A Shift in Thinking: Cellular Thermal Shift Assay-Enabled Drug Discovery

Abstract

A decade has passed since the cellular thermal shift assay (CETSA) was introduced to the drug discovery community. Over the years, the method has guided numerous projects by providing insights about, for example, target engagement, lead generation, target identification, lead optimization, and preclinical profiling. With this Microperspective, we intend to highlight recently published applications of CETSA and how the data generated can enable efficient decision-making and prioritization throughout the drug discovery and development value chain.

Project prioritization and decision-making based on relevant and actionable data are keys to ensure a successful and efficient drug discovery process. Although the end objective, a marketed drug with the required therapeutic effect and favorable safety profile, may appear distant for preclinical drug discovery projects, it is important to stay focused on the target product profile. One crucial part of this process is the assessment of the interactions between a drug and its protein target in a physiologically relevant cellular environment. This step, also referred to as cellular target engagement, is critical for successful discovery and development of compounds with the desired biological and, ultimately, clinical effects. In 2014, Martinez Molina et al. introduced the patented CEllular Thermal Shift Assay (CETSA) for the quantification of target engagement in cells by studying the effect of compound binding on the thermal stability of proteins.¹ As neither the compound nor the protein is modified, CETSA provides reliable measurement of target engagement in native cellular environments under physiological conditions. Biologically relevant target engagement data can be obtained from any type of cellular matrices, such as cell lines, animal tissue, and patient samples, making the method applicable in all stages of the drug discovery and development process.

The basis for all thermal shift assays, cellular or not, is that the thermal stability of proteins can change upon ligand binding. In its basic form, CETSA is carried out on a single-protein basis, assessing target engagement by incubating cells with and without the studied compound and then exposing the cells to a transient heat-shock. Quantifying the amount of protein that remains soluble after the heating and plotting this across the range of temperatures yields the unique melt curve of the protein. If the compound has bound to the target protein, then the thermal stability of the protein is altered, causing a shift in the melt curve, also known as a thermal shift. This could be either a stabilization or destabilization of the protein; both events are caused by a change in the thermal stability of the protein and therefore are indications of cellular target engagement of the compound. Destabilizations can be caused, for example, by the compound interfering with a protein–protein interaction and thereby disrupting of a protein complex or the compound competing with the natural substrate.

By subjecting the cells to various concentrations of the compound and monitoring the degree of target protein stabilization or destabilization at a set temperature, target engagement potencies can be assessed. The CETSA EC₅₀ value is a relative measure of target engagement potency in that it incorporates more than the protein–compound binding affinity—it also incorporates factors such as cell permeability and potential bioactivation mechanisms of the compound, variations of binding site accessibility for assessed compounds, and effects of the initial cellular mechanisms such as protein degradation and subcellular translocation. All of these are factors that make data generated within a cellular system more complex but also more valuable than simplified biochemical or biophysical assay data generated on a recombinant, and often truncated, protein.

The CETSA assay has been applied to a broad range of proteins, including soluble cytosolic proteins, nuclear proteins, mitochondrial proteins, excreted proteins, and even multipass membrane proteins.^2,3 As one requirement for applying CETSA is that the ligand binding must result in thermal stability change of the protein, large proteins with high intrinsic thermal stability and highly disordered proteins with low intrinsic thermal stability can be more challenging to address.

The technology has proven well-suited for small molecules, and recent publications report on CETSA for profiling of promising new modalities, including proteolysis targeting chimeras (PROTACs)⁴ and molecular glue degraders.⁵

The CETSA protocol consists of four key steps: compound incubation with live cells or lysates, heat treatment at different temperatures, separation of proteins that remain folded from the ones that have denatured, and finally the protein detection step. The objective, sample matrix, and application of the CETSA study will determine the most relevant CETSA format to use (Table 1). The first developed and published CETSA format relies on Western blot for the protein detection and quantification step.¹ The publication describing the initial proof-of-concept studies has today been cited more than 400 times when this CETSA format has been utilized for cellular target engagement studies. A high-throughput (HT) microtiter plate-based CETSA format using dual antibody proximity detection technology is the format of choice for the discovery of new chemical starting points, tool finding, or profiling a larger set of compounds.⁶ Al-Amin et al. combined CETSA with the multiplex proximity extension assay (CETSA-PEA) for target engagement assessment of compounds against a set of selected proteins in a multiplex format.⁷ The sensitivity and throughput of this detection format allow for smaller sample volumes and higher throughput compared to the mass spectrometry (MS)-based formats.

Table 1. CETSA Formats.

	Detection method
	Western blot	Dual-antibody proximity assays	Split reporter system	Proteome-wide MS
No. of Compds	1–10	>100K	>100K	1–10
No. of Targets	Single	Single	Single	>7000
Advantages	• Unlabeled target protein	• Unlabeled target protein	• No detection antibodies needed	• Unlabeled target proteins
	• Transferable between different matrices	• Transferable between different matrices	• Automatable	• Proteome wide
		• Automatable	• High throughput
		• High sensitivity	• High sensitivity
Disadvantages	• Detection antibody performance	• Detection antibody performance	• Cloning step	• Low throughput
	• Low throughput	• Medium throughput	• Protein function validation step	• Low-abundance proteins difficult to detect
			• Cell line limitations	• Multipass transmembrane proteins challenging to detect
Applications	• Target engagement assessments	• Primary screening	• Primary screening	• Target identification
	• Target validation studies	• Hit confirmation	• Hit confirmation	• Mode of action studies
	• In vivo target engagement quantification	• Tool finding	• Tool finding	• Selectivity profiling
		• Lead Optimization	• Lead Optimization	• Biomarker discovery

A prerequisite for CETSA to successfully be applied is that the protein can be reproducibly detected and quantified, which makes low-abundance proteins more difficult to study. In these cases, the use of overexpressing cell lines can be an alternative. Another alternative is using a tagged protein for the detection.⁸ One example is the split reporter system, in which the protein of interest (POI) is tagged with a protein fragment of an enzyme that generates luminescence upon binding of a complementary substrate. A potential liability of the tagged format, however, is that the reporter tag can impact the function of the protein, localization, protein–protein interactions, and its melting temperature⁹ and should therefore be validated with orthogonal methods. With CRISPR/Cas9 engineering it is possible to mimic the endogenously expressed protein and avoid some of the potential artifacts that may arise with overexpression of tagged proteins.¹⁰

In contrast to the targeted CETSA formats for quantifying target engagement of compounds on pre-selected proteins, the proteome-wide MS-based CETSA can assess the thermal stability, in an unbiased fashion, of up to 8000 proteins in a single experiment.¹¹ Thermal proteome profiling (TPP), or CETSA MS, is therefore ideal for compound selectivity profiling, target identification, and mode-of-action studies. In the recently developed compressed CETSA format,¹² also known as protein integral stability assay (PISA)¹³ and one-pot,¹⁴ the range of temperature samples are pooled together per compound concentration after the heating step. This format requires less sample material and a shorter running time per sample condition on the MS instrument and allows for automation and easier sample handling. As a result, additional compound concentrations or replicates per condition can be included in the study, which in turn increases the statistical power and accuracy of the data.

Although the ultimate therapeutic target validation does not occur until the positive readout of the proof-of-concept research in phase 2 clinical trials, it is critical to strengthen the target validation and build confidence in the target hypothesis all throughout the drug discovery and development process. Early investigations of the relationship between phenotypic or functional efficacy and target engagement of the compounds will reduce the risk of late-stage efficacy failures and support the translation of the project hypothesis from the bench to the clinic. By combining CETSA data with downstream functional and/or efficacy data, it is possible to strengthen the target validation and connect phenotypic responses and cellular signaling with the initial compound–protein interactions on a cellular level (Figure 1). In a study by Petrilli et al., CETSA in lysate was used to inform about the binding of PROTACs to the target protein PCSK9-LDL, and a degradation assay in intact cells was employed to generate relevant efficacy data, thus confirming the connection between binding and the effect of the investigated PROTACs.⁴ In addition, studying a degrader in the proteome-wide version of CETSA provides information on the binding selectivity, specificity, and mode of action of the compound. In 2020, Chernobrovkin et al. published the proteome-wide CETSA profiling of a set of known immunomodulatory drugs (IMiDs), acting as molecular glue degraders, in both intact and lysed cells.⁵ In this study, CETSA MS profiling of target engagement and downstream pathway profiling were combined with data from unheated samples for assessing the degradation profile at different time points. In these experiments, the E3 ligase cereblon (CRBN) was confirmed as a direct binding target for all compounds. A time-dependent decrease in abundance was observed for several known protein targets and, interestingly, also for some previously unknown protein targets of these molecular glue degraders. In the same study, a cyclin-dependent kinase (CDK) 4/6-selective IMiD-based PROTAC was profiled, confirming the binding to CRBN as well as the time-dependent degradation of its intended target proteins CDK4 and CDK6.

Figure 1.

Timeline of potential cellular responses following initial compound binding.

Generating CETSA and cellular functional data on different compound series can enable prioritization for compound progression and aid in the design of new compounds.¹⁵ This is exemplified in a study focused on finding novel binders of the pain-signaling target human tropomyosin receptor kinase A (hTrkA), using CETSA for the target engagement assessment of allosteric and ATP-competitive inhibitors of hTRkA. The CETSA HT assay allowed for different hTrkA conformational states to be profiled in stimulated neuronal cells. The two types of inhibitors perturbed the thermal stability of the protein differently, which correlated with the binding of the novel allosteric compound to the inactive and not the active form of hTrkA. This finding was supported by biophysical and structural data and shows that CETSA can inform on the binding modes of different chemistries.^16,17

The CETSA HT format has mainly been used for screening of focused compound libraries and in hit confirmation efforts.^18,19 In a recent proof-of-concept study by AstraZeneca, CETSA was used for primary screening of a 500K compound library against the well-validated oncology target RAF proto-oncogene serine/threonine-protein kinase CRAF.²⁰ The developed CETSA HT assay proved to be robust, with a very low frequency of false positives and low susceptibility to PAINS. False positive hits in a CETSA screen could be caused by aggregation or assay interference by the compound, and it is important to carefully triage the hits and include a counter screen in the screening funnel. False negative hits could be due to compound binding without inducing a thermal shift of the protein. Moreover, there is often a log-difference between binding affinity and the compound potency in the CETSA assay. This could impair the identification of compound hits with weaker binding affinity and could be one of the reasons for the relatively low hit rate obtained in the AstraZeneca screen. Nevertheless, the screen yielded both known CRAF binders and novel compounds, which were confirmed as true hits in an orthogonal functional assay. The study proved the CETSA HT format to be a reliable tool for screening of both large and focused libraries for identifying hit compounds. Implementation of CETSA HT in the hit-to-lead strategy can be especially valuable for non-functional protein targets which cannot be pursued using traditional assay tools for screening. In the absence of antibodies for protein detection, the split reporter system could be an option for the screening assay.

Several proteomics approaches can be utilized for target identification, and there are strengths and weaknesses in all the methods. Combining orthogonal methods can be very useful to circumvent the limitations of each method. A comprehensive example of the strength of integrating CETSA MS with other proteomic techniques was recently published by Hendricks et al., where four orthogonal methods were systematically used to investigate the protein targets of a well-characterized CDK9 inhibitor.²¹ One of the methods included in the study, affinity-based chemical proteomics, uses an affinity probe mimicking the CDK9 inhibitor. The transcriptional regulator CDK9 was pulled down by the probe, and as this experiment was performed under mild lysis conditions that preserved multiprotein complexes, several of the molecular partners in the positive transcription elongation factor b (P-TEFb) complex (Cyclin T1, Cyclin T2, and Aff4) were identified with similar levels of competition and concentration responses, indicating that they were pulled down as an intact complex. As CETSA MS experiments in lysate mainly identify direct binding proteins, none of the pTEFb complex proteins showed changes in their thermal stability in the completely lysed cells. In addition to the primary target CDK9, other CDKs and other kinases, such as glycogen synthase kinase 3 (GSK3) alpha/beta, thermally shifted in the lysed cells, which overlapped well with the result observed in the kinase-based approach. In the intact cell experiment, GSK3 alpha and beta were thermally stabilized by the CDK9 compound. Other downstream proteins, such as Forkhead box protein K1 (FOXK1) and GSK3B-interacting protein (GSKIP), known to interact with GSK3 alpha and beta, were also identified as pathway-related protein hits, highlighting the unique opportunity of CETSA MS to discriminate between direct and indirect events by comparing the lysate and intact cell experiments. As the thermal stability of proteins is affected not only by direct binding of the studied compound but also by potential downstream consequences of the initial target engagement, the CETSA technology can be used for studying the pathway initiation and mode of action of the compound (Figure 2).

Figure 2.

A: The compound binds to the primary target affecting the thermal stability. B: Off-target proteins affected by compound binding. C: Compound binding to the primary target affects the thermal stability of associated proteins in the pathway. D: Unaffected proteins.

The strategy to combine CETSA MS data in intact Arabidopsis cells with phosphoproteomics data was employed by Lu et al. to generate novel insights into the downstream proteins of the plant-specific GSK3 inhibitor bikinin.²² These pathway proteins were affected in both their phosphorylation levels as well as thermal stability. The change in thermal stability can be caused by protein phosphorylation but also protein recruitment by a partner protein, translocation, and the general change of the redox state of the cell. Therefore, combining transcriptomics, proteomics, and cellular response with proteome-wide CETSA MS can be a powerful tool to understand the mode of action of a compound. This was the case with a small molecule previously identified as selectively inhibiting growth of human lung cancer cell lines, that was further characterized by Johnson et al.²³ Detailed transcriptome profiling pointed toward oxidative stress and apoptosis as probable cellular mechanisms responsible for the observed cellular response. Based on the results from the CETSA MS experiment, the identified targets involved in oxidation–reduction were selected for further validation in orthogonal methods such as biochemical enzymatic assays and cytotoxicity activity.

Phenotypic screening in drug discovery has the potential to identify hit compounds with an effect on the disease-relevant biology.²⁴ The challenge is to understand which protein targets are responsible for the desired phenotypic effect of the hit compound. The ability to perform the target deconvolution study in the same biologically relevant matrix as used for the phenotypic profiling and with unmodified compounds is an advantage of CETSA MS. In a Parkinson’s disease (PD) model, the natural flavonoid Kurarinone showed a promising result in reducing neuroinflammation via suppressing the activation of microglia.²⁵ For target deconvolution, CETSA MS was combined with another proteomics approach, solvent-induced protein precipitation, in the same PD mouse tissue. Soluble epoxide hydrolase (sHE), a protein previously known to play a role in neuroinflammation, was identified and subsequently validated as a target by Western blot CETSA, in combination with the observation that Kurarinone had no anti-inflammatory effect in sEH knock-out mice.

A caveat with many MS-based proteomics approaches for target deconvolution, including CETSA MS, is detection of membrane-bound protein targets. This is due to the biophysical nature and low abundance of membrane proteins. In a recent publication by Kalxdorf et al.,²⁶ selective enrichment of glycosylated cell surface proteins was combined with CETSA MS to effectively identify compound binding to extracellular membrane receptors. Another approach is the use of detergents to solubilize the membrane-bound proteins and thereby increase the number of membrane proteins detected.^3,27 However, the use of a detergent remains challenging and interferes with the protein stability after the heat step.²⁷ One successful example of the use of detergent in CETSA MS to identify membrane protein targets was reported by Carnero Corrales et al.¹²⁸ In this target deconvolution study of the autophagy inhibitor indophagolin, the purinergic receptor P2X4 was identified as a protein target in cell lysate with detergent. This was confirmed by Western blot-based CETSA. The thermal stabilization was found to occur at an order of magnitude higher than the autophagy inhibitory activity, consistent with the requirement of high compound concentration in CETSA to detect thermal stability changes.²⁰

Wilke et al. discovered a Sigma 1 receptor (S1) antagonist by using a morphological screen based on cell painting followed by proteome-wide CETSA MS for identifying the primary target of the cellular response.²⁸ In the study, 10,000 compounds were profiled in a cell painting assay to identify bioactive compounds. One of the hit compounds was profiled in proteome-wide CETSA in lysed cells, and by combing this data with the morphological fingerprint, the S1 protein was identified as the biology-driving target. Additionally, the group validated the binding of the compound to S1 by a radioligand displacement assay and confirmed cellular target engagement by using the targeted CETSA format (Western blot) in intact cells. The strategy to integrate the phenotypic screen with an unbiased target identification method such as CETSA MS enabled efficient identification of bioactive compounds and the biology-driving target and thereby facilitated swift progression of the project.

The proteome-wide format of CETSA can be used for unbiased selectivity assessment of compounds in the unmodified live cellular setting. In a recently published study by Amrhein et al., the selectivity of cyclin-dependent kinase 16 (CDK16) inhibitors was assessed by proteome-wide CETSA MS.²⁹ Following a differential scanning fluorimetry-based panel screen of approximately 100 kinases, a subset of compounds was profiled in intact cells by CETSA MS. The cellular thermal stability of the primary target CDK16 was perturbed by all compounds profiled, thus confirming CDK16 target engagement. Although the compounds are based on the same scaffold and all engaging CDK16, the number of off-targets varied between the compounds, indicating that small changes in the structure can have a large impact on compound selectivity. These results again stress the importance of using unbiased proteome-wide profiling for relevant assessment of selectivity.

Most reports on the use of CETSA have been in vitro experiments in intact cells and mainly from immortalized cell lines. Although the potential application of the CETSA technology for monitoring target engagement in clinical studies was suggested already in the first paper on CETSA,¹ very few studies have, as of today, been reported on patient-isolated samples. One example of the use of CETSA for assessing target engagement in patients was published in 2020.³⁰ In this study on glutaminase inhibitors, an AlphaLISA-based CETSA assay was developed and used for in vitro target engagement and in vivo target engagement quantification, generating thermal profiles of platelets from treated patients enrolled in the clinical study. This study demonstrates the ability to generate clinical target engagement data by CETSA profiling of samples from patients.

The current effort to expand the drug discovery toolbox has generated new chemical modalities, including protein degraders, biologics, RNA-targeting small molecules, and oligonucleotides.³¹ This is an area in which we believe CETSA will be adding value in the coming years, especially in the targeted protein degradation field, where CETSA could provide valuable information on the cell permeability and the in-cell performance of the degrader. Moreover, novel binders of both the POI and the E3 ligase could be identified in CETSA HT screens. For RNA-targeting therapeutics, although a protein is not its primary target, CETSA could provide a way to study the protein selectivity and the mode of action of the RNA modulators. For these modalities, the downstream protein consequences of the elimination of the target protein could be studied in the proteome-wide CETSA MS format.

The revolution of genomic screening in combination with computational biology to analyze the large datasets generated has led to identification of new potential therapeutic targets. Many of the novel targets identified through this approach are so-called undruggable targets; i.e., they are challenging to pursue in drug discovery efforts for several reasons: their function is often unknown, there is no or little available information on protein structure, and it is challenging to produce recombinant or purified protein samples, rendering both biophysical and biochemical approaches unavailable. The CETSA technology can overcome some of these challenges by enabling lead generation on the endogenously expressed native protein in live cells. Moreover, in the absence of tool compounds, CETSA in the HT format can also be very useful for tool-finding efforts early in the project. There is a large focus in the drug discovery field on using artificial intelligence (AI) and machine learning (ML) in lead generation, and by combining virtual screening with CETSA for hit confirmation, it is possible to rapidly identify chemical starting points and confirm their ability to engage their intended target in a biologically relevant matrix.

Another area of drug discovery where ML is starting to be incorporated is data analysis when large datasets are produced, with the proteome-wide CETSA format being no exception. Computational biology implemented in the CETSA pipeline could enable increased understanding of the relationship between the chemical structures and the changes in protein stability resulting from target engagement of the compounds.

In the coming years, we anticipate more reports of applications of CETSA in clinical research. To enable these applications, there is a need to develop the compatibility of the CETSA technology with more complex matrices such as human whole blood and allow for smaller sample volumes. This would allow for increased use of CETSA for clinical target engagement quantification and thereby better understanding of the required interactions between a drug and its protein targets in the ultimate environment of interest—the patient.

Glossary

Abbreviations

AI

Artificial intelligence

CDK

Cyclin-dependent kinase

CETSA

Cellular thermal shift assay

CRBN

Cereblon

FOXK1

Forkhead box protein K1

GSK

Glycogen synthase kinase

GSKIP

GSK3B-interacting protein

HT

High-throughput

hTrkA

Tropomyosin receptor kinase A

IMiDs

Immunomodulatory drugs

ML

Machine learning

PD

Parkinson’s disease

PEA

Proximity extension assay

P-TEFb

Positive transcription elongation factor b

PROTAC

Proteolysis targeting chimera

MS

Mass spectrometry

S1

Sigma 1 receptor

sHE

Soluble epoxide hydrolase

The authors declare the following competing financial interest(s): I.M.C. and S.L. are employees of Pelago Bioscience AB.

Special Issue

Published as part of the ACS Medicinal Chemistry Letters virtual special issue “New Enabling Drug Discovery Technologies - Recent Progress”.