Spatial Biology: Location, Location, Location!
Complete the form below to unlock access to ALL audio articles.
Spatial biology is the 2- and 3-dimensional study of cells and tissues. The relative position of cells and tissues is critical for biological processes. During development, the relative position of cells dictates tissue layering and patterning. Spatial elements also shape the tumor microenvironment, with implications for tumor heterogeneity, invasion and metastasis. Spatial biology technologies have been developed to uncover cellular and tissue architecture by assessing spatial transcriptomics, spatial proteomics, or spatial metabolomics and lipidomics. Multimodal methods have also been devised that combine spatial technologies with other methods, such as imaging mass spectrometry (IMS) with microscopy. Multimodal methods can generate more information to facilitate a deeper understanding of biological processes. Moreover, advances continue to improve spatial resolution, either through new technologies or computational methods. Enhanced resolution to the single-cell or subcellular level can offer finer-grained spatial information, which may, for example, help identify rare cell types of critical spatial and biological significance or elucidate organellar biology.
This article will review advances in multimodal IMS applications and a computational method to achieve higher resolution from spatial transcriptomics datasets.
Multimodal IMS: Multiple methods to gain more spatial information
IMS is a powerful technology that integrates mass spectrometry with spatial information. Since the seminal publication, which introduced the method to the scientific community in 1997, IMS has grown in scope and in research and clinical applications. IMS capabilities are being further augmented by multimodal applications. “Biological systems are highly complex, which no single modality can fully capture. Multimodal IMS generates more information from a single experiment by analyzing multiple chemical classes or by combining mass spectrometry with other modalities. It produces a single multimodal image or viewpoint that is superior to each modality considered separately,” explained Richard Caprioli, author of the 1997 IMS paper, and the Stanford Moore Chair in Biochemistry and director of the Mass Spectrometry Research Center at Vanderbilt University School of Medicine.
Unimodal IMS can be performed using various ionization methods, which vary in spatial resolution and molecular weight range of ionized molecules. Matrix-assisted laser desorption/ionization (MALDI) is the most common and uses a laser beam to ablate molecules from tissue samples coated with a matrix to aid ionization. Ionized molecules span a broad range of molecular weights, from metabolites to proteins. Spatial resolution is dictated by laser cross-section and distance between pixels, with pixel sizes ranging from 5 to 20 microns in traditional MALDI IMS, but these can drop to around 1 micron with some laser setups, although at a trade-off with fewer ionized molecules. Secondary ion mass spectrometry (SIMS) uses an ion beam to ionize tissue sample molecules. Although it has better spatial resolution than MALDI, it only analyzes lower molecular weight species (< 2 kDa). Desorption electrospray ionization (DESI) employs a stream of charged droplets to ionize tissue sample molecules and is generally of lower spatial resolution than MALDI, but it has the special advantage that it is performed at atmospheric pressure instead of under vacuum.
One of the most promising multimodal approaches is to combine the various spatial IMS ionization methods to enjoy the benefits each has to offer. For example, combining MALDI and SIMS to enhance lipid identification in tissues or using the lower resolution method to guide higher resolution analysis at regions of interest. Multimodal IMS can also capitalize on tandem mass spectrometry (MS/MS) to identify lipids or proteins from the ionized fragments ejected from the tissue sample. “Previously, IMS tandem MS/MS was limited to analyzing fragments from a single precursor ion, i.e., a single molecular species from the tissue sample. Although this may suffice for experiments interested in one single molecule, for example the distribution of a particular drug in a tissue, it is of low throughput and does not allow for an untargeted analysis,” Caprioli explained. “By applying tandem MS/MS to each burst of the MALDI laser during IMS, we were able to concurrently visualize the distribution of multiple phosphatidylcholine lipids in rat brain tissue.”
Analyzing Complex Spatial Data for Deep Insights
Edit or remove this text before publishing content piece.View Application Note
IMS capabilities can also be augmented by integration with microextraction and ion mobility to improve sensitivity. “Microextraction is an excellent way to improve sensitivity by running a preliminary liquid or solid extraction step, which enriches the sample in a specific analyte, for example proteins. When the extractions are applied spatially, such as by liquid extraction surface analysis (LESA) and higher resolution microLESA, we can analyze tissue, such as rat or mouse brain,” Caprioli elaborated on recent research. “The concept behind ion mobility is similar to microextraction, except ions of a specific volume and charge are selected in the gas phase.”
Finally, it is also possible to combine IMS with other modalities, such as microscopy, spectroscopy and electrochemistry. Although still in infancy, it may even be possible to combine IMS with other omics technologies, such as transcriptomics, although, at present, usually without a spatial element for the second technology. “We have done a lot of research combining MALDI IMS with microscopy,” Caprioli continued. “For example, we employed IMS to investigate Staphylococcus aureus siderophore distribution in host mouse tissue, which we integrated with inductively coupled plasma IMS to analyze 56Fe and fluorescence microscopy to analyze expression by S. aureus iron acquisition genes using a GFP reporter assay. The multimodal approach allowed us to visualize the nonuniform distribution of siderophores, suggesting the host may not be uniformly devoid of iron.” Caprioli has also collaborated on the computational challenges of multimodal IMS with Raf Van de Plas, associate professor at the Delft Center for Systems and Control, at the Delft University of Technology, to leverage multimodal IMS to investigate numerous biological processes in addition to host-pathogen interactions, such as pathogen membrane modification and the tissue lipidome.
When asked about future directions for multimodal IMS, Caprioli replied, “Previously, multimodal IMS was combined with tissue sections stained by standard methods, such as H&E. However, we can expect to integrate future multimodal IMS with new highly multiplexed imaging modalities, such as CODEX immunofluorescence microscopy. Combining IMS measurements, e.g., peptide or lipid distributions, with other imaging data that accurately delineates anatomical structures or cell types in tissue can be used to generate fine-grained molecular signatures of cells and their variation across tissue sections.” Van de Plas and Caprioli recently applied this idea in a proof-of-concept multimodal IMS study for automated biomarker candidate discovery.
Van de Plas highlighted the role of computational methods on the horizon for multimodal IMS. “Current multimodal IMS integration uses primarily univariate and multivariate models that can capture linear relationships between IMS-reported molecular species and the properties recorded by other imaging modalities. In the next five years, it will become increasingly common to employ more complex model types to model cross-modality links, making it possible to also capture nonlinear relationships. Additionally, we will need data dimensionality reduction techniques that can reduce the thousands to millions of acquired dimensions in multimodal IMS datasets to a practical number while incurring minimal loss of information. Therefore, I anticipate increased development and usage of advanced signal processing and machine learning techniques to keep IMS analysis practically feasible.”
An All-In-One Solution for Antibody Validation
ChipCytometry™ antibodies leverage decades of histology tradition, using clones validated for flow cytometry and immunohistochemistry. Download this app note to discover a platform that can profile dozens of biomarkers, provides accurate quantification of high- and low-expressing proteins and is compatible with FFPE, fresh frozen tissue and cell suspensions.
View App NoteSponsored Content
Both Caprioli and Van de Plas foresee a dramatic rise in multimodal IMS, “It offers a molecular view of biologically important processes that are otherwise hard to map and its spatial aspects allow it to more fully capture biological complexity. Multimodal IMS accomplishes this in a multiplexed and untargeted manner at a spatial resolution that is very relevant for discovering and understanding biological pathways as well as molecular mechanisms of disease.”
BayesSpace: A computational framework for subspot resolution in spatial transcriptomics
Many methods have been developed to spatially analyze gene expression, which vary in resolution, target analyte (mRNA, microRNA), throughput (i.e., number of analyzed cells or spots), number of analyzed genes, and whether the approach is targeted or untargeted, i.e., analyzes prespecified genes or all sample genes. The methodology also varies, ranging from methods based on microdissection, fluorescence in situ hybridization, in situ sequencing and in situ capture. In situ capture has grown in popularity, especially since commercialization, which made it accessible to researchers.
In situ capture works by laying a tissue section on a slide covered by a grid of uniquely barcoded spots. Each spot contains DNA capture probes, which comprise a poly(dT) tail that hybridizes with sample tissue mRNAs, and a unique barcode coding the spatial information, i.e., where the spot is located on the slide, and hence on the overlaid tissue. After tissue mRNAs hybridize with the probe, a DNA library is constructed and sequenced. Data visualization methods then reconstruct the spatial distribution of transcripts based on the spatial barcode. The spatial resolution is limited by the spot size, which ranges from around 1 to over 20 cells, for most commercially available methods.
In recent years, interest has grown in developing computational methods that can improve the resolution of spatial transcriptomic datasets. BayesSpace is one such approach, a Bayesian statistical method developed in the laboratory of Professor Raphael Gottardo, which takes advantage of both the gene expression and spatial information already present in each spot of the spatial transcriptomics dataset to cluster and enhance resolution. Spatial transcriptomics was a relatively new technology when BayesSpace was developed, with most methods available at the time associated with technologies such as single-cell RNA-seq.
How Spatial Phenotyping Can Uncover Novel Insights in Tissue Biology
Spatial phenotyping allows you to visualize and quantify dozens of biomarkers in a single tissue sample while maintaining cellular and sub-cellular detail. Download this whitepaper to explore a method that can put cellular function in context, account for every cell in your sample and reveal “cellular neighborhoods”.
View WhitepaperSponsored Content
At its core, BayesSpace is a clustering model. Each spot from the spatial transcriptomics dataset can be divided into subspots. Modeling at this higher resolution, BayesSpace can estimate gene expression at the subspot level and identify clusters of subspots with similar expression profiles. By leveraging spatial information, BayesSpace enables enhanced resolution clustering and generates a higher-resolution spatial map.
BayesSpace was validated at the original spot level resolution against three spatial methods, spatialLIBD, Giotto and stLearn*, and three non-spatial methods, Louvain, SC3 and mclust, using sections of dorsolateral prefrontal cortex. The computational methods were used to cluster gene expression from the spatial transcriptomics dataset of the dorsolateral prefrontal cortex sections into the white matter and six cortical layers, which were compared to the manually annotated reference employing a metric called the adjusted Rand index (ARI). BayesSpace had a higher median ARI compared to all other computational methods, indicating it could best predict the cortical layers from the spatial transcriptomics dataset.
BayesSpace was further tested at subspot levels. Since there were no other computational methods available to analyze data at the subspot level for comparison, this was accomplished by performing spatial transcriptomics and high-resolution immunohistochemistry of an invasive ductal carcinoma for CD3 and CD45 to stain T cells and leukocytes, respectively. The anti-CD3 and anti-CD45 signals from immunohistochemistry correlated well with the corresponding enhanced gene expression modeled by BayesSpace from the spatial transcriptomics data, validating the method. Furthermore, BayesSpace refined the boundary delineating immune-rich and immune-poor areas.
Given the importance of spatial information to biological processes, advances continue to improve spatial methods of analysis. This includes multimodal methods, which combine spatial approaches, such as IMS, with other analytical modalities to increase the richness of biological information. Further, technical and computational methods are pushing the envelope of spatial resolution, to derive more granular information on biological systems.
*This article is a preprint that is yet to be peer-reviewed. Results are therefore regarded as preliminary and should be interpreted as such. Find out about the role of the peer review process in research here. For further information, please contact the cited source.
