Drug Discovery Lecture Notes PDF

Topic 12: Structural Biology in Drug Discovery  What is a drug? A drug is a chemical entity that causes a desired therapeutic effect This can be a small molecule “chemical” of natural (e.g. penicillin, tamoxifen) synthetic (viagra, lipitor), or semisynthetic (aspirin,...

Topic 12: Structural Biology in Drug Discovery  What is a drug? A drug is a chemical entity that causes a desired therapeutic effect This can be a small molecule “chemical” of natural (e.g. penicillin, tamoxifen) synthetic (viagra, lipitor), or semisynthetic (aspirin, amoxicillin) origin Alternatively, proteins of natural (e.g. from cadavers) or recombinant origin can be used therapeutically - e.g. insulin, erythropoietin, human growth hormone, various antibody therapies  How do drugs work Most drugs work by specifically binding and affecting a specific biological target Most targets are protein molecules, though other molecules (e.g. DNA, RNA) can also be targeted In order to get a particular effect, it is generally important that of all the biological molecules in the body, only one (or possibly a few) are significantly targeted  The story of the first synthetic drug Hippocrates mentioned in 5th century B.C. Greece that a bitter powder extracted from willow bark was of use as a pain reliever and anti-inflammatory The same compound was used in other medical traditions – e.g. native American The active ingredient was purified in 1828 and named salicylic acid While salicylic acid acts as a mild pain reliever, it tends to cause excessive salicylic acid gastrointestinal distress  Aspirin In 1898 researchers Hoffman and Eichenrung in Germany acetylated the hydroxyl group in salicylic acid to form acetyl salicylic acid Hoffman gave some to his father who suffered from arthritis and found it relieved pain with fewer side effects than salicylic acid Bayer began marketing the compound - the first synthetic drug - in 1899 as “AspirinTM” How does Aspirin work? It is now known that Aspirin exerts most of its effects through inhibiting the synthesis of the hormone prostaglandin This is achieved through inhibiting the enzyme cycloxygenase 2 (COX-2) COX-2 – prostaglandin synthase COX-2 is an inducible oxygenase that synthesizes prostaglandins Prostaglandins mediate inflammation, and, indirectly, some types of chronic pain COX-2 is a monotopic membrane protein, that inserts half-way thorough the membrane  Aspirin irreversibly inhibits COX-2 by acetylating Ser530 Aspirin binds in an internal pocket Salicylic acid binds in this same pocket where it acts as a weak competitive inhibitor Ser530 attacks aspirin’s acetyl group, becoming irreversibly acetylated This explains why acetylsalicylic acid is a much better pain reliever Drugs that covalently modify proteins are often immunogenic Aspirin would probably not be Acetyl-Ser530 approved as a new drug today  What could possibly go wrong? Acetylation can easily be performed on other bioactive natural products 11 days after inventing Aspirin, Hoffman invented a di-acetylated version of the potent poppy-derived pain reliever, morphine morphine This was marketed by Bayer as HeroinTM Initially, Aspirin was far less popular than Heroin, which was thought to be healthier heroin  Drugs are no longer invented this way!! Hoffman’s dad got lucky - Aspirin has few and relatively mild side effects Sticking a random modifying group on a random compound can just as easily end up making an instantly fatal poison - or something more subtle such as a carcinogen or teratogen (causes defects in the embryo), or just a highly addictive drug In a risk-adverse culture, you need to be as certain as possible that the material is at least safe before exposing people to it Drugs must today pass multiple levels of regulatory approval Frist you “discover” your drug candidate You need to prove this compound is as safe as possible (through in vitro and animal trials) before you are allowed to test in a human You then need to prove that it does not have any disproportionately negative effects, initially in healthy volunteers and later in the target patients Finally, you have to prove that the drug is at least as helpful as currently available approaches After all this (and more) data gets put together, a government body of experts approves (often disapproves) the drug Then you can start selling it to patients  Inventing a drug - conceptually Mechanistic A well understanding Systematic search defined leading to a for a chemical “disease” possible route for entity that hits the A treatment state therapeutic target intervention Clinical Laboratory Drug discovery Clinical research research and development intervention  The process of inventing a novel drug Compound(s) One or two Investigative General that show core engineered new drug knowledge elements of the molecules that show filing (IND) Drug of disease/ desired activity a wide range of $$$$ biological “Drug profile desired biochemical process candidate” and biological activities Lead Lead Clinical trials (3 phases) optimization Pre-clinical discovery - very expensive Find preliminary Engineer in affinity, ADME-Tox Tissue Safety and molecules specificity culture, efficacy regulatory Some kind of Chemical synthesis, animal trials, Compound is approval screen with biochemical assays, synthesis scale up administered follow up testing structural analysis and formulation to people of hits etc. The average drug today takes ~15 years and costs ~US$2 billion to develop and approve 80 - 90% of the compounds that make it into clinical trials fail to be approved  Identifying a target Most drug discovery efforts focus on finding compounds that affect a specific molecular target – generally a protein This target is selected so that effective chemical intervention will efficiently produce the desired biological effect only Common targets include the first committed step of an enzymatic pathway, or the receptor of a signal transduction pathway (as subsequent signaling steps amplify this signal) Knowing the structure helps indicate how “druggable” a protein is – an important criterion in target selection  High Throughput screening HTS is the traditional first step for finding a new drug HTS requires that a library of ~105 to 106 drug candidate organic compounds have been previously synthesized For HTS you need to develop a biochemical or cell-based assay that will allow you to reliably identify when compounds bind the target This experiment is scaled down to a < 1 µl volume Robotics is used to run this test in the presence of each drug candidate in ~1000 well plates in a high throughput screen This takes several weeks with a cost ~$1 million Anything that gives a positive result in HTS is deemed a “hit”  Turning “Hits” to “Leads” Hits give a positive result in the assay - but don’t necessarily have the biochemical effect we were seeking The “Hit” could be due to non-specific unfolding or aggregation of the protein, or due to interaction with another assay component (such as coupled enzymes), or be caused by a contaminant Anything intrinsically unsuitable (e.g. too insoluble, too toxic, too hard to synthesize) is eliminated Hits are then tested against a variety of other assays that prove a desirable mechanism of action Many hits are unsuitable because they are non-specific and hit too many related proteins (=> lots of side effects) Compounds that are confirmed to show an appropriate biochemical effect and are worth further investigating are called “Leads”  Lead optimization The initial compound from an HTS will have some desirable properties (e.g binding) but lack many others Lead optimization optimizes the molecule to make something which has the affinity, selectivity, toxicity etc. profile of a drug This means synthesizing many new compounds, and testing their activity, selectivity, toxicity and other properties Leads often have ~10 µM binding constants, development candidates need ~1 nM binding This process takes years of time for teams of ~100 scientists Computational anticipation of the properties of hypothetical compounds is a critical part of keeping the resource requirements of the project within reason  Properties of orally available drugs Drugs need to bind tightly to their target (KD in the low nanomolar range) so that they are effective at low concentrations (minimize cross reactions) Drugs need to be reasonably soluble to be absorbed by the gut and get around the body However, they also need to be sufficiently lipophilic to dissolve into membranes and so get into and out of cells This means that they should have limited numbers of H- bond acceptors (10B Sadybekov, Nature, 2023 drug-like molecules  Virtual fragment screening An interesting recent innovation involves screening the building blocks of a large on-demand library Only a limited number of component fragments are initially tested Only library compounds related to initial hits are then screened This limits the number of docking experiments but gave a 33% hit rate, with a 1 nm inhibitor found for ROCK kinase Sadybekov et al Nature 2021  Virtual screening against Alphafold models AF2 models have been used for docking These models are apo models, in one structural state A recent study found that AF2 models show success rates similar to experimental structures Interesting, different compounds are found vs. the CryoEM structure, reflecting different structural states This result suggests that screening is possible for proteins that have not yet been characterized structurally Lyu et al, Science, 2024  Alphafold 3 and beyond AF3 has the ability to predict protein drug complexes AF3 is claimed to predict up to 93 % of complexes within 2 Å rmsd Training AF3 on large proprietary structural databases from drug companies will likely improve this AF3 and related deep learning tools may potentially Note - Isomorphic Labs has signed revolutionize drug screening and agreements with Eli Lilly and Novartis Total worth $3 billion + future drug revenues optimization  Summary Computational methods allow you to screen drug-like molecules virtually, bypassing expensive HTS Having the structure of a target allows you to “pre-screen” potential compounds computationally to find ones more likely to work It also points out unexploited interactions that you can try and use for additional binding energy AF3 and related deep learning tools may turbocharge structural approaches in drug discovery E.g. by powering reliable virtual screening and virtual affinity optimization (e.g. “evolving” compounds into a pocket) It may also allow you to screen a compound against the genome, checking for possible off-target effects