The application of artificial intelligence to biology and neuroscience.

Over the last decade, the artificial intelligence (AI) has undergone a revolution that is poised to transform the economy, society, and science. The pace of progress is staggering, and problems that seemed intractable just a few years ago have now been solved. The intersection between neuroscience and AI is particularly exciting.

High-throughput sequencing of single neuron projections reveals spatial organization in the olfactory cortex.

In most sensory modalities, neuronal connectivity reflects behaviorally relevant stimulus features, such as spatial location, orientation, and sound frequency. By contrast, the prevailing view in the olfactory cortex, based on the reconstruction of dozens of neurons, is that connectivity is random. Here, we used high-throughput sequencing-based neuroanatomical techniques to analyze the projections of 5,309 mouse olfactory bulb and 30,433 piriform cortex output neurons at single-cell resolution. Surprisingly, statistical analysis of this much larger dataset revealed that the olfactory cortex connectivity is spatially structured. Single olfactory bulb neurons targeting a particular location along the anterior-posterior axis of piriform cortex also project to matched, functionally distinct, extra-piriform targets. Moreover, single neurons from the targeted piriform locus also project to the same matched extra-piriform targets, forming triadic circuit motifs. Thus, as in other sensory modalities, olfactory information is routed at early stages of processing to functionally diverse targets in a coordinated manner.

BRICseq Bridges Brain-wide Interregional Connectivity to Neural Activity and Gene Expression in Single Animals.

Comprehensive analysis of neuronal networks requires brain-wide measurement of connectivity, activity, and gene expression. Although high-throughput methods are available for mapping brain-wide activity and transcriptomes, comparable methods for mapping region-to-region connectivity remain slow and expensive because they require averaging across hundreds of brains. Here we describe BRICseq (brain-wide individual animal connectome sequencing), which leverages DNA barcoding and sequencing to map connectivity from single individuals in a few weeks and at low cost. Applying BRICseq to the mouse neocortex, we find that region-to-region connectivity provides a simple bridge relating transcriptome to activity: the spatial expression patterns of a few genes predict region-to-region connectivity, and connectivity predicts activity correlations. We also exploited BRICseq to map the mutant BTBR mouse brain, which lacks a corpus callosum, and recapitulated its known connectopathies. BRICseq allows individual laboratories to compare how age, sex, environment, genetics, and species affect neuronal wiring and to integrate these with functional activity and gene expression.

High-Throughput Mapping of Long-Range Neuronal Projection Using In Situ Sequencing.

Understanding neural circuits requires deciphering interactions among myriad cell types defined by spatial organization, connectivity, gene expression, and other properties. Resolving these cell types requires both single-neuron resolution and high throughput, a challenging combination with conventional methods. Here, we introduce barcoded anatomy resolved by sequencing (BARseq), a multiplexed method based on RNA barcoding for mapping projections of thousands of spatially resolved neurons in a single brain and relating those projections to other properties such as gene or Cre expression. Mapping the projections to 11 areas of 3,579 neurons in mouse auditory cortex using BARseq confirmed the laminar organization of the three top classes (intratelencephalic [IT], pyramidal tract-like [PT-like], and corticothalamic [CT]) of projection neurons. In depth analysis uncovered a projection type restricted almost exclusively to transcriptionally defined subtypes of IT neurons. By bridging anatomical and transcriptomic approaches at cellular resolution with high throughput, BARseq can potentially uncover the organizing principles underlying the structure and formation of neural circuits.

Whole-cortex in situ sequencing reveals input-dependent area identity.

The cerebral cortex is composed of neuronal types with diverse gene expression that are organized into specialized cortical areas. These areas, each with characteristic cytoarchitecture1,2, connectivity3,4 and neuronal activity5,6, are wired into modular networks3,4,7. However, it remains unclear whether these spatial organizations are reflected in neuronal transcriptomic signatures and how such signatures are established in development. Here we used BARseq, a high-throughput in situ sequencing technique, to interrogate the expression of 104 cell-type marker genes in 10.3 million cells, including 4,194,658 cortical neurons over nine mouse forebrain hemispheres, at cellular resolution. De novo clustering of gene expression in single neurons revealed transcriptomic types consistent with previous single-cell RNA sequencing studies8,9. The composition of transcriptomic types is highly predictive of cortical area identity. Moreover, areas with similar compositions of transcriptomic types, which we defined as cortical modules, overlap with areas that are highly connected, suggesting that the same modular organization is reflected in both transcriptomic signatures and connectivity. To explore how the transcriptomic profiles of cortical neurons depend on development, we assessed cell-type distributions after neonatal binocular enucleation. Notably, binocular enucleation caused the shifting of the cell-type compositional profiles of visual areas towards neighbouring cortical areas within the same module, suggesting that peripheral inputs sharpen the distinct transcriptomic identities of areas within cortical modules. Enabled by the high throughput, low cost and reproducibility of BARseq, our study provides a proof of principle for the use of large-scale in situ sequencing to both reveal brain-wide molecular architecture and understand its development.

Encoding innate ability through a genomic bottleneck.

Animals are born with extensive innate behavioral capabilities, which arise from neural circuits encoded in the genome. However, the information capacity of the genome is orders of magnitude smaller than that needed to specify the connectivity of an arbitrary brain circuit, indicating that the rules encoding circuit formation must fit through a "genomic bottleneck" as they pass from one generation to the next. Here, we formulate the problem of innate behavioral capacity in the context of artificial neural networks in terms of lossy compression of the weight matrix. We find that several standard network architectures can be compressed by several orders of magnitude, yielding pretraining performance that can approach that of the fully trained network. Interestingly, for complex but not for simple test problems, the genomic bottleneck algorithm also captures essential features of the circuit, leading to enhanced transfer learning to novel tasks and datasets. Our results suggest that compressing a neural circuit through the genomic bottleneck serves as a regularizer, enabling evolution to select simple circuits that can be readily adapted to important real-world tasks. The genomic bottleneck also suggests how innate priors can complement conventional approaches to learning in designing algorithms for AI.

Neurons in the primary visual cortex of freely moving rats encode both sensory and non-sensory task variables.

Neurons in primary visual cortex (area V1) are strongly driven by both sensory stimuli and non-sensory events. However, although the representation of sensory stimuli has been well characterized, much less is known about the representation of non-sensory events. Here, we characterize the specificity and organization of non-sensory representations in rat V1 during a freely moving visual decision task. We find that single neurons encode diverse combinations of task features simultaneously and across task epochs. Despite heterogeneity at the level of single neuron response patterns, both visual and nonvisual task variables could be reliably decoded from small neural populations (5 to 40 units) throughout a trial. Interestingly, in animals trained to make an auditory decision following passive observation of a visual stimulus, some but not all task features could also be decoded from V1 activity. Our results support the view that even in V1-the earliest stage of the cortical hierarchy-bottom-up sensory information may be combined with top-down non-sensory information in a task-dependent manner.

Neuroscience Needs Network Science.

The brain is a complex system comprising a myriad of interacting neurons, posing significant challenges in understanding its structure, function, and dynamics. Network science has emerged as a powerful tool for studying such interconnected systems, offering a framework for integrating multiscale data and complexity. To date, network methods have significantly advanced functional imaging studies of the human brain and have facilitated the development of control theory-based applications for directing brain activity. Here, we discuss emerging frontiers for network neuroscience in the brain atlas era, addressing the challenges and opportunities in integrating multiple data streams for understanding the neural transitions from development to healthy function to disease. We underscore the importance of fostering interdisciplinary opportunities through workshops, conferences, and funding initiatives, such as supporting students and postdoctoral fellows with interests in both disciplines. By bringing together the network science and neuroscience communities, we can develop novel network-based methods tailored to neural circuits, paving the way toward a deeper understanding of the brain and its functions, as well as offering new challenges for network science.

Assessing the replicability of spatial gene expression using atlas data from the adult mouse brain.

High-throughput, spatially resolved gene expression techniques are poised to be transformative across biology by overcoming a central limitation in single-cell biology: the lack of information on relationships that organize the cells into the functional groupings characteristic of tissues in complex multicellular organisms. Spatial expression is particularly interesting in the mammalian brain, which has a highly defined structure, strong spatial constraint in its organization, and detailed multimodal phenotypes for cells and ensembles of cells that can be linked to mesoscale properties such as projection patterns, and from there, to circuits generating behavior. However, as with any type of expression data, cross-dataset benchmarking of spatial data is a crucial first step. Here, we assess the replicability, with reference to canonical brain subdivisions, between the Allen Institute's in situ hybridization data from the adult mouse brain (Allen Brain Atlas (ABA)) and a similar dataset collected using spatial transcriptomics (ST). With the advent of tractable spatial techniques, for the first time, we are able to benchmark the Allen Institute's whole-brain, whole-transcriptome spatial expression dataset with a second independent dataset that similarly spans the whole brain and transcriptome. We use regularized linear regression (LASSO), linear regression, and correlation-based feature selection in a supervised learning framework to classify expression samples relative to their assayed location. We show that Allen Reference Atlas labels are classifiable using transcription in both data sets, but that performance is higher in the ABA than in ST. Furthermore, models trained in one dataset and tested in the opposite dataset do not reproduce classification performance bidirectionally. While an identifying expression profile can be found for a given brain area, it does not generalize to the opposite dataset. In general, we found that canonical brain area labels are classifiable in gene expression space within dataset and that our observed performance is not merely reflecting physical distance in the brain. However, we also show that cross-platform classification is not robust. Emerging spatial datasets from the mouse brain will allow further characterization of cross-dataset replicability ultimately providing a valuable reference set for understanding the cell biology of the brain.

The logic of single-cell projections from visual cortex.

Neocortical areas communicate through extensive axonal projections, but the logic of information transfer remains poorly understood, because the projections of individual neurons have not been systematically characterized. It is not known whether individual neurons send projections only to single cortical areas or distribute signals across multiple targets. Here we determine the projection patterns of 591 individual neurons in the mouse primary visual cortex using whole-brain fluorescence-based axonal tracing and high-throughput DNA sequencing of genetically barcoded neurons (MAPseq). Projections were highly diverse and divergent, collectively targeting at least 18 cortical and subcortical areas. Most neurons targeted multiple cortical areas, often in non-random combinations, suggesting that sub-classes of intracortical projection neurons exist. Our results indicate that the dominant mode of intracortical information transfer is not based on 'one neuron-one target area' mapping. Instead, signals carried by individual cortical neurons are shared across subsets of target areas, and thus concurrently contribute to multiple functional pathways.

SYNPLA, a method to identify synapses displaying plasticity after learning.

Which neural circuits undergo synaptic changes when an animal learns? Although it is widely accepted that changes in synaptic strength underlie many forms of learning and memory, it remains challenging to connect changes in synaptic strength at specific neural pathways to specific behaviors and memories. Here we introduce SYNPLA (synaptic proximity ligation assay), a synapse-specific, high-throughput, and potentially brain-wide method capable of detecting circuit-specific learning-induced synaptic plasticity.

Identification of a brainstem locus that inhibits tumor necrosis factor.

In the brain, compact clusters of neuron cell bodies, termed nuclei, are essential for maintaining parameters of host physiology within a narrow range optimal for health. Neurons residing in the brainstem dorsal motor nucleus (DMN) project in the vagus nerve to communicate with the lungs, liver, gastrointestinal tract, and other organs. Vagus nerve-mediated reflexes also control immune system responses to infection and injury by inhibiting the production of tumor necrosis factor (TNF) and other cytokines in the spleen, although the function of DMN neurons in regulating TNF release is not known. Here, optogenetics and functional mapping reveal cholinergic neurons in the DMN, which project to the celiac-superior mesenteric ganglia, significantly increase splenic nerve activity and inhibit TNF production. Efferent vagus nerve fibers terminating in the celiac-superior mesenteric ganglia form varicose-like structures surrounding individual nerve cell bodies innervating the spleen. Selective optogenetic activation of DMN cholinergic neurons or electrical activation of the cervical vagus nerve evokes action potentials in the splenic nerve. Pharmacological blockade and surgical transection of the vagus nerve inhibit vagus nerve-evoked splenic nerve responses. These results indicate that cholinergic neurons residing in the brainstem DMN control TNF production, revealing a role for brainstem coordination of immunity.

BARcode DEmixing through Non-negative Spatial Regression (BarDensr).

Modern spatial transcriptomics methods can target thousands of different types of RNA transcripts in a single slice of tissue. Many biological applications demand a high spatial density of transcripts relative to the imaging resolution, leading to partial mixing of transcript rolonies in many voxels; unfortunately, current analysis methods do not perform robustly in this highly-mixed setting. Here we develop a new analysis approach, BARcode DEmixing through Non-negative Spatial Regression (BarDensr): we start with a generative model of the physical process that leads to the observed image data and then apply sparse convex optimization methods to estimate the underlying (demixed) rolony densities. We apply BarDensr to simulated and real data and find that it achieves state of the art signal recovery, particularly in densely-labeled regions or data with low spatial resolution. Finally, BarDensr is fast and parallelizable. We provide open-source code as well as an implementation for the 'NeuroCAAS' cloud platform.

Network cloning using DNA barcodes.

The connections between neurons determine the computations performed by both artificial and biological neural networks. Recently, we have proposed SYNSeq, a method for converting the connectivity of a biological network into a form that can exploit the tremendous efficiencies of high-throughput DNA sequencing. In SYNSeq, each neuron is tagged with a random sequence of DNA-a "barcode"-and synapses are represented as barcode pairs. SYNSeq addresses the analysis problem, reducing a network into a suspension of barcode pairs. Here, we formulate a complementary synthesis problem: How can the suspension of barcode pairs be used to "clone" or copy the network back into an uninitialized tabula rasa network? Although this synthesis problem might be expected to be computationally intractable, we find that, surprisingly, this problem can be solved efficiently, using only neuron-local information. We present the "one-barcode-one-cell" (OBOC) algorithm, which forces all barcodes of a given sequence to coalesce into the same neuron, and show that it converges in a number of steps that is a power law of the network size. Rapid and reliable network cloning with single-synapse precision is thus theoretically possible.

Cellular barcoding: lineage tracing, screening and beyond.

Cellular barcoding is a technique in which individual cells are labeled with unique nucleic acid sequences, termed barcodes, so that they can be tracked through space and time. Cellular barcoding can be used to track millions of cells in parallel, and thus is an efficient approach for investigating heterogeneous populations of cells. Over the past 25 years, cellular barcoding has been used for fate mapping, lineage tracing and high-throughput screening, and has led to important insights into developmental biology and gene function. Driven by plummeting sequencing costs and the power of synthetic biology, barcoding is now expanding beyond traditional applications and into diverse fields such as neuroanatomy and the recording of cellular activity. In this review, we discuss the fundamental principles of cellular barcoding, including the underlying mathematics, and its applications in both new and established fields.

Selective corticostriatal plasticity during acquisition of an auditory discrimination task.

Perceptual decisions are based on the activity of sensory cortical neurons, but how organisms learn to transform this activity into appropriate actions remains unknown. Projections from the auditory cortex to the auditory striatum carry information that drives decisions in an auditory frequency discrimination task. To assess the role of these projections in learning, we developed a channelrhodopsin-2-based assay to probe selectively for synaptic plasticity associated with corticostriatal neurons representing different frequencies. Here we report that learning this auditory discrimination preferentially potentiates corticostriatal synapses from neurons representing either high or low frequencies, depending on reward contingencies. We observe frequency-dependent corticostriatal potentiation in vivo over the course of training, and in vitro in striatal brain slices. Our findings suggest a model in which the corticostriatal synapses made by neurons tuned to different features of the sound are selectively potentiated to enable the learned transformation of sound into action.

Efficient in situ barcode sequencing using padlock probe-based BaristaSeq.

Cellular DNA/RNA tags (barcodes) allow for multiplexed cell lineage tracing and neuronal projection mapping with cellular resolution. Conventional approaches to reading out cellular barcodes trade off spatial resolution with throughput. Bulk sequencing achieves high throughput but sacrifices spatial resolution, whereas manual cell picking has low throughput. In situ sequencing could potentially achieve both high spatial resolution and high throughput, but current in situ sequencing techniques are inefficient at reading out cellular barcodes. Here we describe BaristaSeq, an optimization of a targeted, padlock probe-based technique for in situ barcode sequencing compatible with Illumina sequencing chemistry. BaristaSeq results in a five-fold increase in amplification efficiency, with a sequencing accuracy of at least 97%. BaristaSeq could be used for barcode-assisted lineage tracing, and to map long-range neuronal projections.

Corticostriatal neurons in auditory cortex drive decisions during auditory discrimination.

The neural pathways by which information about the acoustic world reaches the auditory cortex are well characterized, but how auditory representations are transformed into motor commands is not known. Here we use a perceptual decision-making task in rats to study this transformation. We demonstrate the role of corticostriatal projection neurons in auditory decisions by manipulating the activity of these neurons in rats performing an auditory frequency-discrimination task. Targeted channelrhodopsin-2 (ChR2)-mediated stimulation of corticostriatal neurons during the task biased decisions in the direction predicted by the frequency tuning of the stimulated neurons, whereas archaerhodopsin-3 (Arch)-mediated inactivation biased decisions in the opposite direction. Striatal projections are widespread in cortex and may provide a general mechanism for the control of motor decisions by sensory cortex.

Using high-throughput barcode sequencing to efficiently map connectomes.

The function of a neural circuit is determined by the details of its synaptic connections. At present, the only available method for determining a neural wiring diagram with single synapse precision-a 'connectome'-is based on imaging methods that are slow, labor-intensive and expensive. Here, we present SYNseq, a method for converting the connectome into a form that can exploit the speed and low cost of modern high-throughput DNA sequencing. In SYNseq, each neuron is labeled with a unique random nucleotide sequence-an RNA 'barcode'-which is targeted to the synapse using engineered proteins. Barcodes in pre- and postsynaptic neurons are then associated through protein-protein crosslinking across the synapse, extracted from the tissue, and joined into a form suitable for sequencing. Although our failure to develop an efficient barcode joining scheme precludes the widespread application of this approach, we expect that with further development SYNseq will enable tracing of complex circuits at high speed and low cost.