NEW YORK (GenomeWeb) – A new microfluidics-driven DNA-protein binding assay has its creators dreaming of launching a startup to cash in on several potential applications.
In a paper published in Nature Methods this week, a Switzerland-based team of researchers, led by bioengineer Bart Deplancke of the École Polytechnique Fédérale de Lausanne (EPFL), showcased the assay's ability to determine DNA-binding specificities for transcription factors.
Selective microfluidics-based ligand enrichment followed by sequencing, or SMiLE-seq, combines antibody arrays, mechanical trapping, and next-generation sequencing readouts to provide a new platform for characterizing DNA-protein interactions.
"It allows us to look at transcription factors nobody else has been able to look at," Deplancke told GenomeWeb. "On top of that, what we found is that our microfluidic approach of trapping interactions seems to preserve the quantitative content of this interaction. In terms of predicting where in the genome the transcription factors bind, the models we're able to derive seem to be, in general, actually better than the other assays."
The assay could help accelerate efforts to map where transcription factors bind the genomes of humans and model organisms, Deplancke said. "One of the great missing pieces in our understanding of gene expression is how transcription factors regulate and control it in an orchestrated fashion. One reason we're still lacking in this info is because current technologies have great difficulty working with these proteins."
Bacteria often can't support over-expression needed to get workable quantities of protein for an assay, and in vitro expression often can't synthesize a fully functional transcription factor.
By creating a microfluidics-based assay, Deplancke's team has reduced the amount of sample required. "That gives you great flexibility to parallelize or automate it as we did," he said. "That turned out to be really useful in order to express transcription factors that other assays have trouble working with."
An engineer by training, Deplancke launched his lab about a decade ago and has begun incorporating microfluidics in expectation that it will completely take over the field of molecular biology. "It's too powerful for it not to become mainstream," he said. From single-cell analysis by Fluidigm to Bio-Rad's droplet digital PCR, he said it's primed to be a workhorse technology.
He started working on SMiLE-seq about three years ago, though doubt shadowed the team. Deplancke himself wondered whether this was a scalable technology, or whether it would only be useful to labs interested in transcription factor function. Others, including a colleague at EPFL who developed the key microfluidics technology that eventually made SMiLE-seq work, wondered whether it would work at all.
In 2007, Sebastian Maerkl, now at EPFL, was a postdoc working at Stanford University with Stephen Quake. That year, the two published a paper in Science describing a method to trap low-affinity DNA-protein interactions, specifically those of transcription factors. By physically restraining bound molecules against an antibody array, they capture both transient interactions, which yeast two-hybrid or tandem mass spectrometry methods don't, and weak interactions, which are usually obliterated by the wash in protein-DNA microarrays.
Deplacke called this a "button," and when he pitched his idea to Maerkl, Maerkl wasn't sure it would work.
For SMiLE-seq, what the button captures is interactions between GFP-labeled transcription factors and a library of oligonucleotides. Each oligo has a 30-base-pair randomly generated sequence, flanked by a barcode that allows multiplexing and Illumina adapters. The transcription factors bind to any number of DNA motifs in the library, those interactions are trapped, and then the readout matches up each motif with the transcription factor via the barcode.
To get the readout to work, SMiLE-seq would need to enrich millions of sequences in one step. Whether that was possible wasn't clear at the beginning. According to Deplancke, even Maerkl was skeptical. But when Deplancke showed him, among others, the final data, many doubts were put to rest. "He said, 'Wow, that's remarkable, I never thought it would be that powerful,'" Deplancke said (Maerkl did not respond to GenomeWeb's request for comment).
Deplancke also worked in advanced bioinformatics to derive predictive models for binding interactions. EPFL bioinformatician Philipp Bucher is a co-senior author on the Nature Methods paper. "Deriving good models is its own research domain," Deplancke said. Specifically, the team used a hidden Markov model, which takes enriched sequences from the assay readout and uses that to generate the binding spectrum for that transcription factor. Deplancke said an informatics package will be publicly available soon, though the process is also described in the paper.
To validate the method, the team looked at 58 previously characterized transcription factors across several species and tried to reproduce the results from other assays. "The assay failed for two, but that was because we weren't able to clone them," Deplancke said. "Not only did we find binding models, but ours turned out to be more qualitative in their predictive power. If we were able to clone it, we were successful, not only to validate the binding model, but improve on those models."
The assay could be useful to anyone interested in gene regulation, Deplancke suggested. "Most people who do molecular biology come across transcription factors. It helps to find where it's binding." The assay could also be loaded with transcription factors subjected to post-translational modifications, or complexes of transcription factors. "Phosphorylation could be mimicked by entering a negatively charged amino acid or adding a kinase," he said. "You could see if that changes the binding model." While microfluidics is on a tinier scale, the channels are still enormous compared to the proteins in question.
But what Deplancke is most excited about is the ability for the underlying technology to investigate the chromatin landscape.
"This project made us realize that the enrichment of protein-DNA interactions using microfluidics seems to be so efficient we can apply it to chromatin immunoprecipitation," he said. "In exciting preliminary data, we have evidence that the same principle can be applied to enable [ChIP] that will blow any other protocol out of the water."
Taken together, the transcription factor binding and ChIP applications have a lot of commercial appeal.
To that end, Deplancke has filed a patent application in Europe and is exploring options for commercialization. He's no novice to launching a startup, having co-founded the genomic data management software company Genohm in 2011, and he would prefer this route over licensing the method to another company.
In just over five years, Genohm has grown to more than 30 people, he said, and counts the US Food and Drug Administration and Qiagen as customers.
His experience fundraising for Genhom could influence how he plans to commercialize SMiLE-seq, he said. "We've actually shied away as much as we can from true investment funds," he said, noting that most of Genohm's funding has come from private investors. "Their intentions are not always noble and trying to keep a little control of your own company and having a longer-term vision is not always appreciated."
But launching a biotech company is different than launching a software company. "For what we have in mind for SMiLE-seq, that may not work," he said. "The investment trajectory might actually be quite different."
Several companies have already approached him about licensing SMiLE-seq he said, though he declined to disclose which ones.
"Going through the final steps takes cash, so we would need to find the right partner or right set of investors to make this happen," he said. "The potential is there. I'm quite optimistic we'll go all the way."