August 14, 2024
Last month I had the pleasure of talking about the future of synthetic biology with some smart colleagues at a Chan Zuckerberg Initiative workshop. (For the CZI post on LinkedIn, see here.) While I’ve done a little work in synthetic biology over the years, and have been interested in it since graduate school, I usually don’t think about my current work as ‘synthetic biology’. This was true for a few of the other participants, who don’t typically bill themselves as synthetic biologists. It was a useful exercise for us to rub shoulders with bona fide synthetic biologists and to start thinking again about how basic work in genomics and metabolism aligns with the big goals of the synthetic biology field.
What are the big goals of synthetic biology? For a long time, the field was dominated by two. First, figure out how to design biological circuits from scratch. The pathbreaking 2000 paper by Michael Elowitz and Stan Leibler describing a synthetic “repressilator” has long been one of the paradigm-setting papers for those who want to build synthetic biological circuits. At the CZI meeting, Elowitz presented some interesting ideas about developing biological APIs into which we can plug different synthetic modules. (Here’s a nice explanation of an API for non-computer folks.) As a concrete example, Elowitz talked about his lab’s recent work on modular miRNA circuits called Dosage-Invariant miRNA-Mediated Expression Regulator (DIMMERs), designed to stably tune the expression of transgenes in a cell.
miRNA-based DIMMER. Figure 2B from Du, et al, available on bioRxiv.
The second long-time big goal of the field is DNA synthesis. If you’re going to make sophisticated biological circuits, you need to be able to make a lot of DNA. Hence there has been much work on building synthetic chromosomes over the years. John Glass, of the J. Craig Venter Institute, talked about their efforts to make synthetic chromosomes and minimal genomes. The latter paper was work done with Elizabeth Strychalski, of the National Institute of Standards and Technology (NIST). I hadn’t previously realized that NIST was involved in synthetic biology, but it makes sense that they are.
And speaking of NIST, another important theme at the meeting was standards for the field, something which academics are terrible at thinking about. There is a bit of a collective action problem here. Everyone is incentivized to always focus on developing something new, but in many cases it’s important for a field to step back and collectively decide on standards of performance that any new tool should meet. The ENCODE project did this for basic genomic assays like ChIP-seq and RNA-seq. The American College of Medical Genetics and Genomics is doing this for massively parallel functional assays that aim to help physicians decide whether a patient’s particular genetic variant is causing disease. In the synthetic biology field, there are some opportunities for standards, though, based on what I heard, not a lot of agreement of what those should be. AI models, particularly generative models for design are an obvious candidate for some clear performance standards.
What’s the overall goal of synthetic biology? Perhaps the dominant goal is to cure human disease. With a number of gene and engineered cell therapies already approved by the FDA, the field should be thinking about what we can do to make therapies like these safer, easier, cheaper, and applicable to a broader range of conditions. (Check out this 2022 review of the subject by Rice University’s Caleb Bashor and colleagues.) Alongside these therapies, there is a huge opportunity to use microbes as chemical factories to make drugs, particularly drugs that currently rely on sourcing raw materials from plants that can’t be cultivated at scale. And speaking of microbial factories, I’m surprised we haven’t seen more widespread use of bioplastics. You can envision a future in which plastics are no longer made from petroleum products but custom, microbially-produced materials that can be enzymatically broken down when they are recycled.
One final big theme that came up at the conference: to what degree will AI-based models replace more traditional first principles models? The original repressilator was modeled using a set of differential equations describing the interactions between the circuit components. Making sophisticated models of biological circuits using systems of equations like this has proven difficult, though not impossible. Generative AI, and even simple convolutional neural networks have turned out to be much easier, as long as you have the data to train them. But are there applications where you want to avoid AI and stick with first principles-based systems of differential equations?
Here are a few more links from around the field (not all coming from the CZI meeting) that highlight some interesting work in synthetic biology:
Engineering platelets as drug delivery vehicles: Tara Deans at the University of Utah has a preprint out describing her work to reprogram megakaryocytes, so that they generate platelets with a cargo of therapeutic protein.
Figure 1a from Javdan, et al, available on bioRxiv.
Iterative design of synthetic enhancers: Georg Seelig at the University of Washington has produced one of the nicest examples of regulatory DNA design with deep learning. His groups demonstrates something that is still rare in the field – iterative optimization of models, which takes advantage of our growing capacities to build and assay large libraries of DNA sequences.
Getting drugs into the brain with Toxoplasma gondii: Toxoplasma gondii is a brain parasite that people sometimes pick up when handling cat litter. A new paper out in Nature Microbiology demonstrates how to repurpose this parasite to deliver drugs to neurons, which is hard because the central nervous system is pretty effective at blocking the entry of things that don’t belong there. But a brain parasite knows how to get around the barriers, which could make it a useful vehicle for drug delivery.
The 1000 monkeys with 1000 typewriters approach: Caleb Bashor has a cool preprint on “Ultra-high throughput mapping of genetic design space.” The idea is basically this: throw a bunch of genetic circuit components together into a cells and screen for those that work. Bashor’s group came up with a clever approach called CLASSIC that exploits massively parallel assays to assemble synthetic biological circuits.
Authored by Michael White
