Currently scRNA-seq associates each transcript with a single cell, but information about the location of these transcripts in the tissue is lost; in contrast, spatial transcriptomics techniques know the location of the transcripts but not which cell produced them. Thus, the integration of scRNA-seq with spatial transcriptomics can produce high-resolution maps of cell subpopulations in tissues.
Researchers from the U.S. have published in theNature reviews genetics"Publishing Review Articles.Attempts and efforts to integrate scRNA-seq with spatial transcriptomics technology research, including emerging integrated computational approaches, are reviewed, and avenues for effectively combining current approaches are proposed.
A process model for integrating scRNA-seq and spatial transcriptomics studies
Advances in integrated analysis of scRNA-seq+spatialomics
Integrated spatial transcriptomics and scRNA-seq are now available.data analysisstudies that provide new insights into tissue composition and function. The following table demonstrates the current state of relevant research, including the direction of normal tissue homeostasis and development, the tumor microenvironment, and the microenvironment of other lesions and injuries.
Dissecting scRNA-seq and spatial transcriptome data for research
Integrated analysis of scRNA-seq+spatialomicscalculation method
Given that spatial transcriptomics approaches have not yet been able to produce deep single-cell resolution transcriptome profiles in tissues, analyses that can successfully integrate single-cell and spatial transcriptome data will help to understand the structure of cell-type distributions as well as the putative mechanisms of cell-to-cell communication that make up this structure.There are two main approaches to integrating scRNA-seq and spatial transcriptome data: deconvolution (Deconvolution) and mapping (Mapping)Go. Go.convolutionDesigned to isolate discrete subpopulations of cells from a mixture of mRNA transcripts at each capture site based on single-cell data; mapping is twofold: localization of specified scRNA-based cell subtypes to each cell on the HPRI map and localization of each scRNA-seq cell to a specific ecological niche or region of the tissue.
Strategies for integrating single-cell and spatial transcriptomic data
Deconvolution: separation of discrete cell subtypes from a single capture point.There are two main approaches to deconvolution: inferring the proportion of cellular subtypes for a particular spot and scoring a particular spatial transcriptome spot to determine how well it corresponds to individual cellular subtypes.
Inference-based deconvolution techniques involve estimating the proportion of each cell type at a particular capture point. One approach to this form of deconvolution is the use of statistical regression-based models, with variouslinear regression modelhas been applied to deconvolute bulk RNA-seq mixtures.
A complementary approach to estimating the exact proportion of each cell type in a given capture site is to fit a probability distribution to the gene count distribution of the scRNA-seq data through a Bayesian statistical framework. whereSPOTlight The benchmarking strategy is the most thorough: assessing the accuracy, sensitivity and specificity of the cell type assay as well as the overall relevance to the real situation. In addition, physical validation of the spatial localization of subtypes at higher resolution is available through HPRI.
There are many deconvolution techniques based on enrichment scores, such asSeurat 3.0 and multimodal cross-analysisetc.; deconvolution techniques strategies for solving dataset mismatches, such asSpatialDWLSetc.
Mapping: creates spatially resolved mapping of cell types at single-cell resolution.Just like deconvolution, the first step in mapping is to establish cell subtypes based on scRNA-seq data. The main challenge of mapping is then to assign scRNA-seq-based cell types from the HPRI data to each cell. For the 14 publishedarithmeticPerforming a systematic evaluation, these algorithms implement a batch correction strategy for mapping through a clustering-based analysis that identifies theThree algorithms that most efficiently integrate scRNA-seq data with single-cell resolution spatial data; LIGER, Seurat Integration (from Seurat 3.0), and Harmony.All three algorithms ultimately use different methods to integrate clustering into a low-dimensional space, where cell types are obtained by population detection of clusters.
Incorporating spatial data into the analysis of intercellular communication.Interactions between cellular subpopulations mediate homeostasis, development, and disease in tissues. Spatial transcriptomics data are well suited to assess the reliability of ligand-receptor interactions calculated by scRNA-seq. The standard algorithm for predicting ligand-receptor interaction pairs involved in intercellular communication is primarily a combination of scRNA-seq data and a database of known ligand-receptor interactions. There are many ways to decipher this mechanism of intercellular communication. For exampleGiotto, SpaOTsc algorithmetc. In addition, spatial data can be used to evaluate scRNA-seq mapping reconstructions and estimates of ligand-receptor interactions such asThe novoSpaRc algorithm。
Future directions for integrated scRNA-seq+spatialomics analysis
Other modes of integration
Currently, spatial transcriptomics technologies are focused on detecting mRNA transcription by next-generation sequencing (spatial barcoding) or fluorescent labeling (HPRI). However.Histological images of tissue sections generated by spatial transcriptomics experiments are often not utilized.A number of algorithms have been developed, for example, based on the premise that a large amount of spatial variation manifests itself intuitively at the level of organizational structure, a team of researchers developed aDeep Learning Algorithm ST-Netthat can predict spatial changes in expression of 102 genes per spatial barcode capture point superimposed on the tissue structure. In addition.XFusecombined spatial barcoding and histological sections to predict expression at single-cell resolution. Thesedeep learningThe Saliency map of the model is crucial for extracting new spatial features related to the expression of individual genes in the transcriptome. In addition to improving the algorithms for deconvolution and mapping theOne focus that needs attention is the development of more deep learning models to help distinguish which features of a given spatial transcriptome are most biologically significant.
Defining a three-dimensional spatial transcriptome and real-time cell tracking offer new frontiers for future research.Currently, most studies of the three-dimensional spatial transcriptome use high-density sections to computationally reconstruct or infer the location of scRNA-seq cells from three-dimensional single-molecule fluorescence in situ hybridization data.STARmap and ExSeqis a newly developed method that combines HPRI with the conversion of intact tissue to hydrogels to preserve the 3D alignment of amplicons.
While it is possible to depict the spatial transcriptome throughout the time course of development or tissue pathogenesis, theHowever, spatial transcriptome technology does not allow real-time monitoring of the physical dynamics of cellular subtypes。optical coherence tomography (OCT)have been used to track the migration of tumor-associated bone marrow cells.CellGPShas been used with positron emission tomography to track human breast cancer cells carrying radioisotopes.When combined with spatial transcriptomics, both real-time tracking techniques can be applied to cell types of interest in spatial data to elucidate cellular dynamics in the environment, such as metastatic progression and immune cell dynamics during cancer immunotherapy.
By spatially resolving other biomolecules that are integral to the central laws of molecular biology, going beyond spatiotemporal transcriptome parsing can lead to a deeper understanding of tissue function. DBiT-seq, for example, allows spatial resolution of protein and mRNA transcripts on the same tissue. Genomic sequences ofThree-dimensional in situ imaging, subcellular resolution of RNA, and simultaneous imaging of nucleoli and RNA in three-dimensional chromatin organizationare present at the single-cell scale. Theypromises to be applied to intact tissues and revolutionize our understanding of how central law mechanisms function in the three-dimensional environment of the cell, thereby revealing developmental trajectories and the inner workings of disease (i.e., cancer).
Clinical relevance
Spatial transcriptomics studies with comparative analysis of diseased and healthy tissues have begun to elucidate prognosis, optimal treatment, and potential therapeutic targets.However, such studies are limited in terms of sample size and so far are in the exploratory phase.To speed up data generation, analyses could focus on describing a smaller number of regions of interest that drive disease-related phenotypes. In addition to describing trends in patient prognosis, examining how existing drugs, especially repurposed drugs, affect spatiotemporal gene expression patterns in disease-driving cell types may provide insight into potential therapeutic agents. In this regard, byNASC-seqand other methods to monitor mRNA transcription in response to stimuli may help to better understand how drug interference affects the spatial transcriptome of diseased cells. Once these patient tissue data are integrateddeep learning modelcan help identify the spatial expression patterns that are most relevant to survival outcomes or treatment response, potentially highlighting favorable targets to be reproduced or supplemented with intervention nodes during treatment.
As more spatial transcriptomics analyses are performed, unraveling defined, disease-associated cell types and their gene modules will become increasingly challenging. More and more cell types are being identified and localized in tissues, SeuratIntegrationTools such as Harmony and LIGER may be upgraded to integrate data from different experimental assays to determine if specific cell types are consistently observed in each tissue. In addition, it would be valuable to integrate spatial transcriptomics data for each organ system and disease, such as the SpatialDB database, Allen Brain Atlas, etc. Ultimately, a more defined spatial transcriptome of disease-driven cell types, especially for cases where cellular function is particularly dependent on the in situ environment and neighboring cell populations, may yield more effective biological mechanisms for therapeutic targeting.
Technologies for detecting spatial transcriptomes are rapidly evolving, so there is no single spatial transcriptomics technique that is suitable for all applications. Depending on the biological question posed, experimental approaches can combine any spatial transcriptomics method with scRNA-seq. In addition to the development of enhanced methods, the choice of algorithms for integrating these data is critical because spatial transcriptomics methods do not yet exist for spatially resolving tissues at single-cell resolution, scRNA-seq depth, and whole transcriptome coverage. Such integrated approaches can spatially map specific cellular subpopulations in development and disease and elucidate the mechanisms by which these cellular subpopulations synergistically shape tissue phenotypes.
bibliography
Longo, ., Guo, ., Ji, . et al. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat Rev Genet (2021). /10.1038/s41576-021-00370-8
Images are from references, please contact to remove any infringement.