An industry perspective on drug target validation

KEYWORDS:

• Chemical biology
• chemoproteomics
• phenotypic screening
• target validation

1. Introduction

Drug attrition in clinical trials due to inadequate preclinical target validation is a huge problem facing drug discovery. The issue is compounded by concerns over the reproducibility of published biomedical research, and more worryingly, target validation is particularly vulnerable to confirmation bias, contributing to overconfidence in the specific therapeutic relevance of a new target. Moreover, current approaches to target validation are often slow and costly. To help address these issues, pharmaceutical companies should avoid siloed research disciplines, collaborate strongly with academia, and embrace chemistry at the interface with biology, leading to the advancement of novel technologies that expedite unbiased target validation.

2. Target invalidation and omic profiling

Target-based drug discovery has become the dominant scientific paradigm for the delivery of therapeutic agents into the clinic. However, most clinical experiments fail due to lack of efficacy,[1] suggesting that often the ‘wrong’ target has been selected as the focus of the therapeutic program. More concerning is that the results of a failed in vivo efficacy experiment cannot be understood as it is not possible to confirm that the new mechanism under investigation was effectively tested in human patients (what free drug levels were obtained at the target site, and are these enough to modulate the expected functional pharmacology?).[2] One would, therefore, assume that these observations would put strong emphasis on the need to properly validate a new target in preclinical and clinical research and development. Unfortunately, a culture of conformational bias can exist where many experiments are performed in R&D that perpetuate the therapeutic target under ‘investigation’, rather than approach this issue from the opposite direction – what are the ‘killer’ experiments that can get us to a go/no go decision as quickly as possible. Since most targets do not possess the required balance of therapeutic efficacy and safety, then arguably most research efforts will naturally lead to ‘target invalidation’.

Phenotypic screening approaches in patient-derived cells, which most closely represent the disease state, have been highlighted recently as they can provide confidence that a molecule that perturbs an observable phenotype (and its associated mechanism) will translate in the clinic.[3] In this space, there is an attraction to genetic diseases where the molecular etiology is better defined and where access to patient cells and tissues can be facilitated through a strong network of scientists, foundations, and patient advocacy groups. Unfortunately, the lack of target information often hinders the progress of a therapeutic program, since compound optimization and the delineation of structure–activity relationships (SARs) becomes more difficult in the absence of ligand–protein structures or homology models. Additionally, on-target toxicities cannot be fully understood, and therefore, target and mechanism identification (ID) is an important feature of phenotypic screening approaches.

Chemogenomic libraries contain small molecule probe molecules with well-understood pharmacology. These sets can accelerate target ID since the annotated target(s) of a hit molecule could be the ones involved in phenotype perturbation, facilitating conversion to a target-based project.[4] Target ID methods often rely on chemoproteomics, including the use of activity/affinity-based approaches where the small molecule is armed with a chemically reactive warhead to capture target proteins, and often an enrichment handle, such as biotin, to enable streptavidin-based isolation and analysis using mass spectrometry proteomics. Many chemoproteomic target ID technologies require the use of cell lysate (since affinity reagents are often cell impermeable) but our group, and others, have shown considerable differences in drug–target interactions between live cells and lysate, which emphasizes the need for the development of cell permeable chemical crosslinking reagents.[5] Alternatively, label-free target ID technologies such as the cellular thermal stability assay (thermal proteome profiling) measure the interactions of a drug directly with its target proteins.[6] Transcriptional and metabolomics changes can also be assessed in the presence of the compound to provide pathway and target information for the drug under investigation.[7,8]

However, omic methods such as those mentioned earlier can generate a long list of potential targets, but which of these targets are the important ones involved in perturbing the phenotype? This issue highlights a key difference between simply identifying a putative target versus validation of a therapeutically relevant target, the modulation of which possesses the desired efficacy with therapeutic benefit. Computational techniques (such as the causal reasoning engine) [9] that address issues of confirmation bias and the cherry-picking of data for analysis are important approaches that can generate testable hypotheses for follow-up.

3. Target validation using chemical probes

We have previously described a framework to help ensure candidate targets are adequately validated using highly selective pharmacological agents. The so-called four pillars of target validation are borrowed from our clinical trial experience at Pfizer,[2] and focus on the following questions: Does the drug get to the site of action? When there, does it engage the target? Does that then lead to a functional pharmacology? and Does that lead to a phenotypic effect in a relevant cell system?[10] Emphasis is, therefore, based on the selectivity of the chemical probe, and computational approaches can be leveraged to assess the likelihood of developing a selective pharmacological agent based on the nature of the binding site. However, the intrinsic issue of small molecules not being completely specific for a particular target can directly lead to confirmation bias (off-targets are often ignored if they do not fit the desired hypothesis). It is, therefore, imperative that other technologies, such as RNA interference (RNAi) and cellular thermal shift assays, and computational pathway analyses that take into consideration ‘off-targets’, are used to connect a new putative target to functional efficacy. Another important approach that addresses the off-target issue is the use of structurally similar inactive derivatives – they also provide confidence that phenotypic effects are simply not being driven by the physicochemical, nonspecific attributes of the small molecule.[11]

4. Expert opinion

Successful target validation will increasingly rely on the use of omic technologies applied to pathologically relevant systems and computational pathway analytical tools to generate testable hypotheses in an unbiased manner. Advances in chemical biology technologies in particular are required to convert phenotypic screening approaches into accelerated target-based drug discovery projects. It is likely that several orthogonal technologies are required to provide higher confidence in the therapeutic relevance of a new target, such as high-quality chemogenomic and RNAi libraries.

Another advance to the field of target ID and validation would be the facile generation of chemical biology probes in the medicinal chemistry program. Currently, such chemoproteomic probes (that enable binding protein capture, isolation, enrichment, and analysis) are often made following the medicinal chemistry optimization of the hit molecules from a phenotypic screen. However, we, and others, have shown that target protein affinity and protein reactivity, say through photoaffinity labeling, need not be completely correlated, and these functional SARs may diverge, i.e. the most potent inhibitor of an enzyme may not necessarily be the most efficient photocapture reagent.[12] As a result, synthetic libraries should contain chemoproteomic monomers (fragments enabled for parallel synthesis that contain functionality such as photoaffinity and click handles), such that chemical biology and medicinal chemistry SARs can be generated simultaneously on a project.

Computational techniques will need to be developed that facilitate more holistic, multidimensional, and unbiased pathway analyses that are able to integrate complex data from multiple omic experiments. It should be pointed out that ideally these efforts lead to the ID of a single ‘druggable’ target that possesses the requisite efficacy. However, a more realistic scenario is that small molecules may need to modulate multiple targets to perturb a complex phenotype and therefore small molecule phenotypic screening hits may need to be advanced that possess polypharmacology, for which SARs can be intrinsically difficult to decipher. In the early stages of a phenotypic screening program, it can also be extremely challenging to understand the mechanistic toxicity associated with a particular mechanism of action. As a result, significant advances need to be made through the development of cell-based screening/profiling approaches that de-risk mechanistic toxicity. For example, cell-based omic signatures of toxicity could help triage mechanistic pathways, thus helping to remove molecules that possess intrinsically toxic mechanisms before significant investments are made to advance a molecule to in vivo toxicology assessments.

Due to the complexity of target validation, collaboration is absolutely essential. Groups acting in isolation will fail. Multidisciplinary teams of medicinal chemists, biologists, and chemical biologists are required to address the challenges of target ID and validation. Academia–industry collaborations will also provide new opportunities to develop disruptive technologies and accelerate breakthrough therapies. However, these collaborations should be focused on the delivery of high-value medicines and create a culture that transcends the traditional focus of publishing in high-impact factor journals.

Declaration of interests

LH Jones is an employee and shareholder of Pfizer Inc. The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.