WHITE PAPER

Finding the Balance: Building Physiological Relevance into Early Drug Discovery at Scale

Introduction

High-throughput screening (HTS) plays a fundamental role in the drug discovery process, allowing quick and efficient screening of large compound libraries at scale. It is critical that complex biology and disease-relevant assays are translated into a reduced screenable format, which requires careful consideration and analysis at every step of the screening cascade. But how much physiological relevance can be built into early drug discovery and what are the tradeoffs? In this roundtable discussion, four experts from AstraZeneca’s Global High Throughput Screening Centre share their experience of working on the front line of early drug discovery.

Pursuing the wrong target in early drug discovery can waste significant amounts of time and money, and eventually result in failure of the program. Selecting the right target therefore remains one of the most important decisions and investments that companies can make in these early stages. “At AstraZeneca, we spend a lot of time undertaking target validation and evaluation to make sure that we have the right target,” explained Catherine Bardelle. “After that, we define the best approach to find hits relevant for the target that we are pursuing.”

Companies are increasingly running multiple hit-finding strategies to maximize the likelihood of success. By combining different approaches, such as cellular imaging, affinity platforms, or virtual screening researchers gain more insights and confidence for the decision-making process. However, deciding which platform to choose is a fundamental question in the process. In a recent survey,1 physiological relevance was rated as the most important criteria by nine out of ten of the top 20 pharma companies interviewed, yet the definition for this requirement varied from one company to the next.

“It is important to make a distinction between disease relevance and physiological relevance,” noted Bardelle. “Disease relevance is about determining the right target; for instance, in oncology if the mutant form of the target is important for disease, you want to ensure you are working with that form and not the wild type. The physiological relevance relates to the right tissue, the right cellular background, the right timepoint, the right turnover, and the right level of expression.”

Deciding which platform to use is therefore a fundamental question for the HTS team at AstraZeneca. “We can run complex assays, but we would then be potentially restricting the number of compounds that we can screen,” explained Bardelle. “If we are unable to perform a complex assay at large scale, screening around two million compounds, we may pick a suitable less complex alternative. Physiological relevance is always at the forefront of this decision – we do not compromise on this and will always choose an approach that is relevant.”

The decision of which assay to run is not a choice that the HTS team makes alone; they will always discuss the screening cascade in depth with the rest of the project team. “If we decide that we are unable to do a particularly complex approach, such as using Peripheral Blood Mononuclear Cells (PBMCs) as primary cells, but we can do a simplified version for HTS at scale, for example using a THP-1 human cell line, we always make sure everybody is aligned with the decision and that the approach is relevant,” said Elizabeth Mouchet. “Further down the screening cascade, when we aren’t screening at scale, we aim to build in assays where we can add the physiological relevance. A key criterion for us is scalability for HTS – if we decide to run two million compounds, we also have to consider the cost and feasibility.”

Developments in the Field

Over the last five years, the field of HTS has seen notable developments in technology and processes fundamental to successful drug discovery, including artificial intelligence (AI), machine learning, and CRISPR. “We are now seeing AI and machine learning being applied across all stages of drug discovery,” said Tiziana Monteverde. “Not only can it help us identify the target, it can provide a stronger correlation between a target and disease. In the world of high-content imaging and cell painting, AI and machine learning can help us understand the huge amounts of data produced, which we can then learn from and make predictions. We are now routinely performing two million compound screens using cellular imaging, and advances in data analysis techniques aid interpretation of this complex data.”

She also envisages that advances in machine learning may, in the future, help researchers identify the subsets to screen from the larger library with the ‘right’ pharmacophores and physicochemical properties and reduce the need to screen the entire library. “It might allow us to cherry pick compounds or reduce the dimension of our libraries,” she said. High-content screening (HCS) also helps researchers to address toxicity issues, which are a primary cause of compound attrition, earlier in the screening cascade.

Recently, the ability to image at scale has enabled researchers to gain critical insights into the effects of compounds on cells. “Phenotypic approaches are now much more feasible with cell imaging and allow us to address some of the physiological relevance questions,” said Mouchet. “However, there are often data analysis challenges and I think advances in machine learning will be crucial for multiplexing endpoints and extracting the complex data produced from these approaches.” Safety is another aspect that is now being introduced earlier in the screening cascade. “Safety is part of the process from the start and therefore we will often stop a project much earlier than we used to due to a safety issue,” said Bardelle, adding that this is usually introduced post HTS in the cascade.

Advances in CRISPR technology, which has emerged as a scalable approach to interrogate gene function, has also greatly facilitated early drug discovery. “CRISPR has allowed us to engineer cells and use more relevant cell assays in our screening activities,” said Monteverde. “It has had an enormous impact on research and there are a huge number of applications for this technology; for example, during the pandemic CRISPR was applied as a new method for COVID testing.”

Even with all the given recent improvements in cell line engineering, the process of cell-line generation remains a major bottleneck. She also noted the trend towards label free, which can significantly reduce assay costs. “The growing application of label free and mass spectrometry also means that we can generate assays that are not only cheaper but also specific and sensitive,” said Monteverde. Label free is allowing researchers to tackle more obscure targets.

The Challenge of Complexity

While pharma may be embracing more complex and physiologically relevant assays, numerous challenges remain in this area of early drug discovery, particularly relating to scale up. “One of the main pressures for us is that we need to move rapidly whilst providing the right data,” said Carolyn Blackett. “If we are going to run a screen that isn’t in 1536-well format we have to consider what impact that has and the time it takes. This doesn’t mean that it doesn’t happen, but it biases us towards technologies that can be used at scale as our primary approach.”

When it comes to implementing complex cell models, such as stem cells, 2D/3D, primary cells, or co-cultured cells, challenges often arise. “We have examples where we have implemented these approaches, but we have had to scale back in terms of the number of compounds and go for a targeted screening approach,” said Mouchet, adding that machine learning will likely cut down the number of compounds that are screened in the future. “We need to show that it adds value at HTS scale – that is one of the key criteria for us.” In this context, selecting the right screening model, for example primary cells versus cell lines, may be influenced by user safety considerations. The primary cells are not always well characterized, and the team would need to mitigate these risks to be able to operate at large scale.

Blackett added that approaches have shifted over the last few years and the landscape of tools and technologies that work at the scale they require, or somewhere approaching that scale, is continuing to evolve. “We are still learning when you need complexity and where you need it in the cascade,” she said. “We have to ask ourselves whether to do something relatively straightforward to get that first cut of information and then move into more complex models, or whether to use more complex models upfront but accept that we may not do this at large scale or that it might take longer.”

Maintaining cells in culture for long periods of time can also prove challenging for researchers working with more complex models, as does automating each process. “Even though they are probably more physiologically relevant it has been challenging for us to scale up these complex cell models,” explained Mouchet. “With automation, there are many transfers required for each process and we have issues with the plates and robots being able to handle them.”

Still, the team has successfully implemented CETSA (Cellular Thermal Shift Assay) in a recent collaboration project with Pelago and PerkinElmer, which Mouchet notes has allowed them to perform more physiological relevant assays and has been particularly useful for high-value targets. “CETSA has been totally enabling for us; however, it has huge cell requirements – for a typical cell assay you are looking at 1,000-5,000 cells per well but with CETSA you require approximately 25,000 cells per well. Straight away that gives you the challenge of how you are going to have sufficient cells available to run the assay at scale.”

The Landscape of the Future

Looking forward, the team predicts that label free, imaging, and AI will continue to evolve and further facilitate early drug discovery. “Even in the last couple of years, label free has shown great potential at our scale,” enthused Blackett. “Proving that this has value enables you to put investment into this approach and introduce technologies that allow you to do it.” Mouchet added that the potential of imaging and using AI and machine learning for data analysis is an area she expects to see evolve. “Exploring the data from images and reusing datasets is something I’m excited to see and I think we will learn a lot from these images,” she said.

Bardelle foresees multiplexing becoming more of a reality, stating that multiplexing at the cellular level would be extremely enabling. “If we could multiplex more and have even more readouts that would be fantastic for us,” she said. She also believes that AI analysis will make a significant impact. “It is currently being used on enzyme assays, but I would like to see it at the cellular level in the future.” Estimations are up to 80% of the human proteome cannot be addressed by current therapeutic strategies. A promising approach, targeted protein degradation, using PROTACs or molecular glues can help to address the current undruggable proteome and open up new classes of difficult targets for novel therapies.

Although there are numerous applications and technologies that the team would like to see introduced into the cascade, there are various factors preventing them becoming a reality. “We can always dream of what we want, but most approaches are not available at the moment,” said Bardelle. “For example, now that everyone is moving into new fields and working with more obscure targets, we are going to need different types of reagents in order to pursue those fields. The limitation for novel targets is mainly due to the lack of appropriate antibodies, tools or substrates.

We also need to fully understand how complex we need to be at the start of a primary HTS.” In this context, the team discussed the potential for a new type of compound library, whereby the library would be re-designed based on the phenotypic and/or toxic profiling using cell painting, and not based only on their chemical properties. Here, they believe that HCS and AI have a key role to play in these activities.

Physiological relevance will always fit into AstraZeneca’s 5R framework,2 in which R&D decision-making is focused on five technical determinants – the right target, right tissue, right safety, right patient, and right commercial potential. “I think we also need to consider the right modality as an additional ‘R’ that could be added to the framework,” suggested Bardelle. “COVID has shown us that it is great to have a variety of vaccines available from different modalities because they all have advantages and disadvantages. Commercial strategies in the future may be impacted by different modalities like nucleotides or small molecules versus peptides/antibodies. Therefore, I think that is something that will influence the 5Rs in the future.”

Blackett concluded to say that HTS is always a compromise versus scalability, cost, and feasibility of the assay. “As much as we would always like to run the most disease-relevant assay, the purpose of HTS is to identify initial hits with the desired mechanism of action, and understanding this at the primary screening stage allows us to make an informed choice of assay approach, to give balance between complexity and scale, and hence speed and cost.”

References

  1. Top 20 Pharma Interviews and Insights: Drug Discovery for Small Molecule [Internet]. PerkinElmer. 2021 [cited 21 May 2021]. Available from: https://www.perkinelmer.com/library/WHP-top-20-pharma-interviews-and-insights-drug-discovery-for-small-molecule.html
  2. organ P, Brown D, Lennard S, Anderton M, Barrett J, Eriksson U et al. Impact of a five-dimensional framework on R&D productivity at AstraZeneca. Nature Reviews Drug Discovery. 2018;17(3):167-181.