Spatially resolved transcriptomics and its applications in cancer

Abstract

Spatially resolved transcriptomics (SRT) offers the promise of understanding cells and their modes of dysfunction in the context of intact tissues. Technologies for SRT have advanced rapidly with a large number being published in recent years. Diverse methods for SRT produce data at widely varying depth, throughput, accessibility and cost. Many published SRT methods have been demonstrated only in their labs of origin, while others have matured to the point of commercialization and widespread availability. Here we review technologies for SRT, and their application in studies of tumor heterogeneity.

Introduction

Recent technological advances provide the ability to interrogate tissues and even whole organisms at ‘omics’ scale and with cellular resolution. Single cell RNA sequencing (scRNAseq) studies of hematologic malignancies have yielded an unprecedented view of cellular dynamics driving these diseases and their responses to therapy[1–3]. However, application of scRNAseq methods to solid tumors requires tissue dissociation and therefore loss of spatial context[4–6]. As reviewed, spatial heterogeneity in tumors is driven not only by genotypic diversity that arises during clonal expansion but also by interactions between cancer cells and the immune and stromal cells that comprise the local tumor microenvironment, leading to distinct phenotypes in different regions of a tumor[7]. Understanding the spatial heterogeneity of tumors is clinically consequential, especially in instances where limited physical sampling of tumors is used to guide treatment [8].

SRT encompasses a diverse set of technologies that encode gene expression measurements with information about where in the sample each observation occurred. How this spatial encoding is achieved dictates the capabilities of each technology, including spatial resolution, depth of data per measurement, throughput, cost, and complexity. SRT technologies can require preselection of a panel of transcripts to be profiled, or sample the entire polyadenylated transcriptome. Some SRT methods require costly bespoke equipment and substantial technical support, while others are available as commercial kits requiring almost no specialized equipment. Careful consideration of these properties is required for selection of SRT technologies.

Although this domain of technologies is diverse and rapidly growing, major themes have emerged. Here, we survey currently available SRT technologies for studies of tumor samples, dividing them into 3 major categories based upon how each technology encodes spatial information: 1) In microdissection, light microscopy is used to select the location and shape of each tissue region collected for subsequent ex situ processing; 2) In situ barcoding based methods attach specific DNA barcode sequences to known regions of intact tissue samples and use codetection of those barcodes with tissue derived RNAs during subsequent ex situ sequencing to computationally assign spatial information to expression data; 3) Imaging based methods simultaneously acquire spatial and gene expression information through iterative cycles of fluorescent nucleic acid imaging in intact tissue samples. The resulting images are assembled into a spatially aligned dataset that spans all cycles, and expression data is decoded based on the presence or absence of signal in each pixel for each cycle. Here we describe SRT technologies from each category, summarize their application to cancer research to date, and anticipate future developments.

1: Microdissection based methods

Microdissection and capture of tissue regions or cells of interest is a well-established and robust method for preservation of spatial information during sample acquisition. In laser capture microdissection (LCM), tissue is sectioned onto specially prepared glass slides, stained and imaged for morphology, and regions of interest as small as single cells are dissected using a microscope-guided laser[9]. LCM is compatible with cryopreserved as well as formalin fixed paraffin embedded (FFPE) material. Microdissected samples are amenable to many low input sequencing methods. Equipment for LCM is commercially available and accessible in many core facilities. However, the complexity of sample preparation, collection, and downstream processing limits the potential throughput of LCM.

2: In situ barcoding based methods

Several methods for spatial RNA quantification using DNA barcoding have been developed. These can be subcategorized broadly based on how barcoding is performed. In one group of technolo gies, referred to here as solid phase based capture (SPBC) methods, tissue is delivered to a substrate bearing a pre -arranged set of DNA barcodes. In the second group of technologies, termed selective barcoding (SB) methods, DNA barcodes are either delivered to or collected from selected tissue locations. In situ barcoding technologies from both technology groups have been successfully commercialized.

Solid phase based capture

Spatial Transcriptomics / 10X Visium

In Spatial Transcriptomics (ST) and 10X Visium, tissue is sectioned onto a glass slide bearing an array of spatially barcoded DNA oligonucleotide capture probes arranged in a grid of discrete capture areas. The tissue section is first fixed, stained, and imaged. Gentle permeabilization allows RNAs to diffuse and polyadenylated mRNAs to hybridize to the poly(T) sequence in capture probes. This poly(T) sequence then serves as the primer for reverse transcription, generating cDNA molecules that harbor the spatial barcodes and unique molecular identifiers in the capture probes. The resulting cDNA library is then collected and sequenced via standard Illumina workflows. RNAseq counts are mapped to specific tissue locations by aligning the histological image acquired at the beginning of the workflow to known spatial barcode locations. In ST, spatial resolution is limited by the 100 um diameter and 200 um spacing of capture areas[10,11]. 10X Genomics acquired ST in 2018 and has improved the method by reducing the capture area size from 100um to 55um in diameter, increasing capture area density by hexagonal packing, increasing the number of molecules detected per capture area, and decreasing workflow duration. While the spatial resolution of ST and 10X Visium is insufficient to resolve single cells in most samples, the histological image acquired as part of the workflow enables computational inference of spatial expression at resolutions finer than the physical capture areas [12]. High tissue section throughput and whole polyadenylated transcriptome profiling are distinct advantages of ST and 10X Visium, enabling large scale and unbiased studies of spatially resolved gene expression. These methods are also compatible with FFPE samples[13]. ST and 10X Visium have been used for whole organ atlassing as well as studies of temporal, genetic, anatomic and pathologic variables affecting spatial gene expression[14–17].

Slide-Seq and High Definition Spatial Transcriptomics (HDST)

Two recent methods use beads harboring barcoded DNA oligonucleotide probes to substantially improve the spatial resolution of surface based capture methods. In Slide-Seq, tissue sections are placed onto an emulsion of DropSeq beads for polyadenylated RNA capture and barcoding with 10um resolution, as demonstrated in mouse brain[18,19]. Although its approach to barcoding and subsequent sequencing is similar to ST/Visium, Slide-Seq is currently incompatible with histological imaging and therefore uses sequence data alone to reconstruct tissue morphology. Gene expression data is mapped using known spatial locations of uniquely barcoded beads in each puck. This is accomplished by imaging-based SOLiD sequencing of Slide-Seq pucks prior to tissue sectioning, a step which requires the use of a bespoke fluidics-coupled microscope. HDST uses an Illumina bead array to achieve SPBC with 2um resolution[20]. The HDST workflow is similar to that of ST and 10X Visium, with tissue sectioned directly onto the Illumina bead array, stained, and imaged for downstream localization of sequencing data. The resolution of HDST is so fine that individual beads are binned for single-cell SRT analysis. A substantial fraction of beads produce few reads, sometimes due to being located between cells.

Selective barcoding

Nanostring GeoMX

The GeoMx Digital Spatial Profiling (DSP) system enables multiplex profiling of RNAs or proteins in situ using hybridization of a preselected panel of oligonucleotide probes or oligonucleotide labeled antibodies [21]. These RNA detection probes or antibody tags are then photocleaved from selected regions of the tissue section, collected, and quantified using a NanoString instrument or standard next generation sequencing[21]. Regions of interest can be selected on the basis of tissue morphology or arbitrarily arranged in a grid with resolution as fine as 10um[22].

ZipSeq

ZipSeq uses photo-uncageable oligonucleotides coupled to antibodies or lipids as anchors to selectively barcode cells for subsequent scRNAseq analysis[23]. Samples are first exposed to caged anchors and regions of interest are then subjected to selective photo-uncaging. After photo-uncaging, anchors become available for hybridization to barcoded ‘readout’ oligonucleotides containing an Illumina sequencing handle and a poly(A) sequence. Additional regions of interest can be encoded iteratively in this way. When barcoding is complete, the sample is dissociated and processed for scRNAseq. The poly(A) sequence allows readout oligonucleotides to be captured alongside mRNAs from labeled cells during scRNAseq. Thus, cells and their mRNAs can be localized to a specific region of interest in a sample using a combination of ZipSeq and scRNAseq barco des. This versatile method can be applied to a variety of sample types, including live cells. However, ZipSeq requires costly reagents and bespoke imaging systems, potentially limiting its user base.

3: Imaging based methods

Imaging based SRT methods enable acquisition of data with submicron resolution and are thus particularly well suited for studies of processes occurring at subcellular scales. This resolution comes at the expense of complexity, time, and equipment cost. The sensitivity of imaging based methods is limited by the ability of optical microscopy to resolve densely packed RNAs in situ. Methodological advancements to overcome this challenge result in decreased throughput[24–26]. In general, imaging based methods can be subcategorized into those based upon fluorescence in situ hybridization (FISH) and those that use in situ sequencing (ISS).

Fluorescent In Situ Hybridization (FISH)

Multiplexed iterative FISH is a well-established approach to SRT[24,27–29]. MERFISH, seqFISH+, SABER-FISH and osmFISH use various approaches to overcome the challenge of optically resolving densely packed molecules. These techniques generate single molecule resolution maps of dozens to thousands of preselected RNAs. All remain bespoke methods, though MERFISH is currently being marketed as a commercial product by Vizgen. Due to their technical complexity and the additional challenges of processing and imaging human tissues, the use of FISH based SRT methods has thus far been restricted to studies in cell culture and animal models.

In Situ Sequencing (ISS)

SRT analysis can be achieved by performing sequencing reactions in situ. Several detection strategies and sequencing chemistries for ISS have been successfully demonstrated in the form of FISSEQ, padlock probe based in situ sequencing (PPBISS), and STARMAP[30–32]. All of these methods perform rolling circle amplification (RCA) of target sequences to obtain a strong enough fluorescent signal for reliable sequencing. RCA is difficult to control and generates micron scale amplicons, hindering the sensitivity of ISS methods. Most ISS methods require preselection of target RNAs to be profiled, with throughput ranging from dozens to thousands of RNAs. The maximum number of RNA molecules profiled by each method is limited by the detection strategy used and the short (<20nt) read length reliably sequenceable in situ. Two recent unpublished methods, INSTA-Seq and Expansion sequencing (ExSeq) overcome such limitations by combining ISS with ex situ sequencing[26,33]. FISSEQ reagents and instrumentation have been commercialized by ReadCoor, with specific products for applications in oncology[30]. Similarly, Cartana offers a PPBISS profiling service as well as reagent kits that enable ISS of cryopreserved and FFPE tissues using a standard epifluorescence microscope.

Applications in cancer

SRT methods enable profiling the spatial heterogeneity of cell types and cell states in intact tumor samples. Microdissection based methods have long been used to study the clonal composition of tumors and the cell-type composition of the tumor microenvironment with spatial resolution, but the labor intensive process of dissecting tissue regions of interest has limited the throughput of these methods[34–36]. Solid phase capture based methods (ST and 10X Visium) have also been used to profile the composition and spatial heterogeneity of the tumor microenvironment across tumor types[37–41]. These methods have higher throughput and are suited for translational applications due to their workflow, which combines histopathology and standard NGS methods. Coupled histological and SRT data generated by ST has been used to train machine learning algorithms to predict histopathological annotations based on gene expression data[42] as well as local gene expression based on histopathologic features[43] in breast cancer. Computational developments have also enabled copy number inference and fusion transcript identification in ST data[39,44]. With the improved spatial resolution of bead-based capture and barcoding, HDST has been used to map gene expression with single cell fidelity in a cryopreserved Her2+ breast cancer sample[20]. While solid phase based capture methods enable unbiased sampling of the entire polyadenylated transcriptome with spatial resolution, additional SRT methods confer their own distinct advantages. The Nanostring GeoMX DSP system, for instance, can detect both oligonucleotide RNA probes and oligonucleotide labeled antibodies. It has been used to profile the spatial distribution of 156 proteins in FFPE non-small cell lung cancer tissues in order to identify biomarkers predictive of outcome from PD-1 checkpoint blockade treatment[45] as well as to map the spatial distribution of distinct cancer-associated fibroblast programs and their proximity to immune infiltrates in pancreatic ductal adenocarcinoma[46]. Imaging based methods (Cartana PPBISS, ExSeq) enable mapping pre-selected markers in intact tumor samples with subcellular resolution [23,26,47]. Although not yet demonstrated in intact tumor samples, MERFISH has been used in conjunction with immunolabeling of an endoplasmic reticulum (ER) receptor and nuclear staining to identify RNAs enriched in specific cellular compartments in a human osteosarcoma cell line (U2OS)[48]. Taken together, these studies demonstrate the power of SRT methods for probing the composition and spatial heterogeneity of the tumor microenvironment with various throughputs and across scales.

Future directions

The emergence of commercialized ISS and SPBC technologies for SRT profiling will undoubtedly result in widespread adoption of these methods for research purposes. In the near term, SPBC appears better suited to clinical applications due to its relative simplicity, low barrier to adoption, similarity of workflows to clinical pathology techniques, and compatibility with histological imaging. Indeed, core facilities offering 10X Visium as a service have been established at several universities and hospitals. Currently unpublished studies have demonstrated the utility of combined SRT and ex situ sequencing approaches to increase the throughput of ISS and to profile splicing isoforms with spatial resolution[26,33,49,50]. Such extensions of existing SRT methods will likely continue to add new capabilities to these technology platforms. The recent acquisition of Cartana and Readcoor by 10X Genomics, and Vizgen’s commercialization of MERFISH also indicates that imaging based methods are maturing and likely to become a viable option for non-specialists. Innovation in experimental design and data analysis also offer new possibilities. Combinations of various SRT methods and scRNAseq analyses have enabled deep characterization of cell type composition and local cell states in multiple tissue types[14,27,29,39]. As the catalog of published SRT and scRNAseq datasets expands, including through large scale tissue atlassing efforts, such study designs will continue to be commonly employed. Further, accessible computational methods such as Seurat enable alignment of data between experiments and across SRT and single-cell technology platforms, including from multimodal methods such as CITEseq[51,52]. Clever study designs that combine SRT with mass spectrometry[53–56] or high dimensional immunofluorescence[21,57–60] methods can also provide a multimodal view of tissue heterogeneity, encompassing metabolomics and posttranslational modifications. The rapid pace of progress of SRT and associated methods will soon offer a view of cancer biology that spans the central dogma in the context of the intact tumor microenvironment. Such a view will provide insights into the extent and origins of tumor heterogeneity, informing targeted diagnostics and treatment.

Figure 1:

Comparison of SRT technologies. Y axis represents the demonstrated number of genes detected per experiment, while the X axis represents the number of tissue sections that can be processed in a given amount of time. The marker size represents the minimum resolvable feature size. Marker coloration indicates the relative amount of information each technology generates in a given amount of time. This qualitative measure takes into account experimental throughput, total data acquisition area per experiment, number of sampling areas per experiment, spatial resolution, and number of genes observed.

Table 1:

Comparison of SRT technologies.

Technology	Category	Resolution	Genes Detected	Sampling	Demonstrated Or Marketed In Cancer	Advantages	Disadvantages
LCM	Microdissection	Cellular	10000	Transcriptome Wide	Yes	Well established	Complex, low throughput
Spatial Transcriptomics	SPBC	Many Cells	12000	Transcriptome Wide	Yes	Well established, high experimental throughput	Low resolution
10X Visium	SPBC	Few Cells	22000	Transcriptome Wide	Yes	High sample throughput, commercial support	Moderate resolution
Slide-Seq	SPBC	Cellular	20000	Trai jcriytrme Wide ‘	No	Cellular resolution, many genes detected	Bespoke technology
HDST	SPBC	Subcellular	17 481	Transcriptome Wide	Yes	Subcellular resolution, high throughput	Dependence on custom Illumina bead array
GeoMX DSP	In Situ Barcoding	Few Cells	200	Targeted	Yes	Antibody and RNA detection, commercial support	Few genes detected, few regions sampled
ZipSeq	In Situ Barcoding	Cellula (via scRNAseq)	23000	Transcriptome Wide	Yes	Can be used on living cells, cellular resolution	Bespoke technology, costly reagents, requires scRNAseq
MERFISH	FISH	Super-resolution	10000	Targeted	Yes	Super-resolution, many genes detectible	Bespoke technology, low experimental throughput
seqFISH+	FISH	Super-resolution	10000	Targeted	No	Super-resolution, many genes detectible	Bespoke technology, low experimental throughput
osmFISH	FISH	Subcellular	33	Targeted	No	Subcellular resolution	Few genes detected, low sample throughput
FISSEQ	ISS	Subcellular	8102	Transcriptome Wide	Yes	Subcellular resolution, many genes detected	Bespoke technology, low experimental throughput
PPBISS	ISS	Subcellular	100	Targeted	Yes	Subcellular resolution, many genes detected	Bespoke technology, low experimental throughput, few genes detected
STARmap	ISS	Subcellular	1020	Targeted	No	Subcellular resolution, many genes detected	Bespoke technology, low experimental throughput
ReadCoor	ISS	Subcellular	250	Targeted	Yes	Subcellular resolution, many genes detected, commercial support	Bespoke technology, low experimental throughput, few genes detected
Cartana	ISS	Subcellular	100	Targeted	Yes	Subcellular resolution, many genes detected, commercial support	Bespoke technology, low experimental throughput, few genes detected
INSTA-Seq	ISS	Subcellular	820	Transcriptome Wide	No	Super-resolution, many genes detectible, long reads for imaging based technology	Bespoke technology, low experimental throughput, ex situ sequencing also required
ExSeq	ISS	Super-resolution	23000	Transcriptome Wide	Yes	Super-resolution, many genes detectible, long reads for imaging based technology, antibody detection possible	Bespoke technology, low experimental throughput, ex situ sequencing also required