Chapter 1: Phenotypic Drug Discovery: History, Evolution, Future

Discovery of first-in-class medicines is a monumental and heroic accomplishment. To be of value, medicines must be both effective and safe; not doing more harm than the disease. What makes their discovery extremely challenging is the difficulty of identifying potential therapeutic compounds that will trigger mechanisms to specifically and safely change a disease phenotype. At the core of this challenge is identification of the dynamic molecular actions that trigger the change in phenotype: the right key (medicine) to unlock the right lock (target) that triggers the right phenotypic change (disease) (Figure 1.1), must be identified. Even today, with all the scientific advancements, sufficient knowledge is rarely available to provide a blueprint to rationally design a therapeutic key that both fits and unlocks a specific phenotype. The lack of complete knowledge creates risk and uncertainties. Strategies and methods are needed to manage the uncertainties and mitigate risk. These strategies and methods ideally evolve into processes that can be applied across different drug discovery programs.

Those who practice drug discovery will use all available knowledge to identify a suitable compound starting point and translational bioassay to initiate drug discovery efforts (Figure 1.1). Successful identification of progressible actives requires that the bioassay effectively recapitulates a molecular mechanism that is therapeutically effective and that the compound library includes an active molecule. If these two criteria are not met in the initial screening, then any work is of no value.

What is phenotypic drug discovery and why is it of value? Phenotypic drug discovery (PDD) is a disease-first paradigm to identify new medicines (see definitions, Table 1.1). The emphasis is on assays and strategies that recapitulate disease phenotypes and are agnostic to the molecular mechanisms—how the drugs work. The discovery of medicines now, as well as in the past, has relied upon empiricism to bridge knowledge gaps. PDD has the potential to provide a framework, a paradigm, to address the uncertainties and risk associated with empirical drug discovery.

The goal for this section is to briefly describe the discovery of a chronological sample of important medicines with respect to the role of empiricism in their discovery. The brief descriptions that follow identify the source of the modality, how the disease-modifying activity was identified and the type of knowledge and strategy used as a starting point (Table 1.2). There are many factors that contribute to progressing an initial discovery to a useful therapeutic product. The descriptions given focus on the initial findings while acknowledging that a significant contributor to success was the characteristics of the individuals and organizations that addressed the uncertainties and risk associated with validation of the initial findings and progression to products.

Early remedies were identified empirically, assisted by serendipity and astute observation. As the understanding of medical science has evolved, so did the desire for a more rational approach to invent new medicines. Prior to the genetic revolution medicines were identified primarily by a ‘compound-first’ approach. Pioneers in this era include Paul Ehrlich who invented the first ‘magic bullet’, salvarsan for syphilis and African trypanosomiasis from chemical dyes,⁸  and Sir James Black and Dr Paul Janssen who emphasized to initiate discovery efforts with a ‘pharmacologically active compound’.^9,10  Among Black's discoveries were propranolol in 1964 starting from isoproterenol, an analog of norepinephrine¹¹  and cimetidine, for gastric ulcers, derived from 4-methyl histamine.¹²  Among Janssen's discoveries were loperamide, an antispasmodic^9,13  and fentanyl, an analgesic; these were derived from meperidine.¹⁴  George H. Hitchings Jr, working with Gertrude Elion, emphasized the power of empirical, phenotypic screens when he stated in his 1988 Nobel lecture entitled ‘Selective Inhibitors of Dihydrofolate Reductase’ that ‘Those early, untargeted studies led to the development of useful drugs for a wide variety of diseases and has justified our belief that this approach to drug discovery is more fruitful than narrow’.¹⁵ 

The genetic revolution, with the resulting molecular view of biology, led to the vision that new medicines would be discovered based on new understandings of the role of genes in disease. Since 1990 a ‘mechanism-first’ strategy has dominated drug discovery and has led to the introduction of many new targeted therapies such as vemurafenib, a BRAF inhibitor for melanoma.¹⁶  However, the cost of producing new medicines has far outpaced the ability of the industry to discover new ones. It is argued that there is a gap in the translation of new understanding of disease mechanisms to the invention of new medicines.^7,17  Contributing to this gap is the challenge to identify the molecular mechanisms that trigger a specific response.

As with any new discovery, the discovery of a new drug requires understanding the potential use, awareness to recognize the discovery, as well as a bit of luck or serendipity. Historically, drugs relied primarily on serendipity and empirical observation. Penicillin was discovered by Fleming when cleaning his laboratory; he noticed large yellow colonies of mold overgrowing a culture of Staphylococcus bacteria on an agar plate. Fleming realized that something was killing the bacteria, and he proceeded to experiment with juice extracted from the mold by spreading it on agar plates covered with more bacteria. He found that even when the juice was highly diluted, it destroyed the bacteria. Calling the new antiseptic penicillin, after the Latin term for brush, Fleming had two assistants purify the mold juice, but he performed no tests on infected animal subjects. He published a paper in 1929 discussing the potential use of penicillin in surgical dressings, but went no further. It wasn't until the 1940s that penicillin was taken up by the medical community.¹⁸  Acetaminophen's (paracetamol) discovery was an accident. In the 1880s two French doctors ordered a supply of naphthalene to treat their patients who had a parasitic infection, but instead they received a drug known as acetanilide. They discovered that it reduced pain and reduced fever. A German scientist named Karl Morner discovered that acetanilide became acetaminophen when ingested and metabolized.¹⁸ 

Paul Ehrlich, as noted earlier, coined the terms ‘chemotherapy’ and ‘magic bullet’ to characterize the processes he had in mind.⁸  He had developed an early interest in the specific staining of tissues with dyes, including methylene blue. He reasoned that this might allow the detection of a substance that would specifically bind to and kill microbes without harming human cells. After it had been shown in 1905 that Atoxyl, an arsenical, had some activity against trypanosomes, a series of arsenical derivatives were synthesized. Ehrlich was looking for an agent that could achieve sterile cultures in animals with a single dose. Syphilis was a scourge affecting a significant proportion of men and women in the early 20th century. Routine therapy for the disease had been with mercury, but this was quite toxic. In 1905, Fritz Schaudinn and Erich Hoffmann identified the causative organism of syphilis—a spirochaete—which belonged to the same group of organisms as the trypanosomes. Arsphenamine, synthesized in 1907 and tested on rabbits infected with syphilis, was superior to all the other drugs that had been tested. Arsphenamine, known first by the number 606, as the 606th preparation tested in Ehrlich's laboratory, was named Salvarsan.¹⁸ 

Prontosil was the first drug to successfully treat bacterial infections and the first of many sulfa drugs—forerunners of antibiotics. Finding effective substitutes for natural treatments for bacterial diseases was the goal of Gerhard Domagk and his team at Bayer. In Domagk's view, a drug's role was to interact with the immune system, either to strengthen it or so weaken the agent of infection that the immune system could easily conquer the invader. He therefore placed great stock in testing drugs in living systems and was prepared to continue working with a compound even after it failed testing on bacteria cultured in laboratory glassware. Among the hundreds of chemical compounds prepared by Mietzsch and Klarer for Domagk to test were some related to the azo dyes. They had the characteristic –NN– coupling of azo dyes, but one of the hydrogens attached to nitrogen had been replaced by a sulfonamide group. In 1931 the two chemists presented a compound (KL 695) that, although it proved inactive in vitro, was weakly active in laboratory mice infected with Streptococcus. The chemists made substitutions in the structure of this molecule and, several months and 35 compounds later, produced KL 730, which showed incredible antibacterial effects on diseased laboratory mice. It was named prontosil rubrum and patented as Prontosil.^18,19 

The story of warfarin's discovery begins on the prairies of Canada and the Northern Plains of America in the 1920s.²⁰  Previously healthy cattle in these areas began dying of internal bleeding with no obvious cause. The cattle had grazed on sweet clover hay and the incidence of bleeding occurred most frequently when the climate, and therefore the hay in these areas was damp. The hay would normally have been discarded if it spoiled in storage, but in the financial hardship of the 1920s few farmers could afford to buy supplementary fodder for their cattle, and thus the moldy hay was used to feed. Local veterinary surgeons demonstrated sweet clover disease to be potentially reversible if the offending moldy hay was removed, or if fresh blood was transfused into the bleeding animals. Unclotted blood from dead cows was the link to understanding the cause of the disease. A new in vitro clotting assay using plasma from rabbits was developed to guide chemical fractionation of compounds found in the hay. After some 6 years of intensive work the investigators determined that natural coumarin became oxidized in moldy hay, to form the substance that would become better known as dicoumarol.²¹  Warfarin, a more active soluble form of dicoumarol, was first used as a rat poison prior to the realization it could be used as an anticoagulant.²⁰ 

The history of diabetes mellitus is replete with many therapies, nearly all, including insulin, first given without any knowledge of a mechanism of action. Beginning with early Egyptian physicians who, according to the Ebers Papyrus of 1500 BCE prescribed a mixture of cakes, wheat grains, fresh grits, green lead, earth, and water, up to the more recent introduction of the thiazolidinediones, knowledge of physiologic effects preceded by years or centuries any knowledge as to what the medicament actually did.²²  In medieval times, a prescription of Galega officinalis was said to relieve the intense urination accompanying the disease that came to have the name of diabetes mellitus. G. officinalis is also known as Goat's rue, the French lilac or Italian fitch. The active ingredient in the French lilac that produced the lowering of blood glucose was shown to be galegine or isoamylene guanidine. Metformin, a less lipophilic biguanide was approved for use in the USA in 1995.²³  Bailey and Day noted that there are several ironies about metformin.²³  In our high-tech era of drug discovery and development, this first-line treatment for type 2 diabetes is little removed from an herbal remedy of the Middle Ages. Despite its chemical simplicity and detailed investigation, metformin continues to evade a complete understanding of how it works. While endless pharmacovigilance has monitored the safety profile of metformin, its natural ancestor, G. officinalis (known as professor-weed in the USA) is a class A federal noxious weed in 35 states of America, and appears on the database of poisonous plants. Bailey and Day²³  in their review of metformin thought ‘It is perhaps apt to conclude with a quote from the Swiss born physician Theopharastus Bombastus von Hohenhein (1493–1541), better known as Paracelsus: “The right dose differentiates a poison from a useful medicine”’.

Elion and Hitchings discarded the old ‘magic bullet’ method and applied the basic principles of biochemistry and physiology.¹⁵  Having found that bacteria needed folic acid and purines for DNA synthesis, they were able to develop inhibitors of these pathways. Rather than the traditional trial-and-error method of drug discovery, Hitchings believed in the necessity of a more rational method of research. The recent development of sulfa drugs led him to think that other substances that interfered with the metabolism of microbes—as sulfa drugs had been shown to do—could also be developed as drugs. Hitchings decided to target the synthesis of nucleic acids in cells. Now known as DNA and RNA, the nucleic acids determine the genetic makeup of cells and direct the process of protein synthesis. The idea was to look for differences in nucleic acid metabolism among normal human cells, cancer cells, protozoa, bacteria, and viruses that could be used to develop drugs that would selectively block the growth of cancer cells and of noxious organisms.

Hitchings assigned Elion to investigate the purines, including adenine and guanine, two of DNA's building blocks, and their role in nucleic acid metabolism in cells. They soon discovered that bacterial cells required certain purines in order to make DNA. They reasoned that if they could prevent these purines from being incorporated along the metabolic pathway that leads to DNA synthesis, then they could stop the production of DNA and thereby stop cell growth. They set to work on compounds that did just this by locking up the metabolic enzymes necessary for purine incorporation. This work led to the discovery of azathioprine.²⁴ 

In the 1950s most medicinal chemists started from compounds known to be active in man. James Black summarized his discovery of propranolol as follows, ‘When we started looking for beta-blockers we did not have the luxury of a known compound. All we had was isoproterenol, a selective agonist. The approach of turning agonists into highly selective antagonists seems to work well, but it is invariably slow. All of us who did these sorts of things—it really is a sort of forced Darwinian evolution: we as medicinal chemist would force mutations in a molecule, and then just as in Darwinian evolution, you expose the new entity to an environment asking, “Is its survival better according to our bioassay?”’.²⁵  In 1976, Black used a series of histamine analogs to develop cimetidine—useful in the treatment of gastric and peptic ulcers. Black is noted for saying ‘the most fruitful basis for the discovery of a new drug is to start with an old drug’.

The story of omeprazole describes how in 1972 animal models were used to optimize an active compound to a discover a blockbuster drug class.²⁶  Efforts to identify a drug to block gastric acid secretion were stalled when compounds that worked in rats failed to be effective in humans. The effort was restarted using anesthetized dogs and with a compound that literature search found to have antisecretory properties, but severe acute toxicity. Chemists hypothesized that a specific chemical element was responsible for the toxicity and the new chemical approach identified the lead compound. Subsequently, after much work, the target was identified as H⁺K⁺ATPase and the molecular mechanism of omeprazole determined to be irreversible inhibition by a metabolic intermediate sulphenamide.²⁶ 

Clear insights into the mind-set of drug discovery in the 1970s is described by Jordan in a report on the discovery of tamoxifen for use in breast cancer.²⁷  The strategy of Ehrlich was used, in which a broad synthetic organic-chemistry program is followed by testing of compounds in an appropriate animal model for the disease; compounds with low toxicity are then selected for clinical testing. He states that the ‘Ehrlich strategy of focusing on translational models of relevance to the disease ultimately proved to be an appropriate formula in the 1970s for devising a successful strategy for the treatment and prevention of breast cancer’.²⁷ 

The engine that has driven antiepileptic drug discovery is screening in well-validated animal models of seizures.²⁸  Rogawski notes that the animal models are unbiased with respect to mechanism, providing an opportunity to identify compounds that work in new ways and on novel molecular targets. As clinical experience with a drug increases, it is not uncommon for the new drug to exhibit a personality that differs from that of other marketed antiepileptic drugs. Rogawski reported in 2006 that there were two ways for drugs to enter the in vivo screening process.²⁹  They were submitted for testing in order to characterize their biological activity and serendipitously found to protect against seizures or they were more or less rationally designed based on currently marketed drugs. This is exemplified by lacosamide.³⁰ 

The discovery of cyclosporine provides another example demonstrating the unique characteristics that unite a compound with an assay to provide a progressible observation. Discovery of cyclosporine in 1971 began a new era in immunopharmacology. It was the first immunosuppressive drug that allowed selective immunoregulation of T cells without excessive toxicity. Cyclosporine was isolated from the fungus Tolypocladium inflatum. Cyclosporine was identified as an immunosuppressor as part a program of approximately 50 pharmacological tests performed by different groups in the pharmacological department at Sandoz. Out of all the pharmacological tests in the general screening program only one produced a positive result in the hemagglutination assay for immunosuppression. Borel notes that modification of the hemagglutination assay to account for missed positives was necessary to identify the activity of cyclosporine. Of note is that the compound was much less active in the follow-up assay due to oral dosing and poor solubility. If it had been run in this assay initially, it would have not been pursued.³¹ 

AZT, or azidothymidine, the first weapon against HIV, was a repurposed compound originally developed in the 1960s for cancer; the compound was supposed to insert itself into the DNA of a cancer cell and mess with its ability to replicate and produce more tumor cells. But it didn't work when it was tested in mice and was put aside. Two decades later, after AIDS emerged as new infectious disease, Burroughs Wellcome began a massive test of potential anti-HIV agents, hoping to find an old compound to repurpose. When a resynthesized version of AZT was screened with animal cells infected with HIV, it blocked virus activity.³² 

Artemisinin is the most effective current therapy for the treatment of malaria. The discovery of artemisinin is attributed to You-You Tu, at the Institute of Materia Medica, China Academy of Traditional Chinese Medicine in the early 1970s.³³  An Artemisa extract showed a promising degree of inhibition against parasite growth, consistent with activity that had been reported for this plant species in ‘A Handbook of Prescriptions for Emergencies’ by Ge Hong (Jin Dynasty, AD 284–346). It is reported that the extracts were evaluated against parasites and in mouse malaria model. Artemisinin in an endoperoxide that is activated to a reactive species by heme in red blood cells to provide selective toxicity to all life stages of the malaria parasite.

Vismodegib was identified in a pathway screen using a luciferase reporter. It showed promising antitumor activity in preclinical studies.^34,35  Binding of vismodegib to the transmembrane domain (TMD) of smoothen induces a conformational change that is propagated to an extracellular cysteine-rich domain (CRD) resulting in loss of a cholesterol molecule bound to the CRD-linker domain–TMD interface. Mutations predicted to prevent cholesterol binding impair the ability of SMO to transmit native Hedgehog pathway signals. Vismodegib inhibits Hedgehog pathway signaling by binding to and inhibiting smoothen, a G protein-coupled receptor.

In recent years, PDD has serendipitously identified chemical chaperones that provide a new mechanism to treat genetic diseases with a loss of function. For example, in 2015, lumacaftor was approved for cystic fibrosis. Lumacaftor acts as a chaperone during protein folding and increases the number of Cystic fibrosis transmembrane conductance regulator (CFTR) proteins that are trafficked to the cell surface.³⁶  It is combined with ivacaftor under the name Orkambi. Lumacaftor was discovered with cellular high-throughput screens.^36,37  Another example is provided by migalastat, which was approved for Fabry's disease in 2018. Migalastat, a potent competitive inhibitor of α-Gal A, effectively enhanced α-Gal A activity in Fabry lymphoblasts, when administrated at concentrations lower than that usually required for intracellular inhibition of the enzyme. Migalastat accelerated transport and maturation of the mutant enzyme and elevated the enzyme activity in some organs following oral administration to transgenic mice overexpressing a mutant α-Gal A.^38,39 

Ledipasvir was the first NS5A inhibitor approved for hepatitis C infection.⁴⁰  The initial discovery of this class was with daclatasvir, which was approved subsequently to ledipasvir. The initial active compounds were found using a hepatitis C virus replicon assay. Ultimately the activity was determined to be due to an impurity and the target identified through genetic resistance mapping. Binding was stated to be necessary, but not sufficient for the antiviral effect of daclatasvir.⁴¹  The ratio of NS5A to daclatasvir in a replicon at IC₅₀ concentrations was in excess of 1 : 1000, leading to the proposal that daclatasvir works by association with an oligomeric form of NS5A and effects a distortion that disrupts the function of the entire complex.

These examples highlight some of the trends through the past century in drug discovery. The early part of the century involved more of a ‘compound-first’ approach focused around chemical modifications and experimental pharmacology of compounds active in phenotypic assays. As the century progressed and knowledge of disease increased, biochemical approaches as favored by Hitchings and Elion, and later targeted approaches, became favored in a move to ‘mechanism-first’ strategies. In addition, the contributions of individuals and organizations to recognize need, overcome challenges, and to employ strategies to address risk and uncertainties cannot be underestimated.

History shows that increased understandings of genetics, genomics, chemistry, biochemistry and physiology has decreased the translational knowledge gap. This coincides with drug discovery progressing from Ehrlich's ideas of chemotherapy and magic bullets, the approaches of Black and Janssen to start with pharmacologically active compounds, those of Hitchings and Elion to utilize biochemical understandings, and the use of genetics and genomics to identify drug targets (Figure 1.2).

Despite these advances, it is still challenging to predict the exact molecular interactions that will trigger a specific change in a phenotype. This author feels the need for empiricism is in part due to the challenge to predict these specific molecular mechanisms of action (MMOAs) for first-in-class medicines. The understanding of the specific interactions that will specifically trigger the desired phenotype is usually limited for first-in-class medicines. Many selected drug targets are well studied and characterized in terms of their pathophysiological role,^42,43  but such knowledge may be much more limited for first-in-class medicines, increasing the risk that modulating the target will not have the desired therapeutic effect. In addition, although there are well-established processes for identifying drug candidates that bind to a macromolecule (protein target)^44,45 —which is considered essential for initiating the molecular actions that trigger a system to respond phenotypically—the physiological environment of the target can influence the occupancy of the target by the drug and the translation of the response (for example, through the concentration of competing substrates/ligands, the conformational state, and spatial and temporal location of a target). Furthermore, the biological response to a drug can be affected greatly by other factors, including off-target interactions, as well as drug metabolism and pharmacokinetics.^46–48  Consequently, selective differentiated drug action is generally more complex than described by a simple model in which the drug effects are determined by occupancy of a particular protein target, and can be more effectively addressed using systems biology and network pharmacology to provide a systems-level understanding.^49,50 

This is exemplified in the specific molecular action of aspirin as an antithrombotic agent. The molecular target of aspirin is cyclooxygenase, and the molecular mechanism is irreversible inhibition of prostaglandin formation in platelets.⁵¹  Binding to the target is required, but not sufficient, to explain the activity of aspirin as an antiplatelet drug. The differentiating action is the inability of platelets to synthesize new enzyme, due to their lack of a nucleus. This results in a pharmacodynamic effect that will last the life-time of the platelets (life-span 8–10 days). Both the molecular mechanism of action (irreversible binding) and the physiological context (life-span of platelets) are critical to the therapeutic action. A similar molecular mechanism provides the basis for the utility of eflornithine for African sleeping sickness, a parasitic disease caused by Trypanosoma brucei infections. Eflornithine is an irreversible inhibitor of ornithine decarboxylase that was originally developed as an anticancer therapy, but it is not efficacious in this setting due to the rapid turnover rate of the enzyme in humans (t_1/2 10–30 min). However, it is effective against in T. brucei infections owing to the much slower turnover rate of ornithine decarboxylase in the parasites (t_1/2 18–19 h).^52,53 

Differentiation and specificity of receptors are also dependent on unique ligand- and context-specific conformations and subsequent interactions.⁵⁴  Ligand-induced conformational selectivity in the estrogen receptor system is exemplified by the differentiation of estradiol, raloxifen and tamoxifen. Unique ligand-induced conformational changes due to binding of these molecules to a similar location in the ligand-binding domain results in different therapeutic uses (postmenopausal hormone deficiency, osteoporosis and breast cancer, respectively).⁵⁵  The unique conformations are believed to be associated with distinct patterns of co-activator and co-repressor recruitment, and corresponding tissue specificity.

Many of the medicines that were invented starting with a target specific assay required an additional empirical phenotypic assay to prioritize the actives and identify candidates with functional efficacy. The discovery of gleevec, a c-abl kinase inhibitor that works through stabilizing the kinase inactive state,^56,57  PARP inhibitors such as olaparib,^58,59  that trap the PARP to damage DNA, and maraviroc, a CCR5 inhibitor for HIV infection that stabilizes a unique conformation of the receptor that does not bind to the virus,^60,61  all required empirical assays to identify molecules with effective molecular mechanisms. John Moffat coined the term mechanistic-informed PDD (MIPDD) to account for the need to use empirical assays to identify MMOA in target-based strategies.⁶² 

The utility of empirical assays to identify novel mechanisms of specificity is demonstrated by macrolide antibiotics (e.g. erythromycin) that inhibit protein synthesis by targeting the bacterial ribosome. They bind at the nascent peptide exit tunnel and partially occlude it. Thus, macrolides have been viewed as ‘tunnel plugs’ that stop synthesis of every protein. However, recent evidence demonstrates that macrolides selectively inhibit translation of a subset of cellular proteins and that their action critically depends on the nascent protein sequence and on the antibiotic structure. Therefore, macrolides emerge as modulators of translation rather than global inhibitors of protein synthesis.⁶³  It would have been difficult to envision this mechanism of specificity using a mechanism-first strategy. PDD provides opportunities to identify novel mechanisms that would not have been predicted based on available knowledge.

Renewed interest in empirical drug discovery and its formalization under the name PDD came subsequent to an analysis of the discovery strategies for new molecular entities approved by the US Food and Drug Administration (FDA) between 1999 and 2008.⁶⁴  This analysis determined that most first-in-class small-molecule drugs were discovered empirically, whereas the majority of those that followed were discovered using target-based drug discovery (TDD). This report concluded that the mechanistic knowledge available when a program is initiated is often insufficient to provide a blueprint for the discovery of first-in-class medicines. This includes not only knowledge of the presumed drug target, but also knowledge of how that target translates to a specific, therapeutically useful phenotype—the MMOA.⁶⁵  This knowledge gap was addressed empirically.

The formalization of PDD to a strategy or discipline was subsequently suggested by Eder and coworkers in analysis of first-in-class new molecular entities from 1999 to 2013.⁶⁶  In this work phenotypic screening was defined as testing of a large number of (in most cases randomly selected) compounds in a system-based approach using a mechanistic agnostic assay. Using this different definition Eder and co-workers found that PDD contributed to much fewer discoveries. Despite these difference in conclusions these authors proposed ‘the goal will be to screen phenotypically in an efficient and effective manner and to combine phenotypic screening sensibly and productively with target-based drug discovery’.⁶⁶ 

PDD has the potential to be much more than random screening in complex systems, as defined by Eder and co-workers.⁶⁶  PDD has the opportunity to create further value by providing a strategy to address mechanistic knowledge gaps at any level. This was the thinking and rationale behind the earlier analysis by Swinney and Anthony.⁶⁴ 

In many cases the difference between PDD and TDD are painted as black and white. In reality, there is generally some level of mechanistic knowledge with target agnostic strategies; however, this knowledge does not involve a specific molecular interaction, and for TDD there is a hypothesis regarding the molecular interaction but the mechanistic knowledge is incomplete as it does not include the exact molecular dynamics that will trigger the change in phenotype (as noted earlier). The major distinction between these two strategies is in how they address the incomplete mechanistic knowledge. PDD relies on the translatability of the phenotypic marker in the screening assay while TDD relies on the translatability of the target to the disease. In both cases there is still the risk and uncertainty associated with identification of a specific MMOA that will be therapeutically useful, and this must be derisked in clinical experiments.

Overall, the TDD strategy is a linear process with well-defined and tractable technical milestones.^66,67  Typically, targets representing known ‘druggable’ proteins are selected by their association or ‘validation’ with a particular therapeutic indication. Enablement of primary screens is typically low risk and based on previous industrial experience with members of the molecular target class. TDD flow schemes are principally concerned with enhancing primary target potency/efficacy, achieving biochemical selectivity, and demonstrating cell-based activity upon target engagement, tasks which are informed and generalizable from prior experience with the target class. With these assets in place a drug discovery team can optimize lead compounds for biopharmaceutics properties and safety to provide a drug candidate. Although challenging, discovery and optimization of advanced leads/clinical candidates by TDD follows a process, with predefined milestones and established cycle times. This process is strongly dependent on the validity of the target and is most effective for followers/best in class and monogenetic diseases including many cancers, where a target and MMOA is well validated. This process is not as efficient for first-in-class medicines for complex diseases in which the target and MMOA are difficult to pin-point.

Development of physiologically relevant in vitro disease models are foundational to PDD. As a result, PDD assays are frequently very complex multifactorial cellular systems^68,69  which tend to be unique to the disease model and contrast significantly with TDD assays, which are generally more standardized and process-friendly. PDD flow scheme development is frequently dynamic and utilizes the results of pilot screens and project progression to reveal unwanted cellular processes and signaling pathways, which in turn requires modification of flow schemes to identify undesirable phenotypic mechanisms. Phenotypic actives representing distinct chemical clusters can, in principle, be working through diverse mechanisms; as a result, in vivo proof-of-concept data are frequently desired early in the lead optimization phase to confirm/establish linkage between the in vitro and in vivo systems. These and other factors reviewed by Moffat et al.¹  indicate that PDD projects tend to front-load resources and introduce uncertainty in PDD project milestones and progression metrics, which increases perceived risk for managers.

Subsequent to the report by Swinney and Anthony in 2011⁶⁴  there has been a resurgence in phenotypic screening in industry as well as academia. There have been increased efforts to develop improved disease-relevant assays and to identify new medicines with novel mechanisms of action. A recent report by Haasen and co-workers documents the lessons from 5 years of phenotypic screening at Novartis from 2011 to 2015, detailing a dramatic increase in the percentage of phenotypic screens. Among the many lessons and trends was an increase in more disease relevant models using induced pluripotent stem cells and primary human cells and the use of small-scale screens in flexible lead discovery strategies.⁷⁰ 

Among the strengths of PDD is repurposing to rapidly identify new therapies. This can be particularly important for emerging pathogens, such as COVID-19. Wang and co-workers reported using Vero E6 cells infected with nCoV-2019BetaCoV/Wuhan/WIV04/2019 to screen compounds with broad-spectrum antiviral activity.⁷¹  In this screen, remdesivir and chloroquine were identified as candidate medicines.

Advances in machine learning and artificial intelligence (AI) are poised to make a large impact on PDD. AI will provide new biomarkers and more precise phenotypes. Phenotypic endpoints which translate to the clinic, the chain of translatability,¹  are essential; however, not all disease states are associated with defined sets of disease markers. In these situations, high-dimensional profiles composed of gene expression profiles or cellular morphology features are envisioned to define surrogate disease phenotypes where reversion of the ‘disease state’ profile to a ‘wildtype’ or ‘normal’ representation is an indication of therapeutic efficacy.⁷² 

Cell Painting is a surrogate phenotype approach where cellular morphology is measured by fluorescent labeling of eight cellular compartments and subsequent automated high-content imaging analysis of ∼1500 features per cell.^73,74  The resulting cellular morphological profile reflects general changes in cellular state following chemical^75,76  or genetic perturbation. Morphological changes in cell state can cluster structurally similar compounds or can identify functionally similar but structurally distinct molecules⁷⁷  and has been used to deconvolute the mode of action of phenotypic actives working through non-protein targets.⁷⁸ 

Analysis of these profiles is facilitated by AI and machine learning. Kraus and co-workers report using deep-learning approaches that combine deep convolutional neural networks with multiple-instance learning (MIL)⁷⁹  to facilitate evaluation of morphological changes. They introduced a new neural network architecture that uses MIL to simultaneously classify and segment microscopy images of cell populations.⁷⁹ 

Stokes et al.⁸⁰  have provided proof of concept that deep learning can facilitate the identification of phenotypically active compounds. Using a novel method to describe chemical structures and iterative rounds of deep learning based initially on 2335 diverse molecules and their ability to inhibit growth of Escherichia coli, a predictive model was developed and subsequently used to analyze >107 million molecules from diverse libraries. Subsequent filtering of molecules with high prediction scores but low Tanimoto similarities to known antibiotics identified 23 for testing, eight of which displayed growth inhibition in at least one of five bacterial species (E. coli, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, and Pseudomonas aeruginosa). Significantly, two compounds showed broad-spectrum activity with low Tanimoto similarity to known antibiotics and low predicted toxicity in humans.

Many have high hopes that in the future, machine learning and AI will provide knowledge to bridge the knowledge gaps that will enable the dream of rational discovery of first-in-class medicines. History tells us that a more realistic dream is that they will narrow the knowledge gap to empower drug discovery, not close it. For the foreseeable future there will be a knowledge gap that will need to be addressed empirically by observation of changes to relevant phenotypes.

Recent advances in the implementation and derisking of PDD (for compound library selection, biomarker development, mechanisms identification, and safety studies) and the potential for AI will provide a more formalized process to address the uncertainties and risk as well as help define the path forward. Many of these advances are described in detail in other chapters of this book.

History repeats itself, so the saying goes. Drug discovery history utilized a continued influx of knowledge and technologies to reduce knowledge gaps; however, this rarely precluded a role for empirical trial and error and experimental pharmacology in the identification of first in class medicines. In the future, empiricism will still be required to bridge knowledge gaps to identify novel mechanisms of action and corresponding medicines. PDD provides a strategy to help address risks and uncertainty as well as enable biology to choose effective mechanisms and corresponding medicines. As such PDD provides opportunities to identify novel mechanisms and molecular mechanisms that would not have been predicted based on available knowledge.