Know your target, know your molecule
- Bar color indicates the isoform type and bar size the percentage of the isoform present in the specified tissue. All isoforms annotated contain the catalytic domain. Long forms contain unique N-terminal regions as well as both the upstream conserved region 1 (UCR1) and UCR2 domains; the short form lacks UCR1; and the supershort form lacks UCR1 and has a truncated UCR2 domain. Data was obtained from GTEx (ref. 8) and analyzed at Pfizer. ACC, anterior cingulate cortex; sk muscle, skeletal muscle.
The pharmaceutical industry continues to experience significant attrition of drug candidates during phase 2 proof-of-concept clinical studies. We describe some questions about the characteristics of protein targets and small-molecule drugs that may be important to consider in drug-discovery projects and could improve prospects for future clinical success.
The issue of pharmaceutical R&D attrition has attracted significant interest in recent years. We have argued that the biggest opportunity to enhance overall R&D productivity is to focus on drug candidate survival through phase 2 and the selection of targets for which there is the highest confidence of relevance to human disease1, 2. We have also highlighted the role that chemical biology can play in preclinical target validation through generation of high-quality chemical probes3 and have described the following four ‘pillars’ as a framework for target validation using such probes:
- exposure at site of action;
- demonstration of target occupancy;
- proof of mechanistic pharmacology and
- the use of disease-relevant phenotypic assays2, 3.
This framework can also be used to assess whether a mechanism has been truly tested in the clinic, and establishing the first three of these pillars in human studies has been linked to clinical success4. In this Commentary we build on these themes and explore some key questions to consider regarding the characteristics of biomedical targets and the small molecules that modulate their function. We discuss why we believe the answers to these questions are important in drug discovery and highlight chemical biology techniques that can be used to help provide insight on the questions posed. It is our belief that such knowledge may help increase phase 2 success by aiding the identification of novel drug candidates that are more likely to deliver efficacy and safety in the clinic.
Over recent years, the application of chemical biology has become well embedded within drug discovery at Pfizer. As part of our broader chemical biology program, Pfizer has established a set of questions (Table 1) as a tool to help project teams fully consider the characteristics of their biological targets and small-molecule leads. These questions have also served as a framework for knowledge sharing across the company on how chemical biology approaches have been applied to advance projects in the portfolio. We highlight these questions in tabular form here in the hope that they can serve as a useful resource for others. In the discussion below, we explore a number of these questions in more detail and share some examples from the published literature of their relevance to drug discovery. However, it is important to stress that we are not proposing a set of rules or a checklist that should be addressed for all target-based drug-discovery programs. Instead, we hope that these questions can simply serve as a starting point for conversations within drug-discovery teams. Moreover, it is inevitable that there are many other questions associated with targets and molecular pharmacology that we have not considered here but could be similarly relevant to drug discovery. Alongside the publication of this article, we have thus included all the questions listed here (Table 1) and additional detail and references in the Chemical Biology Wikipedia page (http://en.wikipedia.org/wiki/Chemical_biology). We invite readers to add their contributions to this entry in the hope that it will develop into a useful resource for the chemical biology and drug-discovery community.
What is my target? This fundamental question is not always sufficiently addressed early in a program, particularly if chemical hits are identified via mechanistic or phenotypic screens. Similarly, incomplete annotation of the pharmacology of small molecules identified through target-based screening may lead to erroneous conclusions around target-phenotype relationships. Ensuring that molecular targets are well-defined enables the application of target-based drug discovery to mechanisms discovered from phenotypic screens and accelerates drug development. Key concepts for target identification and validation include (i) the importance of choosing a relevant cell and phenotype with which to screen2; (ii) the use of complementary genetic techniques involving RNA interference, clustered regularly interspaced short palindromic repeats (CRISPR)-Cas systems, transcription activator–like effector nucleases (TALENs) and zinc-finger nucleases (ZFN); (iii) generation of cell lines expressing mutated target proteins; (iv) the use of chemogenomic chemical libraries that are annotated for the targets to which they are known to interact; and (v) the deconvolution of hits from phenotypic screens to reveal the molecular targets.
Target deconvolution of hits from a phenotypic screen is a rapidly evolving field that utilizes approaches from a wide range of disciplines, such as MS, computational science, medicinal chemistry, genetics, biophysics and molecular and cellular biology. A recent illustrative example of target deconvolution was the identification of the enzyme MTH1 as a potential anticancer target of the (S) form of crizotinib5. That study used affinity-enrichment of proteins in the absence or presence of the soluble competitor drug and detected binding proteins through MS proteomics. This work also provides an illustration of the role stereochemistry can have on drug selectivity (the marketed (R)-form of crizotinib is an anaplastic lymphoma kinase (ALK) inhibitor) and highlights the opportunity to utilize enantiomer pairs to better power target-identification efforts.
Another important approach to target validation is to generate a catalytically dead target protein to tease apart the role of the enzymatic and scaffolding functions. This method is often used in kinase validation experiments where the pharmacological performance of a putative inhibitor is tested in cells with overexpressed wild-type protein or its kinase-dead mutant. This approach was used recently to confirm DCLK1 as the functional antiproliferative target of a previously developed leucine-rich repeat kinase 2 (LRRK2) inhibitor for Parkinson’s disease6. Target validation in the kinase arena has also been enabled via the use of analog-sensitive kinase technology7. This elegant approach allows the specific inhibition of a selected kinase in a cellular (or in vivo) model of disease through use of a catalytically active mutant of the kinase that is uniquely sensitive to ATP-site inhibitor analogs that do not inhibit wild-type kinases.
Does my target exist in multiple forms, and does this vary across tissues and species? This question reflects an important and often overlooked biological reality that can have dramatic consequences on the success of a drug-discovery effort. To facilitate analysis of target expression in various tissues, the Genotype-Tissue Expression (GTEx) project has generated a database of mRNA expression levels across tissues8. Here we illustrate the use of this resource (Fig. 1) by highlighting how the phosphodiesterase PDE4B is expressed in various forms across human tissues. Awareness of these different forms has enabled researchers to develop a selective inhibitor that targets PDE4B in the brain by taking advantage of differences in sequences in the long and short forms outside the active site9. It is clear that a protein’s splice variants, tissue expression profiles and interspecies differences are important considerations in target-validation studies. This knowledge can also inform the selection of the appropriate assays and models to use within a drug-discovery screening cascade.
What is the endogenous ligand for my target and its concentration in the diseased state? Identifying the endogenous ligand, whether it is a ligand for a receptor or a substrate for an enzyme, allows one to understand what biological pathways are being modulated (one of the key needs for establishing a physiologically relevant assay) and to consider what will happen if concentrations of the endogenous ligand are increased through inhibition of the desired target. Substrate concentration is also often central to translation from biochemical assays to functional cell assays in enzyme inhibitor projects (the ‘biochemical efficiency’). An illustration of this can be seen in the inhibition of kinases that have a low Km for ATP. If a project uses a biological enzyme assay with an ATP concentration at Km for the kinase, a dramatic shift in potency can occur when the inhibitor is tested in a cellular context, where the ATP concentration is 1–2 mM. This phenomenon was recently demonstrated for the Janus kinase (JAK) family10. JAK1 and JAK3 also nicely illustrate another issue around perceived selectivity, where an ATP-competitive inhibitor of JAK3 that is tenfold more selective than JAK1 when screened at Km (JAK1 Km for ATP is 40 mM, JAK3 is 4 mM) loses this selectivity window entirely when tested in a cellular assay10. For new therapeutic targets, substrate identification and concentration determination in relevant cells and tissues have been facilitated by advances in quantitative metabolomics, peptidomics and proteomics. For instance, clickable chemical reporters enable application of chemoproteomic methods to substrate identification, as shown recently for protein lipidation11.
Is my target post-translationally modified? Post-translational modifications (PTMs) can significantly change the activity, conformation, location and binding partners of a protein and can vary between healthy and diseased tissue. Using the most appropriate form of a target protein, including relevant PTMs, will provide a better chance for translation from biochemical to cellular assays. The best-studied PTM is phosphorylation; however, there are many others, including glycosylation, citrullination, ubiquitination, proteolytic cleavage, palmitoylation, hydroxylation, methylation and acetylation—to name a few. Besides these PTMs, reactive oxygen and nitrogen species, present in many inflammatory diseases, are known to post-translationally modify amino acid residues, often leading to aberrant biochemical signaling and an exacerbation of the pathology. These changes may also affect drug binding to a protein target, particularly if they occur in the binding site of the protein. For instance, many kinases contain a cysteine residue in the binding site that may, under oxidative conditions in the cell, become oxidized to sulfenic acid. The result is redox regulation of the enzyme, as was shown recently for epidermal growth factor receptor (EGFR)12. Dimedone-based probes that chemoselectively react with sulfenic acid residues can be used to ‘canvass’ proteomes for the presence of these modifications.
What is the turnover rate of my target, and is this affected by my compound? The turnover rate of a protein target can affect the options available to a drug-discovery program for therapeutic modality selection. For instance, if a protein has a long half-life in a cell, then developing a covalent inhibitor to this target may be advantageous, potentially allowing lower or less frequent dosing schemes. This strategy was used for Bruton’s tyrosine kinase (BTK), in part owing to its slow turnover13. It can also be important to determine the turnover of mutant proteins, which are particularly common in cancers. For example, some mutations in EGFR result in increased protein half-lives14, probably owing to a concomitant decrease in binding to the Cbl ubiquitin ligase responsible for regulating the degradation of EGFR. This increased half-life could result in a need to change the projected dose or frequency. In addition to mutations altering the turnover rate of a protein, small-molecule drugs can also change the turnover and thus the amount of protein in the cell. For instance, many kinases are known to be clients of the heat-shock protein 90 (Hsp90)–Cdc37 chaperone system, and recent studies have found that some ATP-competitive inhibitors disrupt the ability of Cdc37 to bind to the target kinase and recruit it to Hsp90 (ref. 15). This disruption results in decreased stability of the kinase in cells treated with inhibitor, which presents an interesting potential modality for kinase inhibition. In another approach, ligands were labeled with hydrophobic tags that triggered target protein degradation16. The tag is believed to co-opt protein degradation machinery to the target by mimicking internal hydrophobic amino acids that become exposed to cellular chaperones during protein denaturation.
What potential off-targets are most closely related to my target sequence and function? Knowing the shared features—such as sequence homology or similar pocket shape—of other targets can inform screen sequence design to ensure that the selectivity toward other proteins is understood and evaluated early in a program. It is important to consider targets that are closely related not only in terms of binding site sequence but also in terms of the shape of the binding site. This is often assessed through computational biology and structural bioinformatic analyses of protein sequences and binding pocket topology. Additionally, chemoinformatic approaches using in silico tools to predict potential off-targets on the basis of the structural similarity between a novel small molecule and previously known compounds with different pharmacology are also useful.
What targets and off-targets does my molecule bind? Knowing the biological targets that a small molecule binds is a key step in the characterization of its molecular pharmacology. Target identification enables an understanding of the compound’s mechanism of action and identification of the biological pathways modulated, and it provides the initial rationale for a target-based drug-discovery project. Poor characterization of target binding specificity has led to much wasted effort in drug discovery, with a number of chemical scaffolds having now achieved notoriety owing to their nonspecific assay effects (‘pan-assay interference compounds’, or ‘PAINS’)17.
Numerous techniques to identify binding targets of small molecules have been described in the literature. Activity-based probes (ABPs) that react with one class of biological target can also be used to identify the on- and off-targets of small molecules (activity-based protein profiling (ABPP)). The classic ABP examples are nucleotide acyl phosphate probes (for example, KiNativ) that react chemoselectively with the conserved lysines of >80% of all known kinases18. These probes contain biotin tags that allow for enrichment and identification of all labeled kinases in a cellular lysate. ABPP is a powerful technique that has seen increased use over recent years (Fig. 2). Unfortunately, however, many ABPs are not cell permeable and therefore can be used only on cell lysates, which may not represent the biology of an intact cell. Cell-permeable probes are urgently required to enable protein profiling in living cells. Additionally, although certain protein classes can be profiled using a plethora of probes (kinases and proteases in particular), there are few, if any, ways to assess activity and drug occupancy of other families (such as G protein–coupled receptors, ion channels or solute carriers); therefore, the development of probes in this area should also be a focus of the chemical biology community. ABPs, derived from covalent kinase inhibitors, have recently been used to develop a ‘road map’ to study the targets of kinase covalent inhibitors in living cells19. In the cellular environment, both kinase and non-kinase targets were identified and several electrophiles (unsubstituted and substituted acrylamides) were compared. Using this technology, the on- and off-target activity of the BTK inhibitor ibrutinib (Imbruvica) was characterized in a concentration- and time-dependent manner in cancer cells (Fig. 2).
Proteomics is also one of the main techniques to establish compound binding partners. For example, chemoproteomics was used to assess the multiple binding partners of histone deacetylase inhibitors in protein complexes scaffolded by ELM-SANT domain subunits20. Thermal proteomics, in which researchers look for proteins that show increased thermal stability at higher temperatures owing to compound binding, has been used to identify off-targets of several kinase inhibitors, including the BRAF inhibitor vemurafenib21. In addition, affinity capture, in which a chemical probe is tethered to a solid support, was used to identify cereblon (CRBN) as a thalidomide-binding protein22. Binding to CRBN and inhibition of the associated ubiquitin ligase activity was found to be the cause of the teratogenicity that resulted in deformities in infants whose mothers had used thalidomide during pregnancy.
What is the tissue distribution of my molecule? The tissue distribution of a compound is crucial to defining exposure at site of action (pillar 1)3, 4, as plasma concentrations often do not reflect the actual degree of exposure in the target tissue of interest. Asymmetric tissue distribution can also be used to enhance the therapeutic index of a drug if it localizes in the target tissue. For example, quantitative MS was recently used to determine the concentrations of the chemotherapeutic agent YM155 in various cancer cells23. It was discovered that the solute carrier SLC35F2 was responsible for the uptake of YM155 into cells, leading to increased chemotherapeutic DNA damage in cancer cells expressing this transporter, as intracellular concentrations of YM155 were elevated.
In addition to the use of MS to quantify exposure in tissue slices, simple exposure measurements from pharmacokinetic experiments can also effectively quantify drug concentrations.
What are the functional consequences to my target when my molecule binds?
Once binding targets have been identified, the functional activity produced by this binding can be assessed in a number of ways. Although quinazoline covalent inhibitors are irreversible inhibitors of kinases, reversible interactions between these compounds and EGFR have been shown to correlate with cellular potency24. This nonobvious observation provides a framework for the future design of covalent EGFR inhibitors with low to moderate electrophilic reactivity that may show minimal promiscuity. Another nonobvious observation came from the use of small hairpin RNA (shRNA)-based gene silencing combined with immunoblotting to show that GDC-0879 and PLX4720, which are inhibitors of the serine/threonine kinase BRAF, block mitogen-activated protein kinase (MAPK) signaling in BRAF(V600E) tumors, whereas they activate the RAF-MEK-ERK signaling pathway in tumors expressing mutant KRAS and wild-type RAS-RAF25. This latter effect is due to conformational effects of these compounds on the RAF kinase domain. This paradoxical pathway activation through use of an inhibitor highlights how the molecular pharmacology of a compound can be highly dependent on the cellular context, a subtlety that is often not appreciated.
How much target occupancy do I need to drive my relevant biological phenotype?
Target occupancy allows one to develop confidence in target engagement (pillar 2)3, 4, to potentially link occupancy to on-target mechanistic pharmacology (pillar 3)3, 4 and can be a key enabler to help define efficacious drug concentration. One such study described the use of a clickable covalent probe of fatty acid amide hydrolase (FAAH) to relate target engagement to anandamide elevation and efficacy in models of rodent inflammatory and neuropathic pain26.
Quantification of target occupancy can also be used to validate the relevance of a drug target identified from phenotypic screening.
A clickable covalent probe of the mRNA decapping enzyme DcpS, which used a sulfonyl fluoride warhead to target a reactive tyrosine in the binding site of the enzyme, enabled the assessment of DcpS target engagement of a diaminoquinazoline inhibitor (developed from a phenotypic screen for the treatment of spinal muscular atrophy)27.
This work established a correlation between target coverage and phenotypic effects in living human primary cells, which was not possible using a standard DcpS biochemical assay.
The ultimate translational pharmacology experiment would allow for the quantification of all four ‘pillars’ in a single study. For instance, for therapeutic targets in blood, a single blood sample could theoretically enable assessment of free drug concentrations (through LC/MS), target engagement (through ABPP, for example), functional pharmacology and phenotype perturbation.
Chemical biology sits at the intersection of many disciplines, including medicinal chemistry, cellular and molecular biology, functional genomics, biophysics, computational sciences and systems biology.
Over the past 10 years chemical biology, once principally an academic pursuit, has emerged as a central component of modern industrial drug discovery.
After years of decreasing productivity, the pharmaceutical industry is also now in a period of renaissance. Indeed, the rate of new drug approvals has recently been trending upwards, with 41 US Food and Drug Administration approvals in 2014.
Despite this progress, phase 2 attrition is still far too high, and the overall return on investment for biomedical research needs to be improved. Striving for the deepest possible knowledge about the nature of the targets worked on in drug discovery, and the molecular pharmacology of the small molecules that modulate their function, can help identify novel drug candidates that are more likely to deliver efficacy and safety in the clinic. In this context, it is our hope that the ‘Know Your Target, Know Your Molecule’ questions introduced here can provide a useful starting point for discussions within drug-discovery teams. We also hope that readers with further questions or illustrative examples on this theme will share these by making additions to the Chemical Biology Wikipedia page entry that we have made alongside this Commentary (http://en.wikipedia.org/wiki/Chemical_biology). The power of chemical biology approaches in helping tackle these key questions is already clear.
Over the next decade we anticipate that chemical biology will become even further embedded within the pharmaceutical industry as a core discipline for all drug-discovery projects.
- Nat. Chem. Biol. 7, 335–339 (2011).
- Nat. Rev. Drug Discov. 14, 149–150 (2015). et al.
- Nat. Chem. Biol. 9, 195–199 (2013). , &
- Drug Discov. Today 17, 419–424 (2012). et al.
- Nature 508, 222–227 (2014). et al.
- Mol. Cancer 13, 103 (2014). et al.
- J. Am. Chem. Soc. 135, 18153–18159 (2013). et al.
- The Genotype-Tissue Expression (GTEx) project. Nat. Genet. 45, 580–585 (2013).
- Cell. Signal. 26, 657–663 (2014). , &
- ACS Chem. Biol. 9, 1552–1558 (2014). et al.
- Curr. Opin. Chem. Biol. 24, 48–57 (2015). , , , &
- Chem. Rev. 113, 4633–4679 (2013). &
- Proc. Natl. Acad. Sci. USA 107, 13075–13080 (2010). et al.
- Cancer Res. 67, 7695–7702 (2007). et al.
- Nat. Chem. Biol. 9, 307–312 (2013). et al.
- Nature 487, 308–309 (2012). &
- Nature 513, 481–483 (2014). &
- Biochemistry 46, 350–358 (2007). et al.
- Nat. Chem. Biol. 10, 760–767 (2014). et al.
- Nat. Biotechnol. 29, 255–265 (2011). et al.
- Science 346, 1255784 (3 October 2014). et al.
- Science 327, 1345–1350 (2010). et al.
- Nat. Chem. Biol. 10, 768–773 (2014). et al.
- Proc. Natl. Acad. Sci. USA 111, 173–178 (2014). et al.
- Nature 464, 431–435 (2010). et al.
- J. Pharmacol. Exp. Ther. 338, 114–124 (2011). et al.
- ACS Chem. Biol. 10, 1094–1098 (2015). et al.