I joined the synthetic biology revolution in mid-2011, right before the sector imploded (Figure 1). The survivors, including my employer at the time, Amyris, beat a retreat from biofuels to smaller volume/higher margin opportunities, e.g. flavors and fragrances, cosmetics and nutraceuticals. For some of my colleagues, transitioning from green fuels to beauty products was demoralizing, but I believe that great things happen incrementally and that our mission is critical. The value proposition of synthetic biology is straightforward — replace products synthesized or extracted from nature (often in a brutal fashion) with sustainable and cost advantaged alternatives made via fermentation. But our field is less than 20 years old and we are competing with agriculture, natural product extraction, petrochemicals, and synthetic chemistry. Sober product selection that takes into account our state of technology is critical.
We have made great progress in the last decade, but even with more accessible targets, I have often seen synbio plans born in R&D die in scale-up, failing to achieve the cost or quality targets required for economic feasibility. Serious companies understand that scale-up and manufacturing are a key part of the synbio value proposition, and assess candidate strains under representative bioreactor conditions, initiate downstream purification work early, and plan realistic manufacturing campaign timelines. Even so, the difficulty of moving highly engineered microbes from microliters to hundreds of thousands of liters cannot be overstated and has broken many companies. At Sestina Bio, we are betting that achieving consistent success at scale starts with a new perspective on strain engineering.
Synbio has always been a maximalist discipline, defining itself in contrast to molecular biology. We have measured success by the size of the effort — more synthetic DNA, more complex expression constructs and more radically altered or even completely synthetic genomes. The resulting strains reflect this aesthetic. Dozens of genomic edits are stacked into a microbe to divert central metabolites into production of a target molecule (measured and referred to as “titer”). This is fundamentally at odds with the microbe’s Darwinian imperative: to produce more microbes. From an evolutionary perspective, strain engineering is a long slide down, with each edit trading fitness for titer. These unfit strains often fail at scale because they cannot withstand the rigors of the process and are rapidly out competed by lucky mutants that gain fitness through breakage of the engineered pathway.
So why do we continue to send weak strains to scale? For one, strain engineering isn’t easy! At the theoretical maximum yield for a given pathway design, there is nothing left for growth, i.e. the microbe has a fitness of zero. Hard tradeoffs aside, I also believe we have not been good stewards of fitness. Titer improvements are generally straightforward to measure in high throughput screening and disproportionately guide our trajectory, while fitness is more difficult to quantify and cannot truly be understood until the strain reaches a bioreactor. If fitness is the currency we spend to buy titer, this is like ordering a 12-course meal at a fancy restaurant without knowing the prices or how much cash is in your wallet.
At Sestina, we are addressing this issue by taking a minimalist approach to strain design. We believe that maintaining fitness is paramount, that complexity is the enemy of robustness, and that the manufacturing maxim “start with the end in mind” begins with the strain. We aim to be discerning consumers of edits, spending fitness in a deliberate and efficient manner. The best design usually achieves performance targets with the fewest edits. Paradoxically, achieving this minimal design requires taking a maximal approach to discovering and evaluating edits, inspired by advances in enzyme improvement. Prior to widespread and cheap DNA synthesis, enzyme improvement efforts in strain engineering were either rationally guided and limited to key residues (insufficient) or expensive and resource intensive affairs (rarely done). With full saturation mutagenesis libraries now only a mouse-click away, and significant improvements in analytical assay speed, the situation has changed. A broad distribution of edits can be found in an enzyme in initial rounds of deep screening, often in surprising locations. Next, combinatorial libraries of these edits can be evaluated to find assortments that maximize synergy. The term for this type of synergy is epistasis, and we believe it is also the key to achieving pathway engineering goals while maintaining a high level of strain fitness.
Classically, epistasis describes how combinations of mutations affect fitness, but the principles can be applied to any phenotype, including titer improvement in an engineered strain. Returning to the metaphor of buying titer with fitness, imagine a scenario where two edits that each cost $20 in fitness and each provide a $20 value in titer gain are combined to yield a $40 value in titer gain (Figure 2). In this case, there is no epistasis — you got exactly what you paid for. If instead, $60 in titer value is obtained, you are the lucky recipient of a GREAT DEAL, or beneficial epistasis. More often, though, the titer value obtained is <$40, which is termed antagonistic epistasis. Antagonistic epistasis can arise when two edits are fixing the same problem. Stacking edits that are antagonistic is a waste of fitness. And the chances of this happening are high when engineering efforts are tightly focused on the pathway and proximal genes.
In analogy to enzyme engineering, where early efforts were constrained and focused on the active site, we believe that the solution to minimizing antagonistic epistasis in strain engineering is to zoom out. Instead of operating with a scarcity mentality, where a limited number of pathway proximal edits are greedily stacked into a rigid lineage and never reexamined, we are building an abundance pipeline to support identification of large numbers of edits targeting diverse cellular functions and systems. These collections can then be sampled in combination to identify strains that maximize beneficial epistasis while minimizing fitness loss. As was the case for enzyme improvement, this requires significant upgrades to the DBTL cycle, from library design and build to high throughput test and learn. And while enzyme improvement is focused on a single gene product, our problem spans the genome. To address this challenge, we aim to be early adopters of new technology, importers of solutions from established fields, and developers of entirely new methods. We expect that the resulting platform will greatly reduce scale-up risk for a wide variety of products, broadening the portfolio of commercially feasible products and moving us ever closer to achieving the promise our field was founded on.