The protein design boom is moving beyond pharma. Quercus raised $5M with the conviction that weeds are next

Weeds are one of the biggest problems in growing food.

Anyone who has kept a garden knows this. Most of the work is not planting. It is weeding. It is making sure the plant growing in that patch of dirt is the one you chose to put there. Scale that from a raised bed to eighty million acres of corn and the problem does not change, it just gets more expensive. Weed control is the largest single crop protection expense in row crop agriculture, and herbicides make up more than 40 percent of a roughly $80 billion global crop protection market.

For most of human history the answer was labor, and then chemistry came around and ate the workman's lunch. Starting in the 1940s with synthetic auxins like 2,4-D, growers got compounds that killed the weed and left the crop standing.

Every herbicide works by jamming one specific piece of plant machinery, an enzyme or a process the plant cannot live without. That target is called the site of action. By the 1980s the industry had found about twenty of them, arriving at a rate of roughly one new one every two to three years.

Then it stopped. The first new site of action in more than thirty years, a rice herbicide, is only now making its way to market.

The resistance problem

Weeds evolve their way around weed killers for the same reason bacteria evolve their way around antibiotics. A compound that attacks one specific mechanism works as a filter. Everything it kills leaves the gene pool, and whatever walks away from it inherits the field. Run that on the same acres, with the same product, season after season, and you have built a selection experiment across millions of plants at once. A single mutation that changes the shape of a binding site is enough to pass through it.

The weeds have been winning. There are 548 confirmed cases of herbicide resistance worldwide across 275 weed species, and 21 of the 31 known sites of action have been compromised. You can see the consequence in the spray tank. A field a grower once cleared with one product now takes three to five herbicides in combination to reach the same level of control.

What turns this from a fair fight into a losing one is the clock. A weed population can evolve resistance in three to four years when nobody manages for it. Sulfonylurea resistance has shown up after three to five years of repeated use. Triazine resistance takes seven or more. Chemistry cannot answer on anything close to that timescale. A 2026 study of the major crop protection companies, conducted by AgbioInvestor for CropLife International, put the average at more than 11 years and about $307 million from discovery through registration. Evolution is moving at roughly twice the speed of the industry trying to keep up with it, and it has been for decades.

21 of the 31 known sites of action have been compromised. The first new one in more than thirty years is only now reaching the market.

What the screen could find

Finding a herbicide has always worked about how you would expect, and it looks a great deal like early therapeutic discovery. You take your plant, or the enzyme you think matters inside it, and you push millions of small molecules past it, hoping one binds tightly enough and long enough to do some damage. Whatever kills the plant gets followed up. Then you work backward to figure out what the successful ones were actually hitting.

The method has a blind spot, and it comes down to what small molecules are physically good at. They need a pocket. A deep, well-shaped cleft gives a small molecule something to wedge into and hold onto. Many of the most interesting proteins in any organism simply do not have one. They present broad, shallow surfaces with nothing to grip, and no amount of screening will turn up a hit, because nothing in the library can stick to a surface like that. Those targets never register. They never make it onto the list.

So the list of roughly 20 sites of action describes something narrower than it appears. It is a record of the plant biology that small molecules could reach, which is a much smaller category than the plant biology worth attacking. The method drew the map, and for forty years it was the only map anyone had.

Human medicine ran into the same wall and has spent the last few decades climbing over it. That is the reason antibodies and protein therapeutics took off the way they did. A protein can lie across a broad surface and bind where a small molecule finds no purchase, which is how pharma got at targets that medicinal chemistry had already written off as undruggable. Crop protection borrowed the word druggability from pharma. The tools to expand it are only now within reach.

The list of roughly 20 sites of action is a record of the plant biology that small molecules could reach, not the plant biology worth attacking.

Bringing the pharma toolkit to the field

Jon Lightner saw the opening. He holds a PhD in plant physiology, ran biotechnology at DuPont Pioneer, and served as CSO at Genus plc, where he led development of the first FDA-approved gene editing technology in a food animal. He has spent most of his career inside the industry that stopped producing new sites of action.

In July, Quercus Biosolutions, where Lightner is CEO and co-founder, closed a $5 million seed round led by Serra Ventures. The company's generative design platform was built for human drug discovery, where it was used to design mini-protein binders, and has more than $30 million of development investment behind it. What that lets Quercus do is use AI to design novel protein binders against plant targets.

The obvious question is why a new binder helps if the target is one weeds have already learned to survive.

The answer is that a protein does not have to inhibit an enzyme the way a small molecule does. It does not have to compete for the active site at all. Lightner says that in some of their more advanced programs they can infer the protein is working by a different route entirely, because the active site is physically inaccessible to a molecule that size. It cannot possibly be reaching it. So if a weed's resistance mutation evolved to disrupt small-molecule binding in that pocket, a protein gripping a different surface on the same enzyme is not necessarily affected by it.

Designing for a field instead of a clinic

Biologics worked in pharma partly because pharma could afford them. A therapeutic can cost a great deal per gram and still support an excellent business, and if the molecule needs babying, it gets refrigerated, injected, and administered by a professional.

A herbicide gets sprayed out of a tank pulled behind a tractor, onto a field, in whatever weather shows up. The active ingredient has to be made for a few dollars an acre across hundreds of millions of acres. It needs to be rainfast quickly, which is the label spec for how soon after spraying it can rain without the product failing. It has to be gone before the grower plants a cover crop in the fall, because a residual that is still working in October treats the cover crop the same way it treated the weed. Lightner argues that half the unintended consequences in crop protection history come down to molecules that persisted too long. And it has to kill one plant while leaving the plant beside it alone, with nothing to hit in a butterfly or a mammal that walks through the field.

That list is not news to anyone in crop protection. What is different about how Quercus is approaching it is when the list gets addressed.

The conventional sequence is to find one lead with good activity, then spend development solving everything else about it. The compound works, so now you find out whether you can manufacture it, whether it survives a rainstorm, whether it is selective enough, whether it clears on time. Each of those is a separate program, and each one can kill the candidate after you have already spent the money.

Quercus puts those properties into the design loop instead. Lightner describes it as going back to the model with the problem: this one has good activity, but it takes three days to become rainfast, so design me one that does not. Stability is a dial rather than a goal, and the target is stable enough but not too stable. Specificity gets specified up front, so the protein binds the target organism's pathway and has no corresponding target in a mammal. There is a large body of academic work on inherently stable protein designs from organisms living in extreme environments, and those designs are available to draw on as starting scaffolds.

This is what the AI is actually for. It is less about finding a molecule that works than about generating candidates that already carry the properties a field product needs, so that fewer of them die in development for reasons that have nothing to do with efficacy.

Making it cheap enough

Lightner's read is that manufacturing cost killed roughly half of the biologicals that had good efficacy. So cost of goods is a design input at Quercus rather than a problem handed to a process engineer later.

Size is the clearest case. These proteins get made by precision fermentation and sold by weight, but herbicidal activity is roughly a one to one interaction between inhibitor and target. An inhibitor ten times larger than the job requires costs ten times as much on a molar basis to deliver the same effect. Smaller is cheaper in direct proportion, which makes minimizing size an economic requirement rather than an elegance preference.

Expressibility gets treated the same way. It sits in the design and optimization criteria from the start, favoring candidates that express well and bind at least as well as the alternatives. The production path runs from a high throughput system built on E. coli components, into E. coli itself for early research, and then into commercial production organisms as they scale toward launch, with room to tune the protein further at each stage. At volume, the plan is precision fermentation through industrial contract manufacturers.

This is also where starting from scratch beats starting from nature. Source a protein from an organism and you inherit real bioactivity, because nature selected for it. You also inherit everything else nature selected for, and cost of goods was never on that list. Teams that start from a natural lead spend their development cycles working backward to strip out what they did not want. Quercus starts from simple, expressible designs and uses the model to add the efficacy.

The regulatory advantage

There is one place where biology gets a real structural advantage over chemistry, and it is the reason a company this size can attempt any of this at all.

A new chemical crop protectant takes more than 11 years and roughly $300 million to go from discovery to market. For a protein designed to break down in the field, the regulatory package is much lighter, and in the Americas it can move through established approval processes in as little as two years for less than $2 million in studies. That gap is the business case, and it also decides who Quercus has to sell through.

The company is talking to the majors, which is familiar ground for Lightner after spending two thirds of his career inside companies like them. But a two year, two million dollar approval path is navigable by a small company, which means partnering with a major is an option rather than a requirement. They can also work closer to the farm, through the retail chain and the co-ops. Those businesses sell roughly 90 percent of the crop protection used in the market and apply about half of what they sell, which means a grower is paying them to run the spray rig for the chemistry they recommended in the first place. Lightner's argument is that this end of the chain is underserved by innovation and feels the weed problem as sharply as the growers do.

More than 11 years and $300 million to bring a new chemistry to market. As little as two years and under $2 million in studies to register a protein.

Why row crops

The company is aiming first at weed control for row crop farmers in the American Midwest, and Lightner is explicit that this is a decision rather than a default.

The logic runs one direction. Build something that is economic for a corn and soybean grower and it will also be economic for almond growers in the Central Valley, apple growers in Washington, or grape growers anywhere. Start in a high value niche and you rarely make it back down, because nothing in the cost structure ever forced you to solve for scale. A lot of corn and soy growers are losing $10 to $100 an acre this year, and they will pay $50 to $100 an acre for weed control that delivers. What they expect from that spend is the consistency and efficacy of a chemistry, which is the bar Quercus has set for itself.

The investor list points the same way. Alongside technology and climate funds such as Starshot Capital, Grantham Foundation, and Flyover Capital, the round includes producer-backed funds: Iowa Corn Opportunities Fund, South Dakota Corn's AgriCap Fund, Kentucky Corn, AgVentures Alliance, and Ag Startup Engine. Those are farmer dollars, from organizations that answer to farmers, and they have put the company in front of producers to talk through what weed control actually has to do in the field.

Lightner's read on adoption comes from watching Enlist and dicamba tolerance go from nothing to roughly half of planted acres in a couple of years. When something solves the problem, uptake is fast, because the demand has been sitting there the whole time.

Where this is heading

Quercus is one instance of something larger.

The last few years have produced an enormous amount of protein design capability, and nearly all of it was built for therapeutics. That capability is commoditizing now, and it is starting to move sideways into other industries. Quercus taps an engine built for human drug discovery and points it at weeds. There will be more of this, because the tooling no longer has to be reinvented for each new field, and because the industries that need it most are not the ones that paid to develop it.

Agriculture is a good test of how far the transfer goes. Little about designing a binder changes when you aim it at a plant enzyme instead of a human one. Everything about what that binder has to cost, how long it is allowed to persist, and how it gets manufactured does change. Whether protein design travels well is a question about the second list.

Lightner and co-founder Matt Crisp have been in this industry long enough to know a good many of the people in it. His observation is that, with the exception of growers, almost none of their first investors or partners arrived through those networks. They came from someone running across the company and deciding it fit their thesis. Quercus will be back in the capital markets in 2027. Between now and then, the company is looking for partners along the chain, from the majors down to the retailers who put product on the ground, and for the people who have not come across them yet.

Quercus Biosolutions is based in Madison, Wisconsin and St. Louis, Missouri. Founded in 2024, the company emerged from stealth in March 2025. Matt Crisp, founder and former CEO of Benson Hill, is executive chairman and co-founder.