
NEW YORK – A new method for analyzing gene expression and proteins on tissue slides offers high spatial resolution and sensitivity.
Called deterministic barcoding in tissue sequencing (DBiT-seq), the method uses microfluidic channels to deliver oligonucleotide barcodes as grid-like coordinates on tissue slides, creating so-called "pixels" to help with later spatial analysis. The approach, developed by researchers in Rong Fan's lab at Yale University, offers 10 micrometer spatial resolution and can detect approximately 15 percent of molecules found by single-molecule fluorescence in situ hybridization (smFISH.) For comparison, single-cell RNA sequencing can usually detect between 10 and 50 percent of transcripts in a cell, depending on the method.
In a Cell paper published last week, the researchers said they could, in aggregate, detect 22,969 different genes from across the transcriptome as well as a panel of 22 proteins, using DNA-conjugated antibodies. Moreover, the method captured an average of 2,068 genes per pixel.
"DBiT-seq provides a further step in the evolution of NGS-based spatial transcriptomics," Joakim Lundeberg, a professor at the Science for Life Laboratory in Sweden who developed the spatial transcriptomics and high-definition spatial transcriptomics (HDST) methods, said in an email. "The value of increased granularity and the value of combining modalities are clear from previous work in the single-cell RNA seq world and here the spatial context adds an additional flavor to the understanding of the tissue ecosystem. Overall, this important work should propel the efforts to develop more complete sets of barcoded antibodies for detailed studies of the proteome in situ using next-generation sequencing as readout."
Fan's team has extended the method to work with formaldehyde-fixed tissues slides, detailed in a BioRxiv preprint posted in August. Fan said they have also increased the number of proteins that can be analyzed to more than 500.
"We are interested in pursuing the commercialization of DBiT-seq to democratize it for the broad research community," Fan said, but declined to comment on specifics. He noted that his lab is providing the design files for the microfluidic device as well as the DNA barcode sequences for free to academic labs. The method costs about $150 per sample, much of which goes towards Illumina's Nextera NGS library preparation.
Fan and two of the other lead authors are inventors on a patent application related to the paper. In addition, Fan is a cofounder of Branford, Connecticut-based Isoplexis, which offers the benchtop IsoLight and IsoSpark instruments for analyzing cell proteomes, metabolomes, and secretomes.
The key to DBiT-seq is deterministic barcoding, as opposed to random barcoding. Fan said he came up with the scheme to choose where the barcodes go in space about five years ago and was finally able to put a team together to test out the idea a few years ago.
To start, a microfluidic chip lays down one set of barcodes in parallel lanes that anneal to mRNAs and initiate in situ reverse transcription, creating barcoded cDNAs. This is done again orthogonally, creating "pixels" at the intersections of each lane that contain a distinct combination of barcodes. After the pixels are imaged under a microscope, the cDNAs are recovered, template switched, amplified using PCR, and prepared for paired-end sequencing. To detect proteins, the DNA-antibody conjugates are applied first.
"Everything is compatible with imaging," Fan added.
DBiT-seq joins several sequencing-based spatial gene expression methods, including the 10x Genomics Visium platform, which is based on the spatial transcriptomics method developed at Sweden's Science for Life Laboratory; HDST, a barcoded bead array-based method released last year by SciLifeLab and the Broad Institute; and Slide-Seq, developed at the Broad.
DBiT-seq has two main advantages, Fan said: resolution of 10 micrometers and the number of genes captured in each pixel — more than 2,000. Visium's spots are 55 micrometers across and it captures more than 1,000 genes per spot. High-definition spatial transcriptomics claims resolution of 2 micrometers and a capture efficiency of only 1.3 percent, compared to smFISH; in proof-of-concept studies, researchers were able to identify 186 nucleus-specific genes. Slide-seq offers similar resolution, down to about 10 micrometers, but only captures about 150 genes, Fan said.
With new advances, DBiT-seq will also be able to use more types of histopathology slides, which can be stored for years.
The method also appears to be easy to implement, said Andrew Adey, a professor at Oregon Health & Science University who has developed methods bringing single-cell and spatial analysis closer together, including single-cell combinatorial indexing from microbiopsies with assigned positions, or sciMAP. The chips are the most difficult of the materials to obtain, but aren't hard to fabricate for those who know how, he said, and aside from the DNA barcodes, also easily obtained, other materials are available off the shelf. At about $150 per sample, "that's a steal compared to any commercial platforms," he said.
Fan's lab is part of the NIH Human BioMolecular Atlas Program consortium, under which his team will apply this technology to map human heart tissues at the cellular level. They're also working with colleagues at Yale to apply DBiT-seq to brain tumors and lymphoma tissues, among other tumors.
The new method isn't without limitations, though. "It cannot select, exactly, individual cells to perform spatial sequencing, and some of the square pixels of the 2D grid may cover one cell, two cells, or half a cell," Fan said. But deconvolution algorithms may help separate genes and assign them to cells in images of the slides, he said, adding that this is the same limitation all other NGS-based spatial transcriptomics methods have.
While the method isn't truly single-cell, the data capture clear regional transcriptional patterns that one would expect to see, Adey said, and they integrate well with single-cell datasets, as shown in the paper.
"An increase of the number of lanes for combinatorial barcoding will be an important aspect to increase the number of spatial multimodal measurements per tissue section," Lundeberg suggested. Fan said his lab is doing just that, working to deliver 100 lanes of barcodes at a time, which would give 100,000 pixels per sample.
He and his team are also designing microfluidics that can map between four and eight tissue sections on the same glass slide in a single DBiT-seq experiment and are expanding the panel of proteins it can analyze.
"It would be great to fully automate the reagent delivery," Fan said, adding that it "became a bit annoying" to pipette dozens of barcodes manually, although this is not an uncommon thing for a molecular biologist to do.