The evolution of target validation in drug discovery

by

Dr. Cathy Tralau-Stewart, chief scientific officer at life science company ValiRx, describes the evolution of target validation in drug discovery, ways of tackling project attrition, and the role of human organoid models in future drug discovery strategies.

Key insights:

  • Although there has been developments in patient genomics and AI-driven technologies to identify drug targets, there remains a need to properly validate models of human disease which can be used prior to resource expenditure.
  • The increased number of molecular pharmacological tools available has enabled increased dissection of the role of specific drug targets.
  • Research from AstraZeneca suggests that the application of the 5Rs framework has decreased the overall failure rate in their portfolio.
  • Organ-on-a-chip models, developed an in vivo organ-like microenvironment, are an alternative approach to animal models - however they are still limited as a 2D system.

Drug discovery is a challenging process, with a low probability for any individual project to result in the successful development of a useful novel therapeutic. Very few areas of innovation have such a high failure rate from idea concept through to marketed product. However, success is very worthwhile and can be hugely valuable for human health (as well as in a financial sense).

This low success rate currently dictates that we need more ‘shots on goal’ at the inception of a project. More targets and therapeutic assets to modulate the target need to be designed and assessed for any one successful therapeutic. According to a Bio Report published in February 2021, of 100 assets entering Phase I clinical studies, only eight will reach approval stages. Moreover, McNamee et al. (2017) suggest that it takes on average approximately 29 years from the initial publications around a target to running clinical trials, and 36 years for approval of the drug.

Additionally, the pre-clinical stages of discovery (novel target to candidate IND stages) have their own significant attrition rates, and progress has been made over the last 20+ years to improve this. The reasons for attrition have changed significantly due to improvements in how we select targets and design candidate therapeutics to avoid causes of failure. In particular, failures due to the major areas of attrition i.e., toxicology and drug metabolism/pharmacokinetics, have decreased. This has been achieved by improving systems to analyse these aspects earlier in the drug discovery process and to filter out the poorer compounds. However, recent data suggests that poor prediction of efficacy remains the main cause of attrition and clinical failure.

Research from AstraZeneca suggests that the application of the 5Rs framework (right target, right tissue, right safety, right patient, right commercial potential) has decreased the overall failure rate in their portfolio. The reduction of failure due to other areas (toxicology and DMPK, etc) has increased the overall percentage failure due to prediction of efficacy. Even a slight improvement in our ability to predict efficacy in patients earlier in the process will have a significant impact on the costs, timelines, and overall success of drug discovery and delivery of effective drugs to patients. A net reduction of the failure rate at all stages is predicted to reduce costs (and increase success) significantly.

Target validation

The overall drug discovery process comprises three main activities: (1) identification of a target with a data-supported hypothesis suggesting that it has a key role in the disease process; (2) development and design optimisation of approaches/assets to modulate the target to positively impact the disease; and (3) development of candidate molecules for patient/clinical trials. Each drug discovery program will develop decision-making cascades to move through the stages and focus efforts on the optimal therapeutic entity to assess in the clinic. Each stage is designed to remove poor-performing compounds and this decision is based on the ‘predictivity’ of the models or hypothesis for man. This process begins with a series of investigations to validate any proposed molecular compound and pathway to identify the ‘right target.’

In the past, target validation for drug discovery has used knowledge of the disease biology and known pharmacological tools in organ pharmacology studies systems such as the gut-bath. This involved tissue segments from animals (rat/mouse) and where possible humans to investigate the pharmacology of the proposed disease phenotype. These studies enabled the discovery of beta blockers for heart disease and H2 receptor antagonists for stomach ulcers (Nobel Laurette James Black), in addition to many other pharmacologically discovered drugs. A limited number of compounds would be assessed in these models to derive candidate molecules, and the efficacy element of drug discovery was reasonably well predicted by many of these systems. However, these systems were not high-throughput and required significant pharmacological expertise.

Over the last 20 years, more automated high-throughput models have been used to find hit molecules once the target validation hypothesis has been formed.

Target validation increasingly includes the use of simple and complex animal model systems of disease. The animal models often use stimulators of disease phenotypes and more recently genetic knock-out and transgenic systems (mammalian, worm, zebrafish, and fly) to mimic the disease phenotype. These systems have been increasingly coupled with a greater understanding of genetic-linked disease pathways (particularly for monogenic diseases), target expression in humans, and the molecular pathways involved. The increased number of molecular pharmacological tools (i.e. selective compounds, antibodies, SiRNA, gene editing) available has also enabled increased dissection of the role of specific targets.

The direction of target validation and drug discovery continues to evolve and improve. In the future, it will likely be a multi-factorial process using a variety of models and data for both target validation and compound validation.

Models for decision making

Current decision-making models face many challenges with regard to predicting the efficacy of a novel therapeutic. There is no perfect model of human disease – after all, woman/man is the only true model for woman/man. Human primary cell cultures present expansion and lifespan issues, whereas immortalised cell lines can be cultured repeatedly. However, cell lines tend to evolve away from normal physiology and accumulate mutations that differ from the original lines and cells are studied in isolation, so the interaction with a variety of cells in the microenvironment is lost. These issues decrease the reliability of predicting human disease physiology.

Animal disease models rarely reflect the real causes and phenotypes of human disease effectively. The best models available are for monogenic diseases and these have been used to predict human monogenic ortholog diseases. The transplantation of human cell lines or tumour tissues (PDX) into immunocompromised mice (PDX xenografts) with or without human bone marrow and supplying a more intact immune system still rarely fully reflects the physiology of tumours in humans with complete human immune systems and micro-metabolism. Moreover, these systems are slow and difficult to develop. The growth kinetics of transplanted tumours fail to reflect the usually slower kinetics in humans which are also usually polyclonal, with multiple different drivers of growth.

Transgenic animal models also only partly reflect humans due to the different species' background physiology that they are in. In xenografts, small grafts are unlikely to reflect the heterogenicity of human tumours. Most of the current animal tumour models fail to reflect the evolution of multiple mutations often seen in human tumours. Scannell et al. (2022), suggested that the most predictive models used (i.e. stomach acid secretion) led to effective drugs (H2 receptor antagonists) and therefore were not used further. Whereas the ineffective models (i.e. cancer cell lines) failed to identify good drugs and continue to be used. This has contributed to a gradual decline in the predictivity of models utilised. Scannell outlined the importance of understanding the predictive validity of models and an approach to their assessment.

Human organoid models

Whilst there have been significant developments in how we identify targets to further investigate a drug, which include patient genomics and sub-set analysis and the use of AI-driven technologies to analyse this data, there remains a need to develop and properly validate models of human disease which can be used early in the process and prior to significant resource expenditure. Organoid and tissue slice technologies (organotypic) offer the potential to improve modelling of human normal and disease tissues. They could provide ex vivo platforms which better reflect the phenotype and genotype of the tissues from which they were derived. Furthermore, they offer the potential for multi-cellular, multiple pathway, and mutation models which can be used in target choice and compound evaluation cascades.

There is a lot of work ongoing in this area. Most projects use human cells and tissues (ideally patient-derived) as the starting point for creating models such as small tissue biopsies, multi-cellular spheroids, and organ-on-a chip technologies. Several studies have suggested that three-dimensional organoid culture systems, particularly patient-derived systems, can better model features of tumours and be used to assess drug response in colorectal, pancreatic, liver, breast, and prostate cancers. They have demonstrated that organoid cultures have a higher rate of preservation of the key histologic and molecular traits of the parental human tumours, and drug screening of patient-derived organoids shows high concordance with the matched patient tumour.

Organ-on-a-chip models are an alternative approach to spheroids and are multicellular models of functional organs created in micro-engineered cell culture devices, and which develop an in vivo organ-like microenvironment. These systems enable multiple image-based and other functional readouts for drug screening including immunostaining, viability, cytokines, proteomics, toxicology, etc. The organ models may be able to develop appropriate tissue structures and importantly blood vessel vascularity. However, they are still mainly limited by 2D rather than human-like 3D systems.

Scientists are building databases of organoid-derived information to support the future development of human avatars for AI analyses, with the ultimate aim of supporting the development of partial/full in silico target validation predictive systems.

Next revolution in drug discovery

Many groups and companies are developing organ-on-a-chip systems of human normal and disease tissues for drug development and, when used in combination with increased disease understanding, patient-based genomics, target expression, and AI systems, they hold the potential to significantly improve how we validate targets for drug discovery.

This can ultimately improve attrition due to efficacy in the clinic, as well as the overall efficacy of drug discovery in terms of time to the clinic and fewer but improved assets being tested in patients. The synthesising of data from multiple human (and non-human)-based systems with AI predictions to assess potential targets is the next revolution in drug discovery and holds significant potential to improve the efficiency and cost of the drug discovery process as a whole.