How chemistry can make us healthier

0:00Hey, everybody. It's Farnoosh Charabi from the So Money Podcast. Today's episode is sponsored by NerdWallet's Smart Money Podcast, the show that breaks down financial decisions with a team of trusted journalists. You'll get research-backed insights and clear pros and cons, whether you're planning a big purchase or just want to grow your wealth. They explain the why behind decisions like investing, home buying, and choosing credit cards with clear, research-backed insights. Make your next financial move with confidence. Follow NerdWallet's Smart Money Podcast on your favorite podcast.

0:58podcast app. I'm Rachel Ralph. Today, we ask, how can chemistry make us healthier?

1:16Today, we're examining the role that chemistry plays in medicine and is helping to break the mould in terms of how we predict, diagnose and treat disease in a personalised way. Historically, many assumed that the answer would lie within the DNA code we all carry, and that reading that code would point directly to disease risk factors. But genes rarely work in isolation, and carrying a disease-linked gene doesn't guarantee that that condition will ever rear its ugly head. Because what matters most is how the constellations of genes we all carry interacts with the environment we inhabit and the lifestyle we lead.

1:56This is where the science of phenomics comes to the fore. Here, scientists aim to measure and compare the relative levels of thousands or even millions of molecules and link those to ultimate disease outcomes. The fundamental premise is that the biological milieu inside each of us is dictated by our genetic makeup, working with the raw materials that our living conditions provide. So if we measure our biochemistry, we can infer where our true disease risks are lurking. A pioneer in this space is Jeremy Nicholson at the Hong Kong University of Science and Technology, and he's been speaking with Chris Smith.

2:34What can you tell us in terms of the constraints of our present way of trying to diagnose things or work out who's at risk of things with techniques like genetics and so on? In the case of genetics and genomics, genomics is the ability to measure lots of genes for an individual and to determine the genetic composition. Those technologies have moved on very, very quickly and from an analytical point of view have become extremely rapid in comparison to what they used to be. However, they are still limited in what they can tell you and also in terms of the absolute capacity. So at the moment, the total world capacity for doing human genomics is probably about a million, maybe a million and a half a year, which sounds like a lot until you remember there's nine billion people on the planet.

3:20So it can take thousands of years to do all the genomics with our current technology. But the important thing really for me is that genes don't tell you everything. You get genes at the beginning of your life, of course, and they operate on lots of different pathways. But they're impacted on by the environment. So the so-called gene environment interaction, how you eat, how you live, whether you exercise or not, all of those things impact on your risks of getting disease later in life. None of that is captured by genomics. None of the interactions are. So the genomics just gives you the potential to get a disease based on a set of known genetic variables.

3:58So say I had a gene that meant that I was at high risk for alcoholism. My insurance company would have something to say about it. But if I never went near a drop of alcohol, that would make absolutely no difference to my disease outcome, my longevity expectation, etc. Because it wouldn't have the chance to impact with an environment that would manifest the effect of that gene. Absolutely correct. So given the constraint of that, I mean, if we just read people's genomes, we might get the wrong impression then. How do we fix that? Reading people's genomes is not really practical at scale for our populations.

4:32Genomics is actually quite good in certain very selected areas, looking at the BRCA mutations for things like breast cancer, ovarian cancer, etc. Very, very useful in terms of determining risks for people and also trying to select therapies. We call that targeted genomics, where you're looking for some very specific markers of disease. But in terms of whole genome sequencing, we are literally nowhere at scale for being able to screen people for different genetic markers that indicate likelihood of getting certain diseases. How do we solve the problem then? Well, we have to go more than genetics, adopt technologies that are scalable and translatable to the general population for important diseases.

5:17So one of the ones that is a favourite of mine is cardiovascular disease. It's one of the things that kills most people in the planet and it's a combination of genes and environment, what you eat how much you exercise but genomics is not really very good at predicting that at all but your metabolic features are so when we were doing our covid19 research a few years ago we discovered some new markers for cardiovascular risk in the lipoprotein patterns that you have in in the blood which are affected by the virus that change people's cardiovascular risk and cardiovascular disease is one of the consequences of long covid for instance now those discoveries that we made which were through chemistry we were able to make a simpler test based on what we discovered and scale it down to much smaller instrumentation which we can now deploy i won't say in your home but much closer than a big hospital or a big laboratory so the idea of discovering markers which indicate risk for major diseases which you can then reduce in size reducing cost and scale up so that almost anybody could have one of those tests for you know five pounds or something like that isn't the big problem though that basically we are a biochemical haystack you and i everyone on the planet and there are millions and millions of unknown unknowns so we don't know what to look at so how do you find the needles those markers that you're referring to in those biochemical haystacks to say those are the ones that are important those have got the prediction power about Jeremy or Chris's future cardiovascular risk?

6:52What we call exploratory analytical chemistry. Some people call that fishing, by the way. But you have a sort of shotgun approach where you measure as many things as you possibly can, usually using techniques which measure lots of things simultaneously. Nuclear magnetic resonance, spectroscopy, mass spectrometry being two chemistry examples. And then you have two groups or more defined groups you have something that's normal something that's abnormal and you say well out of all those thousands of things that we measured what are the things that are actually mathematically statistically different that are associated with the disease and that statistically cuts down the the search space and some of those molecules as you quite correctly point out are often completely unknown they may not even be in a chemistry or biochemistry textbook anywhere there's a lot of complexity in humans so what we have to use is chemical technologies including NMR and mass spectrometry to structure those compounds find the exact chemical structure and then we can once we know the structure we can start to figure out how that would relate to the disease process is it a biomarker that's a consequence of that disease process is it a marker that's to relate to the mechanism of the disease process and once we understand that mechanistic and chemical detail that underlies the disease process we can start thinking about whether that process can be druggable whether we can build something that either blocks a pathway enhance the pathway whatever it happens to be.

8:12You were saying that the current bottleneck with doing this genetically is that we just can't do enough genomes to make this realistic but is detecting these biochemicals more realistic then? Is that something we can do in a high throughput way to screen enormous numbers of people, look for enormous numbers of molecules and then marry those molecules up with clinical outcomes to see what predicts what? It's not completely scalable to all diseases because we don't know all the different properties but for many important diseases that will be tractable and the other thing is the technology marches on so chemical technology is always on the move so when we were talking a few years ago about large-scale phenomics we have loads and loads of mass spectrometers and NMR spectrometers measuring thousands of things but there are new mass spectrometers available now which can measure a million samples in a week million urine samples or a million blood samples for hundreds or thousands of molecules that technology has only really come about in the last few years and it's one of the things that's sort of on our list that we need to be testing those technologies for scalability to general populations so it's a combination of things it's a combination of knowledge and the scientific approach to complex systems and understanding interactions between genes and environment and what that means in terms of the body, but also at the same time, you have to have the appropriate advances in technology and also in the computation.

9:40All of this involves a lot of numbers. So artificial intelligence has become an incredibly important part of what we do as a group and will be important in the future for making these technologies more tractable. Jeremy Nicholson at the Hong Kong University of Science and Technology. Chemistry in diagnostics, also known as clinical chemistry or chemical pathology, uses chemical and biochemical analyses of body fluids to diagnose, monitor and manage diseases. A team at the world's largest menstrual fluid biobank is doing exactly this in a bid to transform our understanding of menstrual health. Gemma Sharp, who is leading the cycle track study at the University of Exeter, has been telling Chris Smith all about it.

10:29We are tracking people who menstruate over three menstrual cycles. We're collecting really rich data on all aspects of their lives, what they've been up to recently, their symptoms, how they're feeling, mental health, physical health. And we're also collecting menstrual fluid alongside that information as well. And then we're hoping to use all of that really rich data to try to understand a bit more about menstrual health in general. I think there's loads of questions that we can probably address with this type of data. But our specific kind of first question that we're interested in is, can we identify women who are at risk of iron deficiency or harm from their heavy menstrual bleeding.

11:24This is in essence then a menstrual biobank. You've got all the clinical data and you've got the samples to go with it. Yeah exactly. So we're going to have about 3 ,600 samples alongside all of this really rich phenotypic data, so data that's non-biological. What ages are you going to consider and how are you going to recruit them? One is called Children of the 90s and it's based in Bristol and those individuals have been followed since birth and they were all born in 91 or 92 and then born in Bradford is our other cohort and they're based in Bradford and they are mostly the mums of the birth cohort so they're all women who had a baby around 2007 to 2011 so they're a little bit older but we're focusing specifically on the women who are still menstruating.

12:17So they're around mid-30s to mid-40s. Let's talk about the practicalities for a minute. You're talking about a biological fluid. How are you going to collect this and how are people going to get it to you and how are you going to store it? Yeah, so menstrual fluid is very tricky to work with. It's a combination of blood but also other things. So things like tissue from the lining of the womb. There are lots of enzymes in there that tend to break it down and it's very difficult to transport. So we didn't want participants to have to, you know, get us a sample as soon as they'd collected it and get it straight to our lab.

12:54So we've been working with a company that makes a menstrual pad that's kind of modified to collect a dried menstrual fluid sample. So this means that the menstrual fluid is transferred onto just a filter paper strip that's kind of just hidden inside of the menstrual pad. And then that filter paper strip dries the blood really quickly. They just put it in a little canister and they send it off to our lab. So it comes to us as a dried sample and it's quite easy for participants to do. And then it's easier for us to work with it in the lab as well. How will researchers find that useful? What will they be able to do with that that they couldn't do at the moment with what we have available to us?

13:38so currently if you want to identify whether somebody has heavy menstrual bleeding it's actually really difficult so it sounds kind of easy you just ask them do they have heavy periods however a lot of people don't know we've got really poor understanding of what a heavy period is and it can mean different things to different people and because menstrual fluid isn't just blood you can have quite a high volume of what looks like blood loss but actually not losing that much blood and the opposite way around so losing quite a lot of blood and not not really aware of it what we're really hoping is that with the samples that we're collecting we'll be able to develop a diagnostic tool that helps us to identify which women are at risk of harm because of their heavy menstrual bleeding from their actual menstrual fluid i completely get where you're coming from because I mean when I learned obs and gynae one of the first things we were told is that someone actually did a study in the 70s and they actually asked women to weigh what they collected each month and some women who had enormous amounts of material declared their periods as not that heavy and vice versa so that you're right there does appear to be a lot of different interpretation is it a subjective report from the woman I've got heavy periods and then you get the sample and you'll just see if there are actually differences or will you actually be doing some objective measurements as well to see if someone really does have heavy periods as we would define in terms of volume of blood loss?

15:08Excellent question. So in our study we are using just questionnaire data. Our study is fully remote so we are asking women about their periods and we're not just asking them are they heavy or are they light, we're asking them you know do you have clots, do you pass a large amount of blood in one go? How often do you need to change pads? But we still don't know if those, we're trying to get a more objective measure, but we don't really know if those questions or which combination of those questions is going to be most important. So we're also doing some other work in other cohort studies to look at comparing to the objective gold standard of this is what heavy menstrual bleeding is and comparing that to the questionnaire data that they're also collecting.

16:03So the gold standard approach is to use is an approach called the alkaline hematin test and this is so difficult to administer there's just no way that it would ever be used at scale at the sort of scale that we're using in our study and it certainly wouldn't be used in clinical practice because it basically involves women providing every single tampon that they use throughout a period and then that being taken to the lab immediately and tested to see how much blood it has in it so yeah so we're trying to kind of find the middle ground between that highly objective but very difficult to measure approach and questionnaire-based approach.

16:53There are lots of other things you can presumably probe with this because you're not only getting samples of menstruation but you're also going to get presumably a snapshot of microbiome that's there. You'll also have all the clinical data, people's menstrual cycles, frequencies, changes. So it must be a real goldmine this potentially. It is a real goldmine. I think it's going to be a real game changer for the field. The wider program is called the MIS-Vital Sign and that includes many other research teams that are working on like slightly different aspects of this same problem. and some of those teams are looking at microbiomes so it'll be really interesting to compare what we're looking at in our study we're looking at DNA methylation it would be interesting to compare what we're finding to what they're finding in their studies but there's so many things we're focusing on heavy menstrual bleeding we could also look at you know mood fluctuations across the cycle endometriosis fibroids menstrual pain just a ton of things.

17:55That was Gemma Sharp at the University of Exeter. Gemma and her team already have dozens of participants involved and should have data ready to use by 2028. We wish them the very best of luck with their project. Hey everybody, it's Farnoosh Charabi from the So Money Podcast. Today's episode is sponsored by NerdWallet's Smart Money Podcast, the show that breaks down financial decisions with a team of trusted journalists. You'll get research-backed insights and clear pros and cons, whether you're planning a big purchase or just want to grow your wealth. They explain the why behind decisions like investing, home buying, and choosing credit cards with clear, research-backed insights.

18:35Make your next financial move with confidence. Follow NerdWallet's Smart Money podcast on your favorite podcast app. The Naked Scientist podcast is produced in association with Spitfire, cost-effective voice, internet, and IP engineering services for UK businesses. Find out how Spitfire can empower your company at spitfire.co.uk.

19:03This is the Naked Scientist podcast with Rachel Ralph and Chris Smith. Today, we are examining the role of chemistry in healthcare. Still to come, how computational tools aid in the design of new medicines. But first, chemistry, of course, forms the foundation of modern drug treatments, The process involves designing and optimising chemical compounds that interact precisely with biological targets to cure and manage diseases. Here's Chris Smith with neuropsychopharmacologists and a friend of the programme, David Nutt. So David, when I pop a pill, it's a paracetamol, how does it know where the pain is?

19:43The molecular constituents, i.e. the active ingredient in paracetamol, flowed through the body until it finds something to stick to. And actually, we don't really know what paracetamol does stick to. But one of the interesting theories is that it sticks to an enzyme which breaks down the brain's natural cannabis. And the enzyme is called FAAH. Paracetamol blocks that enzyme, so it boosts the brain's cannabis levels, such as anandamide. And those then dampen down the nerves, which cause the pain. So if we take a different example then, like aspirin, because that also works similarly but differently.

20:21It's another kind of non-steroidal anti-inflammatory drug. And I've got an aching finger. It stops my finger hurting. Now, that can't just be enacted at the level of the brain, because if I stick a pin in the other finger, I still feel the pinprick. So it seems that it's done something locally where my finger was hurting. So how does the aspirin know to do that? Yes, so aspirin floats through the body, and then it finds an enzyme called cyclooxygenase, which it blocks. and by blocking that it stops the production of inflammatory molecules which cause the pain in your fingertip. So in this respect basically these drugs are designed or we have managed to find molecules which happen to engage with targets that are somewhere in the pain pecking order.

21:05Aspirin was a derivative of a molecule called salicylic acid which was used for millennia for pain because it's found in willow bark. Salicylic acid was converted to acetyl salicylic acid by a company called Bayer, sold as aspirin, and it's become the most best-selling drug in the history of the world, I think. Similarly, morphine was discovered to be the active ingredient in opium. People smoked opium or chewed opium for thousands of years to dud and pain. When the active ingredient was found, morphine, Bayer took the molecule and stuck a couple of acetyl groups on it. I changed its name from morphine to heroin, patented it and that became a very popular painkiller as well.

21:49So we can adapt naturally occurring molecules which do all sorts of useful things and turn them into pharmaceuticals which we can patent and then license and then promote. The key thing is though that these molecules go into the body and they interact with a target which is in some way linked to the process that we're seeking to treat. In the case of pain that may be nerve cells conveying pain it may be making chemicals that make nerve cells sense pain but either way we are targeting something with a molecule exactly and the examples i've given you so paracetamol and aspirin they target an enzyme or different enzymes and morphine and heroin they target a receptor and those are the two main targets enzymes and receptors normally we block enzymes and that stops them making things we don't want or breaking down things we do want.

22:44So I mean a really good example of that were the very first antidepressants. They were called monoamine oxidase inhibitors and they block the enzyme monoamine oxidase which chews up serotonin and dopamine and adrenaline. So by blocking that you get more of the neurotransmitter serotonin to lift mood. The morphing receptors they work on nerve cells to calm them down and when we stimulate them with morphine we calm the nerve cells down even more and that takes away the pain so that's through receptors but was a lot of this guesswork and or informed guesswork as in a pharmaceutical company would have just been trawling through nature's medicine chest things that chemists could cook up just hoping for things that might work there was nothing initially rational about this apart from we want to find something that might have an effect the pharmaceutical industry began in the 1800s with, really with aspirin, I suppose, when it was clear that there was an active ingredient.

23:44So rather than having to use willow bark extract, you could extract from willow bark the active ingredient, salicylic acid, and then make that into a different kind of molecule. And the conversion of salicylic acid to acetyl salicylic acid was largely driven by commercial interest but it also did change the molecule you got less gastric irritation so basically the side effects were reduced and in many ways almost all pharmaceutical development until the last 20 years was based on knowing something worked finding out what was in it that made it work and then either changing it to make it work better or making something very similar that worked as well or better.

24:29And so what has been the step change or the shift in the last two decades that has marked now apart from then? So the change was working out what the targets were, whether they were enzymes or whether they were receptors. And in some cases, again, with antidepressants, there are things called uptake sites. So drugs like the SSRIs, drugs like escitalopram, they block the serotonin transporter, the serotonin uptake site. So increase the amount of serotonin in the synapse. And those are the three main targets. There's one other target, which are ion channels, which conduct sodium, potassium and calcium ions.

25:06And there are some medicines that work there. But identifying the target protein allowed people then to develop bench assays, test tube assays, against which they could screen millions, literally millions of compounds, to find ones that did what they wanted. And then they could patent them. That was David Nutt at Imperial College London. Now finally, we're going to examine what the future of chemistry in healthcare might look like. Computational tools like AI could let scientists design new medicines from scratch, rather than discovering them by chance, that famous example of where Alexander Fleming accidentally discovered penicillin back in 1928.

25:48David Baker is an American biochemist and computational biologist, and he won the Nobel Prize in Chemistry in 2024 for his work on exactly this. Chris Smith asked David to tell us how AI is pushing the boundaries of medicine and what might come next. The really exciting thing that has come about in the last few years is the ability now, given a receptor, to design completely new molecules that bind to it to modulate disease. And this has been made possible by the development of deep learning methods, which allow us to now generate new molecules that have pretty much whatever properties we specify.

26:35So in the case of targeting a receptor, we provide these generative AI methods with the structure of the receptor. And much in the way that an image generation program might generate an image, if you give it a text prompt, these AI methods now will generate a new biomolecule that will bind to the receptor. And then once we've done the design on the computer, we can test that design in the laboratory to see whether it binds the receptor and then determine what effect that has on models of disease, first at the cell level. And I'm very excited about this. I think increasingly this will be the way in which new drugs are made throughout medicine.

27:24Is this sort of the molecular equivalent of, I've got a mannequin, I give it to you as a dressmaker, and you stitch me a dress to fit that mannequin. We're giving these systems a molecular mannequin, the receptor or whatever the structure is, and saying, that's the target, make me something that fits that. It's a bit like that. Another analogy is a lock. You give me a lock and ask me to make a key. you actually ask me to make a variety of keys and then I give them back to you and you see which ones fit the best. How on earth does the program do that and why has that been such a tough nut to crack?

28:02Because it sounds very simple. Here's the structure I want you to target, go for it and does it. How's it doing it? Well, the answer has changed over the last 10 or 15 years. Originally, you know, when we first started doing this 20 years ago, we actually tried to construct, using physical principles, a molecule which would serve as a key for the lock. But more recently, really in the last seven years, we've switched over to deep learning methods, as I have mentioned. And there it's the way that the problem is solved is by training on very large data sets like in other AI problems. And the data sets that we're training on are many, many, many examples of keys fitting into locks.

28:50And the program learns the properties a key should have. And so then, after it's been trained sufficiently, you can give it a new receptor and it will generate a biomolecule that will bind to it. But given it's relying on what we already know, how can it know what to do with what we don't know? It's because what we don't know always has some aspects of what we do know in it. It's the same reason that you can ask a language model like ChatGPT or Claude a question that it hasn't seen before, and it can give you a very plausible answer. It's because the patterns that it's learned on the stuff that we already know pertain to most unanswered questions.

29:39Of course, that's not true of everything. and in fact the hardest receptors now to target are the ones which are most different from the ones that we have lots of examples for already. Given that these things do hallucinate from time to time is this one example then where actually the making stuff up might in fact turn out to be a blessing not a curse? Well one of the things that's very different here is that the computer calculation where you're designing the new biomolecule is just the first step. Then in the second step, you go to the laboratory and you experimentally test whether your new biomolecule actually binds the receptor.

30:21So if it were a hallucination that was just completely off base and it didn't bind the receptor at all and it didn't do anything to modulate the disease, you would basically discard it. However, you know, if it was a completely new solution that actually worked, and I think that's what you were getting at, then that could be very interesting because it might be a solution that a traditional drug developer would not have thought of, but the generative AI came up with. Can we do this with any aspect of chemistry? Or are we firmly rooted in the realm of proteins exclusively here? These basically three-dimensional balls of atoms and molecules?

31:03Or can you extend chemical space and just basically do it for anything? It is easiest to do it with proteins for several reasons. First of all, we have a very large number of examples, a very large database of proteins binding to other proteins and receptors. The second reason is that it's very easy to make proteins in the lab. And the way that's done is proteins are encoded in genes by way of the genetic code. So once we've designed a new protein, we can very easily make a synthetic gene that encodes it. And so in a matter of a day or two from when we've done the computer calculation, we can make the new molecule, the new key, and test whether it works.

31:48So in the case of doing this with more general types of molecules, the principles are really very much the same. And we and others are doing this. It's just a bit harder because there are fewer examples and there's a much wider range of possibilities once you get away from proteins, which is both a blessing but also a curse because you don't have as much data. And then second, it's harder to make them in the laboratory because for any arbitrary molecule collection of atoms, you can't simply make a synthetic gene that encodes it. You have to actually put those atoms together using chemical synthesis.