Single cell RNA-Seq (scRNA-Seq) and Imaging Transcriptomics (IT) methods have put a systematic understanding of the brain and brain diseases through comprehensive 3-dimensional map of its constituent cell types within reach. However, scRNA-Seq lacks spatial information, whereas Imaging Transcriptomics methods do not yet collect at transcriptome scale and are hampered by throughput. In particular, in IT methods, samples are passed through multiple rounds of multi-color imaging with single-molecule resolution, and the sequence of colors originating from individual molecules is used to assign each molecule to a gene identity, encoded by a molecularly designed codebook. Currently, analyzing an entire mouse brain with IT would require years of instrument time, while a human brain would require thousands of times longer. Here, we will dramatically scale up the throughput of IT imaging in genes, time, and space, by approaching this problem through the mathematics of compressed sensing. By developing suitable models of the underlying information, we will decrease the number of samples—and acquisition time—necessary to recover the underlying data at transcriptome scale. We will develop methods to compress information along three orthogonal axes. In Aim 1 we will increase the number of genes profiled by exploiting our previous results indicating that full transcriptomic information can be extracted from the activities of a small number of composite measurements of multiple genes simultaneously. We will develop an experimental method to measure these gene composites in situ, followed by a computational decompression to recover the spatial transcriptomic profile of each gene. In Aim 2 we will decrease the number of rounds of imaging necessary to assign gene identity to an RNA molecule by a combined optical and compressed sensing approach to increase the number of color channels recorded simultaneously while decreasing the total imaging and on- stage chemistry time. In Aim 3 we will develop an approach to decrease the number of pixels which must be sampled, allowing imaging at lower magnification. When combined, these three independent aims will result in a ~25,000-fold increase in throughput compared to existing state-of-the-art IT measurements. In Aim 4, we will demonstrate the combined power of these methods by generating a spatially-resolved, full- transcriptome depth atlas of the mouse primary motor and somatosensory cortices. We will use this data set to benchmark the performance of the accelerated pipeline to both current IT measurements and scRNA-Seq. We will work with committed beta testers across BICCN and will share all protocols, code, and reagents openly to BICCN and the broader biological community. Our approach will be applicable across multiple IT methods and will accelerate BICCN and other mapping efforts in health and disease.

To understand the brain and its disease states, it is important to know what kinds of cells it contains and where they are found. Current tools for identifying cell types in brain tissue are far too slow to handle the large number of cells in the brain. We will develop a wide array of software and optics tools to accelerate data generation to meet the scale of this challenge.