippocampus and amygdala circuits.

The Erlich lab is also studying working memory to understand how animals manage competing goals and drives as well processing a barrage of sensory input.

*Neil Burgess is part of our affiliates programme at SWC. 

Expectation, evaluation and motivation

One of the remarkable features of the mammalian brain is its ability to generate adaptive behaviour. This involves learning the value of actions in response to sensory or internal information.

These goal-directed actions involve several different processes including:

• Motivation in pursuit of a goal
• Selection and sequencing of actions
• Execution with the appropriate vigour to generate a coordinated behavioural response

Furthermore, the actual outcome can be compared with the expected outcome to evaluate whether the whole process was successful, and thus drive learning.

At SWC, we are working to develop a comprehensive theory of how neural circuits in the basal ganglia implement these processes.

Current research

Our hypothesis is that each feature of a goal-directed action is controlled by a specific population of basal ganglia output neurons, each regulating either motivation, selection, sequencing, vigour, expectation or evaluation. 

The Stephenson-Jones lab has already identified two different basal ganglia output populations that are responsible for two distinct non-motor aspects of goal-directed behaviour: motivation and evaluating action outcomes.

In collaboration with the Clopath lab, the Stephenson-Jones lab is now working to systematically map and manipulate each population of basal ganglia output neurons, to understand how each contributes to component computations of goal-directed behaviour and build a realistic model that captures them.

The Duan lab is also studying expectation in the context of risk. The team are uncovering the biological basis of economic choices under risk, specifically how the brain computes expected value, compares values of different options, selects an option, and transforms that selection into an action. 

The Duan lab is also working closely with Erlich lab to study how animals’ risk preferences adapt to environmental variability. The long term goal of research in the Erlich lab is to understand how chronic stress influences economic preferences and the link between genes, neural circuits and economic decisions.

Adaptive motor control

The world around us is continually changing and so our motor output must be constantly monitored and adapted in response to changes in our environment, internal state or behavioural goal. This flexibility requires the coordination of sensorimotor circuits throughout the nervous system.

At SWC, research focuses on how the nervous system generates adaptive and flexible movement. We study the neural circuits of the brainstem, which link the learning structures of the neocortex, basal ganglia, and cerebellum on the one hand, and motor execution circuits of the spinal cord on the other.

Our aim is to reveal how brainstem circuits integrate multiple streams of information so we can understand how the nervous system as a whole generates flexible motor actions.

Current research

The Murray lab have developed a novel behavioural assay in which mice receive postural perturbations and must initiate a corrective motor response in a context-dependent manner. 

In collaboration with the Mrsic-Flogel lab the team are developing a conceptual framework for understanding the role of the cerebellum and motor neocortex in adaptive motor control. They suggest that the neocortex implements a robust corrective response to unanticipated perturbations of sensory feedback while the cerebellum learns to provide a robust corrective response to anticipated perturbations. 

They are currently testing this theory using new closed-loop virtual reality and augmented reality behavioural assays within which they can introduce perturbations of sensory feedback – similar to the challenges faced by rodents in their natural habitat – which require the animals to rapidly adjust their motor strategy. 

The teams are looking to understand the underlying mechanism by measuring and manipulating neural circuits across the two structures. In addition, in a collaboration with the Clopath lab, the data will be used to build a model that explains the computations performed by cerebellum and cortex during adaptive motor control.