Bioprospectors mine microbial genomes for antibiotic gold

In March, the journal Nature published the discoveries of two paradigm-breaking antimicrobial compounds in as many weeks: a polyene macrolide antifungal called mandimycin and a lasso peptide antibiotic called lariocidin.

Both these compounds, which use never-before-seen antimicrobial mechanisms, were found using techniques that let researchers look deep into the chemical diversity of microbes—much deeper than a typical antibiotic or antifungal screen might go. And it’s not just these two molecules. Scientists are using the new approaches to discover countless other antimicrobial compounds with the potential to become drugs.

Gerry Wright, a biochemist at McMaster University who led the research group that discovered lariocidin, likens the hunt for antimicrobials to the gold mining operations in his native Ontario, Canada. “In the old days, you’d look for a gold vein and you’d blast out that slag to get as much gold as you could,” he says. “Now, coming back with different tools, you can remine that slag and get more gold or maybe other precious metals.”

The first gold rush

The first bioprospecting operations took place nearly 100 years ago. The antibiotic penicillin was famously discovered in 1928 by Alexander Fleming, who isolated the compound from a fungus that was contaminating his bacterial cultures.

That marked the start of the first “golden age of natural product antibiotic discovery,” according to a frequently cited review paper (Curr. Opin. Microbiol. 2019, DOI: 10.1016/j.mib.2019.10.008). The claim of a golden age is often repeated in other scientific papers, most of which describe the peak of that period as lasting from the 1940s to the 1960s.

Researchers working during that time discovered and developed natural product–derived antimicrobial classes such as tetracyclines, cephalosporins, macrolides, and aminoglycosides. Most of these antimicrobials were found via whole-cell screens in which assortments of microbes—often found in soil samples—were grown alongside pathogens in a laboratory culture. Those cultures were then screened to see if the microbes inhibited growth of the pathogens.

This was a highly effective strategy for finding new antimicrobials, because for hundreds of millions of years microbes have been at war with each other in competition for resources. In that competition, one microbe evolves to manufacture a compound that inhibits the growth of another. And as that process repeats itself over time, microbes create a diverse assortment of drug-like compounds.

After the initial discovery of antimicrobials from whole-cell screens, medicinal chemists created many clinically relevant derivatives of those compounds. Today, most antimicrobials are natural products or are synthesized compounds based on natural products (Med. Res. Rev. 2009, DOI: 10.1002/med.20154).

The rate of antimicrobial discovery started to decline in the 1970s. Whole-cell screens often revealed only previously discovered compounds. Meanwhile, antimicrobial-resistant pathogen strains rendered some earlier drugs obsolete.

The last class of natural product–derived antibiotics to make it to the clinic were the lipopeptides. Daptomycin, a cyclic lipopeptide, was discovered in 1987 and didn’t enter the clinic until 2003.

But in the past decade or so, new techniques have enabled a resurgence of antimicrobial discovery. These discoveries include molecules that use novel ways to target bacteria. In 2020, Wright found a compound that fights bacteria by caging them; he struck gold again with lariocidin earlier this year.

Mining for antimicrobial gold

If whole-cell screens were the crude, antiquated way to find natural products—akin to the early days of gold mining—low-cost, high-throughput genomic techniques are the new tools. While 10 years ago it cost $5,000–$10,000 to sequence a bacterium’s whole genome, Wright says that now “we could do it for 100 bucks, and it comes back a really good quality.” And genome mining, as the process of searching is often called, makes it clear there’s a lot left to be discovered.

To identify drug-like compounds produced by bacteria, scientists take advantage of the fact that bacterial genomes are relatively small and that genes with related functions tend to be clustered close to one another on the genome.

Kai Blin is a computational biologist at the Technical University of Denmark working on antiSMASH (antibiotics and secondary metabolite analysis shell), a genome-mining software that helps researchers identify the secondary metabolite gene clusters that could lead to the discovery of useful natural products. It’s one of many platforms available that assist scientists looking for potential antimicrobials. Blin says discoveries made to date are just the “tip of the iceberg” when it comes to the chemical diversity of microbes.

antiSMASH looks for clusters of genes that code machinery that makes secondary metabolites. It’s these secondary metabolites that the bacteria use to interact with the world around them, and studying the biosynthetic pathways that make the metabolites can reveal a variety of compounds with potentially useful activity.

Blin cites tetracycline biosynthesis as an example. “We know precisely how every step in the tetracycline biosynthesis works. But the fun thing about this is that we also can find everything else that is produced by the same general biosynthesis pathway as tetracycline, and that opens up a big chemical space of completely different structures,” he says.

Searching through that complicated mix of compounds, natural product drug hunters are seeking flashes of antimicrobial gold.

Another way of discovering compounds from warring microbes is by looking for the genes that confer resistance on them. “The organisms that make antibiotics have to protect themselves from the antibiotics,” Wright says. “And so, very often within their biosynthetic gene clusters are hints about what that self-resistance mechanism might be.”

Traditional whole-cell screens for antimicrobials still have their place, Wright says. But that often means going beyond the obvious. His group found lariocidin in bacteria that also produce colistin—an antibiotic discovered in the 1940s. The scientists came across colistin first, and that could have been reason enough to start looking elsewhere.

But now scientists can separate the compounds being produced by a microbe via liquid chromatography or other means and test them individually for drug-like activity. That’s how Wright’s group was able to find lariocidin and distinguish its unique antibiotic properties.

Digging deeper

Artificial intelligence is also being used to make new discoveries in the antimicrobial world. Nikolay Tzvetkov, a biochemist at the Institute of Molecular Biology at the Bulgarian Academy of Sciences, says that “the number of molecules predicted by artificial intelligence is greater than the number of stars in the Milky Way.” And that could be a conservative estimate.

James J. Collins, a synthetic biologist at the Massachusetts Institute of Technology, has been working on developing AI tools to computationally predict new compounds that have antibiotic activity. His lab “trained a deep graphical neural network” on thousands of compounds that were screened for antibiotic activity, Collins says. When fed a novel compound, the team’s AI can look “bond by bond, substructure by substructure" and determine whether that compound will have antimicrobial properties, he adds.

That’s how the researchers in Collins’s group found that a compound originally developed for the treatment of diabetes has antibacterial potential. They dubbed the compound halicin after the computer in 2001: A Space Odyssey (Cell 2020, DOI: 10.1016/j.cell.2020.01.021).

Collins’s model, published in 2020, was trained on only 2,500 compounds, but 5 years later his training dataset has expanded to include 39,000. He hopes to develop this model to make predictions about the other properties of these compounds, including toxicity and bioavailability.

But Blin says AI technologies still have some maturing to do before they exceed the capabilities of human-derived models. Until more training data can be developed, “Rule-based tools still have a leg up,” he says. “The question is not, How do we build the next AlphaFold?” he says, referring to Google DeepMind’s protein structure prediction program. “The question is, How do we build the next PDB [protein database] that we can then use to train the next AlphaFold?”

Designing a microbial drug factory

Finding a compound of interest is a great start, but producing enough of that compound for testing and optimizing it further remains a challenge. And that’s where synthetic biology can shine.

Though Fleming isolated penicillin 1928, it was another 16 years before Pfizer chemists would figure out how to ferment fungi in large tanks. Today, fermentation facilities are producing massive amounts of drugs relied on around the world. But even the most advanced sites have a hard time persuading stubborn bacteria to make more of a compound that they typically produce only in small quantities.

That’s why Wright calls the use of synthetic biology “transformative” when it comes to manufacturing natural products for laboratory testing. “The wild organism rarely tends to do what you want it to do in a lab,” he says.

Researchers will often grab the whole biosynthetic pathway for a specific compound and put it into another organism specifically designed to produce that compound at high concentrations. That move makes studying some of these compounds possible.

But Eriko Takano, a microbial synthetic biologist at the Agency for Science, Technology, and Research, says that process is not so simple.

Taking a biosynthetic gene cluster—such as one found using antiSMASH—and putting it into another organism is just the start of the synthetic biology process. From there, researchers like Takano work to tweak the pathway with different enzymes, either to make production more efficient or to try out modifications of the original molecule. “We can modify our core structure so that it has different characteristics—maybe it’s more active or has less toxicity or more solubility, she says. “There's so much that one can do with synthetic biology.”

AI can also lend a hand here. Let’s say Takano wants to add a carboxyl group to a molecule. Previously, she would have a student dig through the literature and test dozens of carboxylases to find one with the right kind of activity. But now, AI tools can dig through that same literature and make enzyme predictions; liquid-handling robots then perform the routine wet-lab work to make and test the options identified. This automation helps reduce the hours needed for raw experimentation.

Getting an antibiotic approved isn’t easy

While the types of tools able to find new compounds that might have drug-like activity have expanded greatly, validating that those molecules do something interesting is still a major bottleneck. Getting them approved for clinical use is another challenge.

For Wright, Collins, Takano, and many other academic scientists in the drug discovery space, their role tapers off after validation. Generating safety profiles, putting a drug candidate through clinical trials, and—if those are successful—scaling up manufacturing are jobs too big for academic labs to tackle.

The burden of testing new antimicrobials inevitably falls to the pharmaceutical industry, which faces its own hurdles. Antimicrobials offer smaller profit margins than other drugs, and their rate of failure in clinical trials is high. That’s partly why Takano says that only a few out of hundreds of potential antimicrobial compounds will pass testing and make their way to the clinic.

The PASTEUR Act, a bipartisan bill that has been proposed in the US Congress a number of times, aims to appropriate $6 billion to fund antimicrobial development and give the Department of Health and Human Services the ability to enter subscription contracts with companies developing critical-need antimicrobials. But it has been stuck in political limbo for some time and is unlikely to go anywhere soon.

So, despite the novelty of lariocidin’s mechanism of action and its lack of reported toxicity in initial testing, the drug candidate has a long road to the clinic and approval. Yet Wright remains optimistic about the potential for lariocidin and many other recently discovered compounds. “We’re at the beginning of what I would consider a new golden age,” he says.

Only time will tell if all that antimicrobial gold really pans out.