NEW YORK – Researchers from the University of Washington have devised polymerase colony (polony) arrays with pixel-like barcoding features for use in spatial transcriptomics.
The method, called polony-indexed library-sequencing (Pixel-seq), grows polonies, a type of DNA cluster, into barcoded oligo arrays on polyacrylamide gels. Using a sequencing-by-synthesis chemistry and a bespoke instrument, the barcodes are then sequenced to create a spatially indexed map. Then, cryosectioned tissues are placed on the gels, RNA transcripts are converted to cDNA, and those cDNAs are barcoded with the oligo sequences to create an Illumina next-generation sequencing library. Finally, the cDNA reads are mapped back to their positions on the polony gels, providing information on spatial organization.
The researchers, led by Liangcai Gu of the University of Washington, described their method last month in a BioRxiv preprint. In addition to an average feature size of approximately 1 µm, roughly translating to spatial resolution, Pixel-seq offers RNA sensitivity of more than 20,000 unique molecular identifiers (UMIs) in bins of pixels approximating 50 µm features and about 12,000 UMIs for 10 µm.
This level of resolution is a "big improvement" over many existing spatial transcriptomics methods, said George Church, a researcher at the Wyss Institute and a professor of genetics at Harvard Medical School who has done pioneering work on polonies and who helped launch the in situ sequencing startup ReadCoor, acquired by 10x Genomics last year. Gu was previously a postdoc in Church's lab. "Most spatial transcriptomics is low resolution, with many cells with the same barcode," he said. "For a lot of cells, their neighbors are important. A red cell next to a blue cell is not a purple cell."
Data from a single pixel would not be worth much on its own, but "1 µm [resolution] is helpful because it lets you select which features to combine into a cell's worth of RNA," said Andrew Adey, a researcher at Oregon Health & Science University who has developed sequencing-based spatial analysis methods and who was not involved in the preprint. With a 10 µm feature, as provided by Slide-seq, or 50 µm, offered by 10x's Visium Spatial Genomics platform, "you're much more likely to have features that fall across two or more cells," Adey said.
What's more is that the RNA capture is efficient enough that it could get researchers to start thinking about eschewing single-cell RNA sequencing (scRNA-seq) for some studies. Gu said his team has plans to commercialize Pixel-seq once the researchers have established the method in a peer-reviewed paper.
Polonies have been around for decades and their applications in genomics have been significant. Church's lab developed a polony sequencing method in the late 1990s and early 2000s, which for a short while was offered commercially as the Polonator sequencer from Dover, launched in 2009.
That was the same year that Gu started as a postdoc in Church's lab, working with polonies for barcoding and even some spatial analysis. "I learned how to grow these clusters for many years and when I started my lab, I set up the technology to barcode proteins and determine protein interactions," Gu said.
Work on Pixel-seq started five years ago. "The idea is pretty simple, but it took us a while to realize it," Gu said. It shares concepts with several other methods that have come out in recent years, namely Slide-seq, which uses bead arrays to barcode tissue slides, and fluorescence in situ sequencing, also developed in the Church lab. Church lauded Gu's nods to earlier work, including his use of the term "polony."
"This is actually pretty radically different" from using polonies for sequencing, Church said. "He could have just called it in situ transcriptomics."
The key to Pixel-seq, Gu said, is fabricating the gels. "These DNA clusters, they sometimes do pretty weird things going in the gel. You really need to get the right gel structure to get an efficient conversion into a high-quality oligo array." These gels represent the first achievement of continuous, evenly distributed features, he added.
Pixel-seq uses a crosslinked gel substrate, which offers several advantages, Gu said, namely limited RNA diffusion, even compared to a linear gel, and high oligo accessibility. Pixel-seq offers an oligo density of 17,400 molecules per square µm and an average of 85 percent of the substrate surface covered by barcodes. For comparison, Visium's oligo density is about 25,000 molecules per square µm, but only within its barcoded spots, and tissue coverage is about 28 percent. Pixel-seq can get even higher coverage by increasing the template concentration and using a higher resolution sequencing image setup, the authors noted.
To create the barcode array, which Gu said is similar to the random clusters generated for Illumina's HiSeq and MiSeq flow cells, the researchers seeded the gel with the oligos and amplified them into clusters. The 1 µm feature size is due to a happy quirk: when the expanding clusters start to run into each other, they stop and make a division, Church said. He called this phenomenon the "polony exclusion principle," in a nod to the quantum mechanical phenomenon that two electrons cannot occupy the same state in the same system. "It's not something we predicted exactly, but it was among the possibilities we hoped for when we saw our first experiments," Church said.
Pixel-seq will join a crowding field of spatial transcriptome analysis methods. So far, the available techniques offer various levels of RNA capture, multiplexing, and spatial resolution. Transcriptome-wide methods like Visium or NanoString's GeoMx Digital Spatial Profiler have proved popular, while those companies, and others, are also working on more targeted, but still multiplexed, technologies that claim to offer resolution near the physical limit of microscopy.
The new method's closest comparator is probably high-definition spatial transcriptomics (HDST), from Joakim Lundeberg's group at Sweden's Science for Life Laboratory, which is being commercialized as Visium HD by 10x Genomics. When published in late 2019, HDST offered resolution down to 2 µm; however, a 10x spokesperson said Visium HD will have spots approximately 5 µm in diameter with 5 µm gaps between spots, with an oligo density similar to Visium. Pixel-seq, Visium, and even Slide-seq use barcodes and sequencing to map reads to a tissue section.
In proof-of-principle studies, Gu's team analyzed mouse accessory olfactory bulb brain sections, generating data for approximately 23,000 unique genes, and 78 percent of genes mapped to protein coding regions. They reported an average of approximately 1,200 UMIs when binning pixels to get 10 µm resolution. At the same resolution, they said HDST would capture 12 UMIs, and Slide-seq V2 would capture 494. Binning pixels at 50 µm, they captured about 25,000 UMIs, compared to 15,377 for Visium, with 55 µm spots.
This RNA capture sensitivity "is pretty close to single-cell RNA-seq levels from solid, primary tissues," Adey said. Indeed, Gu hopes Pixel-seq can be more than just a mapping tool.
"Right now, for all this tissue mapping, people have to first do scRNA-seq or a single-cell assay for transposase-accessible chromatin by sequencing (ATAC-seq) and then they generate a probe set for imaging-based analysis," Gu said. "We hope we can integrate these two steps into one and people can just directly do Pixel-seq."
Church noted that Pixel-seq, like many spatial RNA analysis methods, only offers spatial information in two dimensions. "However thick your section is, the Z-axis gets collapsed," he said. "Actual in situ analysis can do really thick specimens and get high resolution on the Z-axis."
Adey said he wished Gu's team had tried to aggregate features into single-cell profiles and integrated them with scRNA-seq data. "That is really a major goal of these spatial assays," he said.
"Single-cell segmentation of our data is exactly what we are doing [now] and we fully understand its importance," Gu said, adding that his team is putting together a manuscript for peer review that will also include a biological story. Once they accomplish that, they plan to commercialize Pixel-seq. The authors noted they have applied for a provisional patent covering their technology and the team is in talks with CoMotion, the University of Washington's licensing office, Gu said.
Church said he is not involved with any commercialization efforts for Pixel-seq at this time. The two are coinventors of a patent, US Patent No. 10,858,698, issued in December 2020, for a method of attaching barcodes to polypeptides and detecting molecular interactions. That method uses immobilized barcoded proteins in a polyacrylamide gel.
Both Gu and Church noted that Pixel-seq had applications beyond RNA analysis. "Proteomics is the most obvious," Church said, "but it could be anything you can tag with a nucleic acid."