CHAPTER 1: Introduction to Chemical Probes Free

The last 10 years have witnessed a rapid rise in enthusiasm for developing chemical probes as reagents in chemical biology research. Defined by Workman and Collins¹  as “small molecules to understand the function of genes and proteins and their role in physiology and pathology”, the utility of chemical probes is without question. Frequently, small-molecule discovery is viewed as solely a drug discovery endeavour; however, chemical probes can provide unique laboratory reagents that fulfil a complementary, and often superior, role in understanding the underlying biology of a particular area.

Over 20 years ago, Stuart Schreiber raised the idea of developing a chemical probe for every gene.²  He closed the perspective article with a call to arms that was well ahead of its time in the 1990s: “The original goal of the human genome project was ‘to sequence every gene’. With that goal within sight, I suggest we consider a new goal for this project, one that can only be realized through the creative use of chemistry, ‘to identify a small molecule partner for every gene product’.” Years later, the field was finally on its way toward considering this idea, to the point that attention was refocused from whether this goal was even feasible to developing criteria for what constitutes a truly high-quality probe.

Small molecules provide temporal, spatial and concentration control that is not currently attainable by genetic modifications. Even better, when used in parallel with gene-based experiments, the use of chemical probes allows researchers to avoid being fooled by the off-target effects of either approach. In the case of genetic manipulation, examples exist where knockout mice often have a different, and sometimes the exact opposite, effect of chemical probes.³  Further, genetic knockouts often result in the strong upregulation of related isoforms, which can confound the analysis of the role of proteins in cellular and organismal phenotypes. Thus, chemical probes serve a truly unique research purpose that cannot be accessed by genetic experiments. A challenge in the field has been to identify and use “good” chemical probes. The adage “garbage in, garbage out” applies here: developing a set of criteria for what constitutes a chemical probe, as opposed to, say, a lead candidate has been a major sources of discussion and debate over the last 10 years.

Historically, within the research community, one often sees interchangeable use of the terms “compound”, “small molecule”, “drug”, “tool”, “lead” and “probe”. What to make of these terms: how will we know if we even have a probe for research? Fortunately, many groups have made considerable intellectual contributions to this area and several key papers published in 2009 and 2010 helped propose what constitutes a high-quality probe.^4–7  Importantly (and sensibly), rather than establishing a set of hard-and-fast rules for probes, which can introduce an unhelpful rigidity to the definition, these authors have suggested sets of guidelines for the properties that probes should, for the most part, fulfil.

An important indication that not all probes are created equal was provided by Oprea and colleagues.⁷  This work evaluated the 64 chemical probes emerging between 2004 and 2008 from the pilot phase of the NIH Molecular Libraries and Imaging (MLI) Initiative. The MLI Initiative funded several high-throughput screening centres, which ran 691 assays assembled from investigators throughout the USA. These assays represented 171 biological targets and 29 phenotypic assays. The authors then assembled a set of 11 experts in drug discovery and chemical biology and asked them to rate their levels of confidence in the 64 probes (on a scale of 0 to 10, with 0 representing the highest confidence). Overall, 16 compounds (25%) were deemed low confidence (score 5–6), an additional 16 compounds (25%) were of medium confidence (score 3–4) and 32 compounds (50%) were considered high confidence (score 0–2). In this study, much of the confidence score appeared to be driven by perceived chemical liabilities. Importantly, this paper lays out the evolution of the definition of chemical probes, going from the low bar of “adequate potency and solubility to be useful … in vitro” to starting to incorporate notions of selectivity, toxicity, mode of action and availability to the community. A later analysis of the 375 total probes developed through 2015 in the Molecular Libraries Program highlighted some of the more sophisticated probes developed and emphasized the breadth and novelty of mechanisms discovered through the Program.⁸ 

Contemporaneous work by Edwards et al.⁵  advocated for open collaboration between academics and industry, in a spirit of “open-access chemical biology”. It was particularly noted that avoidance of duplicative studies and the resulting generation of higher quality probes would be a major outcome of such an effort. Some of the key historical lessons cited in the article that were learned from industry included the supremacy of chemical tractability, the success of clever experimental design over brute force, the importance of well-characterized probes and the involvement of the scientific community. As we shall see later, the fact that Target 2035 was launched only in 2019 is a testament to the scope and challenges of this goal.⁹ 

Another perspective on chemical probe criteria, focusing on protein kinase inhibitors, further helped enhance the stringency, and thus the quality, of probes emerging from studies.⁴  These suggested guidelines included several essential criteria, including (1) that the compound must be tested on a panel of at least 50, but preferably 100, protein kinases; (2) at least two structurally unrelated inhibitors should work in cells; (3) the cellular effects should be observed at a reasonable concentration at which the kinase is inhibited; (4) analogues (when available) should rank-order in the same fashion for biochemical activity and cell potency; and (5) publications must include the structure, otherwise results cannot be replicated by other laboratories. Two other desirable criteria, but of much higher stringency, were that (1) the compound should be inactive in cells containing a drug-resistant mutation of the kinase and (2) the compound should be inactive in cells in which the target and related isoforms are knocked out. As we shall see later, the more recent failure of some notable drug candidates on this last “desirable” criterion has the potential to create havoc in cancer drug development.¹⁰ 

In 2010, Frye wrote about “the art of the chemical probe”, with the clear implication that defining a probe was at that point still an art.⁶  He specified five clear principles for chemical probes to follow; these values helped lay the foundation for later efforts to provide more precise definitions. First, there must be sufficient potency and selectivity to link up biochemical and cellular activity. Second, there should be sufficient mechanistic data on the probe's mechanism of action (MoA) to allow biological interpretation of cellular results. Third, a clear identity of the active species must be in place. Fourth, the probe must be useful and be able to address at least one hypothesis of the target's role in cells. Finally, the probe must be available to academic researchers without restriction. Together, this commentary helped advance the conversation about high-quality chemical probes.

A truly comprehensive treatment of probe quality was advanced in 2010 by Workman and Collins,¹  who proposed “fitness factors” for chemical probes, spanning across chemical, biological and contextual properties. Their paper is an essential resource for those interested in developing or using chemical probes, not only because all of the factors are listed in one place, but also because an extensive rationale for their inclusion is provided for the reader. First, the chemical properties of a probe include that the structure is not only revealed, but also can be reproducibly generated, the compound must be stable and membrane permeable and there should be good solubility, generally >100 µM. Importantly, the compound must not be an aggregator, which, as we shall see later, can interfere with many biochemical and cellular assay readouts.

Workman and Collins¹  went on to discuss the biological features that are important for a high-quality chemical probe. The biochemical potency should be <100 nM, with activity in cells in the range 1–10 µM. Of course, analogues with activity are enormously helpful, but perhaps even more useful are inactive analogues, whose lack of cellular activity helps determine selectivity towards the ostensible target. The appropriate pharmacokinetic profile for testing in vivo efficacy in the appropriate tissue is also an important component, but can take additional time to develop. Similarly, having additional chemotypes that show a correlative relationship between biochemical and cellular activity would provide the best evidence when using a chemical probe, but this is not always possible at an early point in probe development. In terms of selectivity, obtaining a target profile against the class of related proteins is nearly a must, especially in the protein kinase world. Venturing outside the target class is even better, but this feature is considered desirable, not essential. Cheminformatics to analyse the chemotype for other activities is also important. Finally, the context in which the probe will be used is a key consideration. For example, these parameters include understanding the biological hypothesis to be tested, the cellular context of the target and whether complementary genetic methods are available for comparison. Perhaps the most important consideration is compound availability, without restriction and in amounts (15–20 mg) to enable follow-up studies. Together, these guidelines are challenging to achieve, but it is better to lay out the highest quality probe characteristics and try to reach them than to easily attain sub-par goals.

The challenges involved in developing chemical probes was revisited in 2015, when a large consortium reflected on the perils and promise of chemical probes.¹¹  As the definition of a probe was becoming more accepted in the field, the next hurdle to be recognized was the enormous cost and effort required to make a high-quality probe. This challenge underscores the need for academic and industry researchers to work together on these probes. Importantly, the authors distinguished the prioritized properties for probes (e.g. MoA, selectivity, availability) from those for drugs (e.g. physicochemical properties), helping to emphasize the fact that the bar for quality is in some ways higher for a probe. As a cautionary note, the authors also provided examples of low-quality probes, including valproic acid and LiCl (both used at too high a concentration), flavones (promiscuous interference compounds) and resveratrol (an assay artefact).¹²  Finally, this group introduced the Chemical Probes Portal, an online resource for sharing data and reviews of various compounds. More on this topic will be discussed in the last section of this Introduction.

As discussed, the work of Oprea and colleagues⁷  provided the field with a good indication of some of the major challenges posed in developing high-quality probes, with one-quarter of the 64 reported probes from the NIH MLI Initiative programme considered highly dubious. With such a high number of questionable probes, what are the modes of failure? One of the most common shortcomings of low-quality probes is promiscuity. Frequently, a compound may be quite potent towards its intended target, but if it inhibits a whole swath of targets, cellular results cannot be reliably interpreted. This phenomenon has been observed for nearly 20 years in the protein kinase field. Philip Cohen and colleagues published some of the first papers describing non-radioactive profiling of compounds across many kinases.¹³  Their work provided some of the first red flags that a number of commonly used compounds, such as rottlerin, quercetin and LY294002, inhibited other targets, sometimes with greater potency than the ostensible target. Despite this important work, poor tool compounds such as LY294002 are still used in research and result in publications;¹¹  that particular compound returns over 10 000 results in PubMed, including papers that have been published at the time of the preparation of this book. A stronger spotlight must be shone on these kinds of results, so that the continued use of these shoddy probes is halted.

Physicochemical properties comprise another major source of poor-quality probes. Of course, poor solubility and permeability will confound any cell-based research performed on a probe candidate, no matter how potent it is in vitro. Chemical reactivity, such as the inclusion of Michael acceptors, can cause instability and promiscuity. The problematic compounds with which most researchers are familiar are Pan Assay INterfering compoundS or PAINS.^14,15  The chemotypes associated with PAINS can cause false-positive results in many assays, due to covalent reactivity, redox activity and aggregation, among other mechanisms. Aggregation in particular is likely to be the major source of bad compound behaviour in cell-free experiments. An important study found that of 50 approved drugs tested, four demonstrated significant promiscuous, aggregation-dependent enzyme inhibition.¹⁶  On that note, the Editors-in-Chief of several American Chemical Society (ACS) journals sounded an alarm that proper benchmarks need to be used to control for artefactual assay activity.¹⁷  For PAINS themselves, many in silico tools exist to help predict whether compounds might cause a problem. PAINS are typically associated with synthetic compounds; the natural products that show several non-specific activities that cannot be explained by common failure modes of synthetic molecules have been termed invalid metabolic panaceas (IMPs).¹⁸ 

Several cautionary tales are worth exploring here; these stories involve the use of poor-quality probes, false-positive results and lack of reproducibility for probe activities. Curcumin, a natural product found in turmeric, is the poster child of a problem compound that has muddied the literature. A mini-perspective exploring the extent of curcumin's use in the literature tells the story of troublesome chemistry and just plain bad logic.¹⁹  Multiple lines of evidence show that curcumin is an unstable, reactive and non-bioavailable compound; nonetheless, more than 120 clinical trials of curcuminoids have been launched, with none of them demonstrating efficacy. The concentrations at which curcumin displays its alleged activities, including inhibition of the histone acetyltransferase p300, histone deacetylase 8, glycogen synthase kinase-3β, tau fibril formation and cannabinoid receptor 1, are in the same range at which aggregation could occur. To add to the confusion, counterscreens were frequently not performed and compound stability was not always considered. These troubling findings result in a waste of time and money and plague the literature with results that are not likely to be real, perpetuating the reproducibility crisis with which biomedical science has been blighted. Importantly, the authors of this overview do not rule out that crude turmeric extract could have activity of interest, which appears to drive the continued fascination with this molecule; it is just not the activity of curcumin alone.

Aside from problematic compounds, the correct assignment of MoA for compounds is an extremely critical part of understanding disease biology. A recent study from Cold Spring Harbor Laboratory and Stony Brook University asked the question of why so many oncology drug candidates fail to advance to approval by the US Food and Drug Administration (FDA).¹⁰  Amazingly, the authors found that the annotated target of 10 compounds tested (spanning five inhibited targets: HDAC6, MAPK14, PAK4, PBK and PIM1) were not essential to cancer-cell proliferation. Using CRISPR-Cas9 to knock out the putative targets, they found that there was no loss of viability of the knockout cells, which was not explainable by upregulation of a homologue. Further, in each case, the knockout cells retained sensitivity to the compounds, proving that the protein in question was not the target. Thus, off-target toxicity is a common MoA for these compounds. The targets had primarily been identified by siRNA or shRNA knockdown, which may explain the discrepancy: the known seed effects of siRNA experiments²⁰  make it all the more important to use well-validated methods to identify new targets. What is all the more remarkable, however, is that these are compounds in clinical trials, meaning they had advanced through many stages of development without these inconvenient facts rearing their heads. We can conclude from these results that prudent science, not just chemical liability filters, must be exercised when developing chemical probes and drugs.

A similar effort was performed to evaluate systematically probes focused on the expanding chemical matter to inhibit the interaction between Nrf2 and Keap1.²¹  The authors collected the 19 small molecules reported in the literature to inhibit this protein–protein interaction (PPI). Six compounds were purchased, four compounds were easily synthesized and nine were synthesized with more difficulty. They then performed a side-by-side comparison of each compound, testing the inhibition by fluorescence polarization, thermal shift assays and surface plasmon resonance. Remarkably, half of the compounds were inactive or much less active than reported and 10 of the 19 compounds showed no evidence of binding at all; these results were independent of the source of the compound. This report should serve as a massive wake-up call to the chemistry and chemical biology communities that reproducibility by other research groups needs to be a key step in truly validating a chemical probe. Even more troubling, practitioners of basic cell and molecular biology will likely not even see this body of literature, resulting in the propagation of research using essentially useless reagents.

Chemical probes have been discovered from a wide variety of chemical resources, including natural products, commercially available screening collections, drug-like libraries and diversity-oriented synthesis (DOS).²²  In the last case, over a decade of research activity has resulted in probes relevant to diabetes, cancer and, perhaps most promisingly, infectious disease.²³  A useful component of the DOS chemical strategy is to synthesize a smaller number of compounds across a larger number of scaffolds. Sampling across these scaffolds, using “Informer Sets”, has yielded relative enrichment in activity in particular chemistries, including in projects focused on identifying inhibitors of insulin-degrading enzyme²⁴  and Cas9.²⁵  These insights, which can be achieved rapidly through pilot screening, enable researchers to screen more deeply in active scaffolds, enhancing the efficiency with which probe candidates are identified.

Of course, the use of chemical probes shows its greatest utility in demonstrating the role of a particular protein target in a cellular phenotype. Activity of a probe in vivo is even better, but sometimes pharmacokinetic properties are not yet sufficient to study without significant medicinal chemistry efforts. With the advent of DNA-encoded libraries (DELs) and the discovery of molecular glues that induce proximity between, for example, enzymes and substrates, exciting new probe opportunities arise. A popular example of this MoA is targeted protein degradation (e.g. PROTAC, degronimid) technologies.^26,27  This increasing focus on small molecules that bind targets, rather than inhibit enzymatic activity,²⁸  may yield probes that enable novel activities, previously accessible only by using genetic fusion proteins. These protein-binding compounds can also be used to exploit modulation of the protein interactome of the target, which can often be more useful in the case of, for example, non-enzymatic targets or allosteric binding sites.

Several groups have also made creative use of high-quality chemical probes as a pilot library itself, for understanding the types of targets that should emerge from phenotypic screening. For example, an overview by Wassermann and colleagues²⁹  discusses a number of collections with well-annotated and characterized compounds that exist for this kind of work (e.g. LOPAC, Oxford collection, Repurposing Hub). They also describe the approach developed at Novartis to a biodiverse screening deck. Similarly, cancer cell-line profiling benefits from the use of a focused library, which must therefore be composed of high-quality probes. For example, a study in 2015 examining the sensitivity of ∼500 compounds across nearly 1000 cell lines focused on compounds that perturb key nodes in cellular signalling.³⁰  Finally, a call for contributions to a comprehensive kinase chemogenomic set (KCGS) aimed to develop a common set that the entire research community could use.³¹  The first such step toward this goal was the assembly of the GlaxoSmithKline (GSK) published kinase inhibitor set (PKIS), a collection of 367 ATP-competitive inhibitors; kinase profiles were posted online and the physical set was distributed to hundreds of investigators.³²  These types of community efforts will improve data quality and enhance the efficiency of generating novel biological insights.

Garbaccio and Parmee³³  provided a perspective from the pharmaceutical industry to highlight the need for greater collaborative efforts between academia and industry. They acknowledge that such efforts will require huge investments, but make the argument that good probes help validate targets and show translatability in drug discovery. These advantages place chemical probes at a unique nexus between understanding fundamental biology and providing new avenues for therapeutic intervention. Overall, these authors exhort the community to provide continued investment and publication of high-quality probes for research use. Some pioneering groups have been at the vanguard of providing, free to the community, important high-quality probes. For example, Jay Bradner and Jun Qi's work to distribute the BET bromodomain inhibitor JQ1 has resulted in an explosion of research, with a more than fivefold increase in the number of patents for bromodomains four years after its first publication.³⁴  More such efforts are required to expand the biological space accessed by chemistry.

On that note, an excellent brief overview of chemical probe quality, literature and resources was provided in the paper “Revisiting the art of the chemical probe”.³⁵  This overview highlights several key online resources, in which information about the chemistry, biology and availability of chemical probes dramatically lowers the barriers to using these probes to validate cellular targets. It is important for researchers to be aware of these resources and use the recommended probes available to the community, rather than perpetuating irreproducible research using shoddy probes.

Probe Miner (https://probeminer.icr.ac.uk) contains a database of over 1.8 million small molecules and provides binding and bioactivity data against 2220 human targets, information on structure–activity relationships and data on whether inactive analogues are available and their structures, in addition to including filters for common liabilities, such as PAINS.³⁶  This first large-scale resource also enables users to evaluate the quality of probes, by providing expert curation and judgement of whether small molecules fulfil minimum chemical probe criteria. Notably, the authors found that of the >350 000 active compounds in their database, only 2558 (0.7%) fulfil these minimal criteria: potency <100 nM, 10-fold selectivity and cell activity <10 µM. Although over half of the compounds possess good potency, the largest drop-off of compounds appears to be in cell activity, with only ∼21 000 (6%) of compounds with cell activity, and even half of those compounds show lower potency than desired for a good probe. From a target perspective, there is a similarly sharp drop-off: of the 2220 targets represented, 1664 (75%) have potent compounds available, but 795 (36%) with selective compounds and only 250 (11%) with high-quality probes. These observations indicate that there does not appear to be much of a problem generating potent biochemical probes, but cellular activity for most probes has lagged. Perhaps the community should take heed of this knowledge and make cellular activity a more upstream requirement for probe generation.

The philosophy of expert curation has been taken to the next level by the Chemical Probes Portal (http://www.chemicalprobes.org), a resource that reports on fewer compounds but aims at high-quality review of each probe.¹¹  This database uses a principle of selective crowdsourcing: there are many invited probe reviewers from all over the world in the chemical biology community. Among the emphases of this resource is experimental reproducibility, which is hampered by poor-quality probes. To that aim, the authors also flag “historic” compounds, which are those that have previously been used as probes but should not be in the future. A comparative analysis of these historic compounds and probes showed that, unsurprisingly, historic compounds were more promiscuous than probes.³⁷  Overall, about half of the compounds were considered target selective, which the authors consider “an encouraging finding”, which speaks to the unsatisfyingly low bar for reporting on active small molecules.

Two resources from the Structural Genomics Consortium (SGC) provide valuable information and, importantly, distribution of chemical probes for the research community to use. First, the SGC Chemical Probes site (https://www.thesgc.org/chemical-probes) provides open-access reagents, with detailed characterization and recommendations on which probes to use in cell-based assays and at what concentrations. The 73 probes on this site, largely focused on epigenetic targets and kinases, all fulfil the following criteria: biochemical potency <100 nM, >30-fold selectivity versus other subfamilies and significant on-target cell activity at 1 µM. This important resource is complemented by the Donated Chemical Probes resource (https://www.sgc-ffm.uni-frankfurt.de) run by SGC Frankfurt.³⁸  An initial consortium of seven pharmaceutical companies have opened up their set of undisclosed probes, which meet a strict set of criteria and are available for use by anyone in the research community. These criteria are similar to those discussed above in earlier literature, but this database also includes proof of target engagement, an increasingly important consideration in cell and in vivo treatment that is not always considered. Inactive control compounds are available for each probe and a variety of profiling data (kinome and GPCR data and select phenotypic assays, among others) is available for each compound. Although this set of compounds is still primarily focused on enzymatic targets, this framework for enabling open-access research to be performed on high-quality probes is a model for how chemical probes will advance our knowledge of target biology in a more rapid fashion.

Another recent contribution to the probes arsenal is the Probes & Drugs (P&D) database (https://www.probes-drugs.org/), developed by CZ-OPENSCREEN.³⁹  This resource contains information on >4000 chemical probes and >12 500 drugs, in addition to tens of thousands of other compounds (e.g. analogues). Further, the P&D database has the most sophisticated filtering, visualization and chemically intelligent tools available. Unlike some of the other databases, however, these probes are not distributed through this group. Therefore, consumers of these data should be cognizant of the advantages and limitations of each resource, with a combination of tools likely to provide the best direction for research.

Finally, a recent call to arms has been raised in the guise of “Target 2035”, proposed by Carter et al.⁹  This global federation aims to harness technological advances to generate reagents for the entire proteome, from chemical probes to selective antibodies to entire chemical libraries. Proceeding from the notion that the natural tendency is for scientists to study genes that are already under study, the authors argue that pharmacological modulators are “the most important reagent for any protein”. It is remarkable to note several lessons that emerge from this effort: (1) the estimated cost for developing a high-quality probe is more than US$2 million, a cost sufficiently large that achieving Target 2035 will require enormous investment from diverse sources around the world; (2) the idea of generating proteome-wide probes is now, 20 years after originally being proposed,²  finally sufficiently ripened to result in such a global effort; (3) the Target 2035 team wisely envisions two phases of the project over time: phase 1 (planned for 2020–2025) entails creating the necessary infrastructure, developing new technologies and self-organizing among chemical biologists, and phase 2 (anticipated to be 2025–2035) involves applying phase 1 resources to create a set of comprehensive modulators for most human protein-encoding genes. Each of these insights reflects the increasingly realistic expectations of the cost, effort and organization required to achieve such a worthwhile goal. An increasing partnership between public and private research organizations is absolutely required for the community to realize the promise of high-quality chemical probes to understand biology, identify and validate targets for therapeutic intervention and generate lead candidates for drug development. Further, greater incentive to study (and publish) the quality of chemical probes needs to be provided to the community, lest the community be saddled with poor probes and irreproducible results.