The history of pharmaceutical development has traditionally been one of exploration on the frontiers of life on Earth. From fungi to molds, we’ve sourced many of our most important drugs from some of the unlikeliest places, and it’s all due to evolution. Nature’s intense competition and selection forces has made it the ultimate developer of pharmaceuticals, with potential cures lying in wait for someone to find them.
Searching nature is expensive though, and thus, pharmaceutical companies re-centered around synthetic chemistry over the past few decades, hoping to realize a more reliable and inexpensive drug discovery model. Unfortunately, we have hit a logjam with such an approach, and the evidence is clear that natural products are often regularly superior to synthetics.
We wanted to dive deeper into the future of biopharma, and so we brought together our own Tess Van Stekelenburg and Elliot Hershberg, the writer of Century of Bio, to work on a new two-part mini-series for the Riskgaming podcast, titled Evolved Technology. It’s an extension of a series of talks that Tess and Elliot (“two crazy bio-optimists”) have conducted in SF, and we hope it illuminates a critical scientific frontier with implications for all of us.
In this first episode, Tess and Elliot talk about the editing of life; why thousands head to the Himalayas to find tiny caterpillars in the dirt; the business history of natural products in pharma; the transition from natural products to synthetic chemistry; the limitations of our current biochem toolkits; and finally, how AI/ML are bringing us back to the search for natural products using higher-order models.
Produced by Christopher Gates
Music by George Ko & Suno
Transcript
Tess van Stekelenburg:
Elliot, thank you so much for joining us today. You and I have both thought and talked a lot at length about this theme of Evolved Technology. The key inside of that is Biology is Technology, and so to lend terms from computer science, biology is a formal language. It has nucleic acid. It's able to encode functions that arrange matter at atomic scales. Because genetic sequences are inherited, many of these functions are not random, but instead have been sculpted and conserved by thousands if not millions and billions of years of evolution for the survival of specific hosts. What started off as random iterations is now entering a unique era, that of intentional search and ultimately design. With new tools like machine learning and sprawling databases at our fingertips, we're starting to decipher the language of life and start to identify these functions in order to leverage them as molecular tools or medicines to weld the world of atoms around us. Instead of programming in a digital realm, these tools allow us to act on matter itself.
And so this is something that you and I started talking about and thinking about both through events that we were hosting, private salons, and it's something that we both wanted to bring into the world. Do you want to give a bit of background on that?
Elliot Hershberg:
I think as two crazy bio-optimists, we had something that was starting on the back of napkin, a set of observations about related companies in this space. And it's evolved and turned into a dinner series where we've brought together like-minded people to think about this, to argue about this. And now I'm excited to put this into a podcast format where our first start is going to be looking at the world of small molecules and natural products, and there's a whole story to also tell in the world of biologics. But Tess, what's your story here on first observing natural products and that this came into your consciousness.
Tess van Stekelenburg:
It actually takes place around five years ago when I was on a very long hike of 15 days in Nepal and I had been hiking at that point for 10 days, acclimatizing, altitude sickness was coming in. As you kept going higher and higher, there's less and less people there. And then at some point we were about to summit a mountain, I think we were like 3,000, 4,000 meters altitude. There was just this huge amount of tents and people and it was like fair that was taking place in the middle of the Himalayan Mountains. And it turns out that there was all these people from Tibet, the Tibetan Autonomous region, that were picking this one fungus called yarsagumba, which is a parasite that grows in the brains of caterpillars and then it dries them out. And so what ends up happening is that they die by digging themselves into the soil and the top of their head sticks out approximately one centimeter above the soil.
And so, these people, it was thousands and thousands of people that were staying there for months intense, can spend their entire day just hours long search for one caterpillar. And the reason they do this is because this ends up selling for more than gold. And so what could possibly be worth more gold per kilogram? Viagra. So inside of this fungus is an active compound that is known as Viagra, or the Himalayan Viagra, and it ends up being the largest income for Tibetan households because they're selling to one of the largest markets in the world. China.
Elliot Hershberg:
What you're observing out in the Himalayas is probably one instance of thousands of natural products that specific communities have already discovered and used as a basis for medicine for hundreds [00:03:30] and hundreds of years. But what's interesting is that this is also the case for western medicine. So there's a fundamental similarity in how development happens across traditional medicine and across the origin of the pharma industry.
One crazy example is statins, which are one of the greatest blockbuster bioproducts of all time. This was initially discovered in the 1970s by a Japanese biochemist named Akira Endo, and he was looking for ways to treat high cholesterol for cardiovascular disease. And he hypothesized that fungi may be producing compounds that inhibit the cholesterol synthesis of competing organisms. What Endo ends up doing is isolating a compound from a microbe called Aspergillus terreus. That's a type of fungi. And this was the basis for the first commercial lovastatin that ends up getting developed by Merck and Co, who's making subsequent statins based on this, that is estimated to have produced over $170 billion in lifetime revenue from this franchise, right? So one of the greatest blockbuster drugs of all time in the West was found in a fungus by a Japanese biochemist in the 70s, right? This is recent history for the development of drugs.
Tess van Stekelenburg:
I think that's one of the highest grossing revenue drugs that we've seen in modern medicine. And that was just one of many. For the longest time in history, whether it's ancient China or Mesopotamia or Egypt, the way that we did medicine was finding these active compounds from fungi, from plants, whether it was the opium poppy where we got pain management from or anti-malaria. Over time as we got new tools like the ability to isolate chemicals to purify them, we saw the birth of modern chemical companies, which ended up turning into the first pharmaceutical companies. In the 19th century almost every major chemical company that was originally producing dyes starts going into this space called natural product. Merck has Vitamin B1 and then ivermectin. Bayer ends up taking aspirin from the willow bark and Pfizer ends up commercializing penicillin.
Elliot Hershberg:
If we've been pulling out things like statins from fungi, why are we still not doing that today? Why does the pharmaceutical [00:06:00] industry not look like that? It would be the end of the 19th century, in the early 20th century, we start to see massive pharmaceutical companies on the basis of early wins and natural products. But by the time they actually have resources and get to scale and become industrialized, in the middle of the 20th century, we see the development of synthetic chemistry and what we would now consider to be industrial drug discovery.
And this ultimately gets brought forward to its logical limit where Joshua Boger, who was a brilliant young scientist at Merck, was seeing the giant sets of libraries that were being produced in the way that drug discovery was currently happening. And he has the fundamental idea that with advances in x-ray crystallography, computer modeling and new approaches, we could actually do full-fledged rational drug design where we're designing small molecules against the actual structures of proteins and it's trying to completely systematize drug discovery, new tools, [00:07:00] computation, synthetic chemistry, that this becomes true engineering and it's more engineering than an art form at this point.
And so it's interesting to think that the same companies that had such a meteoric rise from these blockbuster products that were originally natural products. Once they got to scale, it was more enticing to be able to do this in an industrialized way with synthetic chemistry and more standardized approaches to drug discovery.
Tess van Stekelenburg:
And so in the middle of Merck in 1989, Merck was starting to use new tools to leverage structural biology to do drug discovery. They started having their own research divisions and then a group of scientists there, the whiz kids with computers, ended up spinning out and starting their own company that had a meteoric rise. This company was known as Vertex.
Elliot Hershberg:
This is a $100 billion market cap company that's been one of the generational biotech companies in our lifetime.
Tess van Stekelenburg:
Right. And so we go from natural products, which are largely derived from finding things in nature, to all of a sudden understanding receptors and these targets in the human body and then creating chemistry, synthetically, to go and activate them or modulate them. And by 1997, this process was formalized by Lipinski who came up with the Lipinski's rule of five. It's essentially a set of heuristics to standardize drug discovery and organic chemistry. But what's interesting about natural products is that nearly all natural products that have been approved to date break this Lipinski's rule of five, and yet they're some of our most potent drugs. What's weird about that is that there's almost this paradox, shout out to Enveda Biosciences for an upcoming article where they find that these drugs actually have much higher rates of succeeding than their synthetic counterparts.
Elliot Hershberg:
Why is this? How is evolution producing these structures? They're often so crazy that they can be outside the bounds of what we can synthesize with synthetic chemistry. So there's a company Antheia, which is a synthetic biology company, that is producing natural products in microbes because we can't actually synthetically produce these and so we've historically been sourcing them from the organisms that are producing them in nature, right? So this is outside the bounds of what we can even do with our current chemical toolkit.
So how is that the case? Ultimately, evolution could be the ultimate technology developer. If we think about what evolution is actually trying to do. [00:09:30] Evolution does not care about Lipinski's rule of five. It's not sitting and checking heuristics. It's a giant search algorithm trying to solve problems for biological organisms.
Tess van Stekelenburg:
And so one of these giant problems as you can imagine, is survival. Each of these hosts wants to survive. What the functions that, just to go back to the computer science analogy, the functions that they are encoding in this formal language, which just get optimized over time, are things like solving the problem of competition for resources or being able to attract a mate or being able to go through long periods of fasting and feeding. And so we end up getting these proteins or compounds that are able to do things like break down the cell wall of another organism, statins block an enzyme in the synthesis of sterols, which is crucial for cell membranes, and that reduces the ability for their competitors to grow and survive. We have penicillin, which was derived from mold, which act on bacteria to stop their growth as well.
And so the rules that we're seeing at this genetic level is really all of these conserved motifs that serve a very broad range of biological functions. So for example, in the case of natural product, there's a huge amount of ligand protein binding motifs that are conserved across all of them. And so there's this natural search for a human or a target that humans might already have rather than trying to design something for a target, and we don't know if it's safe, if it works, nature has already come up with something that is highly effective over billions of years.
Elliot Hershberg:
If we think of the set of tools that led to the departure over time in the pharmaceutical industry from natural products, there's this opportunity set in the chemistry encoded by nature. And what's one of the core themes that was the basis for the evolve technology discussions that we've had, the dinners is the fact that there's been this new set of tools that we've seen, this resurgence of companies and science returning to the exploration of natural products.
Tess van Stekelenburg:
Being able to isolate chemicals and purify them led to the rise of the first pharmaceutical companies and chemical companies. And then if X-ray crystallography and computers, the ability to simulate chemicals led to the rise of rational drug design and Lipinski's rule five, what's happening now is that sequencing costs have been plummeting. And so we have this thing called omics. Genomics, proteomics, metabolomics. [00:12:00] Mass spectrometry tools have also increased in their use, and so we can actually look at the metabolites. And so we have this huge amount of databases that are just sprawling on the internet and being generated with data that is far too large and complex for us as humans to look at and understand.
The other thing that's been happening concomitantly with that has been the rise of an architecture of machine learning that's actually able to take in this data. Let humans search through the patterns of it without being able to understand what the actual structure or semantics of it are. And this is transformer architectures.
Elliot Hershberg:
And so let's look at a specific example of this to make it more tangible. One of the companies that's trying to build a platform around natural product discovery is Enveda Biosciences, and they are using a tool called mass spectrometry, which is able to tell you the fragments of molecules that are in a sample. So if they have a plant or a different organism, they can put it through a mass spectrometer and understand the general fingerprints in the chemicals that are represented. And historically when you get data off of this kind of instrument, you do a basic database lookup relative to the fingerprints that are already known and stored in different scientific databases.
What you can imagine is an iceberg where what's represented in these databases relative to what's actually in the sample is just the tip of the iceberg. And so what Envada has done is generate a massive mass spectrometry data set at scale and built a transformer model. So we think of language models that are learning a representation of natural language, and this is a transformer model called ms.to.mol., mass spectrometry to molecule, which is learning an underlying representation of chemistry. And what they're able to do is find a much greater set of the molecules that are present in their mass spectrometry samples and scale the discovery of natural products through technology and artificial intelligence.
Tess van Stekelenburg:
One of these genomic functions that's highly conserved is the ability to act on things like cell division or cell growth or the protective layer of your competitor, which is the membrane and actually damage that. And so, if you look at what a lot of the mechanisms of actions are, of ways that whether it's like yeast and bacteria, the way that they operate on each other, is really by fighting each other. And many of those functions can also be very, very relevant to indications like cancer, where you have rapid cell division, you need to somehow identify cells in your body that are cancerous versus non-cancerous. And so the commonality there is the speed at which they're dividing. And so you can definitely imagine that natural products are a great search space for that because they're already targeting this cell division process or cell growth process.
Elliot Hershberg:
The CEO of Hexagon, Maureen Hillenmeyer, was previously the Director of Genomes to Natural Products at Stanford, and it's the idea that with the sub Moore's law cost curve in genome sequencing, the advance of synthetic biology to actually express clusters of genes and coding molecules, that we can systematize the discovery of natural products for these ADC payloads and other types of molecules by literally being able to go directly from an organism's genome to the chemistry that that organism encodes at scale. And so that's sort of the bet that they're making.
It's a really interesting search space for natural products because there's this new class of modality called an antibody drug conjugate, which is an antibody molecule which gives targeting properties to a specific extracellular protein inside of the body that is linked on to a molecule that you can drag to the site of disease. These are really promising modalities. It's been a huge growth of development, [00:16:00] mergers and acquisitions recently. What's crazy about this story is that the payloads that are typically linked onto these ADCs that are recruited to cancer cells are nearly all based on natural products. So these molecules that are highly toxic to cancers, that are targeting these growth related genes that are coming from the natural world through this process of microbial warfare with fungi trying to evade microbes.
Part of Hexagon's thesis [00:16:30] is with this scale of data to be able to systematically pull out new compounds that are maybe less toxic in a specific instance or more potent with a better safety profile to really expand the set of these molecules that we can use for ADCs in cancer.
Tess van Stekelenburg:
Just to go back to this paradox, we thought that coming up with these highly sophisticated set of parameters of what makes a molecule drug-like, and yet this human rational parameters [00:17:00] that we've set don't end up showing us the most successful based transitions in clinical trials. So if we look at natural products versus synthetics, natural products have a much higher success rate when they go from phase one to phase three in clinical trials versus the inverse trend, which we're actually observing in synthetics.
Elliot Hershberg:
So why have people actually departed from natural products? Again, historically without these new discovery tools, it's been really hard to actually find and isolate natural products, right? It's different discovery [00:17:30] process than working with synthetic chemistry in the lab. It's been challenging to actually screen natural products and to characterize their function. Then there's a whole host of complexity for natural product IP and actually being able to manufacture natural products that companies like Antheia are trying to develop. There's been an enormous amount of productivity from the Vertexes of the world, a huge amount of value created in new medicines. And one of the big theses for this new wave of natural products companies is [00:18:00] that they can actually overcome some of the bottlenecks to natural product development at scale and bring these really safe and potent medicines from nature's chemistry.
Tess van Stekelenburg:
Even though they have all this potency at the start, don't actually end up working as good medicinal chemists starting products. Medicinal chemists don't like working with natural products as much. And there's also questions around IP. It's very hard to patent something that occurs in nature. It's much easier to patent something that is completely synthetically created. And then we go from these rules, we're trying to create this very [00:18:30] rigid list of what makes something drug-like. And it turns out that there might actually be a lot of features that make something drug-like, which we are, as humans, not able to comprehend or understand. And so, this is where having super high-dimensional data and having machine learning architectures that we don't quite understand, pull out those rules for us, where, at this point, I don't think we know still why natural products have the potency that they do. They just work. And so there is something to say about evolution being [00:19:00] the ultimate technology designer, even beyond our own understanding or comprehension today.
Elliot Hershberg:
This counterintuitive technology arc where, as the technology improves, we're actually seeing a return to the natural world and then a move away from synthetic chemistry, is really counterintuitive. And that's the core theme that we're trying to encapsulate and put words around for Evolve Technology that you would potentially expect this indefinite tech cycle in the limit [00:19:30] to be just abundantly producing perfect molecules systematically with synthetic chemistry. But what we're observing in the market and in labs around the world is that as these tools come online, we're learning that there's a chance that evolution is the ultimate technology developer and that we should be pointing our AI algorithms and our measurement technologies at nature's chemistry.
Tess van Stekelenburg:
And so this is what we've seen so far, but we will continue to cover other variations of this theme, particularly looking at [00:20:00] the world of proteins and other types of hosts like viruses and bacteriophages acting on each other. So I'll leave with the question, is evolution the ultimate technology developer?
Elliot Hershberg:
Cheers. Thanks, Tess.