Neural manifold analysis of brain circuit dynamics in health and disease.

Recent developments in experimental neuroscience make it possible to simultaneously record the activity of thousands of neurons. However, the development of analysis approaches for such large-scale neural recordings have been slower than those applicable to single-cell experiments. One approach that has gained recent popularity is neural manifold learning. This approach takes advantage of the fact that often, even though neural datasets may be very high dimensional, the dynamics of neural activity tends to traverse a much lower-dimensional space. The topological structures formed by these low-dimensional neural subspaces are referred to as "neural manifolds", and may potentially provide insight linking neural circuit dynamics with cognitive function and behavioral performance. In this paper we review a number of linear and non-linear approaches to neural manifold learning, including principal component analysis (PCA), multi-dimensional scaling (MDS), Isomap, locally linear embedding (LLE), Laplacian eigenmaps (LEM), t-SNE, and uniform manifold approximation and projection (UMAP). We outline these methods under a common mathematical nomenclature, and compare their advantages and disadvantages with respect to their use for neural data analysis. We apply them to a number of datasets from published literature, comparing the manifolds that result from their application to hippocampal place cells, motor cortical neurons during a reaching task, and prefrontal cortical neurons during a multi-behavior task. We find that in many circumstances linear algorithms produce similar results to non-linear methods, although in particular cases where the behavioral complexity is greater, non-linear methods tend to find lower-dimensional manifolds, at the possible expense of interpretability. We demonstrate that these methods are applicable to the study of neurological disorders through simulation of a mouse model of Alzheimer's Disease, and speculate that neural manifold analysis may help us to understand the circuit-level consequences of molecular and cellular neuropathology.

Adult-born dentate granule cells promote hippocampal population sparsity.

The dentate gyrus (DG) gates neocortical information flow to the hippocampus. Intriguingly, the DG also produces adult-born dentate granule cells (abDGCs) throughout the lifespan, but their contribution to downstream firing dynamics remains unclear. Here, we show that abDGCs promote sparser hippocampal population spiking during mnemonic processing of novel stimuli. By combining triple-(DG-CA3-CA1) ensemble recordings and optogenetic interventions in behaving mice, we show that abDGCs constitute a subset of high-firing-rate neurons with enhanced activity responses to novelty and strong modulation by theta oscillations. Selectively activating abDGCs in their 4-7-week post-birth period increases sparsity of hippocampal population patterns, whereas suppressing abDGCs reduces this sparsity, increases principal cell firing rates and impairs novel object recognition with reduced dimensionality of the network firing structure, without affecting single-neuron spatial representations. We propose that adult-born granule cells transiently support sparser hippocampal population activity structure for higher-dimensional responses relevant to effective mnemonic information processing.

Acute or chronic pulmonary emphysema? Or both?-A contribution to the diagnosis of death due to violent asphyxiation in cases with pre-existing chronic emphysema.

The diagnosis of death due to violent asphyxiation may be challenging if external injuries are missing, and a typical acute emphysema (AE) "disappears" in pre-existing chronic emphysema (CE). Eighty-four autopsy cases were systematically investigated to identify a (histo-) morphological or immunohistochemical marker combination that enables the diagnosis of violent asphyxiation in cases with a pre-existing CE ("AE in CE"). The cases comprised four diagnostic groups, namely "AE", "CE", "acute and chronic emphysema (AE + CE)", and "no emphysema (NE)". Samples from all pulmonary lobes were investigated by conventional histological methods as well as with the immunohistochemical markers Aquaporin 5 (AQP-5) and Surfactant protein A1 (SP-A). Particular attention was paid to alveolar septum ends ("dead-ends") suspected as rupture spots, which were additionally analyzed by transmission electron microscopy. The findings in the four diagnostic groups were compared using multivariate analysis and 1-way ANOVA analysis. All morphological findings were found in all four groups. Based on histological and macroscopic findings, a multivariate analysis was able to predict the correct diagnosis "AE + CE" with a probability of 50%, and the diagnoses "AE" and "CE" with a probability of 86% each. Three types of "dead-ends" could be differentiated. One type ("fringed ends") was observed significantly more frequently in AE. The immunohistochemical markers AQP-5 and SP-A did not show significant differences among the examined groups. Though a reliable identification of AE in CE could not be achieved using the examined parameters, our findings suggest that considering many different findings from the macroscopical, histomorphological, and molecular level by multivariate analysis is an approach that should be followed.

Place Cells in Head-Fixed Mice Navigating a Floating Real-World Environment.

The hippocampal place cell system in rodents has provided a major paradigm for the scientific investigation of memory function and dysfunction. Place cells have been observed in area CA1 of the hippocampus of both freely moving animals, and of head-fixed animals navigating in virtual reality environments. However, spatial coding in virtual reality preparations has been observed to be impaired. Here we show that the use of a real-world environment system for head-fixed mice, consisting of an air-floating track with proximal cues, provides some advantages over virtual reality systems for the study of spatial memory. We imaged the hippocampus of head-fixed mice injected with the genetically encoded calcium indicator GCaMP6s while they navigated circularly constrained or open environments on the floating platform. We observed consistent place tuning in a substantial fraction of cells despite the absence of distal visual cues. Place fields remapped when animals entered a different environment. When animals re-entered the same environment, place fields typically remapped over a time period of multiple days, faster than in freely moving preparations, but comparable with virtual reality. Spatial information rates were within the range observed in freely moving mice. Manifold analysis indicated that spatial information could be extracted from a low-dimensional subspace of the neural population dynamics. This is the first demonstration of place cells in head-fixed mice navigating on an air-lifted real-world platform, validating its use for the study of brain circuits involved in memory and affected by neurodegenerative disorders.

Integrating new memories into the hippocampal network activity space.

By investigating the topology of neuronal co-activity, we found that mnemonic information spans multiple operational axes in the mouse hippocampus network. High-activity principal cells form the core of each memory along a first axis, segregating spatial contexts and novelty. Low-activity cells join co-activity motifs across behavioral events and enable their crosstalk along two other axes. This reveals an organizational principle for continuous integration and interaction of hippocampal memories.

Neural codes - Necessary but not sufficient for understanding brain function.

Brains are information processing systems whose operational principles ultimately cannot be understood without resource to information theory. We suggest that understanding how external signals are represented in the brain is a necessary step towards employing further engineering tools (such as control theory) to understand the information processing performed by brain circuits during behaviour.

Encoding of locomotion kinematics in the mouse cerebellum.

The cerebellum is involved in coordinating motor behaviour, but how the cerebellar network regulates locomotion is still not well understood. We characterised the activity of putative cerebellar Purkinje cells, Golgi cells and mossy fibres in awake mice engaged in an active locomotion task, using high-density silicon electrode arrays. Analysis of the activity of over 300 neurons in response to locomotion revealed that the majority of cells (53%) were significantly modulated by phase of the stepping cycle. However, in contrast to studies involving passive locomotion on a treadmill, we found that a high proportion of cells (45%) were tuned to the speed of locomotion, and 19% were tuned to yaw movements. The activity of neurons in the cerebellar vermis provided more information about future speed of locomotion than about past or present speed, suggesting a motor, rather than purely sensory, role. We were able to accurately decode the speed of locomotion with a simple linear algorithm, with only a relatively small number of well-chosen cells needed, irrespective of cell class. Our observations suggest that behavioural state modulates cerebellar sensorimotor integration, and advocate a role for the cerebellar vermis in control of high-level locomotor kinematic parameters such as speed and yaw.