What are the challenges for medicinal chemists when working with 3D structures? - Interview
In 2011, two scientists from the University of Louisiana speculated about the future role of medicinal chemists in drug design and discovery. In the publication "Medicinal Chemistry for 2020", they anticipated that an increase in structural basis for the design of new drug molecules would affect the medicinal chemists’ role.
In recent years, we truly see more and more medicinal chemists improving their education to understand key areas of biochemistry, structural biology, cheminformatics, etc. It became a highly interdisciplinary and engaging role, especially due to the ambitious development of 3D structural analysis. (Source: What is medicinal Chemistry? – Demystifying a rapidly evolving discipline!)
Considering that, we asked ourselves: What are the current challenges for Med Chems using 3D information?
We invited medicinal chemistry professional Dr. George Sheppard, Senior Principal Research Scientist at AbbVie, for an online interview to give us his perspective. In an informal setting, we discussed the changing attitude of medicinal chemists towards 3D structural information over time and challenges for its usage.
In this article, you will find George's personal experience working with protein structures, his opinion on current advantages and limitations for Med Chems, examples of successful structure-based project outcomes, and advice for enhanced usage of structural data in the future.
Could you introduce yourself, your training/background?
Hi, my name is George Sheppard. I hold a Ph.D. in Organic Chemistry and have had exposure to both methods development and target-directed synthesis during my graduate and post-doctoral studies. Thus, when I joined AbbVie (at the time, Abbott Laboratories) 30 years ago, I had a lot of expertise in how to make molecules, but no training on what molecules to make. That was typical of most new hires to pharma back then and is probably still typically true today.
What is your current position and role in the team?
I am currently a Senior Principal Research Scientist in the Oncology Discovery group at AbbVie. In addition to working as a Principal Investigator on specific projects, I also have a non-management leadership role that we refer to internally as a “Design Governor.” It is a bit of a misleading term, as I don’t really “govern” anything. What it means is that I advocate for best practices in compound design and support my colleagues in Oncology in learning about and using software tools for compound design and chemoinformatics.
What is your personal experience of working with 3D information?
Prior to working in pharma, I was already used to thinking about the conformations of small molecules and how they interacted with each other. In some ways, it wasn’t a huge leap to go from thinking about substrate or catalyst interactions to small molecule interactions with a protein partner, but there were a lot of new things to learn. Over the years I have had to learn how to use various pieces of software to make use of 3D information on proteins of interest. At first, it was just the visualization of experimental structures and models developed by others to generate and refine ideas, then eventually I started to learn how to dock new ideas and so on.
When was the first time you used 3D molecular structures?
Early in my career, I was fortunate enough to get a chance to be part of developing one of the pioneering fragment-based drug discovery approaches, a technique we referred to as “SAR by NMR.” (Literature reference: “Hajduk et al., Discovery of Potent Nonpeptide Inhibitors of Stromelysin Using SAR by NMR” J. Am. Chem. Soc.1997, 119, 5818-5827.). By studying how fragments bound to protein targets using NMR techniques, one can find fragments that bind to proximal sites on your target of interest. If you can link the fragments together properly, you get much more potent binders. I was one of the chemists who got to do the linking part. Prior to that, I had worked on a GPCR project where we didn’t have any structural data, just a 3D pharmacophore hypothesis from QSAR studies. Having NMR structures, then later X-ray co-crystal structures of how the lead compounds bound to their target really allowed me to up my game in terms of designing better molecules.
“Understanding the details of how your compound interacts with the target (or off-target) lets you focus your thinking as you design new and better compounds.”
In your opinion, what is the biggest advantage of using 3D structures?
Every project is different, but I think a common theme would be that understanding the details of how your compound interacts with the target (or off-target) lets you focus your thinking as you design new and better compounds. Drug Discovery involves simultaneously solving a lot of different problems. Once you know what the important features of the protein-ligand complex are, you can make more informed choices about what changes you can make as you try to get all these things right. Where can I put in a solubilizing group? What features do I need to keep if I try a scaffold hop? A detailed 3D understanding of how your compounds interact with a protein really helps you to answer those sorts of questions much more efficiently than just trying known isosteric replacements out of the literature or making things just because the chemistry looks easy.
What is the current usage in your team and in AbbVie in general?
AbbVie has invested heavily in supporting structure-based design, and it is great to have colleagues who are experts in all the disciplines that go into enabling that. There are some projects that are not structure enabled, particularly in the exploratory stage, but the aspiration is there to get our structural biology function engaged with as many projects as possible.
How is it accepted in your team? Who is using it the most?
We have had many success stories, so I really don’t think there are any chemists who wouldn’t prefer to have access to 3D information on their target. The degree to which people interact with 3D data when we have it varies. As you probably know, there is a model in the industry where chemists are specialists in compound synthesis or compound design. AbbVie is pretty pragmatic in our approach. We certainly have people who specialize in CADD or synthesis, but there is always a dialogue with input from everyone on a team. In the Oncology Discovery group, we encourage all the chemists to engage with all three phases of the design, make, learn cycle to get as many minds working on solving the problems as possible.
How are people trained to understand 3D data? How was the learning experience?
I think one of the strongest arguments for a dedicated designer role is that there are so many tools and techniques out there. It is hard for an occasional user to get the most out of sophisticated software and have the level of understanding needed to pick the right approach and understand the choices they are making. A big part of my Design Governor role is to help colleagues learn what tools we have available and do some basic training so they can start exploring them. Demonstrations and tutorials are a good start, but ultimately most people learn best when they dig in on a question that interests them.
What are the current main challenges regarding 3D data (scientifically and technically)?
There have been major advancements in our ability to collect 3D data sets, and as newer techniques like cryo-EM become more widely available the amount of information is going to increase even more. One of the challenges we face is just dealing with all that data! We have the tools for looking at how one ligand interacts with one protein, but it isn’t straightforward to pull together the data to look at how similar molecules interact with related proteins in a comprehensive way. We should be able to extract a lot of valuable information if we can get better at that sort of analysis. Another challenge that I think about a lot is how to account for protein dynamics. Protein crystallography gives you a static picture of how the (often truncated) protein fits in a crystal, and an NMR structure can give you an average of solution conformations (again, often of a truncated protein). Predicting how a protein in a physiological environment might adjust to a new ligand is a challenge. I have worked on projects where protein movement exposed a cryptic binding site that is not evident without the ligand present. There are computational approaches to model protein dynamics and dynamic binding events, but I think we are just scratching the surface. As computational resources grow and become cheaper, we will probably get better at accounting for dynamics in designing compounds.
Are there any strategies you implemented in overcoming these challenges?
The idea of leveraging differences in binding modes between targets and anti-targets seems like it should be straightforward to do but isn’t always easy to implement. An example from my own experience involved getting an NMR structure of a lead compound bound to Human Serum Albumin to inform our efforts to elevate free drug levels (Sheppard et al., “Discovery and Optimization of Anthranilic Acid Sulfonamides as Inhibitors of Methionine Aminopeptidase-2: A Structural Basis for reduction of Albumin Binding” J. Med. Chem.2006, 49, 3832-3849.) What I’d like to see is better tools for assembling and working with sets of data on related ligands and related proteins. Getting actual 3D structure data on multiple targets takes a lot of resources and often won’t be a realistic request. Finding better ways to draw on existing data that is already out there on ligands and or proteins related to the ones you work on to get insights would be faster.
“The insights from structures have been so helpful on numerous projects it is hard to believe the problems would have been solved as quickly without them.”
What is the impact of 3D information on your scientific project outcomes?
I firmly believe that access to detailed information about protein-ligand interactions facilitates Drug Discovery. An example we published recently was the discovery of our BD2-selective BET bromodomain inhibitor ABBV-744. A pair of protein co-crystal structures of a weakly selective compound found by serendipity with both the desired and undesired BET bromodomains gave the team an understanding of the basis for selectivity and guided us in refining that insight into a highly selective molecule. The understanding of the key differentiating interactions also helped us optimize the resulting selective lead into a compound with the attributes needed to advance to the clinic. (If you are curious, the literature references are Faivre, et al., “Selective Inhibition of the BD2 Bromodomain of BET proteins in prostate cancer.” Nature2020, 578, 306-310, and Sheppard et al., Discovery of N‑Ethyl-4-[2-(4-fluoro-2,6-dimethyl-phenoxy)-5-(1-hydroxy-1-methyl-ethyl)phenyl]-6-methyl-7-oxo‑1H‑pyrrolo[2,3‑c]pyridine-2-carboxamide (ABBV-744), a BET Bromodomain Inhibitor with Selectivity for the Second Bromodomain” J. Med. Chem., 2020, 63, 5585-5623.). It is hard to “prove” the co-crystal structures contributed in a rigorous way since you can’t have the same team solve the same problem with and without access to the information, but the insights from structures have been so helpful on numerous projects it is hard to believe the problems would have been solved as quickly without them.
Do you think 3D information is used enough in pharma industry at present? Should it be used more?
I think that Structure-Based Design is well accepted in the industry, but it is unrealistic to expect anyone's approach to be applicable to every project. The technology and infrastructure for determining protein structures is growing rapidly. Likewise, computational resources for modeling, studying protein dynamics, conducting virtual screening, etc. continue to grow. I think the paradigm is going to continue to be widely used and should have an even greater impact as the capabilities increase.
How do you expect this will change in the future? What would you be excited to see?
As I mentioned earlier, I hope we will get better at dealing with protein dynamics in thinking about binding events. I suspect that this could also help us do a better job at identifying possible off-targets and finding ways to avoid them. That is an area that could add a lot of value.
What could be done to increase awareness of MedChems regarding 3D data?
In my opinion, there are a couple of keys to getting more Medicinal Chemists to actively engage with 3D data. The first is to get the success stories out there in journal articles and presentations at conferences. Hopefully, more people will say to themselves, I wonder if we can apply that approach that was tried on Project X to the problem I’m facing. The second is to lower the barrier for individual contributors on the team to start engaging with the data. The learning curve for some of the software can be intimidating. Companies need to be willing to include initial and follow-up user training and support to get the most out of their investment in software.
Wrap Up
As commonly observed, accessing, and using 3D information was and still is challenging for medicinal chemists. However, Dr. Sheppard’s journey is a good example of a steady education and paradigm shift happening within the Med Chem community and proof that such an evolution is possible and beneficial for any drug-discovery project.
The persisting challenges, such as the overwhelming amount of data, considering protein dynamics, or the long learning curve, are areas with room for improvement. However, given the development of the field and its current state, the community can be confident in tackling these challenges during the next years.
