Alexander Schier simply wanted to make sure he destroyed a gene in zebrafish embryos. So like many biologists these days, he turned to the genome-editing system known as CRISPR. But Schier, a developmental biologist at Harvard University, ended up doing much more than knocking out a gene. He and colleagues devised a new way to mark and trace cells in a developing animal. In its first test, described online today in Science, the researchers used CRISPR-induced mutations to reveal a surprise: Many tissues and organs in adult zebrafish form from just a few embryonic cells.
Other researchers are already looking to adapt the method to probe development. “The technique promises to allow the reconstruction of the ‘family tree’ of the cells that compose an animal’s body,” says James Briscoe, a developmental biologist at the Francis Crick Institute in London, who calls it “a creative and innovative use of [the] CRISPR technique.” Other scientists intend to exploit the method to trace the evolution of tumors. And some are racing to develop similar ways of using CRISPR to record a cell’s history—for example, the environmental influences it has been exposed to.
Schier and his colleagues took advantage of what Harvard geneticist George Church calls CRISPR’s “genome vandalism.” In normal CRISPR editing, a so-called guide RNA precisely targets the enzyme Cas9 to a particular site in the genome so that it can break the double-stranded DNA there. In one of CRISPR’s original uses, a template DNA tells the cell’s machinery how to repair the double-stranded break, allowing edits as precise as the changing of a single nucleotide. But if scientists supply no template, the cell cannot precisely repair the break, and the gene usually ends up with a deactivating “scar,” where some nucleotides go missing or some are added that don’t belong.
To make sure the zebrafish gene in his sights was truly obliterated, Schier targeted multiple sites within the gene by introducing several different guide RNAs. But repeating the experiment led to very different outcomes: The size of the deletions varied, and the scars included both small and large insertions. That destructive diversity could be put to use, Schier and geneticist Jay Shendure at the University of Washington, Seattle, realized.
In the genomes of zebrafish embryos, Schier and Shendure inserted a cassette of extra DNA, consisting of 10 different target sequences for CRISPR. Then they injected the single-celled embryos with the Cas9 enzyme and 10 guide RNAs that matched the target sequences. As the embryo developed, the CRISPR system repeatedly disrupted the target DNA in each cell, marking it with a pattern of deletions and insertions—a distinctive barcode. Whenever a cell divided, the daughter cells would start out with the same barcode and then diverge when Cas9 cleaved it at different places. The first changes in the barcode seem to happen in the two-cell stage and then the editing machinery runs out of steam after about 4 hours, when the embryo consists of thousands of cells—after that point, the barcodes that remain will populate the adult animal as cells continue to multiply.
Four months later, the scientists collected organs from the adult fish and isolated more than 1000 different barcodes from about 200,000 cells. Cells with more similar barcodes are likely to have diverged later in development, so the scientists were able to use a computer program to calculate a family tree for the 200,000 cells—essentially a lineage map revealing which cells spawned others. “When you sacrifice the zebrafish at the end of the experiment you've actually got a full time readout of all the cells and where they came from, ” Church says.
One of the most striking findings was how few cells give rise to the bulk of the tissue in any given organ. More than half of the cells in most organs shared fewer than seven barcodes. In every organ except for the brain, 25 different barcodes made up more than 90% of the cells in any organ. “Tissues may be founded by a much smaller group of cells than I would have anticipated,” Briscoe says.
For developmental biologists, the new technique, nicknamed GESTALT (for genome editing of synthetic target arrays for lineage tracing), could help clarify how animals take shape from a single cell. It could also shed light on important questions in cancer research, such as how many precursor cells give rise to a tumor, how the cells in a tumor are related, and how cancer cells that have spread are related to the initial tumor.
The technique has shortcomings, Schier says—for example, it does not reliably mark each new generation of cells. But compared with other ways of tracing cells and their progeny, like dyeing them or relying on natural mutations, a CRISPR-generated barcode is potentially more powerful and easier to use. “I think cancer biologists will start thinking about this because it is a more elegant way of marking cells than we currently use,” says Leonard Zon, who directs the stem cell program at Boston’s Children Hospital. “We definitely want to try it.”
Researchers are already proposing other ways to turn CRISPR into a kind of cellular memory. “I think conceptually that is the most exciting thing, that you can basically record history in the DNA,” Schier says. A team from the Massachusetts Institute of Technology in Cambridge may already have done that. Last week, Timothy Lu and colleagues uploaded a paper to the preprint server bioRxiv.org describing a system called mSCRIBE: mammalian synthetic cellular recorder integrating biological events. Instead of adding a barcode consisting of 10 CRISPR targets, the researchers inserted into cells a single CRISPR target and engineered the site so that it also encodes the guide RNA. As a result, the system targets itself: The guide RNA leads Cas9 to its own source DNA. Cas9 breaks the DNA, leading to a mutation in the sequence, which in turn leads to the production of a mutated guide RNA. That altered guide RNA then leads Cas9 back to the altered target sequence, and the cycle continues, with the DNA sequence and the guide RNA changing in tandem.
By observing how the CRISPR target sequence changed in thousands of single cells endowed with this complicated setup, the researchers estimated how many rounds of Cas9 activity it took to produce certain sequences. (A rough analogy would be estimating how many rounds of the telephone game, also known as Chinese whispers, it would take to scramble the starting phrase “lobster boil” into “losing team.”) In a follow-up experiment, they coupled the activity of Cas9’s gene to the activity of an inflammatory pathway in cells. In cells exposed to more of the inflammatory factor TNFα, the target sequence consequently recorded more rounds of Cas9 alterations
Then they tested their CRISPR-based recorder in mice by implanting the animals with the engineered cells and injecting some of them with an inflammation-provoking molecule. In mice that were dosed with the molecule, the CRISPR target sequence had more changes than those from untreated mice. The method “can be used in vivo to record physiologically relevant biological signals in an analog fashion,” the authors write.
It could also record stimuli that cancer cells are exposed to in the microenvironment of a tumor or track the activity of specific pathways in cells during disease development, Lu and colleagues suggest. Church notes that such methods may also prove valuable in brain studies, for example by recording the activity of pathways involved in memory foundation. “You can take a transient process like the learning of a new task and turn it into a permanent record that's present in the cell body of every neuron in the brain,” he says.
*Update, 1 June, 12:55 p.m.: On Friday, 27 May, another paper describing a similar approach was placed online as a preprint. In it, George Church and two collaborators integrated a CRISPR site engineered to target itself in the genome of human cells and calculated that it can store about 5 bits of information. The researchers argue for using several such sites in parallel, in effect combining the ideas of a self-targeting CRISPR site and a CRISPR barcode. “This measurement suggests that uniquely barcoding the roughly 12 billion cells in a mouse will require at least 7 such [self-targeting CRISPR sites] per cell,” the authors write.