By Kristin G. Beaumont
In cancer biology, the role of the tumor microenvironment is becoming more and more well-appreciated, not only in terms of how it contributes to the progression of disease, but in how different cellular compartments mediate response to therapy and development of therapeutic resistance. Spatial transcriptomics consists of a family of methods and technologies that are used to map gene expression in order to capture spatially localized phenomena and tissue spatial heterogeneity. As this is a rapidly expanding area of scientific interest and technology development (reviewed in 1-6), a wide variety of approaches have evolved, each with its own strengths, limitations and ideal areas for application.
What do we know about the role of the tumor microenvironment?
The tumor microenvironment is a highly complex and dynamic ecosystem that regulates progression of tumor growth as well as response to treatment. It consists not only of tumor cells, but also of immune cells, blood vessels, stromal cells (such as fibroblasts) and extracellular matrix protein scaffolding. Organization of the various compartments within the tumor niche facilitates signaling and regulation between different cell populations but is often highly heterogeneous and dynamic over the course of disease progression and treatment.
In recent years, the tumor microenvironment has been extensively characterized using single cell sequencing approaches. As an overview, this approach involves 1) careful dissociation of the tumor into its component cells, 2) isolation and lysis of single cells plus barcoding of transcripts from each cell 3) reverse transcription, and 4) library preparation and sequencing. Research using single cell sequencing methods has revealed many new biological insights including highly resolved stratification and characterization of novel subpopulations of cells. However, through the evolution of the single cell sequencing field, technical limitations of these methods have become evident. One clear example is that spatial context among neighboring cells is lost during tissue disruption and dissociation. Additionally, certain populations of cells within the sample tissue may be more sensitive to dissociation methods, resulting in increased numbers of delicate cells being lost to damage or death followed by subsequent underrepresentation of these populations in the final sequenced dataset.
Spatial sequencing methods have evolved to address limitations in single cell analysis and permit characterization of intact tissue sections by direct, spatially resolved measurement of RNA transcripts. Even in its early days of use, this method has already revealed exciting tissue architecture and spatially localized biological insights. Some of the earliest efforts include simply evaluating the extent of inter- and intra-tumor heterogeneity (often with a focus on immunosuppressive regions), as high heterogeneity has historically been thought to contribute to poor prognoses across cancers 7-11. Moreover, given the knowledge that localized regions of inflammation or hypoxia can affect tumor growth by influencing cell proliferation, invasiveness and response to stress, other studies have characterized such functionally-significant regions in a variety of different cancers 12,13. Approaches to spatial transcriptomic analysis fall into one of two major categories – low resolution methods that facilitate gross transcriptomic analysis of tissue and high-resolution methods that provide single cell/subcellular-level transcriptomic information–and generating further exciting data depends on selecting, optimizing and implementing the most appropriate experimental methods.
Low-resolution capture methods for high-quality samples
SlideSeq14 and similar commercialized methods (Visium from 10X Genomics, for example) are ideally suited to be a starting point for spatial characterization of tumor tissue, because these are unbiased methods (based on general oligo dT capture strategies) that are used to spatially barcode and sequence any polyadenylated RNA from a tissue section. Relatively large areas of tissue (routinely up to several mm in diameter) can be analyzed at one time, providing a bird’s eye view of gene expression. In brief, the assay captures RNA from carefully permeabilized tissue sections, where capture efficiency and final data quality are dependent on the tissue structure and level of degradation of the resident RNA. Ideally, prior to assay execution, tissues are initially evaluated for quality (both intactness via H&E staining and RNA quality using RIN scoring), then tested to determine an appropriate permeabilization time for RNA retrieval and localized capture. Once optimal assay conditions have been determined, spatially-barcoded libraries are prepared from the captured RNA and sequenced using existing short-read sequencing technology. The current resolution of this approach (defined by the diameter of uniquely barcoded capture regions) is approximately 10-55 m depending on the method, but this is expected to improve with subsequent iterations of the technology. As this area will capture RNA from multiple cells, sometimes integration with scRNA Seq data is leveraged to provide additional inferred resolution15,16.
Practically, this general approach is a logical starting point in understanding spatially-dependent gene expression given that it 1) is based on unbiased capture of polyadenylated transcripts, 2) does not require an instrument and 3) provides data across relatively large areas of tissue.
Low-resolution targeted methods for less-than-ideal samples
Access to fresh frozen tumor tissue is often a clinical luxury, where samples preserved via formalin fixation and paraffin embedding (FFPE) are far more commonly-obtained, particularly for longitudinal studies. Working with such samples is challenging given that the preservation process often causes degradation of RNA within tissue. Even with relatively deep sequencing, short transcript sequences are challenging to align to a given reference genome in order to determine gene expression–a much more sensitive method to extract information from fragmented RNA is to use panels of probes targeted to transcript regions coding for specific genes, where these may be disease/tissue specific or designed to tile the entire transcriptome.
It is important to note that not all FFPE samples are of equivalent quality – sample handling prior to preservation, preservation conditions and storage conditions can all impact resident RNA quality; not even targeted methods can salvage gene expression data from overly degraded samples. As such, thorough characterization of FFPE tissue quality is important, which should include DV200 scoring (which quantifies the percentage of RNA fragments that are > 200 nucleotides) and tissue intactness evaluation.
Commercially, Visium FFPE and Nanostring GeoMx leverage targeting to capture or identify transcripts contained within FFPE samples. Visium FFPE is similar to the Visium assay for fresh frozen tissue, with the exception that targeted probes are used to append polyA tails to transcripts for capture from the entire tissue section onto the surface array. GeoMx is a somewhat orthogonal method that relies on the user to identify regions of interest, from which UV-labile probes are liberated and sequenced. Typically, these regions of interest are identified using histopathological input and are useful for guiding analysis between clearly distinct areas of a tissue sample.
High-resolution targeted methods
Ideally, low resolution spatial transcriptomic methods reveal novel or interesting patterns of spatially-localized gene expression which serve to catalyze further questions about disease pathology. In such cases, hypothesis-driven work can be accelerated using any one of several high-resolution, in situ spatial transcriptomic analysis methods. While technical nuances exist among the individual instruments, the overall premise and workflow remains similar: Fluorescently-resolvable probes (currently up to a few hundred per sample) are bound to transcripts retained within tissue, and these probes are visualized and mapped using high-resolution imaging. Different methods employ somewhat diverse strategies to optimize probe chemistry design (to increase target specificity, total number of accessible probes and signal to noise ratio) and imaging methods (to also increase signal to noise ratio, but decrease overall scan times), both of which affect the number of confidently detected transcripts per cell, and thus, the overall quality of data.
These methods are likely to serve as downstream validation of observations made using lower-resolution spatial transcriptomic approaches for several reasons. First, high resolution methods are typically 1) more expensive, owing to the need for a technically-complex instrument that includes high resolution imaging and 2) lower throughput that their low-resolution counterparts. More importantly, however, insights gained from high-resolution in situ methods are intricately dependent on correct and comprehensive design of the implemented targeted panel, and input from lower-resolution methods can help inform curation of a panel that focuses precisely on the biological questions at hand.
Commercially, there are several instruments that are available for executing these approaches–this field is changing rapidly, thus reporting performance metrics for each instrument in this publication is not likely to be useful. It is, however, worth mentioning that key early offerings in this arena include (in alphabetical order) 10x Genomics Xenium, Akoya Biosciences PhenoCycler-Fusion, Nanostring CosMx SMI, Rebus Biosystems Esper, Resolve Biosciences Molecular Cartography Platform, Veranome Biosystems VSA-1 and Vizgen Merscope.
Special considerations for challenging samples
Assay optimization is a necessity in spatial transcriptomic analysis given the structural and cellular complexity of heterogeneous tumor tissue. Tumor tissue can be soft, fibrous, fatty, necrotic (or some combination), where each of these features can dramatically impact RNA accessibility, transport, capture or labeling and can be encompassed within a single sample. Custom sectioning and permeabilization approaches can be used to optimize RNA capture or labeling from unique tissue provided they are tested first, to prevent reagent interference with downstream processing. As just one example, bone sections require fixation and decalcification prior to RNA retrieval, and optimization of these processes will yield better quality data. Moreover, while standard tissue section thicknesses are usually ~5-10 um, highly acellular tissues may require thicker sections, while hard-to-permeabilize tissues may requires thinner sections when using capture-based approaches like Visium or SlideSeq. Overall, though, a key point is to think critically about tissue structure and properties in order to develop the most robust methods for RNA recovery or labeling in a specific sample.
Approaches to bioinformatic analysis
While low resolution spatial analysis methods yield sequencing data and high resolution in situ methods yield imaging data, both types of analysis consist of 1) matrix generation, 2) cell annotation and 3) network/spatial analysis. Initially, raw data is aligned or mapped to a reference genome and used to generate large matrices that assign spatial coordinates to individual genes, noting the specific location of expression. Once mapped, colocalized gene expression patterns are typically used to annotate and identify specific cell populations. In some cases, canonical gene expression makes this task straightforward, but in other cases, leveraging databases or semi-automated annotation methods can be useful. Finally, more custom methods may be used to evaluate network changes (such as signaling pathway alterations in response to treatment) or transcriptional effects within neighborhoods of cells. There are a wide variety of computational approaches that can be leveraged in this tertiary analysis stage, including those targeted toward mapping cellular interactions and analyzing focused cellular neighborhoods17-20.
Potential translational and clinical applications
As already discussed, spatial transcriptomic approaches have indeed revealed fundamental biological insights about the tumor microenvironment. While this is a relatively new experimental approach, efforts are underway to extend its use into translational and clinical applications and several key use cases have emerged. As one example, spatial transcriptomic data can be used to build machine learning algorithms to better annotate diseased tissue. As one example, this was recently demonstrated in breast cancer biopsy tissue, to inform more confident diagnoses of ductal carcinoma in situ vs. invasive ductal carcinoma21. Similarly, this method was applied in prostate biopsy tissue, to better stratify disease diagnosis based on H&E stained sections22. Additionally, spatial transcriptomic approaches have been used to investigate heterogeneity in patient response to immunotherapy23. These applications likely represent the most early clinical efforts to better characterize heterogeneity in the tumor microenvironment, where wider adoption would be expected as the technique scales to accommodate statistically-meaningful patient sample numbers and the foundation of our published scientific knowledge at this resolution grows.
Upcoming innovations and future directions
The primary areas for innovation in this field are in adding multiomic analysis capabilities and extending analysis to three dimensions. While many of the aforementioned approaches are multiomic in the sense that they permit simultaneous analysis of RNA and a limited number of protein targets, future technology iterations that allow for addition of DNA mutation or modification state, chromatin organization or protein posttranslational modification detection would be very exciting and add substantial breadth to these approaches, where the scientific foundations of some of these methods are already in development within academic labs24,25. Similarly, extending these primarily two-dimensional approaches into three dimensions will permit scaling of spatial analysis to biologically relevant sample numbers; currently, evaluating an intact tissue in three dimensions requires substantial investment via analysis of many sequential two-dimensional sections. Hopefully, future technology iterations that allow for three-dimensional analysis of tissue would permit resources that are currently used to analyze sequential sections from the same tissue to be reallocated to analyzing samples from replicate patients, animals or experimental conditions.
Concluding thoughts
Spatial transcriptomics is an exciting, dynamic and rapidly developing field which has already led to impactful biological insights about the tumor microenvironment. In particular, the coming years should bring innovation via increased resolution of capture-based methods, additional multiomic capabilities and new bioinformatic resources to accelerate discovery and adoption. These approaches offer an entirely new perspective and resolution for understanding the tumor microenvironment which will hopefully translate to better science, diagnosis and treatment.
References
1. Rao, A., et al., Exploring tissue architecture using spatial transcriptomics. Nature,
2021. 596(7871): p. 211-220.
2. Guo, J.A., et al., Refining the Molecular Framework for Pancreatic Cancer with
Single-cell and Spatial Technologies. Clin Cancer Res, 2021. 27(14): p. 3825-3833.
3. Ahmed, R., et al., Single-Cell RNA Sequencing with Spatial Transcriptomics of
Cancer Tissues. Int J Mol Sci, 2022. 23(6).
4. Wang, N., et al., Spatial transcriptomics and proteomics technologies for
deconvoluting the tumor microenvironment. Biotechnol J, 2021. 16(9): p. e2100041.
5. Maniatis, S., J. Petrescu, and H. Phatnani, Spatially resolved transcriptomics and its applications in cancer. Curr Opin Genet Dev, 2021. 66: p. 70-77.
6. Ståhl, P.L., et al., Visualization and analysis of gene expression in tissue sections by spatial transcriptomics. Science, 2016. 353(6294): p. 78-82.
7. Brady, L., et al., Inter- and intra-tumor heterogeneity of metastatic prostate cancer determined by digital spatial gene expression profiling. Nat Commun, 2021. 12(1): p. 1426.
8. Wu, S.Z., et al., A single-cell and spatially resolved atlas of human breast cancers.
Nat Genet, 2021. 53(9): p. 1334-1347.
9. Andersson, A., et al., Spatial deconvolution of HER2-positive breast cancer
delineates tumor-associated cell type interactions. Nat Commun, 2021. 12(1):
p. 6012.
10. Berglund, E., et al., Spatial maps of prostate cancer transcriptomes reveal an unex plored landscape of heterogeneity. Nat Commun, 2018. 9(1): p. 2419.
11. Wu, Y., et al., Spatiotemporal Immune Landscape of Colorectal Cancer Liver
Metastasis at Single-Cell Level. Cancer Discov, 2022. 12(1): p. 134-153.
12. Sun, H., et al., Hypoxic microenvironment induced spatial transcriptome changes in pancreatic cancer. Cancer Biol Med, 2021. 18(2): p. 616-30.
13. Moncada, R., et al., Integrating microarray-based spatial transcriptomics and
single-cell RNA-seq reveals tissue architecture in pancreatic ductal adenocarcino mas. Nat Biotechnol, 2020. 38(3): p. 333-342.
14. Marshall, J.L., et al., High-resolution Slide-seqV2 spatial transcriptomics enables discovery of disease-specific cell neighborhoods and pathways. iScience, 2022. 25(4): p. 104097.
15. Wei, R., et al., Spatial charting of single-cell transcriptomes in tissues. Nat
Biotechnol, 2022.
16. Zhao, E., et al., Spatial transcriptomics at subspot resolution with BayesSpace. Nat Biotechnol, 2021. 39(11): p. 1375-1384.
17. Efremova, M., et al., CellPhoneDB: inferring cell-cell communication from
combined expression of multi-subunit ligand-receptor complexes. Nat Protoc, 2020. 15(4): p. 1484-1506.
18. Jin, S., et al., Inference and analysis of cell-cell communication using CellChat. Nat Commun, 2021. 12(1): p. 1088.
19. Browaeys, R., W. Saelens, and Y. Saeys, NicheNet: modeling intercellular
communication by linking ligands to target genes. Nat Methods, 2020. 17(2): p.
159-162.
20. Satija, R., et al., Spatial reconstruction of single-cell gene expression data. Nat
Biotechnol, 2015. 33(5): p. 495-502.
21. Yoosuf, N., et al., Identification and transfer of spatial transcriptomics signatures
for cancer diagnosis. Breast Cancer Res, 2020. 22(1): p. 6.
22. Chelebian, E., et al., Morphological Features Extracted by AI Associated with
Spatial Transcriptomics in Prostate Cancer. Cancers (Basel), 2021. 13(19).
23. Gohil, S.H., et al., Applying high-dimensional single-cell technologies to the
analysis of cancer immunotherapy. Nat Rev Clin Oncol, 2021. 18(4): p. 244-256.
24. Dhainaut, M., et al., Spatial CRISPR genomics identifies regulators of the tumor mi croenvironment. Cell, 2022. 185(7): p. 1223-1239.e20.
25. Deng, Y., et al., Spatial-CUT&Tag: Spatially resolved chromatin modification
profiling at the cellular level. Science, 2022. 375(6581): p. 681-686.
Kristin Beaumont, PhD is an Assistant Professor in the Department of Genetics and Genomic Sciences and the Associate Director of the Center for Advanced Genomic Technology at the Icahn School of Medicine at Mount Sinai. Specific focal points of her research include investigating tumor heterogeneity in gynecological cancers, as well as understanding the role of heterogeneity in metastatic progression of disease.