Periodic Reporting for period 3 - KiloNeurons (Population Codes for Space in the Mammalian Cortex)
Okres sprawozdawczy: 2024-02-01 do 2025-07-31
The key objective of this project is to identify the fundamental principles of neural computation in the space-coding circuits of the mammalian cortex. The coding is likely to take place in large, distributed and intermixed cell populations. To identify the coding algorithms in these circuits, we need population-wide activity measurements at single-cell resolution, as well as theoretical models to interpret the data. In KILONEURONS, we combine experiments and theory to enable a paradigmatic shift from single-cell to population analysis for a prototypical high-level cortical system, the navigation system of the mammalian medial entorhinal-hippocampal region. In this system, spatial firing correlates of individual cells are so evident that they have been given simple, descriptive names – such as place cells, grid cells, and head direction cells. The wealth of information on the phenomenology of these cells, and the existence of theoretical frameworks that offer strong predictions on their population-wide activity patterns, renders the system perfect for population-level analyses of cortical computation. We are introducing experimental tools to obtain the amount and specificity of multi-neuron data required to decipher neural population codes in freely navigating rodents. Guided initially by theory on attractor network dynamics, we are identifying regularities in firing and connectivity patterns of thousands of simultaneously monitored neurons and use the data to test, refine and develop theoretical models. This is extended to less-understood high-end systems such as lateral entorhinal cortex, where computational operations have remained elusive due to the lack of similar single-cell correlates. The project is transformative in that it is uncovering fundamental and general mechanisms of high-end cortical population coding in mammals.
The project pioneers theory-based neural population analysis of the kind needed to understand the mechanics of cognition in the brain, a goal that must be reached for the community to be able to target dysfunctions in neural function. A better understanding of the fundamental principles of brain computation is necessary if we are to understand and treat complex brain disorders. The studies have particular relevance for Alzheimer’s disease, which is associated with severe destruction of the entorhinal cortex during the earliest stages. With a renewed arsenal of methodologies, we shall be able to determine the mechanisms of normal function as well as dysfunction in entorhinal networks and microcircuits.
The central objective of this project is to test, with novel methods for population recording and population analysis, the neural-circuit predictions of continuous attractor network models for spatial coding, which belong to the most influential theoretical frameworks in neuroscience. With access to activity and connectivity of thousands of neurons in medial entorhinal cortex, we expect to directly visualize the structure and dynamics of the joint activity of grid cells and other space-coding cells in the brain's positioning system.
The project pioneers theory-based neural population analysis of the kind needed to understand the mechanics of cognition in the brain, a goal that must be reached for the community to be able to target dysfunctions in neural function. A better understanding of the fundamental principles of brain computation is necessary if we are to understand and treat complex brain disorders. The studies have particular relevance for Alzheimer’s disease, which is associated with severe destruction of the entorhinal cortex during the earliest stages. With a renewed arsenal of methodologies, we shall be able to determine the mechanisms of normal function as well as dysfunction in entorhinal networks and microcircuits.
The central objective of this project is to test, with novel methods for population recording and population analysis, the neural-circuit predictions of continuous attractor network models for spatial coding, which belong to the most influential theoretical frameworks in neuroscience. With access to activity and connectivity of thousands of neurons in medial entorhinal cortex, we expect to directly visualize the structure and dynamics of the joint activity of grid cells and other space-coding cells in the brain's positioning system.
The project has resulted in several major research developments. They have all advanced our understanding of how information is encoded in large neural populations during experience. Our windows into these computations have been the dynamics of grid cells and other entorhinal neurons during navigation as well as with minimal sensory stimulation during rest and sleep. We have made observations that significantly impact how we think computation is enabled in networks of interconnected cortical neurons. We shall highlight the most important published findings.
Our most recent breakthrough is the discovery of a new form of dynamics in grid cells and other cells of the entorhinal-hippocampal spatial coding system, reported in Nature in February (Vollan et al., 2025). We show that populations of grid cells and place cells encode a position signal that within a 125-ms theta cycle sweeps linearly outwards from the animal’s location into the ambient environment, alternating stereotypically between left and right directions. The decoded sweeps are time-locked to the local theta rhythm. Each theta cycle has one sweep. The sweeps alternate their direction between left and right on successive theta cycles. A striking observation was that the sweeps can extend into never-visited space inaccessible to the animal, and that they persist during REM sleep, all pointing to an intrinsic “hardwired” network mechanism for the phenomenon rather than one determined by specific sensory inputs. The sweeps were accompanied by, and aligned with, a similarly alternating directional signal in a discrete population of direction-tuned cells with putative connections to the grid cells. Sweeps in hippocampal place cells were delayed compared to sweeps in grid cells, pointing to an entorhinal origin. Using simulations in a simple computational model, we show that sweeps represent an efficient mechanism for scanning locations in the ambient environment during navigation. We show that sweep direction can be explained by an algorithm that maximizes cumulative coverage of surrounding space. These findings collectively point to an entirely new form of dynamics, hardwired into the circuit, by which activity systematically scans the surrounding state space, irrespective of whether animals move in a physical environment or not. The Nature paper mostly reports the experimental findings, but we are exploring at this moment whether and how they can be followed up in computational modelling to determine the underlying mechanisms.
In a second paper published during the reporting period, also in Nature (Gonzalo Cogno et al., 2024), we showed that entorhinal cells can self-organize into minute-long periodic sequences that may serve as a scaffold for on-the-fly formation of new sequences. Using both Neuropixels and two-photon calcium imaging, we found that during running in head-fixed mice, neural activity in the medial entorhinal cortex self-organized into oscillations at periods often longer than a minute that involved nearly the entire cell population. During each oscillatory cycle, the neural activity progressed in a stereotyped sequence. Sequences were periodic, with no interruption between oscillatory cycles. The sequences sometimes advanced uninterruptedly for tens of minutes, transcending epochs of locomotion and immobility. The ultraslow oscillatory sequences may have the potential to couple neurons and circuits across extended time scales and may serve as a template for new sequence formation downstream in hippocampus and other regions during navigation and episodic memory formation. Since the publication of this paper in 2024, we have been working on identifying the similar sequences during more natural behaviors, and we are seeking to understand what types of cells participate in the oscillations and sequences.
Our most recent breakthrough is the discovery of a new form of dynamics in grid cells and other cells of the entorhinal-hippocampal spatial coding system, reported in Nature in February (Vollan et al., 2025). We show that populations of grid cells and place cells encode a position signal that within a 125-ms theta cycle sweeps linearly outwards from the animal’s location into the ambient environment, alternating stereotypically between left and right directions. The decoded sweeps are time-locked to the local theta rhythm. Each theta cycle has one sweep. The sweeps alternate their direction between left and right on successive theta cycles. A striking observation was that the sweeps can extend into never-visited space inaccessible to the animal, and that they persist during REM sleep, all pointing to an intrinsic “hardwired” network mechanism for the phenomenon rather than one determined by specific sensory inputs. The sweeps were accompanied by, and aligned with, a similarly alternating directional signal in a discrete population of direction-tuned cells with putative connections to the grid cells. Sweeps in hippocampal place cells were delayed compared to sweeps in grid cells, pointing to an entorhinal origin. Using simulations in a simple computational model, we show that sweeps represent an efficient mechanism for scanning locations in the ambient environment during navigation. We show that sweep direction can be explained by an algorithm that maximizes cumulative coverage of surrounding space. These findings collectively point to an entirely new form of dynamics, hardwired into the circuit, by which activity systematically scans the surrounding state space, irrespective of whether animals move in a physical environment or not. The Nature paper mostly reports the experimental findings, but we are exploring at this moment whether and how they can be followed up in computational modelling to determine the underlying mechanisms.
In a second paper published during the reporting period, also in Nature (Gonzalo Cogno et al., 2024), we showed that entorhinal cells can self-organize into minute-long periodic sequences that may serve as a scaffold for on-the-fly formation of new sequences. Using both Neuropixels and two-photon calcium imaging, we found that during running in head-fixed mice, neural activity in the medial entorhinal cortex self-organized into oscillations at periods often longer than a minute that involved nearly the entire cell population. During each oscillatory cycle, the neural activity progressed in a stereotyped sequence. Sequences were periodic, with no interruption between oscillatory cycles. The sequences sometimes advanced uninterruptedly for tens of minutes, transcending epochs of locomotion and immobility. The ultraslow oscillatory sequences may have the potential to couple neurons and circuits across extended time scales and may serve as a template for new sequence formation downstream in hippocampus and other regions during navigation and episodic memory formation. Since the publication of this paper in 2024, we have been working on identifying the similar sequences during more natural behaviors, and we are seeking to understand what types of cells participate in the oscillations and sequences.
Advances in science tend to occur in a stepwise fashion rather than linearly. In neuroscience paradigmatic shifts are taking place at this time due to the invention of tools for recording and manipulating activity of thousands of neurons at single-cell resolution. This opens the doors to neural computation in circuits, enabling the discovery of the fundamental codes for cognitive activity. In this paradigmatic shift, the KILONEURONS project is playing a pioneering role. We have participated in the development of new Neuropixels silicon probes for large-scale simultaneous recording from thousands of neurons in behaving animals and we have developed the first miniature 2-photon microscope that is small and flexible enough to image neural activity at single-cell resolution with no disturbance of behavior in freely moving mice (a smaller and more powerful version has been developed in our lab recently and is now in use). We have used these advances to test key predictions of neural network theories for spatial coding, finding that neural activity in grid cells operates on a periodic toroidal manifold, in exact agreement with certain continuous attractor network models for spatial coding in grid cells. We have also shown that dynamics on the toroidal manifold follows a stereotypic pattern, consisting of sweeps that alternate in a rigid left-right pattern across successive cycles of the local theta oscillation. Both findings were published in Nature. We are currently extending the testing to include connectivity in the network, establishing whether grid-cell networks exhibit preferential coupling between cells of similar function, and we are testing ideas about how grid cells of different modules might be coordinated through the existence of cells tuned to more than a single module.