Spatial transcriptomics has emerged as a powerful approach in biological research, allowing scientists to map gene activity directly within tissue samples. This technology provides a detailed view of gene expression, preserving its original location within complex tissue architecture. Accurately identifying individual cells and defining their boundaries, known as cell segmentation, is a foundational computational step. This process helps researchers extract information from complex biological images.
Understanding Spatial Transcriptomics
Spatial transcriptomics represents an advancement in molecular biology, measuring gene expression while retaining the precise spatial location of genetic signals within a tissue. Unlike traditional methods that dissociate tissues, spatial transcriptomics maintains the intricate organization of cells in their native environment. This preservation of spatial context allows researchers to understand how cells interact and function based on their position. For example, knowing which genes are active in a tumor cell and its immediate neighbors can reveal insights into disease progression that would be lost if the tissue were disaggregated.
Imaging-based spatial transcriptomics methods directly visualize RNA molecules within tissue sections using fluorescent probes. Technologies like multiplexed error-robust fluorescence in situ hybridization (MERFISH) or in situ sequencing (ISS) detect thousands of RNA transcripts through repeated rounds of hybridization and imaging. These techniques provide subcellular resolution, pinpointing individual RNA molecules within parts of a cell. This detail contrasts with some sequencing-based methods, which capture gene expression from larger “spots” containing multiple cells.
The Necessity of Cell Segmentation
Cell segmentation is a necessary step in imaging-based spatial transcriptomics, enabling researchers to precisely assign gene expression data to individual cellular sources. Without accurately defining cell boundaries, molecular signals could be incorrectly attributed, mixing genetic information from neighboring cells. This admixture can obscure cellular functions and interactions, leading to inaccurate biological conclusions. For example, transcripts from a cancer cell might be mistakenly assigned to an adjacent immune cell, distorting the true molecular profile of both cell types.
Delineating individual cells allows investigators to study cellular heterogeneity within tissues, understanding how different cell types exhibit distinct gene expression patterns. It also facilitates the analysis of cell-to-cell communication, as researchers can determine which cells are physically interacting and how their gene expression profiles influence these interactions. Precise segmentation ensures each detected RNA molecule is correctly associated with its originating cell, which is important for downstream analyses like cell type identification, trajectory inference, and spatial domain discovery. Accurate cell segmentation directly supports the reliability of biological insights gained from spatial transcriptomics data.
Methods for Cell Segmentation
Cell segmentation in imaging-based spatial transcriptomics involves various computational approaches, each with strengths and limitations. Traditional strategies include thresholding, which separates cells from the background by setting an intensity value, and the watershed algorithm, which identifies cell boundaries by finding “ridges” between cellular regions. These methods rely on image intensity and gradient information.
Challenges arise from the diverse nature of biological tissues, which exhibit varying cell densities, shapes, and sizes. Tissues can have densely packed or irregularly shaped cells with overlapping boundaries, making precise demarcation difficult. Different imaging modalities also present unique noise characteristics and resolutions, further complicating segmentation.
To address these difficulties, researchers increasingly employ machine learning and deep learning approaches, which learn complex patterns from annotated images. Convolutional Neural Networks (CNNs) automatically identify features distinguishing cells from their surroundings. Algorithms like Cellpose are generalist segmentation tools effective across various cell types and imaging modalities. Other advanced methods, such as Baysor, optimize cell boundaries by considering transcriptional composition and cell morphology. Newer techniques like ST-CellSeg use multi-scale manifold learning and distance metrics to improve segmentation accuracy and speed.
Advancing Biological Discovery
Accurate cell segmentation in imaging-based spatial transcriptomics advances biological discovery by enabling a precise understanding of cellular functions within their native tissue context. By correctly assigning gene expression to individual cells, researchers can uncover subtle molecular changes contributing to disease mechanisms. In cancer research, for instance, precise segmentation helps identify malignant cells and their microenvironment, revealing gene expression patterns associated with tumor progression or therapeutic resistance. This detailed mapping can lead to the discovery of novel cell types or states previously unidentifiable through dissociated cell analyses.
The quality of cell segmentation influences insights into cell-to-cell communication within complex tissues. Researchers can analyze how specific cell types interact and signal by accurately defining their boundaries and molecular profiles. This capability aids in mapping receptor-ligand interactions between adjacent cells, important for understanding processes like immune response, organ development, and neurological function. For example, understanding how immune cells communicate with tumor cells at a precise spatial level can inform targeted immunotherapies.
Precise segmentation allows for comprehensive analysis of tissue architecture and cellular heterogeneity, aiding in identifying biomarkers and therapeutic targets. Errors in segmentation can confound downstream analyses, potentially leading to misinterpretations of differential gene expression or cell type influences. Therefore, continued advancements in cell segmentation algorithms are important, as they directly enhance the accuracy of spatial transcriptomics data, supporting breakthroughs in understanding health and disease.