NEW YORK – A new approach to CRISPR-based genetic perturbation screening can reduce experimental costs by an order of magnitude or more, a new study says.
"Compressed" Perturb-seq can offer up to a twentyfold reduction in cost by overloading cells during the 10x Genomics single-cell transcriptome sequencing workflow, resulting in fewer empty droplets and more efficient use of reagents.
Alternatively, researchers can achieve up to a tenfold cost reduction by pooling multiple guide RNAs against multiple gene targets in the same cell. In earlier protocols, each cell received only one gene manipulation.
Both avenues to achieving lower costs are possible because of advancements in back-end data analysis. In the case of so-called "cell pooling," the computational pipelines are able to handle doublets — when two cells end up in the same droplet — which increase with overloading. And in "guide pooling," with sufficient randomization, the method is able to pick out the effects of single gene perturbations. Moreover, this approach unlocks the ability to understand combined effects of two or more genes.
"Overall, the more effective approach is guide pooling," said Brian Cleary, a researcher at Boston University and a senior author of a paper on compressed Perturb-seq published late last month in Nature Biotechnology. "As you increase the degree of guide loading in cells, the effectiveness increases." The opposite appears to be true with cell overloading.
"You mostly still want single cells in droplets. But you want each cell to have many perturbations," he said, noting that theoretically both methods could be combined for super savings.
"Great paper!" Christoph Bock, a researcher at the Medical University of Vienna who was not involved with the study and who has developed single-cell, CRISPR-based screening methods, said in an email.
"This is a very nice demonstration of the power of compressed sensing for single-cell CRISPR sequencing, but it remains to be seen if the moderate increase in throughput will outweigh the increased complexity and the increased need for validation when the results are less amenable to manual inspection and quality control," Bock said.
Cleary said the workflow changes to the Perturb-seq method were "minimal." CRISPR-based perturbation screens were demonstrated in a pair of papers published in December 2016 by researchers from the Broad Institute and the University of California, San Francisco, respectively. Bock also posted a preprint about his CRISPR screening method in late 2016. Aviv Regev, then of the Broad Institute, now head of R&D at Genentech, led that group's effort; she is also a senior author on the new paper.
The basic idea is to use CRISPR to knock out or at least interfere with a particular gene and use single-cell transcriptomics as a functional readout for the effects on the cell.
All of the new elements — multiple guides per cell, overloading the 10x Chromium instrument, and even the math for compressed sensing — are not radically different than what has been done before, Cleary said. "What is different is the holistic integration of understanding of how certain algorithmic approaches work and how much data they require.
"We're baking that algorithmic understanding into the experimental design and entire process," he said. "If we know we're going to analyze [certain types of data], we can get away with observing far fewer droplets or samples than we normally would if we generate those in a particular way, which is satisfied with random perturbations."
Some of the analysis takes a little bit more time but isn't "too heinous," Cleary said.
The baked-in mathematics means that now researchers can learn about genetic interactions — nonlinear effects of two or more perturbations. At the time she introduced Perturb-seq, Regev suggested that such interactions were rare. Even the new paper wasn't designed to learn about them, Cleary said, though he noted there are "some preliminary results" on the topic. Still, it has made tractable what was previously an unassailable problem.
There are limitations to guide pooling. Too many CRISPR guides in a cell can cause toxicity and cell death, especially when making double-stranded breaks. But compressed Perturb-seq uses CRISPR interference, rather than gene knockout. "We're often working in immune cells that are resistant to getting tons of viral integrations," Cleary said. "So we're still in a much lower regime than I'd like to be, at least from a mathematical standpoint."
The paper's proof-of-concept studies used about five guides per cell, though others have used more than a dozen. "As you have more and more guides per cell, the individual effectiveness of each starts to reduce a little bit," he noted.
"It's clear that well-done compressed sensing can quite significantly increase scale or cost-effectiveness for large screens," Bock said. "However, it comes at the cost of increased computational complexity, and that may be one factor why compressed sensing in its narrow sense isn't very widely used." The paper's approach simplifies the wet-lab workflow, "but makes the analytics more complex and the data less tangible for the user," he said.
Cleary suggested that the method is a step toward enabling more genome-wide, rather than targeted, genetic perturbation screens. "We're making them exponentially cheaper to run," he said, "which combats exponential growth in pairwise or three-way interactions."