Opportunities and Challenges in Phenotypic Drug Discovery: an Industry Perspective

Home , Drug discovery, Phenotypic screening

PERSPECTIVES

Nevertheless, there are still challenges in OPINION prospectively understanding the key success factors for modern PDD and how maximal Opportunities and challenges in value can be obtained. Articles published after the analysis by Swinney and Anthony have re-examined the contribution of PDD phenotypic drug discovery: an to new drug discovery6,7 and have refined the conditions for its successful application8. industry perspective Importantly, it is apparent on closer examination that the classification of drugs John G. Moffat, Fabien Vincent, Jonathan A. Lee, Jörg Eder and as ‘phenotypically discovered’ is somewhat Marco Prunotto inconsistent6,7 and that, in fact, the majority of successful drug discovery programmes Abstract | Phenotypic drug discovery (PDD) approaches do not rely on knowledge combine target knowledge and functional of the identity of a specific drug target or a hypothesis about its role in disease, in cellular assays to identify drug candidates contrast to the target-based strategies that have been widely used in the with the most advantageous molecular pharmaceutical industry in the past three decades. However, in recent years, there mechanism of action (MoA). Although there is clear evidence that phenotypic has been a resurgence in interest in PDD approaches based on their potential to screening can be an attractive proposition address the incompletely understood complexity of diseases and their promise for efficiently identifying functionally of delivering first‑in-class drugs, as well as major advances in the tools for active hits that lead to first‑in‑class drugs, cell-based phenotypic screening. Nevertheless, PDD approaches also have the gap between a screening hit and an considerable challenges, such as hit validation and target deconvolution. This efficacious drug is often immense and, in article focuses on the lessons learned by researchers engaged in PDD in the our experience, more challenging than for a hit with a known molecular target. Hopes pharmaceutical industry and considers the impact of ‘omics’ knowledge in for PDD to ‘rescue’ the pharmaceutical defining a cellular disease phenotype in the era of precision medicine, introducing industry might also be viewed as an example the concept of a chain of translatability. We particularly aim to identify features of a Gartner hype cycle, in which a peak of and areas in which PDD can best deliver value to drug discovery portfolios and inflated expectations is followed by a trough can contribute to the identification and the development of novel medicines, of disillusionment, before establishing a plateau of productivity. and to illustrate the challenges and uncertainties that are associated with PDD in This article aims to address two order to help set realistic expectations with regard to its benefits and costs. aspects of this situation: first, to illustrate current challenges and uncertainties that In the past three decades, target-based drug of TDD (for example, see REF. 2), and from are associated with PDD to set realistic discovery (TDD) — in which the starting the authors’ experience and anecdotal expectations for benefits and costs; and point is a defined molecular target that is communications, it seems that efforts within second, to identify areas in which PDD hypothesized to have an important role in the pharmaceutical industry to pursue PDD can best deliver value to drug discovery disease — has been the dominant approach have recently greatly increased compared portfolios through the identification and to drug discovery in the pharmaceutical with the years preceding 2011. The power of the development of novel medicines. In the industry, driven by advances in molecular PDD as a tool to address the complexity past two years, conferences organized by biology and genomics. However, in recent of diseases that are poorly understood by the the New York Academy of Sciences9 and the years, there has been a revival in interest scientific community is also clear (see REF. 3 Keystone Symposia on Molecular in phenotypic drug discovery (PDD) for a recent review). Furthermore, there and Cellular Biology, along with the approaches, which do not rely on knowledge have been recent rapid advances in various establishment of a PDD special interest of the identity of a specific drug target or a technologies for cell-based phenotypic group under the auspices of the Society hypothesis about its role in disease. screening, including the development for Laboratory Automation and Screening, This interest has been catalysed in part of induced pluripotent stem (iPS) cell have sustained discussion around PDD and by an influential analysis by Swinney and technologies4, gene-editing tools such its value for the pharmaceutical industry. Anthony in 2011 that highlighted the strong as CRISPR–Cas5, organoids and imaging This article also aims to capture the spirit contribution of PDD to the discovery of assay technologies. Such tools have enabled of these recent meetings by focusing on first‑in‑class drugs1. PDD has since been the development of novel cell-based disease the state of the art in PDD, sharing lessons cited by various authors as a potential models that promise to more realistically learned, and carefully examining the solution to the perceived poor productivity recapitulate human disease biology. opportunities and challenges for PDD.

PERSPECTIVES

We first highlight core concepts in PDD understanding to a mechanistically defined Rare monogenic diseases can also provide and introduce the concept of a chain of effect on a pathway or a biomarker to drug, an opportunity to establish PDD projects translatability for PDD screens, and then and then to a therapeutic effect. with a strong chain of translatability18,19, discuss strategic considerations and as knowledge of the mutation that causes operational aspects for PDD projects, Chain of translatability. The fundamental the disease can be used as the basis for the including library development, hit triage, determinant of the potential success of a development of disease models that are compound optimization and safety PDD effort is the ability of the screening suitable for PDD. For example, pioneering assessments. assay to predict the clinical therapeutic genome-wide expression analysis in patients response to a drug with a specific mechanism with recessive dystrophic epidermolysis Core concepts in PDD of action. This was described by Scannell bullosa (RDEB), a rare disease that is Defining PDD. Drugs typically act by and Bosley as the “predictive validity” of characterized by fragile skin due to mutations engaging a molecular target; however, a discovery model2. Here, we propose the in the COL7A1 gene, showed that the a priori knowledge of that target is term chain of translatability to describe differential expression of genes associated not essential. In the case of PDD, a the presence of a shared mechanistic basis with the transforming growth factor-β ‘physiologically relevant’ biological for the disease model, the assay readout (TGFβ) pathway was responsible for system or cellular signalling pathway is and the biology of the disease in humans, differences in the clinical manifestation of directly interrogated by chemical matter as a framework for developing phenotypic the disease. Based on these results, and the to identify biologically active compounds. screening assays with a greater likelihood of knowledge that the approved angiotensin This target-agnostic approach is the having strong predictive validity. II receptor antagonist losartan attenuates underlying attribute that differentiates PDD projects in the area of infectious both canonical and non-canonical TGFβ PDD from hypothesis-driven TDD10. These disease (seeking antibiotics12, antivirals13 signalling, Nyström et al.20 demonstrated target-agnostic and empirical aspects of and anti-parasitic agents14) often have that long-term losartan treatment of PDD are consistent with its description a strong chain of translatability. Indeed, a COL7A1‑mutant RDEB mice efficiently and usage by scientists in academia and typical PDD assay readout — inhibiting the reduced TGFβ signalling in chronically industry. Use of a uniform definition for replication of bacteria, viruses or parasites injured forepaws and alleviated hallmarks PDD helps to illuminate the impact of PDD on plates — can strongly correspond of RDEB progression. It is possible that on modern drug discovery1,6, and underlines not only to anti-infective activity in a differential phenotypic screen using the importance and impact of empirical in vivo preclinical models, but also to the wild-type and COL7A1‑deficient fibroblasts drug discovery approaches in an era that pharmacodynamic (PD) and the therapeutic could identify compounds that modulate or is dominated by strategies that are based effects sought in patients. For example, that prevent RDEB disease progression to a on molecular target hypotheses1. Although the anti-hepatitis C virus (HCV) drug greater extent than losartan. As in the case primarily an approach for small-molecule daclatasvir was discovered phenotypically of antibacterial and anti-parasitic drugs, discovery, PDD has also contributed to using human cells engineered to express the the chain of translatability for monogenic antibody drug discovery (see the excellent HCV replicon corresponding to a number diseases may extend not only from effects in review by Gonzalez-Munoz et al.11). of clinically relevant genotypes15. Therefore, a cell-based phenotypic assay to effects Most drug discovery projects that are such a system was strongly predictive for in a preclinical animal model, but also to based on a molecular target hypothesis the inhibition of HCV replication in vivo. therapeutic effects in humans, as the same also test active compounds in phenotypic The molecular target of daclatasvir and of mutation found in humans frequently often cellular assays. Although these are clearly subsequent anti-HCV drugs, the HCV NS5A drives a very comparable phenotype, disease not phenotypic or empirical drug discovery protein, was not previously regarded as a natural history and outcome in animal examples, novel and therapeutically target as it then had no known function. models that have, or that are anticipated to important MoAs that differentiate targeted Similarly, PDD projects that aim to have, good predictive validity. drugs can be discovered phenotypically. modulate the production of proteins with Beyond the strong potential for PDD to An example of this ‘molecularly informed either known human pharmacological contribute substantially to the identification phenotypic discovery’ paradigm3 was the activity (for example, insulin) or a highly of therapeutics for patients with some of empirical observation that the oestrogen validated association with human physiology the 7,000 known rare diseases (many receptor (ER) antagonist fulvestrant displayed (for example, PCSK9 (REF. 16)) can have a of which have no specific treatments), such greater than expected efficacy, leading to the strong chain of translatability. Based on efforts could also have implications for elucidation of its ER‑degrading mechanism. human genetics, there was a very strong broader patient populations. Rare diseases Eder et al.6 noted that many drugs rationale that the reduction of PCSK9 are in many cases highly suitable for the discovered by target-unbiased empirical synthesis would have beneficial effects in evaluation, within a very homogenous pharmacology, which could potentially lowering LDL cholesterol levels. A screen (although small) patient population, of a be called ‘phenotypically discovered’ for agents that reduced PCSK9 production novel MoA that might be relevant in other, drugs, were the outcome of serendipitous in CHO cells17 identified a totally novel more prevalent indications that share key observations or of the individual phenotypic molecular MoA — gene-specific ribosome disease characteristics21. characterization of known active stalling. As the compounds act upon such Unfortunately, there are many pharmacophores, rather than a result of a a fundamental and conserved mechanism, therapeutic indications that have neither disease-first lead discovery effort. Going there is a very high probability that, given a a highly predictive animal model nor a forwards, in order to substantially improve molecule with appropriate pharmaceutical quantifiable phenotypic assay end point industry productivity, we argue that PDD properties, the same molecular MoA would that mechanistically corresponds to a must proceed rationally from disease also be active in vivo. causal disease biomarker. Oncology and

PERSPECTIVES

neuroscience are two major therapeutic Building the chain of translatability models28. Thus, capturing human disease areas in which even a chain of translatability Disease understanding. Knowledge at the relevance is the first step in the creation of from in vitro assays to preclinical models molecular level of the causes and drivers a chain of translatability. Therefore, when is often difficult to establish directly, as of the disease is a crucial success factor for starting a PDD programme, we need to efficacy is a measure of a complex response PDD, as it is needed to select and to validate be aware of the molecular mechanistic at the tissue or the organism level. For the best experimental cellular system and information that is available for the disease example, late-stage cancers are frequently readouts to use (FIG. 1). Experience has that we want to replicate in vitro and highly heterogeneous and adaptively shown that assays with generic readouts we should ideally place that molecular independent of the known initiating (for example, viability or apoptosis of cancer information in the context of clinical data. oncogenic mutants, and neuropsychological cell lines) are often not causally related to Are the translational data sufficient to disorders are often completely lacking in the disease biology pathways that we are identify well-validated molecular functional validated preclinical in vivo models. Here, it attempting to modify, and thus are less end points or predictive biomarkers for is much more difficult to correlate clinical likely to be useful in identifying novel and disease modification that can guide the PDD outcomes to a single molecular target efficacious molecular MoAs. effort? If the same molecular functions and (target validation), cellular readout Although there remain diseases for readouts can be encoded in a phenotypic (in vitro model validation) or animal which the molecular-mechanistic disease screening model, then predictive validity is model (in vivo model validation) in the understanding may be insufficient to enhanced. Two key areas that can help to absence of additional studies. undertake effective PDD, this is changing translate disease information into a screen Although these therapeutic areas are rapidly as, for example, next-generation with strong predictive validity are molecular scientifically challenging and fraught sequencing is contributing growing amounts phenotyping and advanced cellular models. with project-management risk, they of genetic data for many disorders23. include clinical indications that represent For some diseases, especially those that Molecular phenotyping. As we mentioned important unmet medical needs and require biopsies for diagnosis or treatment above, whereas knowledge of the molecular business opportunities for the development follow‑up, there may be a large body of drivers of disease has been central to TDD of first‑in‑class therapeutics. To increase genetic and genomic information related approaches, the potential impact has not the probability of the predictive validity for to the baseline condition and to disease yet been fully realized in the context of models that lack a strong mechanistically progression. This is, for instance, the PDD. A major challenge has been the defined chain of translatability, Vincent case in kidney diseases24,25, for which this substantial inability to translate such et al.8 proposed a set of guiding principles. systems biology information has enabled molecular MoA findings in humans in the The authors identified three key features the identification of new mechanisms context of a disease-relevant cell system (rule of 3) of the phenotypic assay — the and targets through the integration of that is appropriate for high-throughput hit assay system, stimulus and readout — large-scale genetic and molecular data identification. Molecular phenotyping29,30 that may enable the establishment of with deep phenotypic information. This — the ability to run high-throughput such a chain. First, the assay system information has already been successfully transcriptome analysis as a secondary must have a clear link to disease (for translated into the identification of novel or even a primary screen — thus holds example, patient-derived primary cells drugs through TDD approaches, such as promise as a technology to fully leverage or iPS-derived disease-relevant cell the JAK2 inhibitor baricitinib26, which has this molecular information. Several such phenotypes) and aim to replicate relevant reached late-stage clinical development for efforts are ongoing in the computational physiological aspects (for example, 3D rheumatoid arthritis. However, attempts to biology30–32, pharmaceutical33,34 and growth or co‑culture systems). Second, use such knowledge to develop biologically toxicogenomics35,36 fields. These efforts the assay readout should be as proximal informed PDD screens for kidney diseases have the common aim of showing that as possible to the disease pathophysiology have not yet been successful because it has the activity of signalling networks can be and clinical end point, ideally with a not been possible to faithfully capture the assessed based on a set of established key high degree of information. Third, the complex human kidney pathophysiology in regulatory and effector genes. Molecular authors concluded that the stimuli used a suitable cellular assay system. Nevertheless, phenotype gene signatures have been shown to induce a disease-like phenotype, which we are optimistic that these challenges can by several groups31,33,36 to consistently are often too simplistic, may result in the be overcome with the adoption of more deliver an accurate pathway-centric view identification of stimulus- rather than sophisticated emerging models, such of the biological system under study. The disease-modifying compounds. Thus, as organ‑on‑a-chip27. modulation of signalling networks identified systems that do not require an exogenous Incomplete disease understanding is a in this way has been shown to be consistent stimulus to induce the biological limitation for the validation of phenotypic with literature or experimental data assessed phenotype are preferred. Used as intended, models and for hypothesis-driven molecular by different technologies. these guidelines aim to increase both the targets. This is illustrated by the challenges In PDD, this means that we have a biological space captured by the assay and of transgenic mouse models of Alzheimer powerful tool to decode the effect of the likelihood of the translation of the disease, which have been widely used, compounds on regulatory pathways in compounds and mechanism identified to but have been unsuccessful in identifying the context of the cellular model adopted. patients. The excellent recent article by clinically effective therapeutics so far. Pioneering work in this respect by Drawnel, Horvath et al.22 provides further detailed Although the molecular driver may be the Zhang and colleagues was recently descriptions and examples of systems that same in both animal models and humans, published37. The authors were able to show, can enhance and validate the translational the resulting pathogenic mechanism does using a 917 human pathway reporter gene relevance of phenotypic assays. not seem to be fully recapitulated in animal signature31, that molecular phenotyping

PERSPECTIVES

a Disease knowledge integration can cluster compounds based on pathway profiles and can simultaneously dissect Physiology-associated pathways associations between pathway activities and disease phenotypes. Molecular phenotyping Disease-associated pathways was applicable to compounds with a range of binding specificities and allowed Physiology Disease false positives derived from high-content signature signature screening assays to be triaged. The approach was used to identify a class of calcium- Disease-model-associated pathways signalling modulators that reversed disease- regulated pathways and phenotypes, which b Incorporation and assessment of disease relevance in cells was validated by structurally distinct compounds in relevant classes37. Missing disease information A similar approach was recently used Disease Disease to discover leptin sensitizers for diabetes. model ± Stimuli phenotype Here, researchers used profiling information on an active compound coupled with a search of the Broad Institute connectivity map (CMAP) to identify withaferin A as a novel leptin sensitizer38,39. We believe that such an approach, used in synergy with molecular disease information that is 40–42 c PDD compound screening and validation being generated by multiple initiatives that are sequencing DNA from patients with various diseases and phenotypically Screening assay Clinical profiling them (for example, Genomics Molecular In vivo proof-of- England41 and Genome Asia 100K43), may phenotype validation concept become a key enabler of future PDD. The critical value of this approach in establishing MoA1 a chain of translatability is twofold. It first Compound 1 offers an unbiased diagnosis of the similarity between the disease state in humans and MoA2 the molecular state of the discovery model (as shown in FIG. 2), and it also provides an Compound 2 evaluation of the extent to which a potential therapeutic modifies the molecular state towards the therapeutically desired state.

Nonspeciﬁc Advanced cellular models. The cellular screening system is a cornerstone of most Compound 3 successful attempts to identify potential novel drugs. Now, technological advances in cell and molecular biology are enabling | Nature Reviews | Drug Discovery Figure 1 Using a chain of translatability in phenotypic drug discovery. a | The first step in the development of models that are likely establishing a chain of translatability for phenotypic drug discovery (PDD) is identifying a disease- to strengthen the chain of translatability associated molecular characteristic or signature (for example, a disease-associated gene expression profile, as shown, or the presence of a particular mutation in a protein) that differentiates the disease even in model systems that have a reduced state (right) from normal physiology (left). When available human genetic or genomic data are insuf- physiological complexity, by more closely ficient to establish the causal components of this signature, animal models of the same disease can modelling the disease-relevant cell or reconstruct earlier biological processes and associated pathway changes, providing a mechanistic cells and tissue, and/or by focusing on the bridge between alterations to normal physiology and the manifestations of the disease. b | Having molecular and mechanistic phenotype. identified the disease characteristics, cellular models aim to reconstruct a cellular phenotype that is In recent years, a broad arsenal as close as possible to the disease condition; for example, by incorporating a specific mutation in the of advanced cellular models44,45 have cells or deriving cells from patients via induced pluripotent stem (iPS) cell generation. Specific become available as microtechnologies disease-relevant stimuli may be required to model the cellular phenotype that is seen in the disease have progressed: microprinted tumour state. The mechanistic similarity of the model to the clinical disease is determined by a comparison spheroids46,47, ‘tissue-on-a-chip’ (REF. 27), of the molecular phenotype signatures. If the signatures are not sufficiently similar, the model is considered invalid. c | Phenotypic screening is conducted using a cellular model validated by the structured co‑cultures and multicellular 48–52 molecular phenotype. Prioritized hits from the primary screen may reveal different molecular pheno organoids . Each system has its strengths types corresponding to different mechanisms of action (MoAs). Only MoAs that affect disease- and weaknesses reviewed in the references relevant pathways will be evaluated for in vivo proof‑of‑concept. Nonspecific MoAs (represented by provided. Complementary advances in compound 3) can be eliminated using molecular phenotype information prior to advancing to in vivo screening hardware, high-throughput proof-of-concept evaluation. cell assay technologies such as confocal

PERSPECTIVES

a Signalling input/output state points for new drugs, reflecting the relatively short period for which iPS phenotypic Model state Disease state screens have been pursued. To address recurring concerns regarding the maturation state of iPS-derived cells, the use of patient-derived primary cells is a possible alternative. Sourcing such cells is still a rate-limiting step, but it is an area in which public–private partnerships and collaborations between biotech and pharma Assay end points companies and disease foundations, as well as direct collaborations with health centres, Disease end points hold great promise56. The translational value of these systems has been exemplified in the development of ivacaftor by scientists b Extracellular environment 3D co-culture 57 3D culture at Vertex Pharmceuticals , including its label expansion based on an additional clinical trial of patients selected based on 2D culture their genotype following compound testing against a wide range of genotypes encoded in patient-derived primary cells58. Fully Assay end points differentiated patient-derived primary bronchial epithelial cells from healthy Disease end points individuals or patients with cystic fibrosis harbouring the CFTRΔ508 mutation have also Predictive validity recently been used by scientists at Pfizer in a PDD project for cystic fibrosis59. In the cancer drug discovery arena, Figure 2 | Predictive validity of disease models as a function of theNature overlap Reviews between | Drug the Discovery mechanisms that drive assay and disease end points. a | The disease model and the disease are repre- results in the past few years from Cancer sented as hidden state-dependent networks, and assay end points reflect the model state. On the Genome Atlas data have highlighted left-hand side, a disease model that depends on a non-physiological external stimulus is shown, for the disconnections between even the which the assay end points are driven by network nodes that have no overlap with the disease state. best-characterized cancer cell line models Consequently, the assay end points are poorly predictive of a therapeutic effect. In the centre, the and patients60–62. Concurrently, technical assay system state shares some common pathways with the disease, and the effects on some, but not advances in generating patient-derived all, of the assay end points may be predictive of a therapeutic effect. On the right-hand side, the net- tumour models in vivo and in vitro are work of the assay model strongly overlaps with the disease, and the effects on assay end points are having radical effects on cancer drug | highly predictive of a therapeutic effect. b Illustration of how increasing the complexity of the assay research that ought to have a favourable model, from 2D cell culture (left) to a 3D culture (centre) and to a 3D co‑culture system (right) might impact on phenotypic discovery models. increase the overlap between the causal networks for the assay and disease end points, as well as the probability that a given assay readout will be predictive. Two of the best-established patient-derived cancer cell culture systems are probably the glioma-derived neurosphere model high-content imaging systems and other derived, used for PDD in the near future. and colorectal cancer-derived organoids. methods for monitoring cell function At an even higher degree of complexity, Both of these models, like the tumours are enabling more complex assays an iPS cell-based model can be coupled to from which they are derived, have a clear to be implemented for high-throughput a molecular readout such as endogenous stem cell component that makes them lead discovery. gene expression. For example, a screen amenable to genome editing63 and scalable In addition to models that address the carried out by Lee et al.55 assessed the ability for high-throughput screening64,65. The anatomical complexities of in vivo tissue of more than 6,000 molecules to restore genetic diversity of tumours is thus not only structures, PDD efforts can leverage the the expression of IKK complex-associated captured but can also be made available predictive potential of iPS cell-based protein (IKBAP), evaluated by real-time through biobanks (for example, see REF. 66). models53,54, which promise to replicate a quantitative reverse-transcriptase PCR However, powerful mechanistically disease in a dish (comprehensively reviewed in cells derived from patients suffering informed cellular models do not always by Avior et al.54). Models using iPS cells from familial dysautonomia, who carry a demand the use of a complex cellular are especially powerful when studying hypomorphic mutation in the IKBAP gene. system — the most important factor is that high-penetrance monogenic disorders Nevertheless, even in the most advanced the molecular mechanism of the disease is that are associated with a clear cellular areas to make use of iPS-derived cell models reproduced in the observed phenotype of phenotype. Breakthroughs in gene-editing — cardiac and neurological disorders — the discovery model (FIG. 2). In a striking technologies such as CRISPR–Cas are also most studies have so far been aimed at example of a ‘rule of 3 breaker’, investigators likely to greatly increase the number and validating the effects of existing drugs in from Roche and Novartis independently the diversity of genetically defined cellular a repurposing effort (see Table 3 in Avior discovered small molecules that correct models4,5, both conventional and iPS et al.54) rather than identifying starting aberrant alternate splicing of the mRNA

PERSPECTIVES

that codes for survival motor neuron 1 expression signatures (Zoffmann, S. et al., paradigm of the past two decades. Without (SMN1), which is the root cause of the unpublished observations; presented at a high-confidence chain of translatability, rare neurodegenerative disease, spinal the 2016 Keystone symposium ‘Modern the risk of clinical failure of PDD projects muscular atrophy (SMA). Importantly, Phenotypic Drug Discovery: Defining the owing to a lack of efficacy is probably both molecules were optimized without Path Forward’). Both approaches indicated similar to that of a TDD project for a poorly prior knowledge of their molecular target the ability to identify cellular pathways that validated target. by screening using generic cell lines that predicted the MoAs of known antibacterial Building a rational, sustainable discovery expressed reporter gene constructs designed molecules. The team at Roche was able to pipeline around PDD presents considerable to detect the alternative splicing of SMN1 further translate the results from RNA-seq challenges. The early (preclinical) stages (REFS 29,67). This was achieved despite a to a higher-throughput bacteria reporter of PDD programmes — from lead-finding widely held consensus that reporter gene strain-based signature, and also reported screen to clinical candidate — tend to assays in engineered cell systems have low that the signature allowed the identification require substantially greater resources than disease relevance and have arguably led to of novel hits with the desired molecular TDD. This is mostly due to the development only a single marketed drug (vismodegib)6, MoA, thus providing novel chemical of higher-complexity screening assays, presumably due to the smaller biological starting points. and sometimes the concomitant use of space that they probe (that is, direct To summarize, we believe that the ability small-molecule and genetic screens, as well transcriptional regulation) and high technical to capture a disease-relevant molecular MoA as more challenging hit validation and target false-positive rates68. The cellular system in in the screening system is a key enabling identification efforts for compound series both cases was very simple but the proximity feature of PDD. Disease relevance can of interest. However, the potential return on to the disease phenotype (that is, the ability to potentially be encoded in a simple cellular this greater investment stems from sampling assess aberrant alternative splicing of a single system, and novel chemical space may be a larger volume of potential target space, gene) unparalleled. This resulted in an assay explored by adopting innovative readouts which may enable the discovery of either that was capable of guiding structure– for which a chain of translatability has been novel targets or unrecognized molecular activity relationship (SAR) studies, as well as established. In the absence of the three MoAs. In light of this potential, it has been producing molecules shown to be effective core elements mentioned above — disease argued that the cost–benefit ratio for a single in disease models29,67, which are currently in knowledge, replication and/or monitoring of PDD project might be more comparable to late-stage clinical development for SMA. the molecular MoA in vitro, and availability several hypothesis-driven TDD projects for Recent antibiotic discovery efforts have of a suitable cellular system8 — the a given disorder70. also provided instructive examples of the probability of a successful PDD programme The perceived risk of advancing application of novel readout technologies is greatly reduced. a compound with an incomplete to a simple ‘classical’ assay system to understanding of its MoA or without drive the exploration of new regions of Strategic considerations for PDD identifying its molecular target varies, chemical space. As discussed above, simple A decision to pursue PDD versus depending on the disease area (including assays based on killing bacteria in vitro TDD strategies requires a multitude of the landscape of existing therapies), the have a strong alignment to the desired scientific, strategic and managerial factors existence of predictive and prognostic pharmacodynamic effect in animal models to be considered. Below, we attempt to clinical biomarkers, safety concerns (see and so are likely to identify hits that can summarize the risks, costs and potential below) and the organizational strategy. In be optimized into leads that show in vivo rewards of embarking on a PDD programme, the absence of a target, the accumulation efficacy. The hurdle faced by antibiotic based on our experience and that of others. of mechanistic information (for example, developers is that such chemical hits are The promised payoffs of PDD include identifying relevant signalling pathways frequently rediscoveries of known chemical the identification of compounds that are or ruling out undesirable MoAs) can help scaffolds and thus are not appropriate for more likely to translate to in vivo and to mitigate safety concerns. The most the development of drugs in novel classes, clinical efficacy studies than TDD-derived compelling argument in favour of advancing which are highly desirable in order to compounds, or to be effective in a disease a PDD programme is the activity of the overcome resistance to antibiotics in existing for which TDD cannot be applied (that is, compound in a disease-relevant in vitro classes. The challenge of identifying novel a disease for which a validated molecular assay and animal disease model. As target chemical starting points can be addressed target is lacking), and/or the identification identification is not always achievable71, the by using discovery models that directly of compounds that act through novel team and management must decide early focus on a molecular phenotype. Two recent mechanisms, potentially increasing the on whether they are prepared to advance a studies have illustrated this concept. Both chances of differentiation from competitor drug candidate into preclinical or clinical groups took the approach of characterizing compounds and existing standard development in the absence of knowledge signatures predictive of the MoA of available treatments. Discovery of a novel target or of the identity of its target. The cost–benefit antibiotics, and searched for novel hits that molecular MoA provides an opportunity to considerations of proceeding in the absence had a similar mechanistic profile but with an develop a first‑in‑class medicine, which is of target knowledge are very different insufficient potency to be found in a classical a most desirable outcome for both patients depending on the indication, the strength of functional antibacterial screen. Nonejuie and an industry facing regulatory and the chain of translatability, the presence et al.69 adopted high-content screening business headwinds for late entrants in an of a causally related predictive biomarker, (HCS) to measure a large number of cellular established target class. the unmet medical need, the competitive features, whereas a second group at Roche PDD is not an easy path or a magic landscape and the risk-tolerance of the extended this idea to the molecular level by bullet, and there are sound reasons why organization. However, the recurring using RNA-seq to identify drug-specific gene TDD has been the dominant discovery myth that the definition of a molecular

PERSPECTIVES

target is required for regulatory approval a phenotypic screen for inhibitors of make up more than 25% of the new chemical should be dispelled (see US Food and PCSK9 secretion delivered an unexpected entities that have been approved since 2010, Drug Administration (FDA) guidance72): molecular MoA in the form of a small an era in which natural products were not the a substantial number of FDA-approved molecule that selectively that stalls PCSK9 focus of pharmaceutical research89. Interest drugs (estimates range from 7%73 to 18%74) protein translation through an mRNA in natural product screening is experiencing lack a defined molecular target. sequence-dependent interaction with the a renaissance (comprehensively reviewed It is generally agreed that the probability 80S ribosome17. Realistically, this molecular by Harvey et al.90) that is contemporaneous of technical success at the stage of the MoA would not have been pursued in a with that of PDD and that is being driven by primary screen or of early discovery traditional TDD context and represents analogous reasons: technological advances programmes is lower for PDD than for a clear example of how phenotypic strategies and the depletion of the ‘low-hanging a typical TDD effort. Technical risks, which can expand our drug discovery horizon. fruit’ in terms of validated and druggable are discussed below, include challenging Taken together, these various target classes. A recent demonstration of assay development, high false-positive considerations, including target validation, the value of a natural-product-inspired hit rate, inability to establish SAR from risk tolerance, breakthrough innovation diverse compound library as a substrate the phenotypic assay, failure to generate versus incremental advancement, cost for phenotypic screening was the discovery a molecule that is suitable for in vivo proof- and competitive status, define a set of of a novel anti-malarial compound that is of-concept validation or the inability to interconnected variables framing a complex efficacious in vivo, works on multiple parasite identify the target of a compound series of decision-making landscape. life stages and has a novel MoA91. interest. This reflects a front-loading Although there are many important drug of risk and a high bar to critical-path Operational aspects of PDD targets that do not require cell permeability, decision-making assay data early in Library selection. The strategic objectives such as receptors, ion channels and secreted the project. Moreover, because PDD of a PDD project should guide not only the enzymes, a diverse phenotypic screening emphasizes biological function it is likely selection of an appropriate screening model, library requires a high probability of cellular that a PDD team will place high priority as discussed and defined above, but also a permeability. Therefore, compounds with on early in vivo activity confirmation (with decision on the chemical matter that is used properties that are incompatible with associated ADME (absorption, distribution, for testing. Although not in the scope of this, such as peptidic compounds, higher metabolism and excretion) and toxicology this article, phenotypic screens of relatively molecular mass (>500–600 Da) and charged studies to deem the molecule safe) following small well-annotated tool compound groups, although capable of providing the identification of compounds with very collections are widely used to identify optimizable lead matter in a biochemical promising in vitro profiles. novel biology for molecular targets that TDD screen, should be deprioritized Conversely, the most critical risk of belong to known drug target classes, and to for inclusion. There are also growing the complementary TDD strategy for explore drug-repurposing opportunities. efforts to define functional diversity using first‑in‑class drugs is target validation. Robust Several recent articles have described the high-throughput approaches to determine target validation is a serious challenge75,76, characterization and applications of such bioactivity signatures. For example, scientists with poor linkage of a molecular target to the ‘chemogenomics’ libraries, with particular from the Broad Institute used a ‘cell painting’ disease biology underlying a given indication regard to understanding the specificity of the method to profile a compound library and to estimated to contribute substantially to the mechanism of action and the curation prioritize compounds that displayed diverse ~50% of clinical failures that are caused by a of nominal target identity83–85. The Chemical bioactivity profiles92. Molecular phenotyping lack of efficacy77. Probes Portal acts as a valuable public approaches, such as the Roche reporter Novelty of target and MoA is the second repository of such information. gene31 and the National Institutes of major potential advantage of PDD. In By contrast, if compounds with novel Health/Broad Institute LINCS (Library addition to identifying novel targets, PDD targets are the goal, then diverse collections of Integrated Network-based Cellular can contribute to improvements over of novel compounds should be screened, Signatures) L1000 panels93,94, are also existing therapies by identifying novel given that an analysis by Santos et al.86 potentially valuable in this regard. Given physiology for a known target, exploring estimated that the molecular targets of a library with a broad range of historical ‘undrugged’ targets that belong to well known drugs and existing tool compounds screening data, ‘biological signatures’ of known drug target classes or discovering constitute only 3% and 6% of the human compounds can be used to select for diversity, novel MoAs, including new ways of proteome, respectively. Such compounds as well as to potentially aid in classifying hit interfering with difficult-to‑drug targets. For could be found in so‑called ‘dark chemical MoAs85. However, prioritizing compounds example, the discovery of fingolimod78,79 led matter’ (REF. 87) — molecules that have been for inclusion in a screening library based on to the recognition of the role of sphingosine frequently screened in TDD projects but that previously observed biological effects is to phosphate G protein-coupled receptors in have not shown any activity. Alternatively, some extent counter to the strategic goal of lymphocyte egress from the thymus, and natural products and their derivatives discovering molecules that have a selective resulted in its approval as a first‑in‑class occupy an orthogonal chemical space to the effect on a disease-specific phenotype medication for multiple sclerosis80. majority of synthetic compounds and are and model with no activity on ‘normal’ Similarly, use of the well-validated maximal frequently considered ‘privileged’, in that cellular physiology, so should not be applied electroshock (MES) model for epilepsy the structures of biologically relevant small too stringently. In fact, the exclusion of enabled the discovery of lacosamide, molecules may be selected by evolution compounds based on promiscuity may which promotes the slow inactivation of to engage protein-binding sites88 and thus be more important than the inclusion of voltage-gated sodium channels rather than may provide valuable chemical diversity for nominally bioactive compounds, based directly blocking them81,82. More recently, PDD. Natural products and their derivatives on the demonstration that the absence

PERSPECTIVES

of a history of hits in TDD projects (dark selective functionally active hits97. An to 10,000 or more compounds can be chemical matter) does not preclude future important factor in making that particular generated. Factors that contribute to the activity against a novel target87. fragment-based phenotypic screen useful and high confirmed hit rates for PDD include As well as being able to access the overcoming the inherent reduced binding the larger molecular target space that is relevant compartment in the body, in most specificity of smaller fragments was that each interrogated, and the potential to generate cases, a useful PDD screening hit needs to compound was also functionalized with a false positives through non-selective selectively bind to and modulate the function photoactivatable crosslinking moiety and an mechanisms. Therefore, rigorous of a macromolecular target, whether it is an affinity tag, and thus any functional hits could follow‑up assays to identify false positives enzyme active site, an allosteric regulatory site, be rapidly profiled for promiscuity of binding are crucial. In addition to assays that or an intra-molecular or an inter-molecular and target identity. monitor cytotoxicity, pleiotropic biological interaction. Therefore, it can be argued that Regardless of the chemical diversity used mechanisms that may mimic efficacy in the key factors in molecular features and in a PDD screen, the degree of compromise the primary assay (for example, general properties of PDD screening collections between throughput and assay complexity secretion inhibition when monitoring levels compared with TDD screening collections are continues to be a challenge. In addition, of a secreted protein) must be anticipated to place a premium on cellular permeability screening a reference library of compounds and the appropriate counter screens are and to have sufficient structural complexity with known targets and MoAs (that is, crucial. In our experience, primary hits from to confer some level of selectivity and a chemogenomics library85) alongside a PDD are rarely target-specific hits. Following low-micromolar affinity on molecular chemically diverse set is a stronglyrecom- the confirmation of a PDD screening hit, interactions. Although these requirements for mended step that will inform the design it is thus important to establish SAR with a selectivity and affinity usually argue against of an optimal hit triage strategy and will series of analogues of the same chemotype the utility of low-molecular-mass fragment facilitate hit MoA and target deconvolution. and to search for correlations, or lack compounds in a PDD screening library, there thereof, with the desired (therapeutic) effect may be exceptions, such as the use of broadly Hit triage. The selection of hits from in order to distinguish target- or pathway- targeted covalent chemical probes for the phenotypic screens for further optimization specific effects from undesired nonspecific identification of targets of certain enzyme is often considerably more complex than effects. If a sufficient cluster of compounds families, such as hydrolases and proteases95,96. hit triaging from a target-based screen for a specific chemotype is not present in A recent paper from the Cravatt laboratory (TABLE 1). Hit rates for phenotypic screens the compound collection, a substantial also demonstrated that low-molecular-mass of >1% are not uncommon (for example, investment in synthetic chemistry will (~250 Da) fragment-like libraries can yield see REFS 67,98), and thus hit lists of 1,000 potentially be required before a chemical series can be fully validated or invalidated. In the case of TDD screens, structurally Table 1 | Comparison of priorities for phenotypic and target-based drug discovery diverse hits often bind to the same site on Phenotypic drug discovery Target-based drug discovery the target, and thus shared pharmacophore Hit triage Counter-screen to remove technical Counter-screen to remove technical features can increase hit confidence, and goals and false positives false positives elements of different pharmacophores may priorities Extensive counter-screening to address Filters for binding, potency, selectivity eventually be combined. However, each undesirable biological mechanisms is and novelty are negotiable depending chemotype arising from a phenotypic screen essential on strategy must be regarded as a standalone starting Cluster hits based on chemical Cluster hits based on chemical point, with a potentially distinct mode of structure, mechanisms of action and structure action and target. Although this is a potential molecular signatures advantage in the long run, it greatly increases – Confirm cellular target engagement the complexity of the early stages of a and modulation of desired phenotypic project. Not every mode of action is suited biology for therapeutic intervention, as some may Recommendation: exclude hits not Sub-optimal profiles can be rescued evoke the desired therapeutic effect while displaying the full phenotypic profile and low-affinity hits can be pursued causing unacceptable side effects that are Lead Potential for different targets and Possible to combine different related to the same mechanism. In addition, optimization mechanisms of action between series pharmacophores based on structural it may be important to distinguish hits with goals and understanding of binding and to a single mode of action from those with priorities evaluate SAR for different properties independently polypharmacology, in which the therapeutic effect is based on the synergistic or the SAR for cellular activity can be – confounded by compound properties additive interaction with multiple targets and off-target pharmacology and pathways99. Although drugs based on 6 Recommendation: molecular profiling – polypharmacology can be effective , such hits to ensure mechanism of action stays may not offer a tractable starting point for a the same, and to start to define target-agnostic lead optimization effort. biological mechanisms Prior knowledge of the activities and Recommendation: prioritize In vivo proof‑of‑concept timing the target classes of chemotypes identified early optimization for in vivo depends on target or mechanistic as hits might help point to targets or target proof‑of‑concept hypothesis novelty pathways and may help to prioritize the hit SAR, structure–activity relationship. list. In this way, the use of well-annotated

PERSPECTIVES

compound collections85 for screening may the majority of resources, but very rarely end from that of the original hit. Therefore, simplify hit validation and prioritization. up being productive. A key lesson over the the application of data-rich mechanistic However, as previously mentioned, a past few years has been that the follow‑up profiling assays, such as a transcriptional reliance on such compound collections will of such suboptimal hits should be avoided, molecular phenotype, is as important for lead probably bias the phenotypic screen towards even if it means stopping the project in the optimization as it is for hit triage. known biology and targets. Alternatively, absence of better matching hits. Another The development of SARs and their unbiased high-content and multi-parameter important success factor is the availability of use to drive improvement in a compound readouts of either the phenotypic screen a translational in vivo model to enable the series is also considerably more complex itself or of the subsequent follow‑up assays accelerated testing of prioritized PDD-derived for phenotypic assays, which are at least (for example, morphological, gene signature lead structures for, first, validating the cell-based, if not tissue-based or whole- and proteomic changes) can greatly facilitate pre-defined in vitro profile, second, enabling organism-based. As a result, additional the clustering and evaluation of hits100–102. the final selection of chemotypes for further variables must be accounted for during However, even with all readout and analytical drug discovery investment, and third, compound design and SAR data tools in place, hit validation and the allowing their efficient optimization. interpretation, such as cellular permeability, identification of relevant signalling pathways metabolic stability, potential polypharma- and molecular targets of phenotypic Compound optimization. In some cases, cology and nonspecific binding to serum screening hits is often a lengthy and complex phenotypic screening hits may provide proteins. Although highly disease-relevant endeavour. Instructive examples for access to several efficacious mechanisms. assays are often more difficult to prosecute, successful hit list triaging are the discovery This allows project teams to evaluate it is crucial that they deliver robust and of an inhibitor of the HCV NS5A protein15 multiple therapeutic mechanisms with reproducible data in order to successfully and Porcupine inhibitors modulating different efficacy and safety profiles, to select enable SAR determination (for example, a oncogenic WNT pathway signalling103. In the best fit for a desired indication (FIG. 3). So, useful rule of thumb is that an assay must both cases, the primary hits were tested for there is the distinct possibility that a team reproducibly detect a threefold difference in specificity towards the inhibition of related may be optimizing several different chemical compound potency to be useful). However, biological systems — the replication of series with different targets. it is important to remember that medicinal other flaviviruses and non-flaviviruses and One important risk when expanding and chemists have proved capable of delivering Hedgehog pathway stimulation, respectively optimizing a compound series in the absence clinical candidates in the absence of target — as well as for overt cytotoxicity. of a direct target engagement assay (whether binding assays and structure-based design81,104. In our experience, if hits that only partially biochemical or cellular) is that new analogues Three key parameters have been match the desired phenotypic profile are not may alter their specificity profile or even proposed to be essential to confidently stringently de‑selected, then they can absorb act through a different target or mechanism test a mechanistic hypothesis in the clinic: compound exposure at the site of action, target binding and expression of functional Target-based screen Phenotypic screen pharmacological activity105. As target engagement information may be out of Target A Target A Target A reach for a PDD programme, the third Target B Target C (MMoA 1) (MMoA 1) (MMoA 2) criterion will have an increased importance in ensuring that the therapeutic hypothesis is effectively tested in humans. Biomarkers Disease model of either compound pharmacology or of the disease itself will thus be crucial for Preclinical safety programme progression70. It is important to note that, as ex vivo assays using Therapeutic patient-derived cells and tissues become proﬁle more prevalent, the need for animal models of disease to provide confidence Magnitude of eﬃcacy Known target–MMoA combination in translation to patients may become less Magnitude of safety liabilities Unknown target–MMoA combination acute. This may represent a considerable benefit for those disorders with poorly 2,106 Nature Reviews | Drug Discovery predictive models . However, the Figure 3 | Leads with diverse mechanisms, efficacy and safety profiles can be derived from a utility of testing compounds in other phenotypic screen. Phenotypic drug discovery approaches will sample a wider range of potential species for safety, pharmacokinetic and therapeutic mechanisms than target-based approaches. With each mechanism possessing a unique pharmacodynamic purposes will remain. efficacy versus safety profile, phenotypic screening offers an opportunity to identify and to evaluate multiple therapeutic options, including those based on previously unknown biology. Thus, for a target- Safety lessons. As PDD inherently identifies based screen (left), the on‑target safety liabilities of the lead compounds (usually evaluated in separate preclinical safety assays) will probably be the same for all lead compounds, as will the magnitude of compounds with unknown targets, there their effect in a disease model. By contrast, a hypothetical phenotypic screen (right) yields compounds are accompanying risks of hits engaging a acting through three different targets, and for one of those targets, there are two distinct molecular target or of having an MoA with significant mechanisms of action (MMoAs), one of which is the known target–MMoA combination shown for the safety liabilities. The safety de‑risking of target-based screen on the left-hand side. Each different target has its own distinct balance of efficacy target-based drug discovery programmes and intrinsic safety liabilities. is based on multiple types of information

PERSPECTIVES

available to researchers, as well as unbiased address specific safety issues early on (for the risk equation for PDD. However, the in vitro and in vivo experimental safety example, detailed cardiovascular studies resource costs of TID and the risk of failing studies (BOX 1). Target-related knowledge, for targets that are expressed in heart tissue to identify the causal molecular target are including known biological roles, expression or the analysis of infection risk for targets still major potential roadblocks, depending pattern, target-deficient or mutant organisms that are important for the adaptive or innate on the discovery strategy. and the existence of closely related proteins, immune system). Overall, the main added At a panel discussion on TID and constitutes a substantial portion of this risk in the absence of a target hypothesis is associated risks at the 2016 Keystone information. If a target or MoA can be the delayed identification of an unacceptable symposium ‘Modern Phenotypic directly identified during a phenotypic screen safety profile and the resulting loss of Drug Discovery: Defining the Path by using an annotated small-molecule library the investment made thus far in a given Forward’, in which both biotech and or a genetic screening approach, the resulting compound series. large pharmaceutical companies were hits can be de‑risked through this general Accordingly, phenotypic projects may represented, there was a general agreement approach. However, phenotypic screens will benefit from earlier and larger investment from both groups that the absence of target usually identify compounds with unknown in safety experiments compared with knowledge is not necessarily a no‑go gate mechanisms, necessitating a different set of target-based programmes. Most, but probably for the hit‑to‑lead activities, as long as SARs safety considerations. not all, of these undesirable mechanisms can are built on a robust phenotypic assay and An additional, underappreciated be rapidly eliminated during the hit triage a highly relevant readout. However, the distinction exists between the efficacious through the thoughtful use of biological discussion did reveal some key institutional mechanisms that are obtained through a counter-screens. Additionally, many of the differences. While for some companies, phenotypic approach and those derived from technological advances previously described TID is a mandatory requirement before known targets in terms of safety prospects. for assay development also enable predictive investing substantial chemistry resources For target-based programmes, scientists can toxicology. These include the application and advancing to lead identification, other first evaluate whether the prospective balance of transcriptomic and proteomic profiling, companies are prepared to conduct SAR of the efficacy and the safety potential is and the use of iPSC-derived models for and TID efforts in parallel, with a view to attractive enough to warrant inclusion in a cardiotoxicity107 and hepatoxicity108 assays. obtaining the clinical candidate, as well project portfolio. Although not fool-proof, Finally, compound-derived versus mecha- as the molecular MoA, before clinical this target-based evaluation is likely to nism-related toxicity can still potentially be studies. An outstanding example of this remove many mechanisms with poor safety resolved by conducting experiments with mindset is the development by Novartis prognostics from consideration. Without pairs of closely related active and inactive of a drug for SMA that corrects an SMN1 this level of review, it can be expected compounds81, similar to what is done for pre-mRNA splicing defect (mentioned that a greater proportion of phenotypic target validation109. Another strategy is, of above), for which there was a willingness to screening-derived mechanisms, although course, to identify the target before large-scale initiate human studies even in the absence efficacious, will have significant safety preclinical investment or clinical entry. of a known target (although the MoA was liabilities (for example, target B in FIG. 3). eventually uncovered before the initiation Two potential issues may further After the screen: is target identification of clinical trials)29. In this instance, the complicate safety de‑risking of a compound essential? PDD and target deconvolution risk tolerance may be attributable to the series in the absence of a molecular target. are inextricably intertwined, both in the combination of an acute unmet medical First, target biology may substantially differ strategic considerations of whether to pursue need coupled with high confidence in the between rodents and humans. An in vivo PDD, and in its practice. Advances in target translatability of the molecular MoA to safety study in a second species, as required identification (TID) technologies, which have clinical efficacy in a genetically defined by the FDA before human testing, can help been extensively reviewed elsewhere110–115 patient population. In other words, all the to address this concern. Second, the lack and which are likely to be further advanced key elements of the chain of translatability of target information leads to a diminished by genomics approaches116, may decrease were in place. ability to design bespoke experiments that the target identification hurdle and change In general, the consensus from the Keystone panel discussion was that smaller biotech companies are prepared Box 1 | Safety aspects of phenotypic drug discovery projects to progress towards clinical trials without Different sources of information inform preclinical and clinical safety de‑risking plans. TID, while following preclinical regulatory safety guidelines. TID is more likely to be On‑target safety risks (assuming the target has been identified) perceived by those in biotech companies • Target expression pattern, known and hypothesized biological functions as diverting limited resources away from • Phenotype of target deficient-model organism the delivery of a clinical asset for which the • Phenotype of human target-specific mutations degree of resource investment needed to Off-target safety risks gain knowledge that is actionable in terms of • Selectivity data against closely related proteins (expression pattern and biological functions) clinical development can be daunting, with 70 • Compound structure (known toxicophores of parent molecule and metabolites, and predicted an uncertain outcome . Conversely, most polypharmacology) large pharmaceutical companies strongly • Compound promiscuity (general and protein family assay panels) emphasize TID as a crucial component of project progression and prioritization as • In vitro safety assays (cytotoxicity and undesirable mechanisms) part of a portfolio that also contains targeted • In vivo safety experiments (unexpected findings) drugs. In this business environment,

PERSPECTIVES

the failure to assign an MoA and/or a the context of clinical development than cellular systems mirroring the in vivo state, target to a new molecular entity (NME) is in the early drug discovery process. This ever-more informative high-content imaging frequently considered a major risk factor would help to explain why successful TID is technologies and sophisticated sequencing- for clinical development and regulatory more likely to be considered important by based methods to investigate disease approval, including the safety challenges large pharmaceutical companies carrying relevance in multi-well plate assays. We discussed above. out phase II/III studies towards regulatory have also highlighted that incorporating the The decisions around TID also have approval than smaller biotech companies precise molecular MoA — even in a simple strategic relevance in the context of the that are more usually focused on reaching cell system, as in the case of SMA mentioned competitive environment. For a disease a phase I/II milestone. above — is at the heart of any successful PDD area with several NMEs in development Another key value of TID to be approach. In summary, we believe that the or with an already available standard- considered is whether it can prompt ability to capture disease-relevant regulatory of-care treatment, can the patients who the inception of a TDD programme. pathways in the screening system is a key will maximally benefit from the NME be For instance, the anti-epileptic drug enabling feature of PDD, and that exploration identified in the absence of a mechanism levetiracetam, which was identified in of chemical space can be carried out through and, therefore, a proximal PD marker? a target-agnostic model (audiogenic seizure- the adoption of innovative readouts, as long Once the NME is approved, in the susceptible mice119) was the precursor as a chain of translatability exists. context of life cycle management, how to several rationally designed follow‑on Rare diseases may represent a ‘sweet can the market be expanded towards new drugs119. These follow‑on drugs were spot’ for the application of PDD in academic indications in the absence of a target or discovered following an impressive TID settings. In addition to a strong chain of MoA? Although there are examples of the effort conducted by scientists from UCB translatability, other elements that are empirical discovery of new therapeutic Pharma, who unravelled the ubiquitous important for success are in place: access to uses of drugs (for example, topiramate117 synaptic vesicle glycoprotein SV2A as the funding provided by governments, charities was initially approved for epilepsy and was target of levetiracetam120. and patient associations121, deep disease later approved for migraine prophylaxis118), molecular knowledge produced by highly this is a valid concern. As highlighted Overall conclusions engaged physician-scientist-led above, TID may have a greater impact in Rather than being viewed as opposing groups, and the ability to generate alternatives in novel drug discovery, PDD iPS-derived and/or gene-edited cell models and TDD should be seen as complementary that can be readily screened in publicly Glossary approaches that can together increase the funded high-quality screening centres. We Chain of translatability odds of discovering and developing drugs have also tried to convey our experiences A molecular-level association between the mechanisms with novel efficacious molecular MoAs. and perspectives of the challenges of that drive the assay phenotype, the preclinical disease model and the human disease. The unique promise of PDD is its ability integrating PDD into organizations with a to exploit a disease phenotype to discover strong target-centric perspective and history. Molecular phenotype novel treatments for diseases for which From the outset, managers and scientists Gene-level and pathway-level ‘omics’ signatures shared by the root cause is unknown, complex or need to understand and accept the different disease model and disease state that correspond to and are predictive of disease state versus normal state. multifactorial, and for which scientific risk/benefit considerations associated with understanding is insufficient to provide valid PDD compared with classical target-based Organoids molecular targets. However, PDD should approaches. PDD is highly likely to In vitro 3D cellular clusters derived from primary tissue or not be regarded simply as an alternative involve greater early-stage resource stem cells that show similar characteristics to the tissue of screening technology or as an easy fix to requirements, greater uncertainty and origin; for example, beating cardiomyocytes. the challenges of clinical attrition rates or to entail longer timelines than TDD. PCSK9 R&D productivity. We referred above to Furthermore, the critical path for project (Proprotein convertase subtilisin/kexin type 9). A secreted PDD as being at risk of undergoing a hype progression and the criteria for validation protein mainly expressed in the liver. Studies of naturally cycle. It is our intention to constructively of chemical matter are more fluid and occurring human genetic variants in PCSK9 provided strong evidence that PCSK9 inhibitors could reduce minimize overly optimistic expectations for empirical. Decision gates are likely to plasma levels of low-density lipoprotein cholesterol and ‘quick wins’ from PDD, but also to provide be different: relevance to the disease reduce cardiovascular risk. advice and encouragement to ameliorate mechanism is crucial, potency is secondary the potential for a trough of disillusionment and advancing to in vivo proof-of-concept Pharmacophores that may arise when organizations are not efficacy studies is a crucial go/no‑go gate Groups of molecular features that mediate interactions between a compound and a particular biological frequently rewarded with first-in-class or to be reached as soon as possible. Based on target macromolecule and trigger (or block) its best‑in-class drugs or even with tractable the input of many practitioners who have biological response. leads from phenotypic screens. shared their experiences, we recommend We have reviewed the relevance of the only undertaking a PDD effort if the RNA-seq Uses rapid sequencing technologies to identify the chain of translatability, a continuum that disease-relevant molecular MoA is well presence and quantity of RNAs in a biological sample links the human disease biology at one end, understood and/or when it is possible to at a given moment in time. through the phenotypic system used for sufficiently establish a chain of translatability compound screening, to therapeutic activity. in a screening model. We surmise that not Rule of 3 Enabling tools and technologies for rational, meeting these conditions is the main reason Three technical factors that influence the probability that a phenotypic assay will identify relevant molecules that mechanistically informed phenotypic assays for the failure of PDD projects. Similarly, we affect relevant disease mechanisms: biological system, are rapidly advancing at all levels: genetically suggest that if a hit is non-selective in terms stimulus and readout. defined cellular models of disease, complex of the desired phenotypic profile, it almost

PERSPECTIVES

certainly cannot be optimized to be selective 5. Fellmann, C., Gowen, B. G., Lin, P. C., Doudna, J. A. & 31. Zhang, J. D., Kung, E., Boess, F., Certa, U. & Corn, J. E. Cornerstones of CRISPR-Cas in drug Ebeling, M. Pathway reporter genes define molecular — an important difference from most TDD discovery and therapy. Nat. Rev. Drug Discov. 16, phenotypes of human cells. BMC Genomics 16, 342 programmes. 89–100 (2017). (2015). 6. Eder, J., Sedrani, R. & Wiesmann, C. The discovery of 32. Zhang, J. D., Schindler, T., Kung, E., Ebeling, M. & A crucial strategic consideration is how first‑in‑class drugs: origins and evolution. Nat. Rev. Certa, U. Highly sensitive amplicon-based transcript strongly to tie project advancement, and Drug Discov. 13, 577–587 (2014). quantification by semiconductor sequencing. BMC 7. Moffat, J. G., Rudolph, J. & Bailey, D. Phenotypic Genomics 15, 565 (2014). perhaps even clinical development, to TID. screening in cancer drug discovery — past, present 33. Zhang, J. D., Berntenis, N., Roth, A. & Ebeling, M. Novelty, unmet medical need, compelling and future. Nat. Rev. Drug Discov. 13, 588–602 Data mining reveals a network of early-response genes (2014). as a consensus signature of drug-induced in vitro and in vivo biology and expected safety margin 8. Vincent, F. et al. Developing predictive assays: the in vivo toxicity. Pharmacogenomics J. 14, 208–216 have triggered clinical decisions in the phenotypic screening “rule of 3”. Sci. Transl Med. 7, (2014). 293ps15 (2015). 34. Moisan, A. et al. White‑to‑brown metabolic conversion absence of a precise molecular target 9. Mullard, A. The phenotypic screening pendulum of human adipocytes by JAK inhibition. Nat. Cell Biol. or molecular MoA for drugs that are swings. Nat. Rev. Drug Discov. 14, 807–809 (2015). 17, 57–67 (2015). 10. Lee, J. A. & Berg, E. L. Neoclassic drug discovery: 35. Nuwaysir, E. F., Bittner, M., Trent, J., Barrett, J. C. & currently on the market or in late-stage the case for lead generation using phenotypic and Afshari, C. A. Microarrays and toxicology: the advent clinical trials, including pirfenidone122, functional approaches. J. Biomol. Screen. 18, of toxicogenomics. Mol. Carcinogen. 24, 153–159 (1999). 123 1143–1155 (2013). its direct competitor nintedanib , and 11. Gonzalez-Munoz, A. L., Minter, R. R. & Rust, S. J. 36. Merrick, B. A., Paules, R. S. & Tice, R. R. Intersection thalidomide and its analogues lenalidomide Phenotypic screening: the future of antibody discovery. of toxicogenomics and high throughput screening in Drug Discov. Today 21, 150–156 (2016). the Tox21 program: an NIEHS perspective. and pomalidomide. Guidance from such 12. Andries, K. et al. A diarylquinoline drug active on the Int. J. Biotechnol. 14, 7–27 (2015). examples can inform decisions about ATP synthase of Mycobacterium tuberculosis. Science 37. Drawnel, F. M. et al. Molecular phenotyping combines 307, 223–227 (2005). molecular information, biological relevance, and whether and when to spend resources for 13. Ma, H. et al. Characterization of the metabolic patient data to improve productivity of early drug target identification and detailed molecular activation of hepatitis C virus nucleoside inhibitor discovery. Cell Chem. Biol. 24, 624–634.e3 (2017). beta‑D-2ʹ‑Deoxy‑2ʹ‑fluoro‑2ʹ‑C‑methylcytidine 38. Lee, J. et al. Withaferin A is a leptin sensitizer with MoA studies. (PSI‑6130) and identification of a novel active strong antidiabetic properties in mice. Nat. Med. 22, PDD is a challenging drug discovery 5ʹ‑triphosphate species. J. Biol. Chem. 282, 1023–1032 (2016). 29812–29820 (2007). 39. Liu, J., Lee, J., Salazar Hernandez, M. A., strategy on multiple levels, but it has 14. Queiroz, E. F., Wolfender, J. L. & Hostettmann, K. Mazitschek, R. & Ozcan, U. Treatment of obesity with a successful track record of delivering Modern approaches in the search for new lead celastrol. Cell 161, 999–1011 (2015). antiparasitic compounds from higher plants. 40. Amos, C. I. et al. The OncoArray Consortium: first‑in‑class drugs. It is a powerful approach Curr. Drug Targets 10, 202–211 (2009). a network for understanding the genetic architecture to exploit the novel biological space of 15. Gao, M. et al. Chemical genetics strategy identifies an of common cancers. Cancer Epidemiol. Biomarkers HCV NS5A inhibitor with a potent clinical effect. Prev. 26, 126–135 (2017). undrugged or unknown targets and poorly Nature 465, 96–100 (2010). 41. Marx, V. The DNA of a nation. Nature 524, understood disease mechanisms, providing 16. Burke, A. C., Dron, J. S., Hegele, R. A. & Huff, M. W. 503–505 (2015). PCSK9: regulation and target for drug development 42. Chambers, J. C. et al. The South Asian genome. PLoS a route to enhance innovation in the for dyslipidemia. Annu. Rev. Pharmacol. Toxicol. 57, ONE 9, e102645 (2014). pharmaceutical industry and to deliver truly 223–244 (2016). 43. Cyranoski, D. Genomics takes hold in Asia. Nature 17. Petersen, D. N. et al. A small-molecule anti- 456, 12 (2008). novel therapeutics for unmet medical needs. secretagogue of PCSK9 targets the 80S ribosome to 44. Nath, S. & Devi, G. R. Three-dimensional culture inhibit PCSK9 protein translation. Cell Chem. Biol. 23, systems in cancer research: focus on tumor spheroid John G. Moffat is at Biochemical & Cellular 1362–1371 (2016). model. Pharmacol. Ther. 163, 94–108 (2016). Pharmacology, Genentech, South San Francisco, 18. Swinney, D. C. & Xia, S. The discovery of medicines for 45. Esch, E. W., Bahinski, A. & Huh, D. Organs-on‑chips at Future Med. Chem. 6 the frontiers of drug discovery. Nat. Rev. Drug Discov. California 94080, USA. rare diseases. , 987–1002 (2014). 19. Swinney, D. C. Challenges and hurdles to business as 14, 248–260 (2015). 46. Ham, S. L., Atefi, E., Fyffe, D. & Tavana, H. Robotic Fabien Vincent is at Discovery Sciences, Primary usual in drug development for treatment of rare diseases. Clin. Pharmacol. Ther. 100, 339–341 production of cancer cell spheroids with an aqueous Pharmacology Group, Pfizer, Groton, (2016). two-phase system for drug testing. J. Vis. Exp. 23, Connecticut 06340, USA. 20. Nystrom, A. et al. Losartan ameliorates dystrophic e52754 (2015). epidermolysis bullosa and uncovers new disease 47. Thakuri, P. S., Ham, S. L., Luker, G. D. & Tavana, H. Jonathan A. Lee is in the Department of Quantitative mechanisms. EMBO Mol. Med. 7, 1211–1228 Multi-parametric analysis of oncology drug screening Biology, Eli Lilly and Company, Indianapolis, (2015). with aqueous two-phase tumor spheroids. Indiana 46285, USA. 21. Fishman, M. C. Power of rare diseases: found in Mol. Pharm. 13, 3724–3735 (2016). translation. Sci. Transl Med. 5, 201ps11 (2013). 48. Friedrich, J., Seidel, C., Ebner, R. & Kunz- Jörg Eder is at Novartis Institutes for Biomedical 22. Horvath, P. et al. Screening out irrelevant cell-based Schughart, L. A. Spheroid-based drug screen: models of disease. Nat. Rev. Drug Discov. 15, considerations and practical approach. Nat. Protoc. 4, Research, 4002 Basel, Switzerland. 751–769 (2016). 309–324 (2009). 23. Alfoldi, J. & Lindblad-Toh, K. Comparative genomics as 49. Katt, M. E., Placone, A. L., Wong, A. D., Xu, Z. S. & Marco Prunotto was previously at Phenotype and a tool to understand evolution and disease. Genome Searson, P. C. In vitro tumor models: advantages, Target ID, Chemical Biology, pRED, Roche, 4070 Res. 23, 1063–1068 (2013). disadvantages, variables, and selecting the right Basel, Switzerland. Present address: Office of 24. Mariani, L. H., Pendergraft, W. F. III & Kretzler, M. platform. Front. Bioeng. Biotechnol. 4, 12 (2016). Innovation, Immunology, Infectious Diseases & Defining glomerular disease in mechanistic terms: 50. Gunter, J. et al. Microtissues in cardiovascular medicine: regenerative potential based on a 3D Ophthalmology (I2O), Roche Late Stage Development, implementing an integrative biology approach in nephrology. Clin. J. Am. Soc. Nephrol. 11, microenvironment. Stem Cells Int. 2016, 9098523 124 Grenzacherstrasse, 4070 Basel, Switzerland. 2054–2060 (2016). (2016). 25. Ju, W. et al. Tissue transcriptome-driven 51. Weiswald, L. B., Bellet, D. & Dangles-Marie, V. Correspondence to M.P. identification of epidermal growth factor as a chronic Spherical cancer models in tumor biology. Neoplasia [email protected] kidney disease biomarker. Sci. Transl Med. 7, 17, 1–15 (2015). doi:10.1038/nrd.2017.111 316ra193 (2015). 52. Hild, M. & Jaffe, A. B. Production of 3D airway 26. Brosius, F. C., Tuttle, K. R. & Kretzler, M. JAK organoids from primary human airway basal cells and Published online 7 Jul 2017 inhibition in the treatment of diabetic kidney disease. their use in high-throughput screening. Curr. Protoc. Diabetologia 59, 1624–1627 (2016). Stem Cell Biol. 37, IE.9.1–IE.9.15 (2016). 1. Swinney, D. C. & Anthony, J. How were new medicines 27. Wang, L. et al. A disease model of diabetic 53. Heilker, R., Traub, S., Reinhardt, P., Scholer, H. R. & discovered? Nat. Rev. Drug Discov. 10, 507–519 nephropathy in a glomerulus-on‑a‑chip microdevice. Sterneckert, J. iPS cell derived neuronal cells for drug (2011). Lab Chip 17, 1749–1760 (2017). discovery. Trends Pharmacol. Sci. 35, 510–519 2. Scannell, J. W. & Bosley, J. When quality beats 28. Keene, C. D. et al. Neuropathological assessment and (2014). quantity: decision theory, drug discovery, and the validation of mouse models for Alzheimer’s disease: 54. Avior, Y., Sagi, I. & Benvenisty, N. Pluripotent stem reproducibility crisis. PLoS ONE 11, e0147215 applying NIA‑AA guidelines. Pathobiol. Aging Age cells in disease modelling and drug discovery. (2016). Relat. Dis. 6, 32397 (2016). Nat. Rev. Mol. Cell Biol. 17, 170–182 (2016). 3. Wagner, B. K. & Schreiber, S. L. The power of 29. Palacino, J. et al. SMN2 splice modulators enhance 55. Lee, G. et al. Large-scale screening using familial sophisticated phenotypic screening and modern U1‑pre-mRNA association and rescue SMA mice. dysautonomia induced pluripotent stem cells identifies mechanism-of‑action methods. Cell Chem. Biol. 23, Nat. Chem. Biol. 11, 511–517 (2015). compounds that rescue IKBKAP expression. 3–9 (2016). 30. Lamb, J. et al. The Connectivity Map: using gene- Nat. Biotechnol. 30, 1244–1248 (2012). 4. Shi, Y., Inoue, H., Wu, J. C. & Yamanaka, S. Induced expression signatures to connect small molecules, 56. Edwards, A. M. et al. Preclinical target validation using pluripotent stem cell technology: a decade of progress. genes, and disease. Science 313, 1929–1935 patient-derived cells. Nat. Rev. Drug Discov. 14, Nat. Rev. Drug Discov. 16, 115–130 (2017). (2006). 149–150 (2015).

PERSPECTIVES

57. Van Goor, F. et al. Rescue of CF airway epithelial cell 81. Choi, D., Stables, J. P. & Kohn, H. Synthesis and 106. van der Worp, H. B. et al. Can animal models of function in vitro by a CFTR potentiator, VX‑770. Proc. anticonvulsant activities of disease reliably inform human studies? PLoS Med. 7, Natl Acad. Sci. USA 106, 18825–18830 (2009). N‑benzyl‑2‑acetamidopropionamide derivatives. e1000245 (2010). 58. Yu, H. et al. Ivacaftor potentiation of multiple CFTR J. Med. Chem. 39, 1907–1916 (1996). 107. Sirenko, O. et al. In vitro cardiotoxicity assessment of channels with gating mutations. J. Cyst. Fibros. 11, 82. Errington, A. C., Stohr, T., Heers, C. & Lees, G. The environmental chemicals using an organotypic human 237–245 (2012). investigational anticonvulsant lacosamide induced pluripotent stem cell-derived model. Toxicol. 59. Berg, A. et al. A phenotypic screen for corrector selectively enhances slow inactivation of voltage- Appl. Pharmacol. 322, 60–74 (2017). discovery using a surface liquid readout in F508del gated sodium channels. Mol. Pharmacol. 73, 108. Mann, D. A. Human induced pluripotent stem cell- primary airway epithelia. Pediatr. Pulmonol. 50, 157–169 (2008). derived hepatocytes for toxicology testing. Expert abstr. 181 (2015). 83. Arrowsmith, C. H. et al. The promise and peril of Opin. Drug Metab. Toxicol. 11, 1–5 (2015). 60. Klijn, C. et al. A comprehensive transcriptional portrait chemical probes. Nat. Chem. Biol. 11, 536–541 (2015). 109. Sweis, R. F. Target (in)validation: a critical, sometimes of human cancer cell lines. Nat. Biotechnol. 33, 84. Wang, Y. et al. Evidence-based and quantitative unheralded, role of modern medicinal chemistry. ACS 306–312 (2015). prioritization of tool compounds in phenotypic drug Med. Chem. Lett. 6, 618–621 (2015). 61. Goodspeed, A., Heiser, L. M., Gray, J. W. & discovery. Cell Chem. Biol. 23, 862–874 (2016). 110. Saxena, C. Identification of protein binding partners Costello, J. C. Tumor-derived cell lines as molecular 85. Jones, L. H. & Bunnage, M. E. Applications of of small molecules using label-free methods. Expert models of cancer pharmacogenomics. Mol. Cancer chemogenomic library screening in drug discovery. Opin. Drug Discov. 11, 1017–1025 (2016). Res. 14, 3–13 (2016). Nat. Rev. Drug Discov. 16, 285–296 (2017). 111. Fetz, V., Prochnow, H., Bronstrup, M. & Sasse, F. 62. Cascorbi, I. & Werk, A. N. Advances and challenges in 86. Santos, R. et al. A comprehensive map of molecular Target identification by image analysis. Nat. Prod. hereditary cancer pharmacogenetics. Expert Opin. drug targets. Nat. Rev. Drug Discov. 16, 19–34 (2017). Rep. 33, 655–667 (2016). Drug Metab. Toxicol. 13, 73–82 (2017). 87. Wassermann, A. M. et al. Dark chemical matter as a 112. Schirle, M. & Jenkins, J. L. Identifying compound 63. O’Duibhir, E., Carragher, N. O. & Pollard, S. M. promising starting point for drug lead discovery. Nat. efficacy targets in phenotypic drug discovery. Drug Accelerating glioblastoma drug discovery: Chem. Biol. 11, 958–966 (2015). Discov. Today 21, 82–89 (2016). convergence of patient-derived models, genome 88. Wetzel, S., Bon, R. S., Kumar, K. & Waldmann, H. 113. Adams, R., Steckel, M. & Nicke, B. Functional editing and phenotypic screening. Mol. Cell. Neurosci. Biology-oriented synthesis. Angew Chem. Int. Ed. 50, genomics in pharmaceutical drug discovery. Handb. 80, 198–207 (2017). 10800–10826 (2011). Exp. Pharmacol. 232, 25–41 (2016). 64. Quartararo, C. E., Reznik, E., deCarvalho, A. C., 89. Newman, D. J. & Cragg, G. M. Natural products as 114. Moore, J. D. The impact of CRISPR–Cas9 on target Mikkelsen, T. & Stockwell, B. R. High-throughput sources of new drugs from 1981 to 2014. J. Nat. identification and validation. Drug Discov. Today 20, screening of patient-derived cultures reveals potential Prod. 79, 629–661 (2016). 450–457 (2015). for precision medicine in glioblastoma. ACS Med. 90. Harvey, A. L., Edrada-Ebel, R. & Quinn, R. J. The 115. Lee, H. & Lee, J. W. Target identification for Chem. Lett. 6, 948–952 (2015). re‑emergence of natural products for drug discovery in biologically active small molecules using chemical 65. Verissimo, C. S. et al. Targeting mutant RAS in patient- the genomics era. Nat. Rev. Drug Discov. 14, biology approaches. Arch. Pharmacol. Res. 39, derived colorectal cancer organoids by combinatorial 111–129 (2015). 1193–1201 (2016). drug screening. eLife 5, e18489 (2016). 91. Kato, N. et al. Diversity-oriented synthesis yields 116. Nijman, S. M. Functional genomics to uncover drug 66. van de Wetering, M. et al. Prospective derivation of novel multistage antimalarial inhibitors. Nature 538, mechanism of action. Nat. Chem. Biol. 11, 942–948 a living organoid biobank of colorectal cancer patients. 344–349 (2016). (2015). Cell 161, 933–945 (2015). 92. Wawer, M. J. et al. Toward performance-diverse small- 117. Maryanoff, B. E. Phenotypic assessment and the 67. Naryshkin, N. A. et al. Motor neuron disease. SMN2 molecule libraries for cell-based phenotypic screening discovery of topiramate. ACS Med. Chem. Lett. 7, splicing modifiers improve motor function and using multiplexed high-dimensional profiling. Proc. 662–665 (2016). longevity in mice with spinal muscular atrophy. Science Natl Acad. Sci. USA 111, 10911–10916 (2014). 118. Silberstein, S. Topiramate in migraine prevention. 345, 688–693 (2014). 93. De Wolf, H., De Bondt, A., Turner, H. & Headache 45 Suppl. 1, S57–S65 (2005). 68. Auld, D. S. et al. Characterization of chemical libraries Gohlmann, H. W. Transcriptional characterization of 119. Rogawski, M. A. Brivaracetam: a rational drug for luciferase inhibitory activity. J. Med. Chem. 51, compounds: lessons learned from the public LINCS discovery success story. Br. J. Pharmacol. 154, 2372–2386 (2008). data. Assay Drug Dev. Technol. 14, 252–260 1555–1557 (2008). 69. Nonejuie, P., Burkart, M., Pogliano, K. & Pogliano, J. (2016). 120. Kaminski, R. M. et al. SV2A protein is a broad- Bacterial cytological profiling rapidly identifies the 94. Liu, C. et al. Compound signature detection on spectrum anticonvulsant target: functional correlation cellular pathways targeted by antibacterial molecules. LINCS L1000 big data. Mol. Biosyst. 11, 714–722 between protein binding and seizure protection in Proc. Natl Acad. Sci. USA 110, 16169–16174 (2013). (2015). models of both partial and generalized epilepsy. 70. Swinney, D. C. The value of translational biomarkers to 95. Matthews, M. L. et al. Chemoproteomic profiling and Neuropharmacology 54, 715–720 (2008). phenotypic assays. Front. Pharmacol. 5, 171 (2014). discovery of protein electrophiles in human cells. Nat. 121. Mavris, M. & Le Cam, Y. Involvement of patient 71. Klotz, J. Phenotypic screening, take two. SciBX http:// Chem. 9, 234–243 (2017). organisations in research and development of orphan dx.doi.org/10.1038/scibx.2012.380 (2012). 96. Backus, K. M. et al. Proteome-wide covalent ligand drugs for rare diseases in europe. Mol. Syndromol. 3, 72. Center for Drug Evaluation and Research (CDER) & discovery in native biological systems. Nature 534, 237–243 (2012). Center for Biologics Evaluation and Research (CBER). 570–574 (2016). 122. Nakazato, H., Oku, H., Yamane, S., Tsuruta, Y. & Guidance for industry content and format of 97. Parker, C. G. et al. Ligand and target discovery by Suzuki, R. A novel anti-fibrotic agent pirfenidone investigational new drug applications (INDs) for fragment-based screening in human cells. Cell 168, suppresses tumor necrosis factor-alpha at the phase 1 studies of drugs, including well-characterized, 527–541.e29 (2017). translational level. Eur. J. Pharmacol. 446, 177–185 therapeutic, biotechnology-derived products. FDA 98. Rottmann, M. et al. Spiroindolones, a potent (2002). http://www.fda.gov/downloads/Drugs/ compound class for the treatment of malaria. Science 123. Roth, G. J. et al. Nintedanib: from discovery to the GuidanceComplianceRegulatoryInformation/ 329, 1175–1180 (2010). clinic. J. Med. Chem. 58, 1053–1063 (2015). Guidances/ucm071597.pdf (1995). 99. Zhang, W., Bai, Y., Wang, Y. & Xiao, W. 73. Drews, J. Drug discovery: a historical perspective. Polypharmacology in drug discovery: a review from Acknowledgements Science 287, 1960–1964 (2000). systems pharmacology perspective. Curr. Pharm. Des. The authors thank A. Subramanian and J. Rudolph for help- 74. Overington, J. P., Al-Lazikani, B. & Hopkins, A. L. How 22, 3171–3181 (2016). ful discussions and comments and acknowledge the invalu- many drug targets are there? Nat. Rev. Drug Discov. 100. Taylor, D. L. Past, present, and future of high content able contributions of the participants of the 2015 New York 5, 993–996 (2006). screening and the field of cellomics. Methods Mol. Academy of Sciences Symposium and the 2016 Keystone 75. Begley, C. G. & Ellis, L. M. Drug development: raise Biol. 356, 3–18 (2007). Symposium on phenotypic drug discovery. The idea to standards for preclinical cancer research. Nature 483, 101. Perlman, Z. E. et al. Multidimensional drug profiling gather and share experiences from multiple companies coa- 531–533 (2012). by automated microscopy. Science 306, 1194–1198 lesced from conversations with many outstanding scientists 76. Prinz, F., Schlange, T. & Asadullah, K. Believe it or not: (2004). during the breaks of those two landmark conferences. how much can we rely on published data on potential 102. Caie, P. D. et al. High-content phenotypic profiling of drug targets? Nat. Rev. Drug Discov. 10, 712 (2011). drug response signatures across distinct cancer cells. Competing interests statement 77. Hay, M., Thomas, D. W., Craighead, J. L., Mol. Cancer Ther. 9, 1913–1926 (2010). The authors declare competing interests: see Web version Economides, C. & Rosenthal, J. Clinical development 103. Liu, J. et al. Targeting Wnt-driven cancer through the for details. success rates for investigational drugs. Nat. inhibition of Porcupine by LGK974. Proc. Natl Acad. Biotechnol. 32, 40–51 (2014). Sci. USA 110, 20224–20229 (2013). Publisher’s note 78. Kiuchi, M. et al. Synthesis and immunosuppressive 104. Black, J. Nobel lecture in physiology or medicine — Springer Nature remains neutral with regard to jurisdic- activity of 2‑substituted 2‑aminopropane‑1,3‑diols and 1988. Drugs from emasculated hormones: the tional claims in published maps and institutional 2‑aminoethanols. J. Med. Chem. 43, 2946–2961 (2000). principle of syntopic antagonism. In Vitro Cell Dev. affiliations. 79. Kovarik, J. M., Schmouder, R. L. & Slade, A. J. Overview Biol. 25, 311–320 (1989). of FTY720 clinical pharmacokinetics and pharmacology. 105. Morgan, P. et al. Can the flow of medicines be Ther. Drug Monit. 26, 585–587 (2004). improved? Fundamental pharmacokinetic and DATABASES 80. Brinkmann, V. et al. Fingolimod (FTY720): discovery pharmacological principles toward improving phase Chemical Probes Portal: http://chemicalprobes.org and development of an oral drug to treat multiple II survival. Drug Discov. Today 17, 419–424 ALL LINKS ARE ACTIVE IN THE ONLINE PDF sclerosis. Nat. Rev. Drug Discov. 9, 883–897 (2010). (2012).