This story has been updated to clarify the per-cell costs of scNT-seq and sci-fate.
NEW YORK – Researchers from the University of Pennsylvania have described a new droplet-based single-cell sequencing method for identifying newly transcribed RNAs.
Built on the Drop-seq single-cell analysis platform, the method uses a chemical reaction to mark new messenger RNAs (mRNAs) with T to C substitutions. In a paper published today in Nature Methods, the investigators reported that they were able to profile transcriptomes in approximately 55,000 cells, delineating between new and old transcripts. The method, dubbed single-cell metabolically labeled new RNA tagging sequencing (scNT-seq), allowed them to analyze both RNA biogenesis and decay and helped resolve cell-state trajectories.
Their technique joins several plate-based methods for analyzing gene expression dynamics in single cells and is the second method published this year to greatly expand the scale on which these experiments can be done, to tens of thousands of cells.
In April, researchers associated with Jay Shendure's lab at the University of Washington described a similar sequencing method based on combinatorial indexing. In a study published in Nature Biotechnology, the authors used single-cell combinatorial indexing and messenger RNA labeling (Sci-fate) to analyze about 6,000 cells. The method has since been used to analyze about 10,00 cells, according to Junyue Cao, the lead author of that study, now a professor at Rockefeller University, and could be pushed to analyze as many as 2 million cells.
These developments "dramatically improve the utility of single-cell RNA-seq by adding a temporal dimension," making it possible to see the temporal RNA dynamics of particular genes, said Hao Wu, a professor at the University of Pennsylvania and senior author of the new paper published in Nature Methods. "Instead of looking at a snapshot of the total mRNAs, you can infer the future behavior of a particular cell."
"It suggests that you can simultaneously identify dynamic and steady-state genetic regulatory networks," he said. "You can use old versus new RNA information to infer the future state of a cell on a timescale from minutes to hours."
The improved scale dramatically lowers costs. For scNT-seq, the cost per-cell, including sequencing, is about $.50 and about $.10 for library prep alone; sci-fate, including sequencing, is about $.10 per cell and about $.05 per cell for library prep costs, but is highly dependent on sequencing depth, Cao said, and could be pushed even lower.
And both methods play well with others: Cao said sci-fate can be layered with other combinatorial indexing-based assays developed by the Shendure lab and its collaborators, including methods to analyze accessible chromatin or methylation. Wu's paper noted that scNT-seq could be paired with genome editing or chemical perturbation screens.
So called time-resolved RNA analysis goes back to the advent of RNA-seq. The simplest way is to stagger analyses so that they capture gene expression at different timepoints. Another popular method used in bulk RNA-seq assays is metabolic conversion, which is employed in both scNT-seq and sci-fate. This approach uses nucleotide analogs to tag newly created transcripts and differentiate them from previously existing ones. Here, that's 4sU, a thymidine analog that results in a T-to-C base flip that happens after a chemical conversion.
Several Europe-based labs have developed plate-based single-cell methods with throughput in the hundreds to thousands of cells, including single-cell thiol-linked alkylation for the metabolic sequencing of RNA (scSLAM-seq), new transcriptome alkylation-dependent single-cell RNA sequencing (NASC-seq), and 5-ethynyl-uridine in single cells sequencing (scEU-seq.) According to the sci-fate paper, scSLAM-seq and NASC-seq cost approximately $11 per cell for library prep.
Incoporporating chemical conversion is a major hurdle that has so far prevented time-resolved RNA-seq from appearing on droplet-based single-cell analysis platforms like 10x Genomics' Chromium and 1Cellbio's platform, based on the inDrop method.
"It's not possible to pretreat these cells without fixation prior to encapsulation on the droplet-based systems so we must perform this conversion after you capture these RNAs," Wu said. "You can't do this conversion after reverse transcription, making it challenging for other microfluidic [single-cell sequencing] platforms, where capture and reverse transcription happens almost simultaneously."
"We only perform the reverse transcription after the droplet breakage, which provides a perfect window of opportunity for on-bead chemical conversion," he said.
The sci-fate team also noted in their paper that in situ 4sU conversion "requires cell permeabilization and, at least in our experience, paraformaldehyde fixation, neither of which is straightforward to introduce on droplet-based single-cell RNA-sequencing platforms such as 10x Genomics."
Another existing approach is to order transcriptome snapshots with algorithms, generally called "pseudotime" methods. But these are best guesses. "It's very important to actually measure how a cell transitions from one state to another," Wu said.
Sci-fate and scNT-seq enable researchers to use RNA velocity, a quantitative measurement of gene expression dynamics(defined as the time derivative of gene expression) that accounts for both RNA biogenesis and degradation. Previous methods couldn't tell whether a high transcript count was due to a high transcription rate or a high level of transcript stability. "Our methods give you this extra level of information so you can tell whether RNA abundance is due to high transcription rate alone or high stability alone, or a combination of both," Wu said.
Sci-fate continues to use an algorithmic approach to calculate RNA degradation. "I think this is a very interesting direction," Cao said.
The research applications include measuring neuronal activity and tracking cell state transitions. Stem cells and organoids are types of cells where the latter application would be useful, Wu said.
Cao added that researchers could use these methods to study how drugs or the environment change gene synthesis and degradation rates in different cell types. "We have very limited knowledge about that," he said. Both methods incorporate unique molecular identifiers (UMIs) to track transcripts. With sci-fate, barcodes could be added prior to treatment, to track which drug a cell was exposed to.
In his newly established lab at Rockefeller, Cao will be looking at cell state trajectories, as well as ways to improve the method. Ultimately, he wants to be able to label cells in vivo, but to do this will require improvements both in chemical conversion efficiency and optimizing combinatorial indexing.
Wu noted that 4sU incorporation is not fully efficient and thus an area for improvement. And bioinformaticians, such as his coauthor Xiaojie Qiu, now at the Whitehead Institute, are still coming up with data analysis methods for incorporating metabolic RNA labeling information into cell state transition analysis at the single-cell elvel. Wu noted that scNT-seq could also work with a second independent chemical conversion using 6-thioguanine, which would lead to a G-to-A conversion, enabling measurement at two time points in a single cell.
Whether the respective technologies will be commercialized in unclear. Cao said combinatorial indexing technology has already been patented; Wu said the intellectual property landscape may not permit patenting the scNT-seq workflow or underlying technology.
"We'll explore the translational or commercial value by identifying biomarkers for therapeutic targets in screens," Wu said.