Innovation in Medicinal Chemistry: Unlocking Unsolved Diseases

Recent innovation in medicinal chemistry has opened up avenues to discover medicines for diseases that we could not have imagined even just a few years ago. At Boehringer Ingelheim, we like to think of our medicinal chemists as molecular locksmiths, searching for ‘locks’ on the surface of proteins implicated in disease and designing drugs or molecular ‘keys’ to fit into these ‘locks’. The lock and key analogy, conceived by the German chemist Emil Fischer over 120 years ago, is still relevant but today’s diseases require medicinal chemists to discover much smarter keys - keys that do more than simply turning off a disease-causing protein. For example, today’s oncology drug molecules need to be highly selective and powerful at the same time while modern CNS drugs need to gently modulate brain activity.

Taking structure-based drug design to the next level

The strength of a drug’s action depends largely on how well it fits in the lock or binding site of the disease-causing protein. To design a perfectly fitting key we need to see, at the atomic level, what happens inside the lock once the key binds. That’s the principle behind structure-based drug design. Seeing, atom for atom, exactly how a potential drug candidate binds, which atoms fit perfectly and which ones do not, enables us to discover the precision quality molecules that are demanded of today’s most important drug targets.

The rigorous application of structure-based drug design has enabled us to make important progress in discovering potential drugs for KRAS, one of the so-called undruggable targets in cancer called ‘cancer’s big four’. The challenge with KRAS is that its two ‘locks’ are very shallow and polar and, as such, are extremely difficult to find keys which fit tightly enough to be a potential drug. This is particularly frustrating because KRAS mutations occur in one in seven of all human cancers, making it the most frequently mutated oncogene which has remained undrugged since its discovery in 1982. Normally difficult projects such as KRAS are only able to obtain protein crystal structures for around 1 in 100 compounds. To finally make progress against KRAS, we took structure-based drug design to a new level and obtained three-dimensional “lock and key” structures for every single KRAS ligand that we synthesized in the laboratory on the way to a drug candidate.¹ This led to a molecule called BI-2852, a KRAS inhibitor that binds with nanomolar affinity to the so-called switch I/II lock on KRAS and triggers antiproliferative effects in KRAS mutant cells.² 

Precision pictures for precision drug design

Because of the transformational power of structure-based drug design we have become expert photographers in obtaining three-dimensional “lock and key” photographs. Unfortunately, obtaining such 3D pictures is not as simple as pressing the button on a smart phone.

There are three main approaches that can be used to “take” pictures for a broad range of proteins and protein assemblies.

1. Nuclear Magnetic Resonance (NMR) spectroscopy is mainly used for very flexible proteins or protein domains in solution with a lower size limit
2. X-ray crystallography uses near-to-perfect crystals of the disease-causing protein in a more static fashion by using X-ray diffraction mainly at very low temperature (100 K)
3. Cryo-electron microscopy (cryo-EM) is a more recent approach which utilizes powerful microscopes also at very low temperature to determine 3D structures of proteins and large macromolecular protein assemblies in solution

To obtain X-ray crystal structures you first need to obtain perfectly pure protein with the help of bacteria, yeast, insect or mammalian cells. Then using robotics, a cold room and patience we test thousands of conditions until we can grow an almost perfect crystal of the protein. Finally, it is bombarded with a beam of
X-rays which locate the position of electrons. Computer algorithms convert these huge data sets into atom positions that lead to the final three-dimensional picture.

X-ray crystallography works well for small or more rigid disease-causing proteins but it struggles with large flexible proteins and complexes containing multiple proteins. This is where the new kid on the block, cryo-electron microscopy excels. Cryo-EM takes 2D projections of the protein with an electron microscope that are reconstructed to a 3D model afterwards. No crystals have to be grown but the samples need to be meticulously prepared and then kept in a thin layer of non-crystalline ice cooled down to very low temperatures (100 K) with liquid nitrogen, or close to absolute zero, using liquid helium.

The destruction approach to drug design

A new class of drug molecules has been recently discovered that promises to significantly expand what is druggable. In contrast to classical drugs which turn off disease-causing proteins, this new class completely degrades proteins. These ‘destructors’ are PROTACS (Proteolysis Targeting Chimeras) that ‘hijack’ the cell’s natural disposal system to shred disease-causing proteins. Hijacking naturally occurring cellular systems to treat disease is a new and exciting way to design drug molecules and is being extended to concepts beyond degradation. To do this a “double key” is needed, one key to bind to the disease-causing protein and a second key to bind to the cellular system you want to hijack. In the case of PROTACs, proteins called ubiquitin ligases - enzymes that tag proteins for destruction by the cell’s proteasome - are hijacked.

We used PROTACs to target SMARCA2, a protein that plays an important role in lung cancer. It is part of a large, almost indestructible complex, called the BAF complex, which has two known “locks”. With PROTACs it doesn’t matter which “lock” you choose, so we chose the easier lock located on the so-called bromodomain which led to the potent PROTAC compound ACBI1. ACBI1 is able to degrade SMARCA2 out of the BAF complex, akin to removing an individual brick out of wall, and effectively kills lung cancer cells in vitro.³

Accelerating drug design with Artificial Intelligence

The key fitting perfectly into its lock is just one of many criteria a drug molecule must fulfil. There are more than fifty parameters to be considered to successfully discover a new drug. This, combined with the enormous size of chemical space (~10⁶⁰ molecules), requires an iterative endeavor which leverages the joint brain power of scientists working together as a team. At Boehringer Ingelheim we are convinced that recent developments in the fields of machine-learning and Artificial Intelligence (AI) can massively accelerate the exploration of chemical space and drug discovery as a whole.

As a successful drug discovery organization with decades of experience, we have more than 300 million data points at our disposal. For every drug discovery project hundreds of data points are measured every week. To master the challenges of modern drug discovery, it is imperative to augment our creativity and scientific expertise with intelligent algorithms that constantly analyse data at large scale, draw conclusions and derive scientific hypotheses.

In the past, medicinal chemists designed, synthesized and tested a few thousand molecules, largely sequentially, to fully assess the potential of each molecule. Digital tools and AI have the potential to transform this process. Using AI we generate millions of molecules virtually and predict their molecular properties using computer algorithms to choose the best molecules to synthesize in the laboratory. It’s like using a navigation system for chemical space to rapidly weigh up different options on which route to take to get to the desired destination in the shortest time.

Our brains also function according to the lock and key principle. Naturally occurring molecules called neurotransmitters, are the keys which fit into locks on neurons to transmit signals. In contrast to oncology where we need powerful molecules that selectively target cancer cells and destroy them, for diseases affecting the central nervous system such as schizophrenia we need more subtle “keys” that are able to selectively modulate brain function. One such class of drugs are negative allosteric modulators (NAM). NAMs are a second key which bind to a different lock at the same time as the naturally occurring neurotransmitter and dampen its effect. Efficient NAMs are extremely difficult to design but with the support of AI we were able to design a NAM for the GABA α5 receptor much faster and in only a few iterations.⁴ GABA α5 is a key protein involved in a wide range of essential processes in the brain including schizophrenia. Our AI algorithms helped us to filter 200,000 virtual compounds down to 200 compounds that most closely matched the required criteria. The accuracy was so good and the predictability so high that we were able to go straight to advanced stages of profiling to identify two drug candidates. With the support of AI this took less than four months, less than half the time of using traditional methods.

Leveraging technologies to innovate for patients

Pushing back the boundaries with new technologies in medicinal chemistry is enabling us to find molecules for targets that previously seemed undruggable with traditional approaches. And this is just the start. Approaches such as structure-based drug design, PROTACS and AI are enabling us to effectively explore the infinite possibilities of chemical space: Locating the right place to start and then guiding medicinal chemists on which direction to take to discover innovative drugs that improve patients’ lives.
 

References

1. Kessler, D., Bergner, A., Böttcher, J., Fischer, G., Döbel, S., Hinkel, M., Müllauer, B., Weiss-Puxbaum, A. and McConnell, D.B., 2020. Drugging all RAS isoforms with one pocket. Future Medicinal Chemistry, (0).
2. Kessler, D., Gmachl, M., Mantoulidis, A. et al. Drugging an undruggable pocket on KRAS. PNAS 2019; 116: 15823-15829
3. Farnaby, A., Koegl, M., Roy M.J. et al. BAF complex vulnerabilities in cancer demonstrated via structure-based PROTAC design. Nat Chem Biol 2019; 15: 672-680
4. Hucke, O., Bieler, M., Larsen, J., Dyhring, T., Jacobsen, T., Nielsen, K., Schauerte, H., Cui, Y., Peters, S., Heine, N., Eickmeier, C., Arban, R. and Montel, F.,2019, August. Leveraging machine learning and the Free-Wilson approach in lead optimization: Efficient discovery of a new chemical class modulating the GABAA alpha 5 receptor. In Abstracts of Papers of the American Chemical Society (Vol. 258). 1155.