Three-dimensional intact-tissue sequencing of single-cell transcriptional states

INTRODUCTION Single-cell RNA sequencing has demonstrated that both stable cell types and transient cell states can be discovered and defined by transcriptomes. In situ transcriptomic methods can map both RNA quantity and position; however, it remains challenging to simultaneously satisfy key technological requirements such as efficiency, signal intensity, accuracy, scalability to large gene numbers, and applicability to three-dimensional (3D) volumes. Well-established single-molecule fluorescence in situ hybridization (FISH) approaches (such as MERFISH and seqFISH) have high detection efficiency but require long RNA species (more than 1000 nucelotides) and yield lower intensity than that of enzymatic amplification methods (tens versus thousands of fluorophores per RNA molecule). Other pioneering in situ sequencing methods (via padlock probes and fluorescent in situ sequencing) use enzymatic amplification, thus achieving high intensity but with room to improve on efficiency. RATIONALE We have developed, validated, and applied STARmap (spatially-resolved transcript amplicon readout mapping). STARmap begins with labeling of cellular RNAs by pairs of DNA probes followed by enzymatic amplification so as to produce a DNA nanoball (amplicon), which eliminates background caused by mislabeling of single probes. Tissue can then be transformed into a 3D hydrogel DNA chip by anchoring DNA amplicons via an in situ–synthesized polymer network and removing proteins and lipids. This form of hydrogel-tissue chemistry replots amplicons onto an optically transparent hydrogel coordinate system; then, to identify and quantify RNA species-abundance manifested by DNA amplicons, the identity of each species is encoded as a five-base barcode and read out by means of an in situ sequencing method that decodes DNA sequence in multicolor fluorescence. Using a new two-base sequencing scheme (SEDAL), STARmap was found to simultaneously detect more than 1000 genes over six imaging cycles, in which sequencing errors in any cycle cause misdecoding and are effectively rejected. RESULTS We began by (i) detecting and quantifying a focused 160-gene set (including cell type markers and activity-regulated genes) simultaneously in mouse primary visual cortex; (ii) clustering resulting per-cell gene expression patterns into a dozen distinct inhibitory, excitatory, and non-neuronal cell types; and (iii) mapping the spatial distribution of all of these cell types across layers of cortex. For validation, per-cell-type gene expression was found to correlate well both with in situ hybridization results and with single-cell RNA sequencing, and widespread up-regulation of activity-regulated genes was observed in response to visual stimulation. We next applied STARmap to a higher cognitive area (the medial prefrontal cortex) and discovered a more complex distribution of cell types. Last, we extended STARmap to much larger numbers of genes and spatial scales; we measured 1020 genes simultaneously in sections—obtaining results concordant with the 160-gene set—and measured 28 genes across millimeter-scale volumes encompassing ~30,000 cells, revealing 3D patterning principles that jointly characterize a broad and diverse spectrum of cell types. CONCLUSION STARmap combines hydrogel-tissue chemistry and in situ DNA sequencing to achieve intact-tissue single-cell measurement of expression of more than a thousand genes. In the future, combining this intact-system gene expression measurement with complementary cellular-resolution methodologies (with which STARmap is designed to be compatible)—including in vivo activity recording, optogenetic causal tests, and anatomical connectivity in the same cells—will help bridge molecular, cellular, and circuit scales of neuroscience.