18 Introduction
18.1 Introduction to imaging-based ST
Imaging-based ST assays have gone from resolving 100s to 1000s of features. Their commercialization by Vizgen, NanoString, and 10x Genomics has made these data increasingly popular. In general, imaging-based ST platforms rely on multiplexed error-robust fluorescence in situ hybridization (FISH), as originally proposed by Chen et al. (2015). Briefly, targets are preassigned a barcode of (multi-color) โonโ and (no-color) โoffโ bits, whereas barcodes are optimized to differ from each other by some bits as to minimize readout errors. After iterative imaging-bleaching, individual reporter binding events are identified computationally (spot calling). Aligning spots along the image z-stack then allows individual targets to be identified based on a predefined barcoding scheme.
1,000-plex CosMx, for example, employs 64-bit barcodes (4 immunofluorescent reports over 16 cycles of imaging-bleaching) with a Hamming weight and distance of 4 (i.e., every target is โonโ in 4 rounds and โoffโ in 12 rounds; each barcode differs from all others by 4 bits as to minimize RNA decoding errors). Combinatorially, a 64-bit barcoding scheme is sufficient to encode a much larger number of targets, namely, ~20,000 protein-coding genes.
Commercially available panels can typically include three types of barcodes:
- RNA targets determine the โplexityโ of the panel.
- Negative probes serve to quantify non-specific binding.
- System controls serve to quantify spot calling errors.
Modeled after synthetic sequences from the External RNA Controls Consortium (ERCC), negative probes contain hybridization regions that are not complementary to the genome or transcriptome of the organism under study; their detection thus corresponds to non-specific ISH probe hybridization events (e.g., in โstickyโ regions of the tissue).
Depending on the barcoding scheme (number of bits, Hamming weight/distance), a number of barcodes will be left unoccupied; these system controls (also referred to as blank or false codes) serve to quantify misidentification of reporter readout.
The following chapters build on this platform overview. Chapter 6 describes common output files and Bioconductor import options, Chapter 19 discusses how molecule locations are assigned to cells, Chapter 20 covers cell-level and platform-specific quality control, and Chapter 25 and Chapter 26 provide complete example workflows. For image-derived features beyond transcript-based analysis, see Chapter 33.
18.2 Platforms
18.2.1 Xenium (10x Genomics)
The 10x Genomics Xenium platform employs rolling circle amplification (RCA) to boost fluorescent signals from reporter-binding events, thereby improving overall spot calling sensitivity.
Xenium gives an imageable area of 12 mm x 24 mm, although slightly less to avoid fiducial markers. Several pre-designed panels are available, including a โPan Tissue and Pathwaysโ for both mouse and human, which both include over 5000 genes and there is a possibility to custom design an additional 100 genes. More focused panels are available for human brain, lung, colon and breast tissue as well as cancer immunological subpopulations; for mouse, focused brain and multi-tissue panels are available.
18.2.2 CosMx (Bruker)
Unlike Xenium, CosMx (He et al. 2022) does not rely on RCA. As a result, CosMx tends to be less sensitive than Xenium (albeit higher-plex at present). Furthermore, segmentation is performed separately for each FOV (without stitching). This results in a variety of technical artifacts, such as fractured and possibly duplicated cells near FOV borders.
The CosMx platform gives an imageable area of 15x20 mm and allows thousands of RNAs and dozens of proteins to be assayed in a spatial context. Available panels include broad discovery panels as well as targeted panels for applications such as cell characterization, immuno-oncology, and neuroscience. In addition, custom panels are available in add-on or standalone formats, and antibody-based protein panels are available for selected biological contexts.
18.2.3 MERSCOPE (Vizgen)
The Vizgen MERSCOPE platform, among the fluorescence-based targeted spatial transcriptomics platforms, allows the highest resolution (โค20 nm), while also providing a high capacity for multiplexing, with (customizable) panels containing up to 1000 genes, and an imageable area of 1 cm2. MERSCOPE was recently shown to be among the most sensitive in a recent head-to-head comparison (Hartman and Satija 2024) using mouse brain tissue.