Drug Discovery: From Medicinal Plants to Computer-Aided Drug Design PDF

Chapter 1 FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN Early Experiences With Plants Drug Discovery: From Medicinal Plants to Computer-Aided Drug Design...

Chapter 1 FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN Early Experiences With Plants Drug Discovery: From Medicinal Plants to Computer-Aided Drug Design Michael K. Gilson and Laurence L. Brunton DESIGNING LARGE MOLECULES AS DRUGS: THE RISE OF BIOPHARMACEUTICALS Drug Discovery or Drug Invention? THE INVESTIGATIONAL NEW DRUG APPLICATION Target Identification Target Validation CLINICAL TRIALS Target Druggability Role of the FDA Beyond Single-Protein Drug Targets The Conduct of Clinical Trials Protein-Drug Binding: Affinity and Allostery Determining “Safe” and “Effective” EXPERIMENTAL APPROACHES TO PERSONALIZED (INDIVIDUALIZED, PRECISION) MEDICINE DRUG DISCOVERY PUBLIC POLICY CONSIDERATIONS Medicinal Chemistry High Throughput Screening The Pharmaceutical Industry Operates in a Capitalist Economy Fragment-Based Drug Discovery Who Pays? Emerging Experimental Technologies Intellectual Property and Patents Bayh-Dole Act COMPUTER-AIDED DRUG DISCOVERY Biosimilars Using Chemical Similarity to Discover Targeted Ligands Drug Promotion Structure-Based Drug Design Concerns About Global Injustice Artificial Intelligence in Drug Discovery Product Liability “Me Too” Versus True Innovation: The Pace of New Drug Development The first edition of Goodman & Gilman, published in 1941, helped to organize the field of pharmacology, giving it intellectual validity and an From Medicinal Plants to Computer-Aided academic identity. That edition began: “The subject of pharmacology Drug Design is a broad one and embraces the knowledge of the source, physical and chemical properties, compounding, physiological actions, absorption, Early Experiences With Plants fate, and excretion, and therapeutic uses of drugs. A drug may be broadly The human fascination—and sometimes infatuation—with chemicals defined as any chemical agent that affects living protoplasm, and few sub- that alter biological function is ancient and begins with our long expe- stances would escape inclusion by this definition.” In practice, of course, rience with and dependence on plants. Because most plants are root- a chemical or biological agent is considered a legal drug only if it has been bound, many produce defensive compounds that animals learn to avoid approved as such by a national regulatory agency, such as the U.S. Food and humans to exploit or abuse. Thus, the prior of an Arabian convent and Drug Administration (FDA) or the European Medicines Agency; came to appreciate coffee (caffeine) after noting the behavior of goats that these approved compounds are the focus of this book. gamboled and frisked through the night after eating the berries of the This first nine chapters of this book, General Principles, provide coffee plant; women sought to enhance their beauty by using an extract the underpinnings for these definitions of pharmacology and drugs by of the deadly nightshade plant, Atropa belladonna (“beautiful lady”), exploring the physiological, biochemical, and molecular mechanisms of enriched in atropine, to produce pupillary dilation; the Chinese herb ma drug action. This section covers drug invention, development, and reg- huang (ephedrine) was used as a stimulant; indigenous people of South ulation, as well as how drugs act in biological systems, i.e., pharmacody- American used curare to paralyze and kill animals hunted for food; and namics, pharmacokinetics (including drug transport and metabolism), the poppy juice (opium), containing morphine (from the Greek Morpheus, the influence of the gastrointestinal microbiome, and pharmacogenetics, with god of dreams), has long been used for pain relief and control of diarrhea. brief forays into pharmacovigilance and drug toxicity and poisoning. Sub- Morphine, of course, has well-known addicting properties, as do other sequent sections deal with the use of specific classes of drugs as therapeu- psychoactive natural products, such as nicotine, cocaine, and ethanol. tic agents in human subjects. The present chapter is an introduction to Note that these drugs did not derive from a search for a druggable target pharmaceuticals, their development, and the activities of the pharmaceu- or any knowledge of a target. Rather, drug discovery in the past often tical industry and government surrounding the discovery, production, resulted from serendipitous observations of the effects of plant extracts or and use of therapeutic agents. The processes of discovery and invention individual chemicals on animals or humans. Drugs were selected based of drugs have changed substantially with the general progress of bio- on effect, with no understanding of mechanism as we use the term today. medical sciences, the advent and improvement of computer-aided drug In the 20th century, the hunt for natural products broadened, driven in design, and technical advances in biochemistry and molecular biology. part by the discovery of antibiotics, such as penicillin and the cephalospo- Some of these new capabilities are reviewed below. rins, which fungi and microbes make to compete with each other. https://ebooksmedicine.net/  4 Abbreviations then use this information to steer the program of chemical synthesis and testing toward increasingly potent compounds. In the 1980s, it became practical to determine high-resolution ADME: absorption, distribution, metabolism, and excretion three-dimensional structures of complex organic molecules and even BLA: Biologics License Application larger molecules such as proteins, using and refining the techniques of CADD: computer-aided drug discovery X-ray crystallography pioneered by Hodgkin, Kendrew, and Perutz in CHAPTER 1 DRUG DISCOVERY: FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN DEL: DNA-encoded compound library the mid-20th century. It was already known that many drugs worked DHHS: U.S. Department of Health and Human Services by binding tightly to a disease-related protein and thereby modulating DMPK: drug metabolism and pharmacokinetics (e.g., inhibiting or activating) its biological function, but the atomic FBDD: fragment-based drug discovery details of these interactions had remained mysterious. As a consequence, the only way to advance a drug discovery project had been by synthesizing FDA: U.S. Food and Drug Administration and testing one compound after another. Now, with the protein’s three- GPU: graphics processing unit dimensional structure in hand, one could finally hope to design a com- HCV: hepatitis C virus pound that would bind with high affinity by fitting snugly into a pocket HDL: high-density lipoprotein in the protein, such as an enzyme’s active site. Thus, protein crystallog- HMG-CoA: 3-hydroxy-3-methylglutaryl coenzyme A raphy enabled structure-based drug design (SBDD), where the three- HTS: high-throughput screening dimensional structure of the drug target is used to guide creation of IND: Investigational New Drug tight-binding compounds, often called ligands. LDL: low-density lipoprotein Around the same time, computer technology began to advance mRNA: messenger RNA rapidly. This accelerated the data processing needed to go from X-ray NDA: New Drug Application diffraction patterns to protein structures (i.e., three-dimensional atomic NIH: National Institutes of Health coordinates) and enabled interactive visualization of complex protein NMEs: new molecular entities structures comprising thousands of atoms. It also opened new vistas in PDUFA: Prescription Drug User Fee Act computer-aided drug discovery (CADD), including the use of molecular SBDD: structure-based drug design simulations to model the physical interactions of compounds and pro- siRNA: small interfering RNA teins, and the development of tools to encode, archive, share, and analyze chemical and pharmacological data. In parallel, automation and minia- turization have dramatically increased experimental throughput, notably through robotic high-throughput screening (HTS), in which hundreds Drug Discovery or Drug Invention? of thousands of compounds can be tested rapidly and at relatively low The conventional phrase drug discovery makes sense for therapeutic com- cost in cellular or molecular activity assays. Today, excitement about the pounds obtained from plants and other organisms. Today, however, only power of artificial intelligence motivates wide-ranging efforts to apply a fraction of the new drugs introduced each year are discovered in nature. these technologies to drug discovery. Instead, most drugs are not discovered, but are totally new compounds, The following section goes into more detail regarding the process of painstakingly optimized against many criteria through an interplay of drug discovery, focusing on so-called small-molecule drugs, organic com- design and experimentation. In that sense, today’s new drugs are more pounds with molecular weights typically less than 500 Da, which have invented than discovered. traditionally been the most common type of drug. Subsequent sections The current paradigm for drug development grew out of synthetic introduce biological drugs, such as antibodies and other engineered organic chemistry, which arose as the dye industry in the late 19th biomolecules. century and has continued to flourish. Dyes are colored compounds with selective affinity across various biological tissues. Study of these Target Identification interactions stimulated Paul Ehrlich to postulate the existence of Today, most small-molecule drug discovery projects grow out of basic chemical receptors in tissues that interacted with and “fixed” the dyes. research that implicates a specific macromolecule, usually a protein, as a Similarly, Ehrlich thought that unique receptors on microorganisms key player in a disease and, further, suggests that a small molecule which or parasites might react specifically with certain dyes and that such binds this macromolecule could be used to treat the disease. The macro- selectivity could spare normal tissue. Ehrlich’s work culminated in the molecule thus becomes a candidate drug target. Many small-molecule invention of arsphenamine in 1907, which was patented as “salvarsan,” drugs are inhibitors (antagonists), which work by reducing the activity of suggestive of the hope that the chemical would be the salvation of their macromolecular target. Examples include the statins, which reduce humankind. This and other organic arsenicals were used to treat syph- cholesterol synthesis by binding and inhibiting the enzyme 3-hydroxy-3- ilis until the discovery of penicillin. Gerhard Domagk demonstrated methylglutaryl (HMG) coenzyme A (CoA) reductase, and β-lactam antibi- that another dye, prontosil (the first clinically useful sulfonamide), otics, which kill bacteria by inhibiting enzymes involved in the synthesis of was dramatically effective in treating streptococcal infections, thereby bacterial cell walls. However, some small molecules are activators (agonists) launching the era of antimicrobial chemotherapy. The collaboration of rather than inhibitors. Activators frequently target proteins whose normal pharmacology with chemistry on the one hand and clinical medicine role involves cell signaling, such as hormone receptors. For example, the on the other has been a major contributor to the effective treatment of asthma medication albuterol dilates bronchi by binding and activating β disease, especially since the middle of the 20th century. adrenergic receptors on bronchial smooth muscle, thereby mimicking the Early on, new compounds could be tested for their activities only in effect of adrenaline (epinephrine; see Chapter 10). whole organisms. This is how the nonsteroidal anti-inflammatory drug Candidate drug targets have been identified in many ways indomethacin was discovered, for example (Brune and Hinz, 2004). In (Hughes et al., 2011). For example, the enzymes targeted by the β-lactam the past 70 years, researchers have begun to understand in considerable antibiotics were unknown in advance and were discovered precisely detail the cellular and molecular mechanisms of disease. As a result of this because they are bound by these naturally occurring antibiotics. In con- basic biomedical research, it is possible to do initial testing of compounds trast, the target of the statins, HMG-CoA reductase, was identified by in vitro (“in glass”), using cellular and molecular assays. For example, elucidation of the pathways of cholesterol synthesis (Tobert, 2003), and one could look for the cellular responses due to inhibition of a protein this information was used to help discover the first statins. Similarly, as involved in a disease process. In this scenario, by testing enough appro- researchers have determined the regulatory functions of human protein priately chosen compounds, one could develop at least a partial under- kinases—enzymes that change the activities of other proteins by cova- standing of which types of compounds are most likely to be active and lently attaching phosphate groups to their hydroxyl-containing side chains—specific kinases have been targeted for small-molecule drug target validation aims to “de-risk” a project by lowering the probability 5 discovery (Cohen et al., 2021). Many kinase inhibitors are anticancer that a compound carefully developed to hit the targeted protein will fail agents that work by inhibiting protein kinases that accelerate cell prolifer- in clinical trials, whether because hitting the target does not influence ation. Some of these targeted kinases carry abnormal, cancer-associated the disease as expected or because the compound generates unanticipated mutations that make them hyperactive, so inhibiting them returns their toxicity, termed on-target or mechanism-based toxicity. SECTION I regulatory activities toward normal. The pioneering example of this sce- There are no absolute criteria for target validation, nor is there a single nario is the drug imatinib, which inhibits a cancer-associated mutant method. One approach is to use a chemical probe, a small molecule that protein kinase, the Bcr-Abl tyrosine kinase, and is used to treat chronic binds the target, and study its biological effects (Quinlan and Brennan, myelogenous leukemia (Buchdunger et al., 2002). 2021). This approach requires that such a probe be available, and the In recent years, technological advances enabling genome-wide experi- fields of chemical genetics (Stockwell, 2000) and chemogenomics (Bredel mentation (omics) have opened new approaches to identifying candidate and Jacoby, 2004) aim to create selective chemical probes for as many targets (Lindsay, 2003; Paananen and Fortino, 2020). Fast, inexpensive proteins in the human genome as possible. Alternatively, one may use GENERAL PRINCIPLES genome sequencing facilitates genome-wide association studies, in which gene silencing via small interfering RNA (siRNA) to block production variations in the susceptibility to a disease across many people are cor- of the target protein, thereby mimicking the effect of an inhibitor of the related with variations in specific genes, leading to suggestions for gene protein’s activity. Additional insight into the biological role of a candidate products (i.e., proteins), that may be suitable drug targets. The growing drug target may sometimes be obtained by studying genetically modified availability of patient genomic data in the context of patients’ electronic mice, including knockout mice, in which the gene coding for the target medical records will likely open new opportunities for data mining in has been disabled entirely, and transgenic mice, in which expression of support of target discovery in the coming years. It has also become routine the target’s gene is placed under the control of a promoter that can be to measure the quantities of messenger RNA (mRNA) transcribed from turned on by feeding the animals a specific compound, such as tetracycline thousands of genes simultaneously (the transcriptome) and to quantify (Lindsay, 2003). thousands of translated proteins (proteomics). By comparing such data between, for example, cancer cells and normal cells, one can identify pro- Target Druggability teins transcribed or present at elevated or depressed levels in the disease It is important to know whether the candidate target is drug- state. Mining data about these proteins from sources such as biomedi- gable, that is, whether it can, in principle, bind a small molecule with suf- cal databases, scientific articles, and patents, and integrating it with the ficient affinity. If the protein has been the target of a prior drug discovery omics data, may suggest certain proteins as candidate drug targets. effort, there may be informative small-molecule binding data in a public A totally different approach starts with the use of high-throughput database, such as BindingDB (Gilson et al., 2016), PubChem (Kim et al., instrumentation and robotics to test a large collection of small molecules 2021), or ChEMBL (Gaulton et al., 2012), or in an article or patent not (a chemical library) for biological activity in a phenotypic screen (Swinney yet curated by one of these databases. One may also check the Protein and Lee, 2020), which might use automated microscopy and image anal- Data Bank (Berman et al., 2000; Berman and Gierasch, 2021) for a crystal ysis to determine which compounds produce desired biological effects, structure of the target, which may assist in locating a suitable binding such as the activation of a desired gene in cultured human cells or the pocket for the small molecule to be developed as a drug. This is frequently death of a parasitic microorganism in culture. Various methods may then true for metabolic enzymes and receptors that have evolved to bind small be used for target deconvolution (i.e., to determine how the active small substrate and transmitter molecules. Many proteins belong to families, molecules work). For example, candidate targets of compounds found to such as the protein kinases, whose members have similar properties kill the malarial parasite Plasmodium falciparum were identified by cul- (e.g., an ATP binding pocket), so that if one member of a family is drug- tivating these organisms in gradually increasing concentrations of the gable, then the others probably are also. In contrast, receptors for proteins compound to select for resistant protozoa and then using omics methods often have large, relatively flat binding surfaces, rather than small binding to determine which genes had changed. The proteins encoded by these pockets suitable for a small-molecule drug, and are thus less likely to genes may then become candidate drug targets (Flannery et al., 2013). be druggable and influenced by small molecules. Efforts are under way to systematically search for all druggable targets encoded by the Target Validation human genome (Nguyen et al., 2017; Finan et al., 2017; Hopkins and After a candidate drug target has been identified, additional research is Groom, 2002) and to gain traction against targets hitherto considered usually warranted to validate it by seeking stronger evidence that a small undruggable (Dang et al., 2017). molecule that binds and modulates it will actually treat the disease (Jones, The ultimate validation of a candidate target is the successful devel- 2016; Lansdowne, 2018; see Box 1–1). For example, the fact that a protein opment of a novel drug that works by binding to it. Such a novel drug is is more abundant in cancer cells than normal cells by no means proves termed first-in-class. A first-in-class drug is a true innovation and may that it is a suitable drug target. Instead, this might be a correlate rather represent a medical breakthrough, so one might expect first-in-class to than a cause, so further research is needed to assess its role. Accordingly, be the goal of every drug discovery project. In fact, however, pharma- ceutical companies often engage in less innovative, more predictable projects by developing me-too drugs against old targets that are already BOX 1–1 Target Validation: The Lesson of Leptin fully validated by a first-in-class drug. Such projects aim to improve on the first-in-class drug through, for example, greater potency, reduced Biological systems frequently contain redundant elements or can side effects, or more convenient dosing (e.g., oral instead of intravenous), alter expression of drug-regulated elements to compensate for the and ideally to produce a new drug considered best-in-class. For example, effect of the drug. In general, the more important the function, the Merck’s lovastatin broke ground as the first statin, the first in a class of greater the complexity of the system. For example, many mechanisms drugs that lower cholesterol by inhibiting the enzyme HMG-CoA reduc- control feeding and appetite, and drugs to control obesity have been tase (see Chapter 37); but other statins, such as atorvastatin, have also notoriously difficult to find. The discovery of the hormone leptin, achieved enormous commercial success. which suppresses appetite, was based on mutations in mice that cause loss of either leptin or its receptor; either kind of mutation results in enormous obesity in both mice and people. Leptin thus appeared to be Beyond Single-Protein Drug Targets a marvelous opportunity to treat obesity. However, on investigation, A number of drugs, whether by accident or by design, hit multiple pro- it was discovered that obese individuals have high circulating tein targets, a phenomenon termed polypharmacology (Peters, 2013). concentrations of leptin and appear insensitive to its action. This phenomenon is particularly common when the target is a mem- https://ebooksmedicine.net/ ber of a family of proteins with similar binding sites. For example, the  6 full physiological effect of an adrenergic antagonist is determined by The hydrophobic effect, in which nonpolar or “greasy” parts of the its actions across the family of adrenergic receptor types and subtypes. drug and protein associate with each other to reduce their energeti- Similarly, many protein kinase inhibitors inhibit multiple kinases, each cally unfavorable exposure to water, much as oil droplets coalesce in to a different degree. There are instances where hitting multiple targets salad dressing is fruitful, such as inhibiting sequential reactions in a series. Modulating Dispersion forces—the attractive part of van der Waals interactions— multiple proteins in a single biochemical pathway or signaling network short-ranged attractive interactions between the instantaneous elec- CHAPTER 1 DRUG DISCOVERY: FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN overcomes the evolved redundancy of a robust biological system and trical dipoles that result from the constant fluctuations of negatively hence leads to greater efficacy than modulating only one protein. A single charged atomic electron clouds around positively charged atomic compound may, alternatively, hit two entirely different targets in different nuclei pathways, although this is more challenging to achieve without going to larger compounds. The analysis of complex molecular systems in relation These attractive forces need to overcome the entropic tendency of to drug action is termed systems pharmacology. the drug and protein to wander apart, due to thermal energy. There are Polypharmacology is not always beneficial, and indeed, it can lead to also, inevitably, forces that oppose binding and that must be overcome by toxicity. Some of the unintended effects of a drug will be termed side the attractive ones. For example, there is an energy penalty for stripping effects or even major adverse drug responses. For example, a number of water from polar chemical groups of the ligand and protein as they come initially promising compounds have proven to bind and inhibit hERG, together to bind. Thus, the overall affinity of a drug-protein interaction the K+ channel in the heart that mediates repolarization (the IKr current; reflects a delicate and hard-to-predict balance of attractive and repulsive see Chapter 34); inhibition of hERG can lead to potentially fatal arrhyth- interactions. mias. The hERG channel has, therefore, become a notorious anti- Small-molecule drugs do not bind to the relatively smooth, exterior target that must be scrupulously avoided by drug discovery projects surfaces of their protein targets, but instead are enfolded by binding pock- (Garrido et al., 2020). ets in the protein (see Figure 1–4). This structural arrangement makes it Some small-molecule drugs do not bind to proteins at all. For exam- possible to form the extensive, short-ranged, physical interactions that ple, platinum anticancer drugs, such as carboplatin, kill cancer cells by are needed to hold the two molecules together tightly. Druggable binding binding covalently to DNA; the aminoglycoside antibiotics block bacte- pockets (i.e., ones that enable small-molecule binding) usually are avail- rial protein synthesis by binding to RNA within the bacterial ribosome; able in enzymes whose substrates are small molecules and in receptors and antiviral nucleoside analogues are incorporated into viral DNA in that bind small-molecule hormones and transmitters. However, many place of normal nucleosides and then block DNA replication. The drug proteins lack a concave pocket and therefore are difficult or impossible sugammadex has both an unusual purpose and an unusual mechanism. to drug with a small molecule. In such cases, one may instead consider Surgical patients often receive not only general anesthesia but also the developing a protein therapeutic, such as an engineered antibody that nondepolarizing neuromuscular blocking agent rocuronium, which targets the protein of interest. Because proteins are large, they can form prevents involuntary movements of skeletal muscle during surgical extensive, short-ranged, physical interactions even with the relatively flat procedures (see Chapter 13). Sugammadex, a larger, cup-shaped mole- exterior surface of a targeted protein, and thus can achieve adequate bind- cule, binds and sequesters rocuronium. Thus, injection of sugammadex ing affinity where a small-molecule drug cannot. These considerations rapidly reduces the concentration of unbound rocuronium in the blood also help explain why it is difficult to develop a small-molecule drug that and promptly reverses paralysis when a procedure is complete. will block a protein-protein interaction: protein-protein binding usually involves a large number of interactions on a relatively flat binding inter- face between the two proteins, and a small molecule cannot get sufficient Protein-Drug Binding: Affinity and Allostery purchase on such a flat surface. A successful drug with a protein target must bind to its target with high Note that a drug must not only bind to its target but also have the affinity so that even a small dose of the drug will yield a blood concen- desired effect upon it. If the goal is to inhibit an enzyme, then a drug tration high enough to bind a large fraction of the targeted protein. If the that binds in the active site should easily accomplish this by simply affinity were low, then a high concentration of drug would be needed for blocking association of the enzyme with molecules of substrate. In a substantial fraction of the target sites to be occupied, and a large dose contrast, when a cell-surface receptor is the target, a small molecule of drug would need to be administered, leading to inconvenience and might interact at the agonist binding site but without inducing an acti- an increased risk of side effects. The affinity of a small molecule for a vating conformational change and thus might function as an antag- protein is generally given as the dissociation constant, the concentration onist or inverse agonist (see Chapter 3). A drug may also inhibit the of free drug molecules in solution at which 50% of the targeted protein function of a protein by binding in a pocket outside the active site, and has bound drug; the lower this concentration, the higher the affinity (see thereby modifying the three-dimensional conformation of the targeted Figure 3–3). Drug design projects typically aim for a dissociation con- protein; this is an allosteric effect. Such a drug must not only bind in stant on the order of 10–9 mol/L (1 nM); such a “nanomolar drug” is typi- a suitable pocket but also induce the desired conformational change. cally dosed in milligrams to grams per day. A successful drug should also Efavirenz and nevirapine, used in treating HIV-AIDS, are nonnucle- exhibit a high degree of specificity for its target protein, meaning that the oside reverse transcriptase inhibitors that act allosterically to inhibit drug does not interact with other proteins that could lead to undesired viral transcription of viral RNA to DNA (see Figure 65–5). Similarly, side effects and toxicity. In some cases, the effectiveness of a drug may a number of ligands interact with allosteric sites on GABAA receptors be influenced by not just the affinity but also the kinetic rate constants (see Figure 16–11) and other Cys-loop receptors to modulate receptor/ for drug-protein binding and dissociation, which determine the drug’s channel function. Allostery can also offer a sophisticated strategy to residence time at its receptor (Copeland, 2016). target a single enzyme from among a family of similar enzymes. Thus, Most drugs bind their targeted proteins via attractive, intermolecular inter- in designing a drug, one might take advantage of the fact that, even actions that do not involve a covalent chemical bond. These noncovalent within a family of related proteins with similar active sites, the members interactions typically include: will likely have other regions of their structure that are more variable and possibly unique. Designing a small ligand that binds to such a site Hydrogen bonding, in which an electronegative atom with a bound might produce an agent that is a quite selective allosteric modifier of hydrogen atom, such as a hydroxyl group, partly shares its hydrogen enzyme function. This approach is being used to target selected protein with an electronegative atom on the other molecule phosphatases (Mullard, 2018). Attractive electrostatic interactions between atoms of opposite charge, A few small-molecule drugs react chemically with their protein targets such as between a negatively charged carboxylic acid belonging to the to form irreversible, covalent bonds, rather than relying entirely on the drug and a positively charged arginine side chain of the protein noncovalent attractions discussed above. Such covalent drugs bond to a specific chemical group of the protein target, often a relatively reactive Medicinal Chemistry 7 amino acid side chain within an enzyme’s catalytic site. In principle, Synthetic organic chemistry remains at the heart of small molecule drug covalent drugs should require smaller, less frequent dosing, because a discovery, where it is specialized and known as medicinal chemistry. covalently bound drug will not dissociate from the protein as the con- Medicinal chemists typically are part of a project team that includes, centration of free drug dwindles over time following a dose (but note among others, biologists, assay specialists, and computational chemists; that some boron-containing compounds form reversible covalent bonds SECTION I their role is to reduce chemical concepts to practice by synthesizing and [Diaz and Yudin, 2017]). Drug developers have tended to avoid cova- purifying compounds that may ultimately lead to a new drug. In addition lent drugs because they necessarily possess chemically reactive groups to providing the expertise needed to synthesize compounds of interest, that risk reacting not only with the desired target but also with other they also help guide the design and selection of the compounds to be proteins and biomolecules, with the potential for causing undesired made. A key consideration is the complexity of a compound’s synthesis, biological effects. However, selectivity can be achieved by specific non- or “synthetic accessibility”, which must be balanced against the level of covalent interactions between the drug and the protein that pull the GENERAL PRINCIPLES interest in the compound. For example, it can be difficult to generate compound into a location and conformation where it is poised to form pure stereoisomers of compounds with multiple chiral carbon atoms, and the desired covalent bond. certain chemical structures can by synthesized only via demanding, mul- Covalent binding has been used to successfully target and inhibit ti-step syntheses. A compound that is too difficult to make or purify will a member of the RAS GTPase family, KRAS G12C, which had been not only slow down the research effort but may also lead to a drug that is viewed as virtually undruggable. As a result of such targeted posi- too costly to manufacture. tioning, the cancer drug sotorasib gains both potency and specificity Medicinal chemists also inform the drug design process by providing by forming a covalent bond with a cysteine side chain present in an insights into the properties of various chemical groups that might be oncogenic mutant form of KRAS but not in normal KRAS (Lanman incorporated into a drug, such as the attractive or repulsive interactions et al., 2020). they may form with the targeted protein, their susceptibility to metabolic changes following administration, their potential to spontaneously form undesired covalent bonds with biomolecules, and their influence on the Experimental Approaches to Drug Discovery compound’s ability to cross the blood-brain barrier (which may be desir- Given a validated target, the next major milestone in a drug discovery able or undesirable, depending on the goal of the project). This expertise project is arrival at a clinical candidate, a small molecule that binds the comes into play, for example, when a compound binds the target well but target with high affinity and specificity, has the desired effect on it, is rapidly metabolized by the liver into an inactive product. In this setting, and meets a range of other criteria for a safe, efficacious drug (Hefti, the medicinal chemist may try substituting the part of the compound that 2008). Some of these criteria relate to pharmacokinetics: How well will is metabolized with a “bioisostere”, a different chemical group with a sim- the compound be absorbed if given orally? How well does it distribute ilar shape and ability to interact with the protein but with reduced suscep- to the targeted organs and tissues? How rapidly and by what mecha- tibility to metabolic modification. More broadly, decades of experience nisms is it eliminated? Is it metabolized to an active metabolite? These have led to a number of rules of thumb for what makes a compound properties are often lumped together as absorption, distribution, “drug-like”, such as the “rule of five” (Lipinski, et al., 2001). These may metabolism, and excretion (ADME) or drug metabolism and pharma- be useful guides during drug discovery projects, but there are also many cokinetics (DMPK). exceptions to the rules (Zhang et al., 2007). It is also essential to confirm that the compound does not show evi- dence of toxicity. Both pharmacokinetics and toxicity can be initially High-Throughput Screening studied in vitro. For example, there are in vitro methods that examine If nothing is known about the structure of the target protein and what the ease with which the compound enters cells (see Chapter 4) and the small molecules can bind it, it is common to turn to HTS, in which thou- likelihood that liver enzymes (see Chapter 5) will chemically modify the sands or millions of compounds are tested using automation and robotics compound. Compounds also can be evaluated in vitro for evidence of (Wildley et al., 2017). Tiny samples of each compound are drawn from toxicity and mutagenicity. However, in vitro studies cannot fully model a stored chemical library and deposited into multiwell plates for testing. the complexities of a living organism; animal studies are still required to Substantial effort often must be invested to devise an assay that works reli- minimize the chances that a compound will be problematic when first ably in miniature and without user intervention. Most provide an optical given to human subjects. For example, toxicity is usually assessed by long- readout, such as a change in luminescence, fluorescence, or color, as these term monitoring of the health of two species of animals, generally one can be efficiently measured with an optical plate reader. The compounds rodent (usually mouse) and one nonrodent (often rabbit), when dosed screened can range from part of the vast, in-house compound collection with the compound. A good clinical candidate should also meet some that a major pharmaceutical company has assembled over the years to a nonbiological criteria. In particular, it must be amenable to large-scale smaller set purchased from a commercial vendor. A screening library is synthesis and high-grade purification at acceptable cost, and it should be often designed for the particular application. For example, one can pur- possible to create a formulation (e.g., a tablet or injection) that is suffi- chase libraries tuned for activity against protein kinases, libraries with ciently water soluble and stable. reactive groups that can form covalent bonds to the protein, and libraries Sophisticated technologies have been developed to speed the process designed to sample a wide range of compounds through high chemical of generating a clinical candidate. These mainly focus on the discov- diversity. A compound chosen at random from a screening library has a ery or design of compounds that will bind the protein target with high very low probability, typically 0.1% or less, of being active against a given affinity (potent ligands). Less progress has been made toward designing target (Shun et al., 2011), and HTS measurements are subject to experi- in safety and favorable pharmacokinetics. These properties pose more mental error. Therefore, many of the compounds that appear active on an complex challenges, because they go far beyond how a small molecule initial screen (hit compounds) are false positives, so careful data analysis and a protein interact with each other and instead involve the interac- and confirmatory testing are essential. tions of the small molecule with thousands of different biomolecules in Even the confirmed hits from a high-throughput screen are far from a living system. The technologies for ligand discovery are both experi- being drugs. Their affinity for the target usually is orders of magnitude mental and computational, and different methods are applicable in dif- too weak, they may lack the desired specificity, and they do not meet ferent settings. The following subsections touch on broad approaches DMPK or safety criteria. However, they offer an initial toehold on the but are not comprehensive. Note, too, that various approaches can challenge of finding a potent drug candidate. The next step is to purchase be used in combination, so the distinctions made here are ultimately (analogue by catalog) and/or synthesize (medicinal chemistry) similar somewhat artificial. compounds that ultimately give a picture of how various changes in https://ebooksmedicine.net/  8 1 O Compound ALDH1A1 ALDH2 ALDH3A1 O N 1 0.02 82 7.7 2 0.06 2.1 16 O 2 Cl 4 O CHAPTER 1 DRUG DISCOVERY: FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN 3 0.58 2.1 69 O N O 4 0.07 3.5 0.45 N 5 0.07 >100 0.31 O 6 2.0 0.05 18 3 Br 5 O O N O N 6 O Br O N Figure 1–1 Structure-activity relationship: scaffolds and substituents. Five inhibitors of the aldhyde dehydrogenase family of enzymes have a common chemical scaffold (black) while having different chemical substituents at two positions (red, green). The table lists the IC50 (μM) of each compound for three members of the aldehyde dehydrogenase family of enzymes: ALDH1A1, ALDH2, and ALDH3A1; i.e., the concentration of compound needed to provide 50% inhibition of each enzyme. The lower the IC50, the more potently the compound inhibits the enzyme. Focusing first on compounds 1, 2, and 3, one can see that adding an increasingly bulky halogen atom (Cl, Br) on the six-membered ring tends to reduce the compound’s potency against ALDH1A1 and ALDH3A1 but to increase it against ALDH2. Focusing next on compounds 1, 4, and 5, one can see that adding increasingly bulky, nonpolar, aromatic substituents at the nitrogen modestly reduces the potency against ALDH1A1, initially improves but then destroys potency against ALDH2, and consistently improves potency against ALD3A1. Such patterns can guide the design of new compounds with desired potency and selectivity. For example, the substituents in compounds 3 and 4 each reduce potency against ALDH1A1 while increasing potency against ALDH2, so it is not surprising that compound 6, which combines both substituents, has particularly low potency against ALDH1A1 and high potency against ALDH2. Note, however, that this kind of reasoning can only offer guidelines; its predictions are not always borne out by experiment. Data drawn from Kimble-Hill et al., 2014. the chemical structure influence activity against the target (structure- to the protein. This information can be used to stitch together designed activity relationships, or SAR) and other properties (Figure 1–1). This compounds that place the appropriate fragments at the right places in the information is used to guide the synthesis of often hundreds of com- protein’s binding pocket (fragment linking) or to optimize and expand one pounds with gradually improving properties. The most promising early selected fragment (fragment growing). In this way, FBDD avoids the com- molecules (lead compounds) serve as starting points for further improve- binatorial explosion of possible compounds made from various chemical ment (lead optimization), ultimately generating, hopefully, a clinical components and allows researchers to focus quickly on compounds made candidate, potentially accompanied by several backup compounds in case from only a productive subset of chemical components. The drug vemu- the leading candidate fails. rafenib, which targets an oncogenic mutation of B-Raf kinase and was developed with a fragment-growing strategy, is usually referenced as the Fragment-Based Drug Discovery first FBDD success story (Bollag et al., 2012). Even a large-scale screen can fail to provide useful hits (Keserü and Makara, 2009). This result becomes understandable when one recognizes Emerging Experimental Technologies that the number of stable, drug-sized, organic compounds is on the order The difficulty and cost of drug discovery, coupled with the market and of 1060 (Reymond et al., 2010), so a screen of even 106 compounds scarcely human need for new medications, have driven ongoing innovation in touches the vastness of chemical space. This vastness results from the com- drug discovery technologies. For example, DNA-encoded compound binatorial explosion of ways of connecting various chemical substruc- libraries (DELs) dramatically expand the number of compounds that tures, such as benzene rings, hydroxyl groups, and cycloalkanes. To be can be tested, relative to conventional HTS (Halford, 2017). Unlike a a good binder, a compound has to get multiple substructures positioned traditional HTS compound library, where each compound is kept in its so they all form favorable interactions with complementary groups in the own separate container or well, a DEL is a mixture of compounds in a targeted binding pocket. If it has two chemical components suitable for single container and can include far more compounds—into the billions binding the target but a third that is inappropriate or in the wrong place and even trillions. Each unique compound in the mixture is covalently on the compound, it may fail to bind the target. This perspective moti- bound to a corresponding unique short DNA molecule, which serves as vates another method of discovering binders, fragment-based drug dis- an identification tag. Such libraries can be synthesized and tagged with covery (FBDD) (Erlanson, 2012; Lamoree and Hubbard, 2017). In FBDD, the methods of combinatorial chemistry, where a mixture of compounds one conceptually breaks down drug-sized compounds into their sub- is split into multiple portions, each portion is modified with a different structures (fragments) and tests simple substructures against the target. chemical step and its DNA tags modified accordingly, and the portions Although such fragment-like molecules can bind only very weakly, such are mixed again. This process is iterated until the synthesis is complete. studies can, nonetheless, identify a small set of chemical substructures To screen the DEL for active compounds, one may immobilize the target that are suitable for the target, and one can then buy or synthesize larger of interest on a solid surface, expose the surface to the DEL mixture, and compounds assembled from these components. When either X-ray crys- then wash the surface to remove all the DEL compounds that have not tallography (Patel et al., 2014) or nuclear magnetic resonance spectros- bound tightly to the target. The binders are then removed from the target copy (Shuker et al., 1996) is used to detect or analyze fragment binding, by more aggressive washing, and the active compounds in the wash are specific information is usually available about where each fragment binds identified by sequencing the DNA tags they carry.  Another emerging technology, sometimes termed clinical trials in a dish instead computes tens or hundreds of quantitative descriptors for each 9 (Alpeeva et al., 2017; Fermini et al., 2018; Strauss and Blinova, 2017), aims compound. Examples include simple descriptors, such as molecular to predict the effects of a compound in humans more accurately than is pos- weight or number of aromatic rings, and more complex descriptors such sible with standard cell culture or animal models. This approach involves as electrical dipole and quadrupole moments. If one imagines descriptors creating specific cell types of interest from human pluripotent stem cells and as Cartesian coordinates in a multidimensional space, one can then quan- SECTION I using them to create three-dimensional organoids in culture (Fligor et al., tify the similarity of two molecules in terms of how close they are in this 2018; Liu et al., 2021; Sato and Clevers, 2013) or artificial tissue architectures descriptor space (Wale et al., 2008). via three-dimensional bioprinting (Ferrer and Simeonov, 2017). These rela- Similarity metrics such as these enable virtual screening, a fast, inex- tively intricate in vitro constructs promise to better recapitulate the proper- pensive, computational alternative to experimental HTS (Figure 1–3). ties of the corresponding in vivo tissues and may be used to test compounds In this approach, every compound in a chemical library—a large set of for activity, DMPK properties, compound metabolism, and toxicity. compounds that are available or synthesizable—is assessed for its simi- larity to one or more known ligands of the protein target. The most sim- GENERAL PRINCIPLES ilar compounds are tested in an experimental assay, and confirmed hits Computer-Aided Drug Discovery become candidates for further chemical optimization. This approach is most relevant when the three-dimensional structure of the targeted pro- The rise of information technology has enabled the research community tein has not been determined. When the structure is known, powerful to store and move large quantities of information, to write and maintain structure-based methods become applicable. complex software, and to do calculations at unprecedented speed and scale. These continually improving capabilities are used in a variety of ways to support and accelerate drug discovery. Thus, chemical informat- Structure-Based Drug Design ics enables compact databasing of information on hundreds of millions The detailed three-dimensional structure of a targeted protein opens up a of compounds and rapid recovery of chemical data for a specific com- range of additional computational methods for designing a small molecule pound and/or chemically similar compounds (Willett et al., 1998), while that binds the target with high affinity (Figure 1–4). The applicability of the Internet makes chemical (Gaulton et al., 2012; Gilson et al., 2016; Kim such SBDD methods has grown continually, due to rapid increases in com- et al., 2021), macromolecular (Benson et al., 1994; Berman et al., 2000; puter power and the development of technologies that make determining Berman and Gierasch, 2021; UniProt Consortium, 2015), biomolecu- protein structures easier and faster. One example is the use of synchrotrons lar pathway (Croft et al., 2014; Ogata et al., 2000; Oughtred et al., 2021; (e.g., the Advanced Photon Source at Argonne National Laboratory) to Wishart et al., 2020), and other databases readily accessible to researchers generate high-quality X-ray beams for use in protein X-ray crystallography. worldwide. These data are useful in their own right and also support the Another is the development of methods to solve the structures of mem- development and evaluation of computer models used in drug discovery. brane-bound proteins, such as ion channels and cell-surface receptors. In parallel, exponential increases in computer speed, measured as These can be high-quality drug targets because a drug does not need to the number of mathematical operations executed per second, have enter the cell to access them and because they regulate many cellular pro- made more and more detailed molecular simulations feasible. Ideally, a cesses. However, their structures were virtually impossible to solve until computational chemist could design a compound, hand the design to a methods were developed in recent years to grow three-dimensional crystals medicinal chemist to synthesize, and the compound would prove to bind of them. Since at least the 1980s, the promise of advances in SBDD methods the target with nanomolar affinity. When this level of accuracy becomes has inspired the founding of multiple companies. feasible, one might go further and compute the affinity of a candidate The field of physical chemistry tells us how to compute the binding drug to all known human proteins in order to check for unwanted inter- affinity of two molecules in water (Gilson and Zhou, 2007). Ideally, actions. This level of accuracy is not possible today, but existing methods one could use numerical solutions of SchrÖdinger’s equation to obtain have predictive value, and growing computer power may make this vision the electronic wave function for the compound, the target protein, achievable in the coming years. and the aqueous solvent, for any given conformation of the system Approaches to predicting the interactions of a small molecule with (i.e., given the Cartesian coordinates of all atoms). From the wave func- a protein may be broadly divided into ligand-based and structure-based tion, one could then compute the instantaneous force on every atom. approaches, as explained below. Given this method of computing atomic forces, one could simulate the system at atomistic detail, computing the reversible work of gradually Using Chemical Similarity to Discover pulling the compound out of the protein binding site as all the atoms wiggled, jiggled, and shifted due to thermal motion (Feynman et al., Targeted Ligands 1963). This reversible work would equal the free energy of binding, DGo, If the targeted protein is an enzyme with a small-molecule substrate which is directly related to the dissociation constant, KD: or a receptor for a small-molecule transmitter (e.g., histamine), then compounds chemically similar to the substrate or transmitter may be ∆ G o = RTlnK D (Equation 1–1) active against the target and thus useful starting points for drug design (Figure 1–2). For some targets, more extensive information about ligands This would be a prohibitively massive calculation with existing com- for the target may be available from prior drug discovery efforts and may puter technology. However, researchers have created fast approximations be used to guide a new project. As noted above, even if a drug has already to such an ideal calculation, each with its own strengths and weaknesses been developed against the target, there may still be room for a me-too in terms of accuracy, range of applicability, and the computer power drug with better properties, such as less frequent oral dosing or reduced required (Figure 1–5). side effects. Large quantities of data to support this ligand-based drug dis- An important approximation used in molecular modeling is the covery approach are available in the scientific literature, patents, and public force field or potential function, a mathematical model for the atomic databases (Gaulton et al., 2012; Gilson et al., 2016; Kim et al., 2019). forces that can be evaluated orders of magnitude faster than solving Metrics of chemical similarity abstract the detailed chemical structures SchrÖdinger’s equation (Dauber-Osguthorpe and Hagler, 2019). Force of compounds into characteristics that can be computed and compared fields often contain adjustable parameters fitted to give agreement with across molecules. One approach computes a compound’s molecular reference solutions of SchrÖdinger’s equation. With a force field in hand, fingerprint, which indicates whether various molecular substructures are it becomes practical to use molecular simulations to estimate protein- present (Muegge and Mukherjee, 2016). Other similarity metrics jetti- ligand binding free energies (Tembe and McCammon, 1984; Kollmann, son such details and, instead, compute and compare the overall shapes 1993; Gilson et al., 1997; Simonson et al., 2002). Such free energy methods of the two molecules and the electrical fields they generate (Bajorath, are among the most accurate approaches available to predict protein-li- 2017). In a third approach, even molecular shape is set aside, and one gand binding affinities (Schindler et al., 2020), and their use by the drug https://ebooksmedicine.net/  10 A. Statins O OH O OH O OH O OH HO HO HO HO O OH OH OH OH CHAPTER 1 DRUG DISCOVERY: FROM MEDICINAL PLANTS TO COMPUTER-AIDED DRUG DESIGN OH F F F N N O N O HN Mevastatin Fluvastatin Cerivastatin Atorvastatin B. SGLT Inhibitors SGLT IC50 (nM) IC50 (nM) Relative selectivity HO OH OH inhibitor at SGLT1 at SGLT2 for SGLT2 (col2/col3) Phlorizin 290 21 ~14 O O HO O Canagliflozin 710 2.7 ~260 HO OH Dapagliflozin 1400 1.2 ~1200 OH Empagliflozin 8300 3.1 ~2700 Phlorizin Ertugliflozin 2000 0.9 ~2200 F Cl O S O O HO HO HO OH HO OH OH Canagliflozin OH Dapagliflozin O O Cl Cl O OH O O O HO HO OH HO OH OH Empagliflozin OH Ertugliflozin Figure 1–2 Using chemical similarity to develop ligands. A. Statins. Statins inhibit 3-hydroxy-3-methylglutaryl coenzyme A reductase (HMG-CoA reductase), the rate-limiting enzyme in cholesterol synthesis. These inhibitors are widely used to lower blood levels of cholesterol (see Chapter 37). Mevastatin is a natural product that inspired development of the three FDA-approved statins shown here. Each compound has a polycyclic lower part linked to a common hydroxyacid moiety, which can also exist as a cyclic lactone. B. SGLT inhibitors. Sodium-glucose cotransporters (SGLTs) facilitate glucose ingress in the gastrointestinal tract (SGLT1) and the kidney (SGLT2). The natural product phlorizin inhibits both SGLTs to varying extents. Modifications of the phlorizin structure led to the four FDA-approved relatively specific SGLT2 inhibitors, the gliflozins, shown here. Gliflozins reduce renal reabsorption of glucose, thereby lowering blood sugar concentrations, and thus are used to treat type 2 diabetes (see Chapter 51). Each compound has a glucose moiety (except ertugliflozin, which has a glucose-similar moiety), sensible for compounds that interact with transporters that bind glucose. Phenyl-containing moieties endow each inhibitor with varying activities against each of the two protein forms, as shown in the table. Activities are given as IC50, the concentration of drug (nM) that reduces the transporter’s activity by 50%. Data adapted from Fediuk et al. (2020) and Wright (2021). discovery community has been enabled by the acceleration of molecu- Another computational approach, molecular docking (Guedes et lar simulations on graphics processing units (GPUs) (Salomon-Ferrer al., 2014; Huang, 2010; Meng, 2011), is fast enough to substitute for R, et al., 2013). Even with GPUs, though, the simulations are too slow (or supplement) a large-scale experimental high-throughput screen. to replace an experimental high-throughput screen of millions of com- In docking, most or all of the protein is held rigid, and the software pounds. Instead, simulations are most commonly used to help medici- tries a vast number of different locations and conformations—poses— nal chemists decide which chemical variations on a promising starting of a small molecule in the target’s binding site, searching for the one compound are worth synthesizing and testing. Fast molecular simula- that is lowest in energy and hence most stable. Because docking leaves tions also are used to explore the various conformations that a protein out so many known contributions to the free energy of binding (e.g., can adopt. For example, if a simulation shows that a new binding pocket protein flexibility and entropy), the energy model usually must be could form as a result of thermal protein motions, it may be possible to tuned against experimental binding data to make it more predictive. design a drug that will bind this hitherto unrecognized site. The resulting model is often called a docking score, to differentiate  A. Screening based on concepts of chemical similarity 11 Compound Library Known Ligands SECTION I Similar? GENERAL PRINCIPLES Candidate ligands Optimization (Med chem, Assay crystallography, modeling)