Identifying functional molecules with the potential to be developed into new therapeutics is often the sole aim of phenotypic drug discovery.
However, the approach also provides valuable opportunities to uncover previously unknown, disease-specific drug receptors that can then be exploited through target driven means. Although target deconvolution has a reputation as a major bottleneck in phenotypic drug development, advances allowing for rapid identification of specific receptors for both biologics and small molecules are increasing the pace of phenotypic drug discovery while also providing access to the significant IP generated by the discovery of novel drug targets.
Phenotypic drug discovery has enjoyed a resurgence of interest over recent years with its relative advantages and disadvantages compared to target-based methods being extensively discussed and debated. Traditionally, phenotypic drug discovery focused on the use of small molecule effector libraries to induce a desired phenotype. However, the principles of phenotypic discovery are now widely applied in the field of biologics, with several established and newly-emerging strategies now being employed for antibody based phenotypic discovery. These approaches can range from: evoking an anti-tumour antibody response through immunisation of animals with tumour tissue (as opposed to single antigen immunisation), through to differential panning of normal/disease cells with antibody phage libraries to select molecules that meet disease-relevant functional endpoints (1,2).
At what point is the target relevant?
By following a ‘hypothesis-free’, step-wise screening approach, it is only once a desired phenotype has been induced does the focus switch to attempting to understand the underlying mechanism of action of the phenotypic molecule. Success at this stage can not only uncover novel receptor targets but can also validate existing targets. However, with only a 10-15% chance of successfully deconvoluting the target using conventional methodologies, coupled with the threat of the key finding being obscured by false positive data, it is little surprise that the exercise is often approached with trepidation.
Recent developments in cDNA expression-based target deconvolution have largely overcome that hurdle, boosting the chances of finding specific receptor targets to around 70% for compatible molecules and drastically reducing false positives. Reliable and efficient target deconvolution, in turn, opens up further possibilities for phenotypic discovery. For example, screening the antibodies involved in the human immune response to cancer – or those present in patients that have displayed a good response to anti-cancer treatment – is now being used to identify novel and important anticancer drug targets.
Due to the nature of antibodies and other large molecules and the way that they have been selected (for example, by panning for binding to cancer cells) there is a high likelihood that the critical target receptors for newly-discovered phenotypic biologics will be cell surface proteins. Although there are various approaches for target deconvolution, the most challenging and poorly-served area is the human cell surface protein. Standard protein arrays involve generating proteins (often in yeast or insect cells), purifying these and then spotting them on to substrates to then probe with the molecule of interest. This provides a high throughput and relatively low-cost way of screening for targets. However, the arrayed proteins are not localised in a plasma membrane, are not likely to be appropriately folded (and are therefore not presenting conformational epitopes for binding) and have not been subject to the normal post-translational modifications that would occur in human cells. This results in low rates of success in identifying biologically-relevant ligand-receptor interactions.
The other standard approach to target deconvolution is the use of pull-down assays coupled with mass spectrometry. This provides a source of proteins from a relevant cell or tissue type, however, solubilising these proteins from membranes will have its own disruptive effect. In addition, the screening process can be time- and resource-intensive, again, with limited rates of success.
Using physiologically-relevant receptors as prey for phenotypic ‘baits’
The development of cDNA expression arrays now provides a powerful way to deconvolute the targets of both phenotypic antibodies and small molecules by focusing exclusively on cell surface, plasma membrane targets. Membrane protein-encoding cDNA expression vectors in a lipid complex are arrayed on to specialised slides. Human cells are then seeded on the slides. Cells that grow directly over each of the vector ‘spots’ become reversetransfected resulting in over-expression of the membrane protein in a cellular context in situ.
A phenotypic molecule or other ligand of interest is added to the slides and allowed to bind to its over-expressed target. The primary receptor and any potential secondary target binding of the phenotypic molecule is detected using fluorescence imaging, typically after the application of a fluorescent secondary antibody. Radiolabelled small molecules can also be screened using the arrays, with target binding detected by phosphorimaging.
Currently, the largest library of cDNA expression vectors that encode full-length, untagged, unfused human plasma membrane proteins for receptor identification exceeds 4,500 clones. This is estimated to cover around 75% of the plasma membrane proteome. Importantly, there is no bias within the library towards any particular sub-class of human membrane protein. As a result there is representation from all major sub-classes including GPCRs, receptor kinases, Ig superfamily receptors, ion channels and GPI-anchor proteins, among others. As the technology uses both human proteins and human cells it provides the necessary environment for proteins to be appropriately localised in the plasma membrane in the correct conformation, having also been subject to the normal post-translational modifications which can be critical in mediating interactions (3). Use of the highly-transfectable human HEK293 cell line results in good levels of over-expression, providing favourable ratios for exogenous:endogenous expression that allows background binding to untransfected cells not to cause an issue for the vast majority of molecules screened.
The utility of cell microarrays for target deconvolution
A cDNA expression microarray target deconvolution screen typically involves three distinct phases of activity. The first phase ensures that there is low or negligible binding of the phenotypic molecule to the host cells (in this case, untransfected HEK293 cells). If levels of endogenous binding are high enough to compete with the interaction signal then the screen becomes unfeasible. However, more than 90% of molecules that are tested have low enough levels of background binding to be deemed compatible with the technology.
Phenotypic molecules with sufficiently low background binding are progressed into the second phase where they are screened for binding against cell microarrays that are over-expressing the full library of membrane proteins. This identifies primary ‘hits’ which need to be confirmed as specific and reproducible in the final phase of the experiment. Primary hits are identified by their position on the microarray and then sequenced to confirm. Expression vectors encoding these primary hits, plus positive and negative controls are spotted on to custom slides and the HEK293s cells are reverse transfected. Binding of the phenotypic molecule is then confirmed and non-specific hits weeded out using appropriate controls and competition experiments. Each stage of the cell microarray screening involves duplication at multiple levels, for example each vector is spotted in duplicate on each slide and each phase is tested on duplicate slides.
This rigorous procedure maximises the chance of identifying specific, reproducible membrane interactor(s) for the test molecule and generates a very low incidence of false positive results. This makes it relatively straight-forward to follow up and validate the cell microarray results using orthogonal approaches.
Can knowing the target alter the route taken?
Recently a study by Sandercock et al1 demonstrated how powerful the combination of effective phenotypic screening coupled with efficient target deconvolution could be in opening up new avenues for drug development. The group focused on ScFv antibodies and designed ankyrin repeat proteins (DARPins), isolating functional candidates by phage display selection against primary cells from non-small cell lung cancer patients. Cells were grown in multiple formats, including 3-D cultures, and were monitored to detect the anti-proliferative and pro-apoptotic activity of each molecule. Multiple antibodies that demonstrated a desired activity were isolated and the specific antigens that they targeted were identified using cDNA-expression cell microarray technology. Notably, a subset of antibodies that had distinct phenotypic effects all bound to CUB-domain containing protein 1 (CDCP1) – a cell surface transmembrane protein that is upregulated in many tumour cells and is associated with invasive and metastatic phenotypes. These strong functional effects led the group to test an anti-CDCP1 IgG antibody from their panel for the potential to inhibit tumour growth in vivo using a mouse xenograft model. Although the antibody treatment alone was not efficacious, significant enhancement of tumour growth inhibition and increased survival times were observed when the antibody was co-administered with cisplatin, compared to cisplatin alone at the same dose. This identified CDCP1 as a potential target for cisplatin combination therapy.
cDNA expression microarray technology has been used in other phenotypic screening programmes to identify new, critical receptors. A study set up to search for cancer immunotherapy targets utilised phage display using DARPins to select antibody-like molecules that showed preferential binding to human regulatory T (Treg) cells. The cDNA expression target deconvolution that followed rapidly identified TNFR2 as the primary receptor for all molecules screened. Subsequent tests on specific molecules that were categorised as TNFR2agonists resulted in tumour inhibition in mouse models, highlighting the importance of TNFR2 as an immunotherapy target (2).
The recent advances in target deconvolution is providing researchers with the necessary tools to support the development of promising phenotypic molecules while also uncovering novel receptor targets that can provide a real competitive advantage.
Strengths of cDNA microarray technology
The physiological relevance of the system and the vast cDNA library of unmodified (untagged) proteins is the core strength of the cell microarray technology. Numerous target deconvolution projects with industry partners have allowed the rate of success in identifying the membrane target of a compatible phenotypic antibody to be quantified at around 70%. This is significantly higher than the success rates of standard protein arrays and proteomics-based approaches, with results delivered in as little as two weeks.
An additional advantage of cDNA cell microarray screening is that, as well as identifying a primary target, there is the opportunity for concomitant identification of potential off-targets for molecules tested. A low incidence of off-target binding among phenotypic antibodies may be anticipated; however, detecting any potential cross-reactivity at this early stage can provide valuable additional information that can help guide lead selection and optimisation.
The sensitivity of the technology does depend on the level of expression of the target receptor relative to endogenous expression in HEK cells. Although specific target receptors are identified, the technology does not quantitate the affinity of interactions that are observed. However, without any amplification strategies, a known antibody:target interaction with predicted Kd of 10uM has been detected using cell microarrays, indicating the sensitivity of the technology.
Cell microarray screening is extremely versatile and suited to a wide variety of molecules including human and non-human antibodies, ScFvs, DARPins, proteins, peptides and small molecules, among others. Beyond its applications in phenotypic screening, the technology is also being used to great success in identifying the human cell surface receptors of disease relevant orphan ligands (4).
There is a very high likelihood of detecting an interaction with the test ligand among the >75% of membrane proteins that are already represented within cDNA-expression cell microarrays. It is inevitable that, even despite rigorous transfection controls, some receptors may be poorly expressed to the extent where interactions may not be readily detectable. In these cases, strategies to amplify the binding signal or create an avidity gain for the test molecule have been shown to increase the sensitivity of the technology and help to minimise the threat of false negative results.
The library of membrane proteins continues to expand; however, not every membrane protein – nor every protein isoform – is currently represented. In addition, although the over-expression in HEK293 cells provides a native human system, in some cases it may not result in the correct protein conformation, the particular post-translational modifications or the disease-relevant epitopes that would be necessary to facilitate binding in vivo. In some cases – for example, ion channels – ligands may only interact with multi-subunit receptors. Therefore, over-expression of a single subunit may be insufficient to provide a fully functional protein within the system. Despite these theoretical limitations, there is evidence that the human cell expression system can, in some instances, make allowances for single subunits through the interplay with endogenouslyexpressed receptors and the provision of the additional co-factors where necessary. This has been observed in a target deconvolution study involving integrin receptors where the target was identified despite the underlying cDNA vector only encoding one monomer of a heterodimeric receptor. A possible future direction for cDNA expression cell microarrays could be a hypothesis-driven, focused effort on creating multi-subunit arrays through co-transfection of multiple constructs in order to ensure full representation of receptor complexes and multi-subunit entities.
Cell microarrays are very well-suited to screening antibodies and protein ligands using a wide variety of detection systems from direct fluorescent labelling through to using specific secondary antibodies and protein tags (such as His-, Flag-, Fcand biotin, among others). However, small molecules do need to be radiolabelled which can prove impractical in early screening. Development of label-free detection strategies for small molecules is currently under way.
In fewer than 10% of cases, high background binding of the phenotypic molecule to the host HEK293 cells will be observed. In all probability this is caused by high endogenous expression of the target which renders the molecule unsuitable for screening. This issue could potentially be addressed by optimising the technology for alternative cell types.
cDNA expression cell microarrays now provide a very powerful approach for rapidly identifying the membrane targets of phenotypic molecules which has removed the traditional deconvolution bottleneck from phenotypic drug discovery. Due to success rates which far surpass other methodologies, the approach has been widely adopted across the industry and has garnered considerable interest among academic groups.The power of the technology to facilitate phenotypic drug discovery will only increase as libraries continue to expand and provide further coverage of the membrane proteome, incorporating both new membrane proteins as well as additional isoforms of proteins that are already represented.
As phenotypic drug discovery continues to advance, the accompanying target deconvolution efforts will undoubtedly expand our collective understanding of the critical receptors that are implicated in disease processes and that mediate the therapeutic response.
Dr Jim Freeth co-founded Retrogenix in 2008, developing a world-leading technology for identifying both the receptor targets and off-targets of antibodies, proteins, small molecules and viruses. Retrogenix now works with 14-15 top pharmaceutical companies, numerous drug discovery companies and leading academics. Prior to Retrogenix, Jim spent more than 10 years in management within the biotech and pharmaceutical industry. A biologist by training, Jim obtained his PhD at Manchester University, UK in 1997.
Dr Elizabeth Kingsley provides scientific support to the business development activities at Retrogenix, working closely with research groups at global pharmaceutical companies to plan projects aiming to identify the specific cell surface targets of biologics and small molecules. She also works more widely with biotechnology companies on PR communications and marketing strategies. Elizabeth has a PhD in molecular biology from the University of Manchester, UK.
Jo Soden co-founded Retrogenix and was instrumental in developing and commercialising the cell microarray technology. Jo leads a talented team of scientists, overseeing all internal technology R&D activities and managing Retrogenix’s portfolio of commercial projects. She has contributed to numerous peer-reviewed scientific publications and her background includes 17 years of research experience at the University of Manchester, UK, where she also obtained her MPhil.
1 Sandercock, AM et al. Identification of anti-tumour biologics using primary tumour models, 3-D phenotypic screening and image-based multi-parametric profiling. Molecular Cancer 2015; 14:147
2 Williams, G et al. Phenotypic screening reveals TNFR2 as a promising target for cancer immunotherapy. Oncotarget 2016; 7(42): 68278-91.
3 Salanti, A et al. Targeting human cancer by a glycosaminoglycan binding malaria protein. Cancer Cell 2015; 28(4):500-14.
4 Turner, L et al. Severe malaria is associated with parasite binding to endothelial protein C receptor. Nature 2013; 498:502.