Drug Discovery World
Phenotypic Screening - advancing the science from discovery to regulatory science
By Dr Ellen L. Berg
Fall 2017

Phenotypic screening is transforming drug discovery as advances are made in development of human-based physiologically-relevant in vitro systems.

Integration of phenotypic drug discovery approaches with target-based screening can improve success rates as assays can be leveraged for the discovery of novel mechanisms and for building knowledge of disease biology.

Pharmaceutical productivity continues to stagnate despite advances in genomics, assay technologies and automation that have enabled high throughput screening of millions of compounds in target-based assays. Fewer than 10% of drug discovery programmes entering the clinic make it all the way to regulatory approval. Indeed, in 2016 only 22 novel drugs were approved by the FDA, a six-year low. As the limitations of target-based drug discovery (TDD) were becoming apparent earlier this decade, Swinney and Antony (1), with their unexpectedly-impactful analysis of historical drug approvals, initiated a shift within the industry to reconsider phenotypic screening approaches or phenotypic drug discovery (PDD) for identifying novel drugs. Despite the challenges with these approaches, phenotypic screening is poised to make an impact.

What is phenotypic screening?

Phenotypic screening uses live biological systems (in vitro cell cultures or animal models) and was the primary means employed to discover new drugs prior to the genomics era and large-scale identification of potential drug targets. Target based screening utilises simpler assays, often biochemical-based, to identify molecules that interact with a specific protein target and is preferred when the disease biology is well understood and the relevance of a target to that disease has been validated with high confidence, since target-based assays can be run at very high throughput-toscreen multi-million compound libraries. However, as Swinney and Antony (1) in their 2011 paper (updated here (2) in 2014) observed, despite heavy investment in target discovery and validation, phenotypic approaches have continued to provide a significant fraction of first-in-class drugs. This suggests that although the human genome project was successful in identifying large numbers of prospective drug targets, too few of these have led to new therapies. The methods used to validate these targets (and our knowledge of disease mechanisms) have been inadequate to yield the number of drug approvals needed to support the industry, contributing to the unsustainable high cost of drugs. Using decision theory analysis of the pharmaceutical R&D process, Scannell and Bosley (3) identified ‘predictive validity’ of screening models as a driving factor for improving pharmaceutical productivity. Thus, as the industry begins to adopt phenotypic approaches in drug discovery, it will be important to take into consideration that successful phenotypic programmes will also require validation to ensure that the assays used effectively recapitulate the disease biology of interest. The need for better understanding of disease biology remains true whether target-based or phenotypic approaches are taken.

Designing physiologicallyrelevant phenotypic assays

There have been several recent perspectives written about the design of physiologically-relevant assays for phenotypic drug discovery. In 2015, Vincent and co-workers at Pfizer (4) published their ‘rule of 3’ covering: (1 Selecting physiologically-relevant cell types and formats; (2) Choosing assay stimuli that optimise disease relevance; and (3) Using assay endpoints that are proximal to clinical endpoints. Horvath and colleagues (5) described a set of 17 principles to be considered when designing physiologically-relevant assays encompassing assay objectives (context of use), cell types and media selection, as well as micro-physiological characteristics and selection of clinically-relevant doses. Most of the efforts in phenotypic discovery are focused on in vitro systems rather than animal models, as in vitro assays provide higher throughput and can be developed with human cell types, improving translation. Incorporating greater physiological relevance through the use of human iPSC-derived cells or primary cells in phenotypic screens has been made possible through the development of methods to culture and differentiate these cells. While primary human cells are considered the gold standard, iPSC or stem cell-derived cell types are highly attractive due to the potential for genetic screening or gene editing through new technologies such as Cas9/CRISPR6. Differentiation protocols to produce more adult-like phenotypes continue to evolve (7). The use of co-cultures, incorporating more than one cell type, is another option to improve physiological relevance within the confines of high throughput screening paradigm. Coculture systems, if appropriately designed, have been shown to increase the biological relevance and detection of disease-relevant mechanisms with minimal impact on assay performance (8).

While developing assays with high physiological relevance is a goal, there can be practical considerations for high throughput primary screening that limit assay design choices. For example, the use of patient-derived cells may not be feasible due to excessive variance caused by patients being at different stages of disease or having different drug exposures. Throughput is also a consideration. Although advanced cellular systems with complex formats such as organs-on-a-chip, bio-printed tissues and organoids have been proposed as a solution for their perceived greater physiological relevance (5,9), many lack the requisite throughput required for large-scale screening. Indeed, initial studies using these systems have generated much excitement (10), even though validation of these for many contexts of use remains to be established. The validation process for complex in vitro systems to establish reproducibility and disease relevance is not trivial. As the complexity of an assay system increases, so do the number of variables and the potential for reproducibility problems. Validation requires sufficient throughput to perform the extensive evaluation needed to fully characterise the biology represented by the system and clinical relevance (11). Since sampling the largest expanse of chemical matter is the goal in primary screening, it is preferable to employ the simplest assay that covers the desired biological mechanism. Thus, the most valuable application of these highly complex, physiologically-relevant systems may be in basic research, building our understanding of disease biology through testing known drugs or genetic modifications, for new target identification, or for characterising small numbers of lead candidates at later stages of discovery.

Phenotypic drug discovery – practical lessons learned

As adoption of phenotypic approaches in pharmaceutical discovery research has expanded, reports of successes as well as lessons learned have begun to emerge. A recent perspective9 written by investigators from Genentech, Pfizer, Eli Lilly, Novartis and Roche emphasised the need to have sufficient understanding of disease biology to develop a successful screen. They propose a ‘chain of translatability’ to build confidence in the assay system to ensure effective modelling of the disease of interest. This chain of translatability connects the endpoints measured in the assay system through to the clinical outcome, a framework that is in some ways analogous to the adverse outcome pathway concept being developed for chemical toxicity risk assessments (12). The strength and confidence in this connection is important given that the ability of the screening assay to predict clinical therapeutic response is a fundamental determinant of success. This is more straightforward for some therapeutic areas (eg infectious diseases where the infectious agent itself is causative) than for others (eg chronic inflammatory diseases and CNS disorders) where the causes may be unknown, and intervention points can be far downstream. Thus, a more comprehensive understanding of the cell types, biological processes and tissue responses involved in the disease is required to design successful screening assays. A group from Novartis recently shared their experiences with phenotypic screening over a five-year period, from 2011-15 (13). In 2011 ~5% of the discovery screens at Novartis were phenotypicbased, rising to a peak of nearly 40% in 2014. This experience led the Novartis team to recommend prosecution of small pilot screens of 1,000-5,000 compounds on multiple assays representing different aspects of biology in order to select and refine a desired phenotype. Sampling and testing the biology this way allowed them to advance their understanding of the disease biology, exploring and refining the therapeutic hypothesis before final assay selection.

Prosecuting phenotypic drug discovery programmes – process considerations

The prosecution of target-based drug discovery programmes, having been the subject of intensive process improvement efforts over several decades, follows a more-or-less predictable path, from initial screen and hit identification to lead selection, optimisation, preclinical assessment and clinical candidate selection. The framework for prosecuting phenotypic discovery programmes is less wellestablished. Unlike target-based programmes, hits from phenotypic screens may act through any one of a number of known or unknown targets, and so require mechanistic characterisation to progress. This characterisation is needed to triage hits based on undesirable mechanisms as well as to group hits into similar mechanism classes. Developing an appropriate assay flow scheme of secondary assays for counterscreening for a phenotypic programme requires greater upfront investment than for target-based discovery programmes, and the flow scheme can change as the programme progresses as new information is incorporated.

The first tier of secondary assays is typically identified during assay development from in-depth characterisation studies performed on the selected phenotypic screen. Assay characterisation is initially performed to support the relevance of the system for the purpose of predicting the clinical outcome of interest, for example, through the use of translational biomarkers or disease signatures that can be matched to clinical samples. Once this is established, further interrogation of the assay is performed via perturbation studies using drugs and chemical probes as tool compounds, to reveal known targets and pathways that are active in the assay and need to be excluded. Secondary assays for these targets can then be run following the primary screen. Following this triage, the remaining hits must be further classified and prioritised. This mechanistic classification can be accomplished using unsupervised multi-parameter profiling approaches such as transcriptomic, proteomic or high content analysis within the screening assay itself, or by employing more extensive phenotypic profiling across additional assays. The usefulness of these unsupervised methods is increased by coverage of biology and the availability of reference signatures of compounds with known mechanisms to which hits may be compared. This effort can reveal additional mechanisms or identify new assays that should be incorporated into the screening funnel, and can also help support target deconvolution.

Target deconvolution is an important activity that must also be considered for phenotypic discovery programmes. While many drugs have been approved without a known molecular target, knowledge of the target is highly valuable for the discovery of follow-up compounds, and for derisking the potential for unexpected toxicities in vivo (9). Known targets can be eliminated by testing in target-based cellular or biochemical panels, such as those established for safety pharmacology or for broad coverage of particular target classes, such as comprehensive GPCR or kinase panels. For the identification of truly novel targets, there have been advances in methods to identify targets using chemical proteomics techniques such as affinity pull down, functional genomics methods or molecular profiling. These can be time-consuming and resource intensive, and often multiple approaches are required. Thus, for most discovery research groups, unless the chemistry lends itself to such an effort, target deconvolution activities are only initiated late in discovery. Phenotypic profiling studies that leverage the use of bioactivity fingerprints across standard panels, ideally that cover a broad range of biology to classify compounds based on mechanistic similarity, can be used to prioritise candidates prior to initiating these efforts.

Future – the integration of phenotypic and target-based discovery approaches

A common theme emerging from pharmaceutical research is the need for more effective integration of human biology into the drug discovery enterprise. Years of investments in ’omics technologies that measure the cellular components of biological systems and permitting the generation of pathway networks, has built our knowledge of intracellular signalling mechanisms, but has not led to comparable advances in our understanding of disease biology. This is particularly true at higher levels of organisation, including how information is communicated between cells and tissues to effect human outcomes. The regulatory mechanisms at these higher levels of organisation are closer to clinical outcomes and therefore are more tightly associated with and predictive of therapeutic efficacy. Indeed, it is at this level that the differences between animal models and humans is the greatest. Small differences at the molecular or pathway level, can become amplified at the next higher level of organisation, a consequence of modular, hierarchical complex system architectures. The focus on subcellular pathway regulation may explain why we have not been more successful in target validation. In highly networked systems, feedback mechanisms exist to keep the system balanced and functioning in the face of external stimuli and stress. Understanding the feedback and communication mechanisms that operate at these higher levels of organisation will not only help us identify phenotypic screening assays with higher predictive validity, but also validated drug targets.

Effective integration of disease biology into drug discovery requires a multi-pronged approach that combines both phenotypic and target-based methods. This involves taking a holistic approach to the disease, mapping out the known biology, and then exploring multiple phenotypic as well as target-based screens for their applicability to the disease. Discovery efforts in a particular disease should begin with a theoretical framework of the disease on to which existing knowledge of cell types, cellular interactions, biological processes, known mechanisms and clinical biomarkers are mapped. Given the limitations of animal models for human translation, it is best to develop these theoretical frameworks primarily from human clinical information, such as genetic associations, drug effects and biomarker studies. Then, as phenotypic assays representing various cell responses and interactions within the framework are developed and piloted, and the effects of genetic and compound perturbations on potential biomarkers are measured, the resulting information can be used to refine the model, fill in knowledge gaps and build support for particular therapeutic hypotheses. Advanced cellular models, such as organs-on-a-chip or microphysiological systems and bioprinted tissues have an important role to play here as well, to test hypotheses and evaluate novel targets or biological processes in a more complex setting, as they provide a useful bridge from simpler, higherthroughput assays to the clinical situation. At this point, potential targets may emerge and decisions to initiate target-based screening can be made. The information generated is also used to select the most suitable phenotypic assays for high throughput screening. Although this process can be timeconsuming, the phenotypic assays used for primary screening to discover hits with novel mechanisms, can also be applied to support the validation of compounds identified from target-based screens. For a single disease, an organisation will typically be prosecuting multiple assays and mechanisms. While this takes greater resources, it also increases the odds of success, and builds the organisation’s internal expertise in phenotypic assays and disease biology – a competitive edge.

There are additional advantages of taking this integrated approach, despite the larger initial investment required. As phenotypic assays are prosecuted over time, the data from these assays can be combined to build large chemical biology datasets that can yield valuable insights into chemical structure-function relationships and also address the problem of drug safety. Pharmaceutical discovery groups are familiar with the value that has been gained from mining target-based screening data to support in silico-based computer-aided drug design which permits the prediction of target activity from chemical structure (14). In this way, chemical space can be sampled more efficiently without having to synthesise every possible chemical analogue. Large phenotypic assay data sets can be mined too for valuable insights, not only on chemical structure-function relationships but also for the discovery of novel efficacy or toxicity mechanisms. These discoveries are enabled by annotation of tested compounds for clinical indication or adverse effects, from known effects of the drugs or tool compounds evaluated during assay characterisation; or for target interactions from information that is available from compounds in the screening deck that have been run in historical target-based screening campaigns.

The potential of this approach to address the problem of clinical failures due to unexpected toxicities represents a strategic advantage and can benefit an organisation across all therapeutic programmes. An estimated 30% of programmes fail in clinical testing due to safety issues (15). The mechanisms of most toxicities are not known and animal testing has shown limited predictivity. The ability to mine large chemical biology data sets for correlations to known clinical activities, including adverse effects, as well as target mechanisms, helps build knowledge of toxicity mechanisms. Indeed, data mining of well-annotated bioactivity databases for drug associations with adverse effects has been useful for discovering novel mechanisms of toxicity (11,15-17). Analysis of phenotypic profiles from a standardised panel of human primary cellbased assays has revealed a novel mechanism involving autophagy in vascular endothelial cells that contributes to thrombosis-related side-effects in humans (11), and has also successfully uncovered mechanisms underlying skin hypersensitivity in non-human primates (17). As the number of toxicity mechanisms that are identified increases, and either phenotypic or target-based assays for these become available, it will be possible to rely more and more on in vitro tests for predicting safety risks. For drug pharmacokinetics, it is already standard practice to use a combination of in vitro and in silico models to predict human exposures. As the ability to predict human outcomes from these approaches improves, the dependence on animal testing will decrease while programme success rates increase.

For these approaches to be incorporated into regulatory decision-making, given the primary remit of the FDA, EMA and other drug approval agencies to protect patient safety, significant investment in data-driven, performance-based validation studies will be needed. If we are to move the industry forward, open shared data on publicly available drugs will be needed, as the quality of predictive models depends on the amount of data that is used to develop the model. There is tremendous potential value from incorporating human-based in vitro data, whether phenotypic or target-based, in regulatory assessments. These data address human-specific mechanisms and so have the potential for better predictivity, and their use supports the goal of promoting non-animal-based methods. The addition of phenotypic screening for improving the accuracy of drug safety assessments contributes to the modernisation of regulatory science, bringing more efficacious and safer drugs to market at lower cost. While phenotypic screening today is already making an impact in pharmaceutical R&D, the greatest benefits may be yet to come.


Dr Ellen L. Berg is Chief Scientific Officer at DiscoverX (now a part of Eurofins Pharma Discovery Services) and leads the scientific direction of the BioMAP® phenotypic profiling platform. Dr Berg was co-founder and CSO of BioSeek (acquired by DiscoverX) and prior to that led a research team developing biopharmaceuticals at Protein Design Labs (now AbbVie). She received her PhD from Northwestern University and was a postdoctoral fellow at Stanford University, where she was a fellow of the American Cancer Society and a Special Fellow of the Leukemia Society of America. Dr Berg has served in various positions with and is a Fellow of the Society of Laboratory Automation and Screening (SLAS). She is a board member of the American Society for Cellular and Computational Toxicology (ASCCT), and a member of the Society of Toxicology (SOT) and Inflammation Research Association organisations. Her research interests include human-based in vitro systems, chemical biology, drug and toxicity mechanisms of action, phenotypic drug discovery and predictive methods for human efficacy and safety outcomes. Dr Berg holds several patents in the field of inflammation, and has authored more than 80 publications.



1 Swinney, DC and Anthony, J. How were new medicines discovered? Nat Rev Drug Discov, 2011. 10(7): p. 507-19.

2 Swinney, DC. Opportunities for phenotypic screening in drug discovery. Drug Discovery World, 2014. Fall: p. 33-42.

3 Scannell, JW and Bosley, J. When Quality Beats Quantity: Decision Theory, Drug Discovery, and the Reproducibility Crisis. PLoS One. 2016. 11(2): p. e0147215.

4Vincent, F, Loria, P, Pregel, M, Stanton, R, Kitching, L, Nocka, K et al. Developing predictive assays: the phenotypic screening ‘‘rule of 3’’. Sci Transl Med 2015. 7(293). 293ps215.

5 Horvath, P, Aulner, N, Bickle, M, Davies, AM, Nery, ED, Ebner, D et al. Screening out irrelevant cell-based models of disease. Nat Rev Drug Discov 2016. 15: p.751–69.

6 Hockemeyer, D and Jaenisch, R. Induced Pluripotent Stem Cells Meet Genome Editing. Cell Stem Cell. 2016. 5;18(5): p. 573-86.

7 Avior, Y, Levy, G, Zimerman, M, Kitsberg, D, Schwartz, R, Sadeh, R, Moussaieff, A, Cohen, M, Itskovitz-Eldor, J and Nahmias, Y. Microbial-derived lithocholic acid and vitamin K2 drive the metabolic maturation of pluripotent stem cellsderived and fetal hepatocytes. Hepatology. 2015. 62: p. 265-78.

8 Berg, E, Hsu, YC and Lee, JA. Consideration of the cellular microenvironment: physiologically relevant coculture systems in drug discovery. Adv Drug Deliv Rev. 2014. 69-70: p. 190-204.

9 Moffat, JG, Vincent, F, Lee, JA, Eder, J and Prunotto M. Opportunities and challenges in phenotypic drug discovery: an industry perspective. Nat Rev Drug Discov. 2017. 16(8): p. 531-543.

10 Ingber, DE. Reverse Engineering Human Pathophysiology with Organson-Chips. Cell. 2016. 10;164(6): p. 1105-9.

11 Berg, EL. Phenotypic chemical biology for predicting safety and efficacy. Drug Discov Today Technol. 2017. 23: p. 53-60.

12 Edwards, SW, Tan, YM, Villeneuve, DL, Meek, ME and McQueen, CA. Adverse Outcome Pathways-Organizing Toxicological Information to Improve Decision Making. J Pharmacol Exp Ther. 2016. 356(1): p. 170-81.

13 Haasen, D, Schopfer, U, Antczak, C, Guy, C, Fuchs, F and Selzer, P. How Phenotypic Screening Influenced Drug Discovery: Lessons from Five Years of Practice. Assay Drug Dev Technol. 2017. 15(6): p. 239-246.

14 Cheng, T, Hao, M, Takeda, T, Bryant, SH and Wang, Y. LargeScale Prediction of DrugTarget Interaction: a DataCentric Review. AAPS J. 2017. In press.

15 Ledford, H. Translational research: 4 ways to fix the clinical trial. Nature. 2011. 28;477(7366): p. 526-8.

16 Maciejewski, M, Lounkine, E, Whitebread, S, Farmer, P, DuMouchel, W, Shoichet, BK and Urban, L. Reverse translation of adverse event reports paves the way for derisking preclinical off-targets. Elife. 2017. 6. pii: e25818. doi: 10.7554/eLife.25818.

17 Boland, MR, Jacunski, A, Lorberbaum, T, Romano, JD, Moskovitch, R and Tatonetti, NP. Systems biology approaches for identifying adverse drug reactions and elucidating their underlying biological mechanisms. Wiley Interdiscip Rev Syst Biol Med. 2016. 8(2):104-22.

18 Shah, F, Stepan, AF, O’Mahony, A, Velichko, S, Folias, AE, Houle, C, Shaffer, CL, Marcek, J, Whritenour, J, Stanton, R and Berg, EL. Mechanisms of Skin Toxicity Associated with Metabotropic Glutamate Receptor 5 Negative Allosteric Modulators. Cell Chem Biol. 2017. 24(7): p. 858-869.e5.