NICK OUZOUNOV

Chief Technology Officer, Geltor, Inc., San Leandro, CA, USA

While fermentation has been used for the production of products like beer, yogurt, and bread for millennia, most consider the start of precision fermentation (PF) to be in the late 1970s with its transformation of medicine. Genentech’s recombinant human insulin was one of the first major PF products. In 1978, Genentech genetically engineered bacteria to produce human insulin, replacing the old method of extracting insulin from pig and cow pancreases. This biosynthetic insulin (marketed as “Humulin”) proved to be as effective as insulin from the human pancreas while eliminating the allergic reactions caused by animal-derived insulin (1). The new process also solved supply issues—previously it took >50,000 animal pancreases to make one kilogram of insulin (2)—by enabling production at scale via fermentation tanks.

Beyond pharmaceuticals, precision fermentation has rapidly expanded into other markets like the beauty and personal care sector (3). Biotechnology now offers a way to “brew” high-value ingredients in microbes providing an alternative to sourcing them from animals or petrochemicals. Fermentation derived cosmetic actives are growing in demand as brands seek natural, sustainable, and high-performance ingredients. Perhaps the most notable example of this change is Hyaluronic acid (HA), valued for its hydration and viscoelastic properties in skincare. HA was originally obtained from animal tissues, notably rooster combs (4). For decades, companies extracted HA by processing tons of rooster combs, which contain the highest levels of this molecule found in animals. However, this method had downsides: it was costly, yielded limited quantities, and raised ethical and safety concerns (animal parts can introduce impurities or pathogens). HA from animal sources can also vary in quality and carries a risk of immune reactions if not purified thoroughly. Since the early 1990s, most HA has been produced via precision fermentation using microbes, eliminating the need for animal harvest (4, 5). Typically, strains of Streptococcus bacteria are engineered to secrete hyaluronic acid during fermentation. This shift to microbial production was driven by the need for a more controllable, scalable process—it’s easier to brew HA in tanks than to depend on slaughterhouse byproducts.

As with HA, precision fermentation has recently been used to produce proteins of interest such as collagen for personal care applications. Animal collagen production is resource-intensive, and can be linked to environmental degradation and deforestation, particularly in regions such as the Amazon, where livestock farming drives significant ecological damage. Investigations have revealed direct links between collagen sourcing and deforestation practices associated with major industry players, spotlighting an urgent need for sustainable alternatives (6, 7).

On top of the sustainability challenges, commercial collagen for food and cosmetics typically consists of only two fibrillar collagens types: type I and a small fraction of type III (8). For example, type I collagen constitutes around 85-90% of the organic content of skin, followed by type III collagen at less than 10%. These two types (I and III) are valued for their skin-plumping and structural benefits, but they represent just the “structural scaffold” collagens. This is because traditional methods of boiling skins and acid/enzymatic extraction cannot easily isolate other collagen types as those exist in much lower quantities or in hard-to-extract tissues (9, 10). In practice, this means nearly all collagen ingredients available until recently have been the same few types (mostly I, occasionally II or III), but most of the 28 collagens (which account for less than 3% of the total collagen pool) are not accessible in useful quantities from conventional animal sources.

Type 21 collagen is one of the more recently discovered collagen types. It was identified through the Human Genome Project in 2001 when researchers mining genomic data found a gene (COL21A1) encoding a collagen protein (11). Expression studies showed COL21A1 (type 21) is present in many tissues–including heart, skeletal muscle, stomach, skin, kidney, and placenta—and it is described as helping to maintain the integrity of the extracellular matrix (12, 13). A proteomic study of human dermis across ages found COL21A1 is highly expressed in toddler and young adult skin but drastically drops off with age. This reduction could contribute to the reduced dermal elasticity and impaired wound healing seen during aging (14). While Type 21 collagen is present at extremely low concentrations and nearly impossible to isolate from tissues, precision fermentation makes it possible to produce it in the quantities and purity needed for beauty and personal care applications regardless of its natural scarcity.

In conclusion, precision fermentation has fundamentally transformed ingredient sourcing for personal care by providing alternatives to traditional sourcing that can have limitations such as ethical concerns, environmental impacts, and ingredient scarcity.

Early protein engineering was very laborious. It was heavily dependent on sourcing naturally occurring sequences, followed by extensive chemical mutagenesis and the screening of thousands of variants to identify beneficial changes. The first protein created entirely from scratch (de novo-designed protein) without relying on natural sequences was a four-helix bundle protein created by rational design in 1987. It was a landmark development that highlighted the initial potential of computational methods even though it was limited by computational constraints at the time (15, 16). In the 1990s, the landscape changed significantly with directed evolution, pioneered by Frances Arnold, who introduced iterative mutation and selection methods to evolve enzymes with enhanced or novel functionalities. Arnold's pioneering work was awarded the Nobel Prize in Chemistry in 2018, recognizing directed evolution as a powerful tool for enzyme and protein development (17).

Traditional computational approaches to protein engineering remained limited by predictive inaccuracies. Initial computational attempts at rational protein design typically had low success rates (often below 1%), requiring significant experimental validation (18). Over the subsequent decades, substantial improvements in computational accuracy were achieved, particularly with the development of sophisticated software like Rosetta, which facilitated early successes in protein structure modeling and design (19). Despite these advances, it was the introduction of artificial intelligence, especially deep-learning techniques, that dramatically changed the trajectory of protein engineering, significantly increasing the speed and accuracy of computational protein design.

Recent breakthroughs in AI-driven protein design, notably recognized by the 2024 Nobel Prize in Chemistry, have profoundly reshaped the biotechnology landscape. AlphaFold, developed by DeepMind, has provided scientists with a transformative tool capable of accurately predicting three-dimensional protein structures from amino acid sequences. Nobel Prize laureates Demis Hassabis and John Jumper revolutionized structural biology by developing AlphaFold, which provides researchers with computationally accurate structural predictions rapidly, and at an unprecedented scale (20).

Complementary advances such as RoseTTAFold, developed by David Baker's laboratory at the University of Washington, similarly harness deep-learning approaches to facilitate accurate predictions and de novo protein design (21). Baker’s work, also recognized with a Nobel Prize, allowed the design of novel proteins with specific functions and folds never previously observed in nature (20). These tools have dramatically improved success rates in protein engineering, from initial success rates of around 1% to contemporary design success rates approaching 10%–100%, depending on the specific protein library and design goals (22).

Moreover, innovative AI platforms such as BindCraft have further refined the capability of computational design by accurately predicting protein-ligand interactions and significantly enhancing specificity and binding affinity. BindCraft leverages the predictive power of AlphaFold2, enabling the rapid design of highly specific and potent protein binders without extensive high-throughput screening (22). Collectively, these developments have transitioned protein engineering from a predominantly empirical discipline to a highly precise computational science, unlocking previously unimaginable innovations in functional protein design.