TL;DR
This paper introduces pciSeq, a novel method for mapping the spatial organization of individual cell types within intact tissue. By combining single-cell RNA sequencing (scRNA-seq) data with in situ sequencing, they successfully identified and mapped 16 distinct inhibitory interneuron subtypes in the mouse hippocampus, revealing their precise laminar distribution. This work bridges the gap between transcriptomic cell classification and spatial localization, offering a powerful tool for understanding brain circuitry.
What Problem Does This Paper Address?
Understanding how complex tissues like the brain function requires knowing not only what cell types are present but also where they are located and how they are organized. While single-cell RNA sequencing (scRNA-seq) has revolutionized our ability to classify cells based on their gene expression profiles, it destroys the spatial information of the tissue. This leaves a critical gap: we can identify many subtle cell subtypes, but we don't know their precise locations within the intricate architecture of the brain. The hippocampus, a region crucial for memory, is known to have a complex organization of inhibitory interneurons, with subtle differences in their laminar distribution. This paper aims to address this by developing a method to map these fine cell types spatially within the hippocampus.
Methods
The core of this paper is the development and application of probabilistic cell typing by in situ sequencing (pciSeq). This method integrates two key components:
- Leveraging scRNA-seq data for gene selection: Instead of randomly selecting genes, they used existing scRNA-seq datasets of hippocampal and cortical inhibitory neurons to identify a minimal set of ~99 marker genes that are sufficient to distinguish cell types. This selection process involved downsampling expression levels to account for the lower efficiency of in situ methods and manual curation to exclude broadly expressed genes.
- Multiplexed in situ sequencing: They adapted a barcode-targeted in situ sequencing approach. This involves:
- Designing padlock probes for the selected marker genes. These probes have recognition arms that bind to specific mRNA sequences and a barcode for identification.
- Applying these probes to tissue sections, where they hybridize to target mRNAs.
- Using a ligation enzyme to circularize the probes that perfectly match.
- Performing rolling-circle amplification (RCA) to generate many copies of the circularized probe's barcode, creating sub-micron DNA molecules (RCPs).
- Identifying these RCP barcodes using multi-color fluorescence imaging over several rounds.
- A final round of hybridization is used for genes with very strong signals (like Sst and Npy) to avoid swamping the signal.
- Probabilistic cell assignment: A custom pipeline analyzes the detected barcodes. It uses a variational Bayesian algorithm to assign each detected read to a specific cell and then assigns each cell to a type based on a probability model derived from scRNA-seq clusters. Crucially, this assignment algorithm does not use laminar location, allowing it to be used as an independent validation.
Datasets
- scRNA-seq datasets: The authors utilized existing scRNA-seq data from hippocampal and cortical inhibitory neurons (specifically referencing work by Zeisel et al. and Tasic et al.) to inform their gene panel selection and build their probabilistic cell typing model.
- Allen Mouse Brain Atlas: This atlas was used for verifying the spatial distribution of the identified genes and cell types.
- Newly generated data: The study generated new in situ sequencing data from 14 coronal sections of a P25 male CD1 mouse hippocampus, covering different parts of the dorsal hippocampus. This dataset contains approximately 15.4 million reads passing quality control.
Key Insights and Ideas
- Bridging the transcriptomic-spatial divide: The most significant insight is the successful integration of scRNA-seq's high-resolution cell classification with the spatial information preserved in tissue. This overcomes a major limitation of scRNA-seq.
- Minimal gene sets are sufficient: The study demonstrates that a relatively small panel of ~99 carefully selected genes can accurately distinguish a large number of closely related cell types, challenging the notion that comprehensive transcriptomic profiling is always necessary for spatial mapping.
- Probabilistic assignment is robust: The use of a probabilistic model, informed by scRNA-seq, allows for confident assignment of cells to types, even when dealing with subtle differences and potential noise in the in situ data.
- Validation through laminar organization: The fact that the identified cell types were found in their expected laminar positions within the hippocampus provides strong validation for both the pciSeq method and the underlying scRNA-seq classifications.
Key Results
- Identification of 16 inhibitory interneuron classes: The pciSeq method successfully identified 16 distinct inhibitory interneuron classes in the hippocampal CA1 region.
- Spatial mapping consistent with ground truth: The spatial distribution of these identified cell types closely matched the known laminar organization of the hippocampus, as validated by comparing their locations to established neuroanatomical knowledge.
- High accuracy with a limited gene panel: The method achieved high confidence in distinguishing cell types using only 99 genes, demonstrating the efficiency of their gene selection strategy.
- Low false-positive rate: The authors emphasize that a key to their method's success is a very low false-positive gene detection rate, which is crucial for accurate cell typing.
- Demonstrated applicability to other cell types: Beyond inhibitory neurons, the method also confidently distinguished two types of pyramidal cells and non-neuronal cells.
Limitations and Future Work
- Limitations:
- The current study focused on the hippocampus. While the method is presented as broadly applicable, its performance on other brain regions or tissues would require further validation.
- The method relies on existing scRNA-seq data for gene selection and model training. The quality and comprehensiveness of this prior data are critical.
- While the gene panel was optimized, some cell types were too rare for reliable quantification or laminar distribution analysis (e.g., Sst/Nos1 cells).
- Future Work:
- The authors envision applying this method to map cell types across the entire brain and other tissues, leveraging the growing availability of large-scale scRNA-seq datasets.
- The pciSeq approach's requirement for only low-magnification imaging, unlike some FISH-based methods, suggests potential for high-throughput applications and the creation of body-wide spatial cell type maps.
- The development of an online viewer for reads and probabilistic assignments indicates a commitment to making the data and tools accessible for further research.
Summary updated