Which Small Molecule? Selecting Chemical Probes for Use in Cancer Research and Target Validation

Small-molecule chemical “probes,” most commonly inhibitors, are chemical tool molecules characterized by meeting quality criteria to warrant their use in target validation studies and thus have become invaluable research reagents that complement the use of molecular biology and genetic technologies (for detailed definitions of chemical tools, probes, and related categories, see Box 1). A major role of chemical probes in fundamental biological research is to explore the functions of proteins in normal cellular processes and disease mechanisms. As well exemplified in the field of nuclear receptors, protein kinases, and epigenetic regulatory proteins, the discovery and widespread availability of high-quality probes can stimulate research on more neglected protein family members and hence shed light on the “dark proteome” (1, 2). Furthermore, in translational research, high-quality chemical tools are invaluable for target validation (3).

However, despite the rapidly increasing number of chemical probes (4), further accelerated by initiatives such as the Target 2035 project (5), we continue to see the publication of misleading results caused by poor selection and inappropriate use of small-molecule reagents (6, 7). This is exemplified by a very recent systematic analysis that showed that compliance with well-established best-practice criteria—using the recommended concentration range and inclusion of orthogonal active and inactive controls—was as low as 4% (7). Expertise in the design of high-quality chemical probes and their best-practice use mainly resides in chemical biology and medicinal chemistry/drug discovery teams. Because of this compartmentalization of expertise, interdisciplinary dialogue and collaboration between biologists and chemists are deemed critically important to ensure the selection of the best-suited chemical probes (8).

In this review, we juxtapose examples of poor chemical probes with those of high quality and suggest best practices for the selection, validation, and use of the latter in cancer research. We also note the complexity associated with tools for novel drug modalities, exemplified by protein degraders, and provide advice and resources to facilitate the identification of appropriate small-molecule tool compounds by researchers.

Several publications and organizations have described quality criteria for probe molecules or target validation tool compounds (6, 9, 10), and these criteria warrant repetition. At a high level, they encompass selectivity, potency, and evidence of functional target engagement and modulation in a cell-based assay rather than a cellular phenotype. Specific criteria established by the Structural Genomics Consortium and their collaborators include (i) evidence of target binding or modulation of functional activity, IC₅₀ or K_d <100 nmol/L; (ii) evidence of selectivity within the target protein class, with at least 30-fold selectivity in a similar binding or functional assay, ideally with broader profiling; and (iii) evidence of direct target engagement in a cellular assay or target modulation using a proximal pharmacodynamic biomarker, IC₅₀ or EC₅₀ <1 μmol/L (9, 10). Similar criteria are recommended by the Chemical Probes Portal (vide infra; ref. 11).

However, these criteria encompass only some of the attributes of a quality chemical probe. The probe should exhibit stability in storage and in relevant assay media with sufficient solubility to enable testing over a suitable range of concentrations without precipitation. In addition, an approved drug may have beneficial polypharmacology for patients but could confound target validation experiments. Vendor catalog descriptions of a compound may not reflect the evolution of data that reveals polypharmacology and should never be used as a surrogate for a deeper dive into the literature before ordering and using a compound. Examples of unsuitable tool compounds are found in Table 1 and described below, with chemical structures in Supplementary Fig. S1.

Iniparib exemplifies how chemical reactivity can confound data interpretation. This compound progressed to phase III trials in 2009, described as a PARP1 inhibitor for the treatment of non–small cell lung cancer (NSCLC; NCT01082549) and triple-negative breast cancer (NCT00938652) under the sponsorship of Sanofi. Unfortunately, the phase III studies did not recapitulate the progression-free and overall survival benefits as observed in an earlier phase II study (12). The nitro and iodo groups present in iniparib are atypical features of drug molecules, and it has been rationalized that the active form of the drug is a reduced, nitroso metabolite (13). Researchers at Abbott examined the compound more closely because it was distinct from other PARP inhibitors in both its structure and mechanism of action (MoA; ref. 14). Detailed cellular studies demonstrated that the C-nitroso metabolite reacted nonspecifically with cysteine residues when incubated with cell lysates, as well as when incubated with human breast cancer cells; none of these proteins were PARP isoforms. A key lesson from this molecule is that its reactivity and MoA were not well characterized in early studies, leading to consequential and unfortunately misleading clinical trial results (15). Despite claims in vendor catalogs, it should not be used as PARP inhibitor, and other probes are available.

Although useful as an early tool compound to inhibit PI3K and extensively cited, LY294002 is in fact a poor probe compound due to its polypharmacology and modest potency. It was initially identified as a PI3K inhibitor in 1994, with the claim that it “completely and specifically abolished PtdIns 3-kinase activity (IC₅₀ 1.40 μmol/L) but did not inhibit PtdIns 4-kinase or tested protein and lipid kinases” (16). It should be noted that this testing encompassed only 10 kinases, as large, commercial kinase panels were not available at the time. In cells, it was tested at a single concentration of 50 μmol/L in human neutrophils and rabbit aortic smooth muscle cells, and it inhibited the uptake of BrdUrd in a smooth muscle cell proliferation assay (IC₅₀ 32 μmol/L). However, despite its modest potency and lack of selectivity (17), nearly 3,000 publications indicate that researchers continue to cite or employ LY294002, even though it has been superseded by more potent compounds whose PI3K selectivity is explicitly characterized—for example, BYL719 (alpelisib; PI3Kα selective), CAL-101 (idelalisib, PI3Kδ selective), and GSK2636771 (PI3Kβ selective; ref. 18). The key lesson from LY294002 is that citation frequency does not reflect compound quality, and a critical survey of more recent literature is essential to understand the shortcomings of historically used compounds and to find superior probes.

Crizotinib reflects an evolving understanding of poly­pharmacology. It was initially patented as a potent inhibitor of c-Met kinase (K_i 3 nmol/L; ref. 19), but the first detailed account of its efficacy reported near-equivalent activity against nucleophosmin (NPM)-ALK expressed in human lymphoma cells (20), and noted its selectivity in a panel of >120 kinase biochemical assays. The clinical development strategy evolved to take advantage of its ALK activity, and its approval in 2011 was for the treatment of patients with metastatic NSCLC whose tumors were ALK- or ROS1-positive as detected by an FDA-approved test. Subsequent discovery and development activities led to the identification of compounds such as alectinib and tepotinib, which are more selective for ALK (21) and c-Met (22), respectively. Although poly­pharmacology can lead to clinical success, it can potentially complicate preclinical target validation, and again, a deeper look at emerging data can be beneficial.

Like crizotinib, the understanding of sorafenib pharmacology has matured over time. This compound was designed and originally labeled a RAF kinase inhibitor. However, during its clinical development, sorafenib was found to be ineffective as a RAF inhibitor in cells. Pharmacodynamically, it turned out to be most effective and clinically most useful based on its “off-target” activity against VEGFR2 and other receptor tyrosine kinases (23). With a better understanding of the activation mechanism of the RAF kinase, significantly more effective and selective RAF inhibitors, such as vemurafenib, have been discovered and are better chemical probes for this target (24).

Natural products often demonstrate polypharmacology, which can confound data interpretation. One example is the quassinoid obtained from the seeds of Brucea sumatrana, brusatol. Despite a report in 2015 noting that it is an inhibitor of general protein synthesis (25) and not an NRF2 inhibitor, nearly 100 journal articles, reviews, and patents have been published since 2015 with the attribution of its activity to NRF2 inhibition, and some chemical suppliers continue to describe it as “an inhibitor of NRF2.” Unfortunately, no chemical probe is available for the inhibition of NRF2, although KI-696 is a well-characterized molecule that inhibits the NRF2–KEAP protein–protein interaction (26). This example again underscores the challenges of biological characterization and due diligence in confirming vendor catalog claims.

Lastly, some compounds can be described as “pathfinders” (9). A pathfinder is commonly not as potent or selective as a high-quality chemical probe molecule but enables a degree of initial hypothesis validation while indicating the druggability of a selected target and is often superseded by higher-quality reagents as the field matures. This applies to LY294002, as previously mentioned. Allosteric MEK inhibitors represent another example of such a compound. An early pathfinder compound, PD-098059, was identified in a kinase cascade assay of the MAP kinase RAS/RAF/MEK/ERK signaling pathway, and subsequent experiments confirmed MEK as its target (27). CI-1040 (PD-184352) is a more potent MEK1/2 inhibitor resulting from a follow-on screen facilitated by PD-098059 as a positive control (28). It provided in vivo proof of concept preclinically (29) and was advanced to clinical trials (30). A poor pharmacokinetic profile hindered its clinical success, but the positive results paved the way for other allosteric inhibitors of MEK1/2. Four such inhibitors (trametinib, cobimetinib, binimetinib, and selumetinib) have been approved by the FDA, and all can be used as chemical probes in vitro or in preclinical in vivo experiments.

Although iniparib represents an extreme case of compound misuse with harmful clinical results, it is easy to imagine how probe mishandling/misinterpretation also occurs in basic research settings. In a frequently encountered scenario, a researcher who has identified a particular protein of interest (POI) will attempt to support a hypothesis by selecting commercially available compounds advertised as inhibitors of the POI. The researcher may then test one compound at a single concentration with little consideration of the intrinsic potency of the chemical tool, possibly “overdosing” the protein. Off-target effects can also confound these results. Although the pressure to generate publishable results will continue to drive such practices, we strongly believe that awareness and use of relevant resources (10, 11) will greatly benefit the cancer research community.

Here, we provide three distinct stages of a research workflow to enhance the use of small-molecule chemical probes: (i) a researcher is exploring a particular biological target and would like to identify one or more tool compounds to interrogate it, (ii) a researcher has identified a tool compound and would like to gain confidence that its use is appropriate, and (iii) a researcher has a validated small-molecule probe in hand and would like to employ it in the most impactful manner.

Fortunately, useful, publicly accessible online resources can make the process of identifying and choosing tool compounds less daunting despite the vast, disparate (and potentially misleading) amount of information that is freely accessible on the internet (31). These resources have been assembled by experienced biologists and chemists and provide accurate information to help guide the identification of the best compound for the proposed experiment (Table 2). They span a continuum from indexing highly characterized molecules to capturing all known compounds associated with a given protein. Representing the former end of the continuum, the Chemical Probes Portal, freely available via ChemicalProbes.org, is a user-friendly resource specifically developed to provide advice on the appropriate selection and use of small-molecule chemical probes (10, 11). In practice, we suggest that researchers first consult the portal website to identify and evaluate potential chemical tools, particularly for highly druggable targets with many known ligands. If suitable chemical tools are not listed by the Chemical Probes Portal, databases such as Probe Miner (https://probeminer.icr.ac.uk/#/; ref. 32), Probes & Drugs (https://www.probes-drugs.org/home/; ref. 33), PubChem (https://pubchem.ncbi.nlm.nih.gov/), ChEMBL (https://www.ebi.ac.uk/chembl/; ref. 34), and canSAR (https://cansar.ai/; ref. 35) can then be used to cast a wider net to identify possible ligands for use as small-molecule tool compounds. These sites, ideal for identifying known ligands of a comprehensive range of targets, may require additional effort by researchers to triage the most suitable tool compounds to address a specific question. If these sites fail to identify suitable compounds, the ZINC chemical database (36) may enable the development of novel ligands of interest.

A very useful feature of the Chemical Probes Portal is the collection of “Unsuitables,” previously termed “Historical Compounds” (https://www.chemicalprobes.org/unsuitables). This list features compounds that have previously been used as chemical probes but are no longer deemed suitable because of lack of potency, selectivity, or even target misattribution. At the time of writing, the collection contained 250 compounds, along with a note and reference to explain the categorization. When there is doubt about a chemical probe, it is recommended to enter the name of the probe planned to be used in the search field of the Chemical Probes Portal website. If it is part of the “Unsuitables” list, the search result reflects this. All inadequate tool compounds in Table 1 are part of the “Unsuitables” collection.

In addition to these publicly available resources, excellent reviews (2, 3, 6, 8–10) not only guide the selection of potential chemical tools but also enhance the appreciation of the benefits and limitations of various small-molecule tools to interrogate biology. In 2015, insights and output from the decade-long Molecular Libraries and Imaging Program initiative of the National Institutes of Health (NIH; ref. 37) represented an important step in the conceptual development of chemical probe molecules. Our collective appreciation for best practices in the selection of chemical probes has continued to evolve (4), and we recommend the inclusion of at least two different chemical probes from different structural classes (chemotypes) and negative control compounds (38), which are useful for reducing the risk of off-target effects.

Finally, to identify tool compounds that may be suitable for interrogating a particular biological hypothesis, there are two additional resources beyond those described above. The first is a collaborative effort among a group of pharmaceutical companies and academic institutions to make high-quality probe molecules available for general use. This resource provides guidance on selectivity profiles, use, and availability (39). The second is an initiative to develop selective probe molecules targeting every member of the human proteome, called Target 2035. This community-based initiative is an ongoing effort and represents an important vehicle for identifying newly discovered high-quality chemical tool compounds for a POI (5).

If a potent, selective, and cell-permeable molecule is identified from a resource such as the Chemical Probes Portal, the cancer researcher is now in an excellent position to deploy it in subsequent studies. Unfortunately, in more instances than not, compounds against a particular POI are identified, but insufficient knowledge is available to use them with confidence. Therefore, it is advisable to conduct further validation studies to confirm their utility. Some of the recommended activities that will help ensure that the compound(s) being used as tools are well-suited for biological studies under consideration are described below and compared in Table 3.

• Confirm direct target engagement and/or activity in more than one assay format. Often, the literature data for compounds are generated from assay formats that lack relevance to cellular biology. For example, binding assays may not consider the relevant protein conformation, concentration of cofactors such as ATP, or presence or necessity of accessory cellular proteins. Fortunately, established contract research organizations have strong expertise in generating robust reagents and data for biochemical assays for a wide range of targets/classes. Whenever such resources are used, it is strongly advised to include known control compounds in the assays for comparison and quality control. In addition, understanding the exact assay protocols (e.g., ATP concentrations for a kinase radiometric assay) and their potential relevance to cellular or in vivo settings is critical. Obtaining IC₅₀ or EC₅₀ values by assaying across a full range of concentrations (preferably eight or nine points) is essential for progressing probe molecules in cellular or in vivo settings, with confirmation that the compound is soluble at the highest assay concentration. For enzymatic targets, target engagement studies are strongly complemented by using cellular assays that measure the intracellular turnover of an enzyme substrate. A range of assays are available to measure target engagement and modulation in living cells (40, 41). Although resource intensive, performing single-point mutations of enzymes on active sites and comparing to the phenotype exhibited by a compound presumed to be an inhibitor can be very informative, as can CRISPR knockout of the target. Mutational analyses can also be applied to proteins at the interface of protein–protein interactions if a compound is hypothesized to work through disruption of such an interaction.

  For many targets relevant to oncology, particularly those that do not possess enzymatic activity, target engagement can be demonstrated using biophysical assays. Examples of such techniques include isothermal titration calorimetry, differential scanning calorimetry (DSF), and surface plasmon resonance (SPR). These three assays can measure the direct physical association of the ligand with the POI and are frequently used in drug discovery as routine screening tools. Another commonly used technique to measure the binding of a ligand to a target protein within the cell is cellular thermal shift assay (CETSA). A recent excellent example demonstrates the application of various biophysical techniques to invalidate the putative target of a series of compounds thought to be β-catenin inhibitors that were discovered through phenotypic assays (42). Downstream target–based modulation should also be assayed using suitable biomarkers to establish a “pharmacological audit trail” (43). Lastly, for definitive proof of not only whether but also how a molecule binds to a target, X-ray crystallography, cryogenic electron microscopy (cryo-EM), and/or nuclear magnetic resonance studies remain gold standards to demonstrate target–protein binding.

• Conduct orthogonal confirmation of on-target acti­vity through genetic studies. In cellular studies, demon­strating that the observed phenotype from compound treatment can be recapitulated by genetic studies is a powerful method for validating the suitability of a compound for further studies. Genetic knockout/knock-in using CRISPR is now commonly used and often supersedes short hairpin RNA (shRNA) or siRNA studies in which incomplete knockout may occur. An excellent resource to help uncover the relationships between genetic perturbation and compound sensitivity is the DepMap portal (https://depmap.org/portal/depmap) from the Broad Institute. If a particular compound of interest is included in this database, a wealth of information can be obtained to complement future studies. It should be noted that small-molecule drugs typically alter the specific functions of proteins and therefore do not necessarily phenocopy knockdown/knockout at the gene level. Other important genetic studies include demonstrating loss of the phenotype when the putative target of a chemical probe or drugs are knocked down/out or a mutant protein resistant to the small mole­cule is expressed (44). In contrast, small-molecule targeted degraders (vide infra) act with mechanisms that, at least in principle, better correspond to genetic knockdown.

• Examine the correlation between biochemical and cellular activity. Having multiple structurally related compounds in the same chemical series with a range of biochemical activities can also validate a tool compound. A strong correlation between the concentration–response of the phenotypic effect and the respective biochemical potencies for several compounds can provide strong support for a biological hypothesis and help demonstrate that the observed phenotype is indeed “on-target.” Correlation studies of this type also provide additional guidance regarding ideal concentrations that should be used in cell-based studies. Importantly, it should be noted that a cellular assay measuring either direct target engagement or downstream molecular consequences of activity against a known direct substrate of the protein target of interest (i.e., a “proximal” biomarker) offers the most compelling validation for a tool compound. An excellent example highlighting the approaches used to evaluate a series of known DYRK and CLK inhibitors as tool compounds has recently been published (45).

• Conduct selectivity screening against potentially confounding targets. It is also critical to determine whether activity against other biological targets will obfuscate the interpretation of an experiment. This information is often available in the literature for well-characterized probe compounds. However, for ligands for relatively new biological targets, additional selectivity screening is strongly recommended. Unfortunately, cost limitations can be a major issue for academic researchers. Focused screening against closely related targets in the same protein family can often provide an indication of greater overall selectivity. Alternatively, one can screen against commercially available representative panels of promiscuous targets.

• Evaluate compounds for potential structural alerts/physical property limitations as early as possible and follow-up with confirmatory assays. Certain chemical structural classes (e.g., rhodanines, chalcones, and certain hydrazones) are known to possess inherent liabilities that can lead to false-positive activity across a range of assays and gene families and can create misleading results. These compounds have been given the descriptive acronym “PAINS” (pan assay interference compounds; ref. 46) and are also referred to as “nuisance” or “promiscuous” compounds. Those compounds with such inherent potential liabilities should be evaluated early using appropriate assays. A relevant example outside of oncology can be observed in the flurry of activity seen with repurposing drugs for COVID in early 2020 (47). A closer examination of some of the reported hits revealed that under certain assay conditions, the effects of many of these “active drugs” were attributed to artifacts such as chemical and redox reactivity (48) or aggregation rather than genuine inhibition (49).

Obtaining as much of the above data as possible prior to using a tool compound increases the chances of delivering impactful results. It is certainly appreciated that many labs do not have the financial resources to gather such data or experience to interpret it. Therefore, we strongly advocate collaboration with individuals or institutions that can share costs, openly sharing data whenever possible and seeking advice from drug discovery practitioners to alleviate these issues. Reaching out to contacts within well-known academic groups or institutions with a strong drug discovery component is often a good first step toward gathering the information that ensures a compound is fit for its intended purpose.

Appropriate use of tool compounds is critical for their application in supporting biological studies. We offer several recommended best practices to avoid common pitfalls and ensure that the tool compounds are used in a manner that leads to meaningful results.

• Use the tool compound within the recommended concentration range. The tool compound must be present at a sufficient concentration to engage with its cognate target(s). Care should be taken to use the tool compound at a concentration that minimizes the engagement of collateral targets that could confound biological studies. Well-characterized tool compounds have an experimentally supported concentration range at which they can be effectively used. Among other approaches, the availability of established methods to develop target engagement assays (40, 41, 50) provides a basis for the recommendation of concentration ranges in cellular systems. There is particular concern with these issues in studies related to cancer therapies in which the phenotypic endpoints are commonly so-called down readouts, often inhibition of cell growth or promotion of cell death, which can occur through nonspecific or off-target effects (51).

• Use negative control compounds along with tool compounds. Negative control compounds have chemical structures closely related to the tool compound of interest but have greatly reduced activity for the intended target(s). Negative control compounds are paired with active tool compounds: Both should be characterized to the same extent, and the pair should be used in parallel in biological studies to account for potentially confounding off-target activity of the negative control. This comparative characterization will help ensure that the phenotypes imparted by the negative control compound did not stem from engagement with the cognate target(s) of the tool compound. Therefore, negative control compounds can provide greater confidence in the MoA of their pair members (38).

• Use more than one structurally distinct tool compound. Although not guaranteed, a structurally distinct, “ortho­gonal” tool compound of a different structural type (chemotype) with activity on the same target likely possesses a nonoverlapping collateral target profile. Orthogonal tool compounds must also be highly characterized and used in parallel with tool compounds. An orthogonal tool compound increases confidence that induced phenotypes are caused by the engagement of the intended targets (9).

• Determine the degree of target engagement/modulation in your own model system. As mentioned earlier, evidence that the chemical probe binds to and modulates the target is important, and this should be checked in the same model system used, as effects may vary between models.

A detailed discussion of the in vivo use of probe compounds in animals, usually mice in cancer research, is outside the scope of this review, but additional factors must be considered for experiments in animal models (43, 52). Useful, albeit brief, guidance is available at the Chemical Probes Portal (https://www.chemicalprobes.org/info/guidelines-animals). In contrast to cellular studies in which the effective compound concentration can be readily controlled, in vivo animal studies require information on the compound's pharmacokinetic and pharmacodynamic profile (52) and data on the plasma protein binding from which the free (unbound) probe concentration can be obtained (53, 54). Again, selecti­vity is pertinent:

• Be cognizant of off-target effects and how they can affect the interpretation of studies—for example, dose used and pharmacokinetic/pharmacodynamic relationships and tolerability. It is important to understand free probe concentration in the context of off-target activity.

• Off-target effects in animals can stem from target families unrelated to the target family of focus (e.g., a kinase inhibitor that hits a G protein–coupled receptor) and may affect tolerability.

Thus, small-molecule tool compounds can provide powerful support in biological studies, provided that the tool compounds are selected and used properly, although real-world scenarios may not allow all the best practices described above to be followed. A visual flowchart (Fig. 1) is provided to assist in experimental planning as a function of the available information.

The fraction of the proteome targetable by small-mole­cule drugs that exert pharmacological effects through rever­sible binding is limited, leading to the exploration and advancement of “expanded” drug modalities. These include MoAs that proceed through “event-driven” pharmacology, with ubiquitin–proteasome-dependent molecular degradation being the prime example. In contrast to the “occupancy-driven” pharmacology of traditional inhibitors, the degradation event, rather than functional inhibition, is driving the pharmacological effect. This means that, in practice, the ligand binding site does not need to be functional. In addition, stronger pharmacological responses can be achieved than with an inhibitor, because of the catalytic nature of the process, with protein turnover controlling the duration of action. Furthermore, the involvement of a ternary complex between the target protein, an E3 ligase, and the degrader molecule is a significant factor that may lower the potency and selectivity requirements of the ligand to its target protein, thus enabling pharmacological action on poorly tractable targets. These points should be considered when deciding whether to pursue a degradation approach for a given target, as opposed to a traditional inhibition approach. Undoubtedly, the exploration of degraders in discovery and clinical settings is on the rise (55), and an extension of the probe definition to this class of compounds is deemed timely (56).

Here, we focus our discussion on the two most prominent and advanced classes of targeted degraders, bifunctional proteolysis-targeting chimeras (PROTAC) and molecular glue degraders, both of which have gained significant attention not only because of their clinical potential but also because of their value as biological tools. In many ways, targeted degradation represents the chemical equivalent to genetic knockdown, allowing the removal of a protein after translation rather than after transcription; thus, molecular degraders have become exceptionally valuable tools to interrogate protein function (57–59). However, as with traditional reversible small molecules, the use of molecular degraders for pursuing biological questions warrants careful selection and use. Parallel to the requirements for reversible small-molecule probes (potency, cellular target engagement, and selectivity), the three key requirements for small-molecule degraders are measures of induced target depletion, cellular degradation that is on-target, and proteome-wide selectivity. These recommendations are summarized in Table 4.

• Reproducible compound-induced depletion of the desired protein. Compound-induced degradation can be measured by determining protein levels in cells; the most common and straightforward method is Western blotting. However, this method requires specific antibodies, is low-throughput, and provides only a limited dynamic range (59). Quantitative methods include immunoassay, immunofluorescence, and enzyme-fragment complementation assays such as HiBiT (high binary technology). Many systems are commercially available, and the protocols are detailed in the literature (60).

  Quantitative degradation assays provide information on the maximum achievable reduction in protein levels (D_max) as well as the concentration necessary to achieve 50% reduction in protein levels relative to the maximum achievable level (DC₅₀). Similar to genetic knockdown studies, it is desirable to achieve complete or near-complete target degradation, and we propose a D_max of 80% as a desirable cutoff for degrader probes. This is an empirically derived value representing the authors’ experience based on what is typically practically achievable through a moderate amount of compound optimization and often found sufficient for obtaining a phenotype representative of the complete depletion of the POI population in a given cell. However, even when reaching high D_max values, this may still not prevent residual protein levels from upholding the cellular function of the protein in question (61). Partial degradation thus represents a risk that the experimenter should be aware of. To address this, data-driven approaches of obtaining meaningful D_max targets for degraders can be used—for example, gene knockdown or knockout titration experiments that monitor associated cellular phenotypes (62) or the use of the dTAG system (63). We recommend a DC₅₀ <1 μmol/L, with at least a 10-fold window to general cytotoxicity in the same cell line. This may mean that using a cell line that is highly dependent on the presence of a POI for survival is not a recommended best practice for identifying target-specific degraders.

• Validation of on-target and on-mechanism degradation. PROTACs and molecular glue degraders operate via proximity induction and, specifically, by recruiting a ubiquitin E3 ligase to the POI. It is the ternary complex between the three components that enables productive ubiquitin transfer to a surface-exposed lysine of the POI, ultimately leading to its proteasomal degradation. All three events, proximity induction, ubiquitin transfer, and proteasomal degradation offer distinct intervention points that can be leveraged for mechanistic validation (58).

  PROTACs are heterobivalent degraders containing a distinct target ligand and an E3 ligase ligand; therefore, a straightforward way to prove dependency on binding to either entity is to demonstrate that degradation can be abolished via ligand competition by adding either an excess of the target ligand or the E3 ligand. If there is no clarity that the downstream effect is driven by an enzymatic effect of the target ligand that is part of the PROTAC or the PROTAC degradation effect, E3 ligase inactive control compounds that are unable to degrade the target can be very useful. In the case of PROTACs engaging cereblon (CRBN), the use of N-methylated immunomodulatory imide drugs (IMiD; ref. 64), and in the case of VHL PROTACs, the use of the inactive diastereomer of the VHL hydroxyproline (65), both devoid of affinity to their respective E3 ligases, are commonly used tools for this purpose. Furthermore, to demonstrate that degradation occurs via the ubiquitin–proteasome system (UPS), cotreatment with inhibitors of the UPS machinery, for example, a proteasome inhibitor such as bortezomib, or an E1 inhibitor such as TAK243, is recommended (58). The toxicity of bortezomib and related control compounds requires short incubation times and may not be compatible with slow degradation systems. Most ligases co-opted for degradation, such as CRBN and VHL, represent cullin-RING ligases (CRL), which specifically require activation via NEDD8 conjugation for their activation. To prove CRL involvement, a neddylation inhibitor, such as MLN4924, is frequently used in degrader studies (66). The use of the above measures is typically sufficient for mechanistic validation of PROTACs.

  The mechanistic validation of molecular glues can be more complex, and we recommend different strategies. Although the field is rapidly developing, most molecular glues used in tool compound studies are still CRBN/IMiD ligand based, and these compounds typically have measurable affinity to CRBN; however, in the absence of CRBN, they have often immeasurable affinity to the target. Hence, there is typically no available target ligand that binds competitively to the glue molecule. The most stringent way to demonstrate a direct degradation effect of glues (apart from proteomic specificity) is ligand-induced proximity between the E3 ligase and the POI, as determined either by biochemical or biophysical methods [e.g., time-resolved fluorescence energy transfer (TR-FRET) or SPR] or cellular assays (e.g., NanoBRET; ref. 58). CRBN-type glue molecules are prone to poly­pharmacology (67), and extra care needs to be taken to demonstrate that the observed depletion of POI is caused by a direct and not an indirect mechanism. A frequent off-target of IMiD-type degraders, glues, and PROTACs is the translation termination factor GSPT1, a globally essential gene (DepMap, www.depmap.org). Its degradation is known to deplete levels of many proteins by blocking new protein synthesis (68). An important diagnostic for direct target degradation is a rapid onset of degradation (within 4–6 hours), while indirect effects often take >12 hours.

• Adequate selectivity across the proteome. Demonstration of the selectivity profile of a degrader probe molecule is necessary to rule out off-target effects as major contributors to the observed cellular phenotype. Two levels of selectivity should be considered: degradation and binder selectivity. With respect to degradation selectivity, the best practice is to provide a whole-cell proteomics dataset for a molecular degrader probe in the same cell line used for the quantification of degrader potency. We recommend incubation with the degrader at a concentration 10-fold above DC₅₀ and at an early time point sufficient to reach D_max. All proteins depleted more than 2-fold with a P value of ≤0.05 should be reported and considered when interpreting pharmacological profiles. Different cell lines can provide different degradation profiles, and the selection of a cell line model that is relevant to the application of interest should be an important consideration.

  It is important to note that ligand binding independent of inducing protein degradation can contribute to cellular phenotypes. For example, many PROTACs that target kinases are derived from kinase inhibitors. While constructing PROTACs from such kinase inhibitors may provide a level of degrader selectivity, the resulting PROTACs may still inhibit many more kinases than they degrade (69). Therefore, PROTACs derived from functionally active promiscuous small molecules may not be suitable as molecular degrader probes for targeted biology studies.

In addition to the summary of these recommendations in Table 4, a modified decision tree for the selection of a degrader probe can be found in Fig. 2.

As it is still early days for degrader drug discovery, only a limited number of fully characterized PROTACs are in the public domain that could be considered high-quality molecular degrader probes, as often one or more of the characterization data types have not been reported. However, the SMARCA2 PROTAC, ACBI2, jointly developed by the Ciulli and Boehringer Ingelheim labs (70), is well characterized and accessible to the scientific community via Boehringer Ingelheim's OpnMe portal (71). Examples of potent and selective molecular glues include an IKZF2 degrader, NVP-DKY709, from Novartis (72), although SALL4 degradation has also been reported, and a GSPT1 degrader, MRT-2359, from Monte Rosa Therapeutics (73). The Chemical Probes Portal has started to include degrader probes in its collection, and at the time of this writing, 31 reviewed PROTAC and glue degraders were listed, and another eight compounds were under review.

Small molecules can be powerful reagents for the elucidation of biological targets and pathways. However, compound selection based on vendor catalog descriptions or citation frequencies has limitations. Citation metrics and brief catalog entries may not reflect the entirety of a compound's pharmacology, and a deeper dive into the literature is warranted to fully understand its activity. Investigations based on historically used, inadequate small molecules can create momentum for research that follows unsound hypotheses, wasting time and resources, and diverting funding from inquiries that are supported by more scientific rigor.

We propose that the appropriate selection and use of tool compounds should be recognized as a crucial aspect of any research effort that includes small molecules. In the scientific literature or research proposals, data adequate to support any associated tool compound usage should be easily available, whether in the article itself, supporting information, or in a linked reference. Scientists, grant reviewers and funders, publishers, and manuscript reviewers can all benefit from the growing availability of online resources such as the Chemical Probes Portal and Probe Miner.

Although researchers are encouraged to apply as many as possible of the approaches described throughout this review, to maximize the robustness of the findings, we suggest that the minimum requirements for the validation of a small-molecule probe compound in a new protein target are:

• Confirmation of direct target engagement and/or activity in more than one assay format (i.e., for an enzymatic target, obtain biochemical activity and assess binding affinity by at least one biophysical method, and measure target engagement in a cell-based assay)

• Orthogonal confirmation of on-target activity through genetic studies (i.e., confirm that the phenotype of compound treatment in a cellular context is recapitulated through knockout or knockdown studies)

• Characterization of selectivity against potentially confounding targets (i.e., for an enzymatic target, evaluate the biochemical activity from targets within the same gene family).

• Use of the chemical probe in the recommended concentration range.

• If available, inclusion of a matched inactive control compound and an orthogonal active compound from a different chemotype.

Potency and selectivity criteria associated with these requirements were outlined above.

Although we hope that the criteria, guidelines, and resources highlighted here will aid cancer researchers in the selection and use of tool compounds, we encourage collaboration and conversation with medicinal chemists seasoned in the use of small molecules for both target validation and drug discovery, particularly when thoroughly vetted compounds are not available.

M.M. Mader reports other support from Eli Lilly and Company, Pfizer, and Relay Therapeutics outside the submitted work, as well as a patent for WO 2023288242 pending, a patent for WO 2021222556 pending, and a patent for WO 2020231990 pending. I.V. Hartung reports personal fees from Merck Healthcare KGaA outside the submitted work. J. Rudolph reports that he is a current employee of Genentech, Inc. and a former employee of Bayer. D. Uehling reports personal fees from the Cancer Prevention & Research Institute of Texas and Tavotek, Inc. outside the submitted work, as well as a patent for WO2022226668 pending to the Ontario Institute for Cancer Research (OICR), a patent for WO2022226665 pending to OICR, a patent for WO2022226666 pending to OICR, and a patent for WO2022226667 pending to OICR. P. Workman reports personal fees from Alterome Therapeutics, Black Diamond Therapeutics, CHARM Therapeutics, Epicombi Therapeutics, Nextech Invest, Storm Therapeutics, and Vividion Therapeutics, other support from CV6 Therapeutics and Chroma Therapeutics, grants and personal fees from Astex Therapeutics, Merck KGaA, and Nuvectis Pharma, and grants from AstraZeneca, BACIT, and Sixth Element Capital/CRT Pioneer Fund outside the submitted work, and is Executive Director of the nonprofit Chemical Probes Portal. W. Zuercher reports that he is a current employee of F. Hoffmann-La Roche AG and former employee of GSK.

The authors are members of the American Association for Cancer Research Chemistry in Cancer Research (CICR) Working Group, and discussions among its members have influenced our thinking and experiences, as well as our desire to share these concepts more broadly within the cancer research community. P. Workman thanks colleagues at the Chemical Probes Portal (www.chemicalprobes.org/) for valuable discussions, and acknowledges funding for the portal from a Wellcome Biomedical Resource and Technology Development Grant (212969/Z/18/Z) and support for his research from Cancer Research UK (Program Grants C309/A31322 and C309/A11566; Strategic Award C35696/A23187; and Infrastructure Award C309/A27413).