An infant arrives wailing into the world, at once, the most alone he’ll ever be. Sure, he’s got parents, grandparents, and nurses at his beck and call. And someday, he’ll live in a bachelor apartment miles away from family and friends. But when he slurps down top ramen in his twenties, he’ll also feed trillions of microbes. When he’s comforted with his first-ever meal, he’ll have only a few bugs he picked up from his mother on his way out.

The health of our microbiome has been linked to everything from obesity and immune issues to Alzheimer’s and depression. Yet infants are born without a microbiome. That means the first days, weeks, and months are critical for microbiome development. His mother’s body knows that. She’s been busy building a whole buffet of sugar-like compounds specially formulated to feed not the baby but his tiny passengers.

Human milk oligosaccharides, or HMOs for short, were discovered relatively recently. Scientists got curious about the composition of breast milk and why so many molecules in breast milk were seemingly undigestible by humans. It turns out they aren’t meant to feed humans at all. HMOs show up in very high concentrations in colostrum, the milk produced immediately after a baby is born, and they play important roles in developing the infant microbiome.

But what if that doesn’t happen? What if his mother is unable or chooses not to breastfeed? Then he’s given a formula that, for all the science, effort, and regulation behind it, is orders of magnitude less complex than breast milk.

“Some formula brands are adding HMOs,” said Steven Frese, Assistant Professor of Nutrition at the University of Nevada Reno, “but the amount is very small.” Human milk contains as much as 15 grams per liter of hundreds of different HMO molecules. Even the most sophisticated infant formulas contain at most a handful of compounds at concentrations up to a few grams.

The main barrier to solving the HMO problem is cost. These molecules are complex. They are synthesized in multi-step processes with various building blocks that each have their own manufacturing cost. Formulas supplemented with HMOs have only been on the market since 2016, and they’re currently produced using fermentation.

In fermentation, bacteria or yeast cells are engineered to produce a molecule of interest. That molecule is then extracted, purified, and safety tested. The problem is those cells have interests of their own. Despite our best engineering efforts, the majority of energy poured into these tiny fermentation factories goes toward cell growth. So, the process is not efficient.

What if we could take the cell out of the equation? Cells are just the environment. Enzymes are the real workhorses turning basic compounds into complex molecules within the cell.

“Enzymes don’t really require energy to survive,” explained Gideon Lapidoth, Co-Founder and CEO of an emerging enzyme engineering company called Enzymit.

But enzymes are not used to doing their jobs outside of cells. They do require a certain co-factor called ATP. “You can’t use ATP in cell-free reactions because it’s too expensive,” said Lapidoth.” So we designed an enzyme that doesn’t need ATP.” Lapidoth estimates conservatively that their cell-free system can reduce the cost of HMOs by 50%.

Designing an enzyme that doesn’t use ATP—or performs any specialized function, especially outside of a cell—is much easier said than done. Fortunately, we now have a tool that helps turn saying into doing for all sorts of applications—artificial intelligence.

“It’s like chat-GPT for enzymes,” said Lapidoth of their AI-based enzyme engineering platform.

Enzymes work like a lock and key mechanism, so an enzyme’s structure is critical to its function. But resolving the structure of an enzyme is challenging. Even the gold standard tools, X-ray crystallography, and nuclear magnetic resonance, leave blind spots in our understanding of key parts of many enzymes.

Artificial intelligence is really good at taking a ton of incomplete or disorganized information and finding patterns. In this case, the information is partially resolved enzyme structures paired with their function. AI can use what we know about enzymes to predict what structures will perform known or novel functions. And these predictions turn out to be quite good.

“AI-guided tools for enzyme design are proving to be very effective,” said J. Casey Lippmeier, Senior Vice President of Innovation at Conagen, an enzyme engineering company that’s well-established in the space. “The next iteration is to type on your keyboard and say, ‘give me the structure of an enzyme that will perform this chemistry,’ we’re not quite there yet, but we’re getting close.”

“It’s a special time right now in terms of computational tools and experimental tools coming together,” said Joseph Jacobson, Head of the Molecular Machines Group at MIT and scientific advisor to Enzymit. “Enzymes are really the most precise things we can engineer, and if they [Enzymit] and others are successful in this space, it’s going to be a real step forward for how the world builds chemistries.”

Besides cost, there are additional benefits to using a cell-free approach. “We developed a cell-free bioconversion process that enables us to synthesize larger HMOs,” said Lippmeier of Conagen’s HMO product.

The confines of a cell also limit the size and complexity of the molecules we can produce. HMOs aren’t naturally engineered by microbes (or else they wouldn’t need us to cook them for them), and some of the larger molecules, if you can get the bugs to build them, are difficult to get out of the cells and purified in-tact.

So where does all of this leave us on the path to infant formulas that mimic breast milk? Not very close. The first question is whether or not we’re even focused on the right molecules.

Gut bacteria produce a compound called butyrate in response to HMOs, so its presence following feeding is used as an indicator of whether the HMO’s impact is clinically relevant. “But what we found in our clinical studies is that there’s not a lot of butyrate in healthy, breastfed infants,” explained Frese. Another byproduct of HMO fermentation is lactate, and lactate and butyrate compete for the same transporter. Frese personally suspects that lactate is more important.

And HMOs are just one class of molecules that are unique in breast milk. There are a variety of other compounds in breast milk that are not present in formula, including anti-oxidants and certain microbe-inhibiting compounds. Conagen is working on producing some of these for potential use in food preservation, as well as infant formula.

Another company in the space, Heleina Bio is focused on proteins in breast milk that are relevant to immunity. They’re using fermentation to produce these proteins with the intention of first using them to fortify nutrition products for adults and the elderly.

“Our ultimate goal is to be able to combine multiple approaches together, so potentially HMOs along with these proteins along with the good bacteria,” said Carrie Zanon Malinczak, Head of Nutritional Biology & Safety at Helaina. “If we can supply all of these things together, it will act synergistically to create a much better nutritional environment to help people really thrive.”

But before we can even figure out which breast milk compounds are the most important and at what level, we have to be able to produce them affordably.

“If they can cut the costs [of HMOs] by half right out the gate, that’s going to be a big leap,” said Frese of Enzymit’s cost-saving projections.

“500 years from now, we’re going to have an infant formula that looks very much like breast milk,” said Lippmeier in jest. The goal, he clarified, is not to replicate breast milk perfectly because that would be impossible. It even varies from mother to mother and over the life of a single infant.

The goal is to fine-tune our understanding to build better nutritional products. Adding AI-based enzyme engineering and cell-free production are important steps in that direction.

‍