The Pharma Lab Show Crystal Structure Prediction and Drug Development

"A real revolutionary change will happen as we apply Crystal Structure Prediction to the lead optimization phase of pharmaceutical discovery and to the materials discovery in general." - Chandler Greenwell

Crystal Structure Prediction (CSP) has been a vital part of the mission of XtalPi, an artificial intelligence and quantum physics based pharmaceutical technology company aiming to improve the speed, scale, novelty, and success rate of drug discovery and development.

Dr. Chandler Greenwell, an applications scientist at XtaPi, shares how CSP is revolutionizing the field.

Simon: Welcome to The Pharma Lab Show; the PODCAST that explores the partnership between analytical and the Drug Development Life Cycle. I am Simon Bates, the host of today's show, and our guest is an application scientist from XtalPi, Chandler Greenwell. He is here to talk about crystal structure prediction and polymorphism. Welcome, Chandler.  

Chandler: Hey, Simon. It's a pleasure to be here and I'm really looking forward to our discussion.  

Simon: Chandler, could you please introduce yourself to the listeners and tell us a little bit about XtalPi?  

Chandler: Yeah, sure.  So, my name is Chandler Greenwell.  And I am the solid-state application scientist at XtalPi. XtalPi is an artificial intelligence and quantum physics based pharmaceutical technology company with the mission to revolutionized drug discovery and development by improving the speed, scale novelty and the success rate. And crystal structure prediction is a vital part of that mission.  

Simon: So, if my understanding is correct, crystal structure prediction relates to polymorphism. Can you discuss polymorphism in pharmaceutics and how it relates to crystal structure prediction?  

Chandler: Yeah, I'm happy to do so. So, let me first define both concepts for the listeners. In the context of solid-state chemistry and material science, polymorphism refers to a molecule that has more than one crystal form. So, polymorphs often have different physical and chemical properties like solubility, which is one reason why pharmaceutical companies are interested in polymorphs. Crystal Structure Prediction, or CSP, is a field of computational chemistry and material science that seeks to predict all the stable crystal structures corresponding to an organic molecule. A CSP calculation starts with the 2D diagram of the compound and then, by combining the optimization algorithms, quantum mechanical calculations, AI and molecular dynamic simulations, you can predict all the possible 3D packing arrangements, rank the lattice energy of those structures and you can calculate the free energy of the most plausible crystal structures as a function of temperature. So, it's a lot of work, but the end result is a theoretical landscape of all the possible polymorphs and how close together they are in terms of their thermodynamic stability.  

Simon: Yeah, that sounds really interesting. Before we get into the details, could you please let the audience know how you first became interested in crystal structures?  

Chandler: I think, like a lot of chemists, my interest in molecules and crystal structures probably started from Legos. But my first exposure to polymorphs happened before I did my PhD, when I worked as a quality control chemist in a stability lab at a pharmaceutical company in Salt Lake City. So, one of the quality tests I performed involved viewing a generic version of a rotigotine patches through a microscope. So, the test was to visually check for the appearance of dendritic crystals on the surface of the patch. And I didn't know it at the time, but those dendritic crystals were associated with a less soluble polymorph of rotigotine. So, the patch formulation is used to stabilize an amorphous form of rotigotine. The presence of those dendritic crystals would indicate that form II of rotigotine, which is the thermodynamically stable polymorph, had crystallized on the surface of the patch. So, the crystalline form would no longer be biologically available for the patients. So, it's crucial when you're performing solid formulations to ensure that a pharmaceutical product is stable over a wide range of conditions. And oftentimes the thermodynamically stable polymorph is the safest bet. In terms of, when I really started digging into crystal structure prediction and polymorphism as concepts; that's when I was in graduate school, where I was working on new quantum mechanical bottles, and polymorphs presented a great test system for the accuracy of the methods, because polymorphs typically differ in energy by less than two kilo joules per mole.  And long story short, the methods worked pretty well. And I've been studying CSP and polymorphism ever since.  

Simon: Two kilo joules per mole sounds like a rudely small amount of energy. What does this effect really mean in terms of percentage energy change for a typical structure?  

Chandler: Yeah, so, I'm not sure about percentage energy changes; but to give you some context, a typical O-H-O hydrogen bond contains about twenty-one kilo joules per mole of energy. And polymorphs typically differ by a tenth of that amount. So, we're talking about a very delicate competition between inter- and intra-molecular interactions.  

Simon: Well, so, a tenth of the energy of a typical hydrogen bond I guess under ambient conditions; that's really, small energy difference between these polymorphs. You know 

Chandler: Yeah, I mean that's the key challenge for the quantum mechanical methods. They have to be highly accurate in order to discern those energy differences.  

Simon: Yep, I consider absolutely.  So, from what you're saying, Chandler, Crystal Structure Prediction seems like it could be a useful technique for the characterization of pharmaceutical actives.  

Chandler: Yeah. You're absolutely right. So, at XtalPi, I get to work directly with solid-state chemists, solid formulation experts and seeing professionals from different pharmaceutical and agrichemical companies. And they're deeply engaged in not only characterizing the solid-state landscape for a given compound, but also in de-risking selection of a stable solid form for the manufacturing process. So, it's a huge responsibility because there's enormous time pressure to make a decision. But any mistakes can result in product recalls and expensive reformulation efforts. So, our clients don't want any unexpected surprises and CSP has become a valuable tool for providing insight. So, we can perform the CSP calculations while they're performing their characterization experiments and setting the solid form landscape. It’s this beautiful synergy between experiment and computation.  

Simon: You mentioned insights. Could you go through these specific insights that CSP gives for de-risking the polymorph landscape?  

Chandler: Yeah.  So, our typical client will explore different solid forms in order to optimize the material properties. So, these can include the free base forms, the hydrate forms, salt forms and co-crystals. They'll also thoroughly characterize the structure and properties of these forms. So, they have to determine which forms that can converge, which forms have good properties like solubility, which forms readily form solvates, and which forms are polymorphic. So, CSP provides valuable insights in several ways. We can perform CSP on the free base form, the salt forms, hydrate forms, co-crystals or even a solvate; and the energy landscape and the free energy calculations for a given solid form reveals the thermodynamic relationships between the polymorphs. So, sometimes the energy landscape makes it clear that the experimental structure is in fact the most stable structure in the CSP landscape. However, other times the energy landscape can indicate that there are many possible low energy structures. In this case, a client may want to expend additional effort during solid form screening to search for additional polymorphs and rule out undesirable form transformations.  

Simon: So, we talked about, well you talked about de-risking. What additional extended information can we extract from the CSP?  

Chandler: Yeah, so, in addition to de-risking solid form landscape, the predicted structures can serve as the basis for further modeling. For example, CSP structures can provide insight into disorder, desolvation, dehydration, atropisomerism, zwitterion formation, ISO energetic forms and many other structural phenomena that solid state chemists encounter in their day-to-day workflow.  

Simon: So, disorder. That sounds like an interesting problem, as I guess disorder can raise the polymorph energy, which would directly impact the de-risking landscape. How do you handle the presence of disorder in the simulated structures?  

Chandler:  Yeah, so that's a great question. So, in terms of disorder, there is static disorder and there's also a dynamical disorder in different structures. One thing, so, let's take for example, like a functional group that can differ in a disordered structure by a 180-degree rotation. oftentimes we'll find that both orientations of the molecule will be predicted in crystal structures that are landscape. So, if we say there's like a major component of the disorder and a minor component, oftentimes we’ll have a structure that corresponds to those major and minor components. So, we can compare the energy difference between those two components. But we can also take that molecule and we can take super cells of it and calculate the energy. And that lets you see how the lattice energy, or the relative lattice energy of the crystals changes as you incorporate more or less conformations of the major or minor component.  

Simon: That's really interesting. You know, another challenge we routinely run into in The Pharma space is that during solid form screens, polymorphs screens, we observed data corresponding to completely unknown crystal forms. How can CSP help us in those situations?  

Chandler:  Yeah, so, as you know, it's not always easy to solve a crystal structure; But CSP can help because it provides many plausible crystal structures; including the lattice parameters, their z-prime, and also the conformation of the molecules in the unit cell. So, if you have experimental PXRD patterns, you can compare those directly to the computed PXRD patterns and that can help you determine likely crystal structures for a hard to resolve form.  

Simon: You know, I must admit when I first came across crystal structure prediction, I was really blown away by the application possibilities in solid form screening and solid forms selection. And almost everyone I know from The Pharma world that’s looked into crystal structure prediction has had the same reaction. It really is a no brainer that we go down this pathway.  

Chandler:  Yes, I mean I think you're absolutely right about that. I mean I'm biased, but crystal structure prediction is a truly amazing application of computational chemistry. See, exhaustive solid form screening requires hundreds of different experiments to explore the solid form landscape, screening for polymorphs of each new solid form and thoroughly characterize each new form. But the hardest part is that there's no rule of thumb for when you've done enough experimental screening to constitute an exhaustive search. So, in that respect, each form for which you calculate a CSP energy landscape is kind of like a map. So, you still have to do experiments to reveal which solid forms can be obtained, but the CSP energy landscape gives you an idea of where to look and what you can expect.  

Simon: So, looking more to the future, how do you see crystal structure prediction developing and where might this enhanced performance benefit the development of solid forms? Are we looking at a revolutionary or evolutionary change? 

Chandler:  Yeah, I would say that CSP has developed incrementally as a field for the last several decades. So, it's taken a lot of contributions from many scientists across the globe to bring us to where we are in our understanding of crystal structures and polymorphism in general. So, QM methods had to become more accurate, computing power to increase and searching algorithms had to become sophisticated enough to handle large degrees of freedom. Within the last few years, I think we've really reached a point where CSP can be routinely applied to solid formulations and product development. So, at XtalPi, we're focused on decreasing the overall computational effort involved in CSP, without sacrificing the accuracy of predictions. Lower computational cost will lower the barrier for researchers or companies to engage with the CSP vendor to study the energy landscape of their solid forms. So, I think a real revolutionary change will happen as we apply CSP to the lead optimization phase of pharmaceutical discovery and to the materials discovery in general. So, CSP at the lead optimization stage will allow drug discovery scientists to factor in solid state material properties as they select lead compounds for further development and testing. For example, CSP can service the basis for accurate solubility calculations. From the materials discovery standpoint, we have a number of ongoing projects where CSP is used for co-crystal screening. So, CSP is used to predict theoretical low energy co-crystal structures of a possible API and coformer repairs. And stay tuned on that, because some of the predicted low energy co-crystals have later been obtained during experimental screening. So, this same logic could easily be applied to solve forms and hydrates. And I personally think that the ultimate goal would be to predict a CSP energy landscape for every small molecule development. That would be a real revolution in our understanding of the solid form and polymorphism in general.  

Simon: Yep. Absolutely. So how do you see these changes impacting bench campus involved in the hands-on solid form generation?  

Chandler:  Yeah, Oh, okay. So, another potential revolutionary breakthrough would be the ability to suggest how to crystallize predicted structures that you get from a CSP calculation. So, there are two challenges here: There's the overprediction problem, and then you need to simulate nucleation and growth. So, in terms of the overprediction problem, there's always more theoretical polymorphs than are ever obtained from experiment. There's this molecule ROY, which has twelve well characterized polymorphs and is by far the most polymorphic molecular organic molecule to date. But if you perform CSP on ROY, you're going to get hundreds of potential polymorphs in your landscape. So, at XtalPi we've developed some methods that help us reduce the number of structures in the CSP landscape. So that's the first step. The next thing you need to do is to determine the likelihood of crystallization for the remaining low energy structures. And we have a few open R&D projects that we hope will provide our clients with valuable insight into how to obtain those predicted polymorphs.  

Simon: That's great. So, Chandler, for the listeners who want to dig a little deeper into CSP, where would you recommend that they focus their efforts?  

Chandler:  Men, I'd so, I think the best part of being a CSP researcher, is that you have a great community that's so bright and so enthusiastic about the technology. And there really are so many must read papers, and there's way too any that I could name here. But I'll mention just a couple that I think can serve as a starting off point for anyone who's interested. So, the first is a chemical society reviews article called facts and fictions about polymorphism. Another great paper is called why don't we find more polymorphs? And that one deals with the triumphs and the challenges of crystal structure prediction. And if you want to catch up to the state of the art in the field, you should read the report on the six blind test of organic crystal structure prediction methods. 

Simon: Has XtalPi actually published any papers in this area.  

Chandler:  Yeah, so, XtalPi is very actively engaged in ongoing research projects related to crystal structure prediction. And I'm happy to talk about a couple of those papers. So, the first one I should mention is called Harnessing Claud Architecture for crystal structure prediction calculations. This one details XtalPi’s protocol, a CSP protocol, which can be paralyzed on up to a million cores. Another one of our papers is called prediction of relative free energies of drug polymorphs above zero Kelvin. This one addresses the difficult task of calculating finite temperature effects. So, when you perform CSP, the quantum mechanical calculations that give you the lattice energy don't include the effect of temperature. So, if you want to compare the relative stabilities at higher temperatures, you need to calculate the free energy as a function of temperature. So, to do that we use molecular dynamic stimulations; and this allows us to routinely and accurately calculate the free energy as a function of temperature. So, finally, I think I should mention three XtalPi papers that demonstrate how CSP can be used for material discovery or material optimization. So, the first entitled selecting a stable solid form of remdesivir using microcrystal electron diffraction and crystal structure prediction, demonstrates the power of synchronizing your experiments with CSP calculations. So, in thirty-three days we were able to solve crystal structures of remdesivir forms IV and form II directly from the powder. So, there was no effort invested to obtain single crystals. Also, simultaneously we are able to predict both forms as the first and second most stable polymorphs in the energy landscape respectively. One thing you should know, though, is that form II of remdesivir is the more thermodynamically stable form at room temperature. So, when we computed the free energy, we actually saw that form II became the more stable polymorph at room temperature. So that result was an excellent agreement with the experimental data. And I think the final two papers I want to mention both showcase a less computationally intensive CSP protocol that allows us to consider more compounds. So, the first is called virtual coformer screening by crystal structure predictions: the crucial role of crystallinity in pharmaceutical Cocrystalization. So, this one showcases a CSP based co-crystal screening method and the high accuracy predictions from this method can be used to guide co-crystal screening experiments. The second, called guiding lead optimization for solubility improvement with physics-based modeling, explores the use of CSP to generate low energy crystal structures for compounds in two different lead series. So, by calculating the enthalpy of fusion with CSP and the Gibbs free energy of hydration with COSMOtherm, we provide physics-based insight into how functional group changes affects solubility of a lead series.  

Simon: So, thank you, Chandler for sharing your opinions and experience with crystal structure predictions with the listeners today. It was a great pleasure speaking with you.  

Chandler:  Likewise, Simon, thank you very much for having me on the show and I really appreciate you giving me this platform to talk about CSP.  

Simon: As a reminder, Chandlers contact information and LinkedIn profile can be found in the show notes for this episode, along with links to the papers that Chandler mentioned; and I'd like to thank you all for listening and encourage you to catch other episodes of The Pharma Show. The podcast explores the partnership between analytical and the Drug Developmental Life Cycle. Until the next time, keep well.  

Pharmaceuticals have the power to change the world for the better, but before they can ever do that, they need to be proven safe and trustworthy. Here at Rigaku, we strive to make this a reality as the leading Global Scientific Analytical Instrumentation Company, specializing in X-Ray and thermal analysis and Raman spectroscopy. To learn more about how we do that, visit us at Pharma.rigaku.com. You've been listening to The Pharma Lab Show. To ensure you never miss an episode, subscribe to the show in your favorite podcast player. Thanks for listening. Until next time. 

• Facts and fictions about polymorphism
• Why don’t we find more polymorphs?
• Report on the sixth blind test of organic Crystal Structure Prediction methods
• Harnessing Cloud Architecture for Crystal Structure Prediction Calculations
• Prediction of the Relative Free Energies of Drug Polymorphs above Zero Kelvin
• Selecting a stable solid form of remdesivir using microcrystal electron diffraction and Crystal Structure Prediction
• Virtual Coformer Screening by Crystal Structure Predictions: Crucial Role of Crystallinity in Pharmaceutical Cocrystallization
• Guiding Lead Optimization for Solubility Improvement with Physics-Based Modeling

For more insights into the pharmaceutical industry, subscribe to the show on Apple Podcasts, Spotify, or wherever podcasts are found.
Check out the Rigaku Pharmaceutical Technologies Showcase for more information on Rigaku's commitment to the pharmaceutical industry

Listening on a desktop & can’t see the links? Just search for The Pharma Lab Show in your favorite podcast player.