If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
1 Present address: Sygnature Discovery, Bioscience, Alderley Park UK. 2 Present address: Roche Pharma Research and Early Development, Therapeutic Modalities, Roche Innovation Center Basel.
The cellular thermal shift assay (CETSA®) has increasingly been used in early drug discovery to provide a measure of cellular target engagement. Traditionally, CETSA has been employed for bespoke questions with small to medium throughput and has predominantly been applied during hit validation rather than in hit identification. Using a CETSA screen versus the kinase CRAF, we assessed 3 key questions: (1) technical feasibility – could the CETSA methodology technically be applied at truly high throughput scale? (2) relevance – could hits suitable for further optimisation be identified? (3) reliability – would the approach identify known chemical equity. Here, we describe the first large scale AlphaLISA SureFire based CETSA (Alpha CETSA) approach allowing us to screen a large library of almost 0.5 million compounds. We discuss the issues overcome in automating and executing the screen and describe the resulting screen output.
High throughput screening (HTS) cascades have evolved to identify high-quality hits utilising assays applied to the testing of large compound collections. Often, primary screens focus on the identification of modulators of biological function, with direct target engagement assays placed further down the cascade [
]. Where technologies capable of interrogating target engagement are employed, for example, direct binding biochemical methods, indirect methods such as probe-displacement assays, or biophysical methods such as differential scanning fluorimetry (DSF), they are usually dependent upon the availability of purified protein. Additionally, these approaches are undertaken in the less physiologically relevant setting of the isolated protein interrogated in buffer [
The ability to demonstrate compound-target engagement in a cellular context, is therefore, attractive, and important when identifying and progressing hits. Furthermore, the use of cellular assays during primary screening allows compounds eliciting modulation of the target in its native state to be identified. The advantage this provides over isolated protein assays is that compound effects can be observed when they require interactions of the target protein with biological partners in a physiologically relevant network. However, cell-based screens often can be complex and may be prone to off-target effects, thus requiring additional, time-consuming target deconvolution studies for active compounds. As it requires no time-consuming cell line engineering, the Cellular Thermal Shift Assay (CETSA®) provides the potential to use more complex cell models or primary cell lines allowing target engagement to be monitored in these cellular settings [
]. The CETSA® approach relies upon the thermodynamic stabilising effect of ligand-binding to a target protein in disease-relevant cells and provides a format that often is more straightforward than other cell-based methods.
By utilising a modification to the original CETSA method, employing an isothermal approach, where cells are heated to a defined temperature previously demonstrated to achieve around 90% of target unfolding, more rapid screens can be established. In this approach, unbound target protein would be expected to unfold and precipitate upon this heat-shock, whereas ligand-bound protein will remain folded in the supernatant [
]. Monitoring thermal stabilisation of a cellular target in order to detect ligand binding therefore requires a quantitative and scalable endpoint to measure soluble, thermostable protein in cell lysates, following application of a heat-shock to live cells. Previous reports have demonstrated the use of split reporter systems to enable and scale such protein detection [
A high-throughput dose-response cellular thermal shift assay for rapid screening of drug target engagement in living cells, exemplified using SMYD3 and IDO1.
In the approach exemplified here, the stabilized protein is detected via interaction of a pair of anti-species antibodies in an AlphaScreen® system, offering the advantage of application to unmodified target proteins in any cell background [
]. This high throughput (HT) CETSA® format offers the potential for large numbers of compounds to be tested in a HTS setting. Here, we report the development and application of an AlphaScreen-based CETSA HT (Alpha CETSA), delivered through a collaborative effort utilising CETSA expertise from Pelago Bioscience, reagent development from PerkinElmer and screening capability at AstraZeneca. The screen tested almost 0.5 million compounds for their ability to bind to cellular CRAF. This library of compounds was a subset of the full AstraZeneca collection representing diversity of the full collection and containing both core (Mw: 325–750 Da) and low molecular weight compounds (Mw: 250–325 Da). Our data demonstrates that CETSA HT can be used as a primary screening approach, identifying relevant binders in cells from large screening collections. This is particularly timely given the expanding interest across the drug discovery industry in new therapeutic modalities, such as PROTACs (PROteolysis TArgeting Chimeras), where proteins lacking a catalytic function may now potentially be targeted [
]. Approaches such as CETSA HT will be important in identifying ligands for the protein of interest (POI), that are demonstrated to bind in the cellular setting, facilitating PROTAC development [
THP1 suspension cells were cultured in phenol red-free Roswell Park Memorial Institute (RPMI) 1640 media + 1% L-Glu+ 10% foetal bovine serum in a humidified incubator at 37 °C, 5% CO2. Cells were short tandem repeat (STR) DNA-profiled and confirmed mycoplasma free.
Compounds
Dabrafenib (20 µM, 50 nL from 4 mM DMSO stocks) was used as positive control in every plate.
The negative (neutral) control was DMSO (50 nL from 100% DMSO stocks).
Single point screening
Compounds were tested at 50 µM by firing 50 nL of 10 mM stock solution in DMSO using an Echo 555 Liquid Handler (Labcyte).
Concentration response screening
Compounds selected for follow – up screening were tested as 10-point concentration response curves ranging from 125 µM to 0.015 µM. Compounds were fired from the Echo 555 and backfilled with DMSO to create desired concentrations, while keeping total DMSO volume to 50 nL.
Preparation of cells
Cell density and viability of THP1 cells grown in 10-layer stack was measured. If cells were acceptable (>90% viability), cell solution was decanted into a 1 L flask and spun at 400 g for 4 min. Supernatant was then decanted, and pellet resuspended in warm Hanks Balanced Salt Solution (HBBS), cells were counted and diluted to appropriate concentration in HBSS. Cells were kept at 37 °C by keeping in a warming jacket and kept in suspension by stirring with a magnetic stirrer at 100 rpm.
Compound incubation
Cells were seeded in BioRad hard shell 384 well PCR plates at a volume of 10 µL/ well using a MultiDrop Combi (ThermoFisher) at standard speed (standard cassette) to yield 20,000 cells/well. A tick time of 1 min per dispense was used to ensure incubation times remained constant throughout the heating process. Plates were shaken for 5 s to avoid air gaps due to conical wells. Plates were then sealed with clear sealing foils (Agilent 16985-001) on automated PlateLoc (Agilent) (130 °C for 1 s). Plates were incubated at room temperature for one hour. Following incubation plates were heat-shocked using a LightCycler480 (Roche), heating to the target specific Tm (CRAF = 47 °C) and holding for 3 min, then cooling to 20 °C before removing from LC480. The rate of cooling was set to the maximum permitted by the LC480 instrument, resulting in cooling from 47 to 20 °C taking less than 1 min. Immediately after heat shock, 10 µL of 2 x PerkinElmer lysis buffer was dispensed using a Certus Flex (Gyger). Plates were then sealed with aluminium foil on a PlateLoc (150 °C for 1 s). Plates were incubated at room temperature for a minimum of 10 min before freezing at −20 °C. The introduction of the freeze-thaw step allowed for greater throughput and aided gentle cell lysis, however the distortion of PCR plates upon thawing led to issues with robot handling and would not be recommended for future high throughput CETSA protocols.
Automated lysate transfer from PCR plates to AlphaPlates
Prior to loading onto the HighRes Biosolutions CoLAB Microstar automation platform, the sealed microtiter PCR plates were thawed for a minimum of 30 min at room temperature and then loaded into an automated SteriStore incubator (HighRes Biosolutions) at 4°C. Utilising Dispenser and VPrep CoLAB Flex carts, the PCR plates were shaken for 3 s at 1000 rpm (BioShake, QInstruments) and then centrifuged at 300 rpm (Microspin, HighRes Biosolutions) for 1 s, de-sealed (XPeel, Nexus) and then mixed (VPrep, Agilent) by pipetting 10 µL lysate up and down ten times at slow pipetting speed (aspirate and dispense pipette speeds of 50 µL /sec). Cell lysate (5 µL) was transferred to an AlphaPlate using the VPrep, dispensing 1 mm above the base of the plate well, again at slow speed. Increasing the number of mix cycles had a positive impact on the lysis step, but after 10 cycles there was no further improvement in cell lysis. The VPrep tips (30 µL Maximum Recovery, Axygen) were then washed with 10 cycles of 10 µL of 10% ethanol in water solution, with an additional 10 µL rinse in Hanks Buffered Saline Solution (HBSS), tips were exchanged for new after 30 uses, no traces of carry-over were observed. This was assessed by initially testing with brilliant black in DMSO to visually check for residue left on tips. Following this, concentration response plates with potent compounds were run and no carryover effects seen on following plates were observed.
ALSU (AlphaLISA SureFire) detection
Total RAF1 (CRAF) AlphaLISA SureFire reagents were sourced from PerkinElmer (ALSU-TRAF-A50K). Acceptor bead solution (2 µL, PerkinElmer) was added to all wells of the AlphaPlate using a Certus Flex (Gyger). Plates were then covered with black lids to ensure protection from light and incubated for 1 h at room temperature. Following incubation, 2 µL donor beads were added to all wells using a Certus Flex in dimmed conditions (<100 Lux). Alpha plates were then sealed with clear foils (130 °C for 1 s) and stacked on an Envision plate reader (PerkinElmer). Following an overnight incubation, plates were read on the Envision using an AlphaScreen module.
Data analysis
Primary screening hit identification and 4-parameter logistic concentration-response curve-fitting data analysis was carried out using Genedata Screener version 16 (Genedata AG). Data were normalised using a two-point normalisation, setting Neutral DMSO Control at 0% and the Stimulator (Dabrafenib) control at 100%.
Both the % inhibition and the Robust Z-score were calculated for the compounds within the Screener session. For the routine normalization of well data with Neutral and Inhibitor control well groups as reference, the following equation was used to normalize the signal values to the desired signal range in Screener Assay Analyzer:
Where x is the measured raw signal value of a well, <cr> is the median of the measured signal values for the Central Reference (Neutral) wells on a plate, <sr> is the median of the measured signal values for the Scale Reference (Inhibitor) wells on a plate, CR is the desired median normalized value for the Central Reference (Neutral), SR is the desired median normalized value for the Scale Reference (Inhibitor).
Screener also performs a modified or robust Z-score normalization using the following equation:
Where ≪cr≫ is the median absolute deviation (MAD) = median (|x -<cr>|)[
Iglewicz B, Hoaglin D. 1993. How to detect and handle outliers. Volume 16. In The ASQC Basic References in Quality Control: Statistical Techniques, E.F. Mykytka, Ph.D., Editor, ASQC Quality Press.
Active compounds were labelled A when the Z-score was greater than or equal to 4 and they demonstrated at least 30% activity.
Results
Initial generation of a CETSA HT assay to measure target engagement with CRAF
Development of the AlphaLISA involved initial experiments to identify optimal buffer conditions and cell volume and were performed using the AlphaLISA toolbox assay approach [
]. This identified a pair of mouse-derived and rabbit-derived antibodies to the target and an anti-mouse secondary antibody conjugated to Alpha Donor beads as well as an anti-rabbit secondary antibody conjugated to AlphaLISA Acceptor.
Following the AlphaLISA toolbox development, the endpoint was adapted to an AlphaLISA Surefire (ALSU, Perkin Elmer) utilising a biotinylated CRAF antibody coupled to Streptavidin-coated Alpha Donor beads and an alternative antibody to CRAF coupled to AlphaLISA acceptor beads using CaptSureTM technology (TGR Biosciences).
Direct coupling of the Donor and Acceptor beads to the anti-CRAF antibodies was found to significantly increase the AlphaLISA emission signal and the ALSU was used to perform a CETSA melt curve experiment in live THP1 cells. This demonstrated loss of ALSU signal consistent with thermal unfolding and aggregation of the CRAF protein between 46–48 °C. In addition, cells treated with the pan RAF-kinase inhibitor Dabrafenib demonstrated a thermal stabilisation of CRAF (Fig. 1). However, we did not observe significant thermal stabilisation of CRAF in the presence of other reported CRAF inhibitors PLX5568 [
Fig. 1Thermal melt curves generated against CRAF using the AlphaLisa Surefire kit in the presence of DMSO or control compounds. Melt curves confirm stabilisation of approximately 4.8 °C in the presence of Dabrafenib, but no stabilisation in the presence of other reported RAF inhibitorsPLX5568 or Compound 10d. Data represents the mean ± SD n = 3.
Initial analyte titrations were completed to ascertain optimal cell density. In a 17 µL assay volume, unheated ALSU signal plateaued at 15,000 cells/well, with no further increase in ALSU signal with increasing cell numbers (Fig. 2a, top left panel). In a miniaturised 9 µL assay volume, a hook effect was observed whereby reduced ALSU signal was observed at higher cell densities above 10,000 cells/well, (Fig. 2a, bottom left panel). Cell density had little impact on ALSU signal from heated samples (Fig. 2a, top right and bottom right panels) Since the signal to background (signal following heating to 70 °C) was sufficiently high for further optimisation at a cell density of 10,000 cells per well in 9 µL volume, allowing a 2.5-fold reduction in reagent consumption, these conditions were used to investigate bead concentrations. We found a linear reduction in unheated (RT) ALSU signal upon dilution of the ALSU antibody-bead conjugates, and reduction in total ALSU signal (RT and 70 °C) when both beads and buffers were diluted (Fig. 2b). These experiments enabled miniaturisation of the ALSU assay reagents to 5 µL of cell lysate and 2 µL of each of the ALSU acceptor and donor bead mixes, while generating sufficiently high signal values.
Fig. 2Exploration of reagent requirements. A: Analyte cell density titration of THP-1 cells (RT) and heated THP1 cells (70 °C) in both initial detection volume (17 µL) and miniaturised volume (9 µL). B: Reduction of reagent concentration at a fixed analyte concentration to ascertain acceptable dilution levels for signal generated. In both cases the signal for the heat-treated cells is around 2 orders of magnitude lower than the un-heated signal, demonstrating a wide assay window.
Using the optimised and miniaturised ALSU detection conditions to quantify thermostable CRAF, an isothermal dose response protocol was performed, heating cells to a single temperature of 47 °C and comparing compound signal to DMSO controls. Dose response curves were generated for the same tool compounds described in Fig. 1. Dabrafenib was identified as being capable of thermally stabilising CRAF in a dose dependent manner with an EC50 of 1.2 µM (Fig. 3). As a result, Dabrafenib was used as a positive control for HTS optimisation and validation experiments, as well as an on-board control for primary and follow up screening. We observed some weak activity in the high µM range with PLX5568, but no observable stabilisation of CRAF by compound 10d. Both PLX5568 and compound 10d have been reported as CRAF inhibitors, which could indicate some limitations of the CRAF CETSA assay to compound binding modes or potency.
Fig. 3CETSA isothermal concentration response curves using ALSU detection volume of 9 µL. Dabrafenib, previously identified as demonstrating a CETSA thermal shift in Fig. 1, demonstrated dose dependent thermal stabilisation of CRAF in THP1 cells. Data represents the average ± error n = 2.
Optimisation of the CRAF CETSA HT assay for high throughput screening
The cell density titration was repeated following transfer of the assay onto the instrumentation in the High-throughput screening laboratory to identify optimal cell number per well while keeping cell requirements low. The signal of both positive and neutral controls increased with cell density, with the resulting signal:background and Z’ values showing a tendency to plateau once cell density surpassed 5000 cells/well. A cell density of 5000 cells/well was selected for screening (Fig. 4).
Fig. 4Positive control (Dabrafenib treated) and Neutral control (DMSO treated) signal following 47 °C heat-shock of THP1 cells, obtained at increasing cell densities. RZ' values
To increase the ease of use for high throughput screening the use of assay ready cells from cryovials was investigated in comparison to cells from continuous culture. We found that the signal from continuous cultured freshly plated cells was around 10-fold higher than from cryovials and that cryo-preserved cells assayed immediately after thawing showed a rapid signal drop off over time. For these reasons it was not possible to use cryo-preserved cells throughout an entire screening day (∼ 8 h) (Fig. 5). The final screening protocol is summarised in Supplementary Fig. 1.
Fig. 5Comparison of signal stability for cells from continuous culture and assay-ready cryovial cells. A: Positive control (Dabrafenib), Reference (Dabrafenib at half concentration, 10 µM) and Neutral control (DMSO) signal from continuously cultured cells seeded into plates at various times following harvesting and resuspension in HBSS. RZ’ values for 0, 1.5, 2.5 and 4.5 h, respectively are: 0.57, 0.77, 0.68, 0.82. B: Complementary signal from assay ready cryovials seeded at various times following harvesting and resuspension in HBSS. RZ’ values for 0, 1.5, 2.5 and 4.5 h, respectively are: 0.62, 0.13, −1.10, −0.20. Data shown is mean ± SD n = 3.
Following optimisation of the assay, several validation experiments were carried out to ensure the assay remained reproducible and robust in the presence of compounds and during scale up. A well annotated set of nuisance compounds termed the undesirable mechanism of action (uMOA) collection exists within the AstraZeneca compound collection. Represented in the set are examples of redox cycling, colloidal aggregate, thiol reactive, cytotoxic, chelating, DNA-binding and surfactant type molecules in addition to common metal contaminants [
]. It is utilized within AstraZeneca to assess and address the susceptibility of targets and assay systems to these common causes of false positive and undesirable hits in HTS screening. Testing of the uMOA compound collection in the CRAF CETSA HT assay revealed a remarkably low level of susceptibility to nuisance compound interference, indicating a low propensity for identifying these types of compounds during screening (Supplementary Fig. 2).
Additionally, a collection of approximately 1500 compounds, representing diversity across the AZ compound collection were tested in two separate runs, with compounds being placed in different positions on the plates. This was used to explore variability between runs and assess the robustness and reproducibility of the assay. Fig. 6A demonstrates that the assay was reproducible. This experiment predicts an extremely low hit rate, with no compounds from the diverse 1500 set producing greater than 11% effect. This may correspond with the low level of susceptibility to nuisance compounds. Further evaluation of the assay using a small subset (∼800) of compounds predicted to interact with kinase targets was undertaken to explore the potential for identifying kinase hits. In this case, compounds showed good reproducibility between separate screening runs for most compounds identified as ‘active’ (showing > 30% effect in both runs) although a few outliers were observed (Fig. 6b). Based on the screening results for these small collections, the hit rate for a larger screen was predicted to be low for most of the collection screened (<1%) but enriched for kinase inhibitors (∼1.8%).
Fig. 6CETSA assay reproducibility. A: Activity of 1500 diverse compounds tested on 2 separate occasions (X axis and Y axis). B: Activity of 800 kinase inhibitors tested on 2 separate occasions (X axis and Y axis). Solid lines represent a 1:1 correlation of % activity across the two runs and indicate the 30% activity cut-off for each experiment.
The CRAF CETSA assay was scaled to HTS screening and a total of 462,888 compounds from the AstraZeneca collection were screened for their ability to thermally stabilise cellular CRAF at a screening concentration of 50 µM. This was accomplished across 18 experimental runs, where each experiment processed 80–100 screening plates. Assay parameters across screening runs are shown in Fig. 7. Hits were defined as compounds that increased CRAF ALSU signal to >30% of control and significantly outside the deviation of the on-plate controls as calculated by a robust Z-score ≥ 4.
Fig. 7Screening metrics displaying the range of key assay parameters across the 18 primary screening runs. A: number of compounds screened per run. B: variation of signal:background for each run. C: robust Z’ calculated for each run. D: calculated hit rate per run.
Consistent with initial experiments, a low hit rate of approximately 0.4% was observed. As expected for a kinase target, these hits were strongly enriched with known kinase pharmacophores, within which the hit rate was 1.7%. In general, the experiments maintained acceptable assay parameters over the course of the screen, with RZ’ (where RZ’ is the robust Z prime value – also see Figs. 4 and 7) [
] routinely over 0.4, even where individual experiments show a reduced signal-to-background. This suggests that automated high-throughput CETSA screens can be robustly and reliably applied at this scale.
CETSA screening identifies relevant chemistry
Following completion of the screening of 0.46 million compounds at a single concentration, 2058 compounds were defined as hits. These compounds were collected and expanded by near-neighbour analysis to supplement with similar chemical structures from within the AstraZeneca collection. This produced a collection of 2263 compounds of interest which were subsequently tested in a concentration response form of the CETSA HT assay. Cherry-picked compounds selected for concentration response follow up were assessed for purity via LCMS, with 155 compounds found to have less than 80% purity, and so were discarded from further testing. Triage assays aimed at identifying metal contaminants [
] and redox cyclers were also undertaken. Of 1541 compounds sent for metal triage testing, 22 (1.4%) were found to have metal contaminants (3 copper, 11 palladium, 8 unclassified). 29 hits (1.9%) were found to be redox cyclers.
634 of the tested compounds demonstrated significant concentration-dependent thermal stabilisation of CRAF in the isothermal concentration response CETSA assay, though many showed weak activity and partial curves. 247 compounds were identified with EC50 < 10 µM. The most potent hit compound identified from the screen produced a CETSA EC50 of 0.08 µM. The majority of these 247 compounds displayed pharmacophores which were common amongst kinase inhibitors and could broadly be categorised into 5 groups based on structures common to inhibitor classes (Fig. 8).
Fig. 8Primary screening output. A: Hits grouped by inhibitor class. B: Exemplar structures from each. inhibitor class.
Additionally, 196 of these 247 originated from historical AstraZeneca kinase programmes, with 99 of these having demonstrated activity against the related kinase BRAF, indicating a propensity within the identified hits for activity against kinases and the related BRAF kinase. The distribution of origin of the CETSA hits is shown in Fig. 9. This suggests that the CETSA-HT assay was capable of identify chemical equity expected to have potential for activity against CRAF.
Fig. 9Distribution of CETSA-HT hits grouped according to target class versus which they were originally synthesised. 37.6% BRAF inhibitors, 14.1% TEK inhibitors, 19.4% non-kinase inhibitors, 28.9% other kinase inhibitors.
CETSA identifies genuine CRAF ligands with minimal false positives
To further investigate the susceptibility to false positives from the CETSA-HT screen, we performed artefact testing on 603 available compounds from the 634 hits which had shown concentration dependent thermal stabilisation of CRAF ALSU signal. A control plate was used with Max/Min responses against heated THP1 cells in order to define the ALSU signal within the experiment that represents the 30% effect cut-off applied to identifying hit compounds (Fig. 10A). In parallel, 603 hit compounds were tested at 50 µM in an artefact assay where no THP1 cells were present, to explore the effect of compounds on the ALSU signal alone within the assay. In this setting where compounds were incubated only with the HBSS buffer prior to heating and ALSU endpoint, 9 compounds were identified which increased ALSU signal above 30% and are therefore artefacts having an effect upon the ALSU endpoint, with a small number of additional effects just below 30% (Fig. 10B).
Fig. 10ALSU artefact assay for 603 hits from the CRAF CETSA screen. A: THP1 cells treated with Max (Positive Inhibitor control) or Min (Neutral DMSO control) were used to define 30% activity (dashed line) within the experiment, the cut-off used to identify hits from screening. B: Across two screening plates, 603 hits from the screen were tested for their impact on ALSU signal by performing an identical screening run in the absence of THP1 cells. Compounds were incubated with HBSS, heat-shocked at 47 °C and ALSU endpoint developed. 9 compounds were observed to artificially increase.
As described above (Figs. 8 and 9), the majority of hits from the screen showed elements of known kinase inhibitor pharmacophores and evidence of activity against kinases. From the 634 hits, a selection of 17 compounds were identified which were considered more novel pharmacophores and were therefore further tested for genuine activity against CRAF. These 17 compounds underwent further testing in isothermal dose responses and complete CETSA melt curves. Upon more thorough analysis by performing full CETSA melt curves, fourteen of these 17 compounds appeared to be false positives, demonstrating the utility of performing CETSA melt curves to differentiate true hits from artefacts (Supplementary Fig. 3), and suggesting as expected that false positives are enriched outside of compounds showing kinase pharmacophores.
Outside of the kinase related structures, three compounds were identified representing true positives with more novel pharmacophores, though the precise binding mode and activity of these compounds will require further study. Such compounds demonstrated small but significant thermal shift events in CETSA melt curve experiments (Supplementary Fig. 3). The most potent of these novel hit compounds had an isothermal CETSA EC50 of 15.5 µM (Fig. 11A) and demonstrated activity in a commercial CRAF activity assay (CRAF Y340D Y341D Z-Lyte® SelectScreen® Kinase Profiling, LifeTechnologies) with an IC50 of 0.049 µM (Fig. 11B), further strengthening the conclusion that the CRAF CETSA screen identified CRAF ligands.
Fig. 11CRAF CETSA screening identifies CRAF inhibitors. A: CETSA HTS hit compound in isothermal dose response experiments (EC50 15.5 µM). B: the same compound tested in CRAF Y340D Y341D Z-Lyte kinase inhibition assay (LifeTechnologies) (IC50 0.049 µM). The example hit compound demonstrated activity as a CRAF inhibitor in an isolated protein system. Some of the difference in mid points may be explained due to the difference in assay format (biochemical versus cell).
CRAF is a proto-oncogene serine/threonine-protein kinase, which is part of the ERK1/2 pathway and a MAP kinase (MAP3K) that functions downstream of the RAS subfamily of membrane associated GTPases. RAF kinases are downstream effectors of the small GTPase RAS. Mutations in RAS and RAF are associated with a variety of cancers and genetic disorders. Of the three RAF isoforms, CRAF is most frequently involved in tumour initiation by RAS. Cytosolic RAF is auto-inhibited and becomes active upon recruitment to the plasma membrane.
CRAF was selected for this proof-of-concept study, as many documented inhibitors of CRAF and closely related targets were already known [
], allowing us to compare the screen output with previously identified compounds. As demonstrated above, this CETSA screen was able to identify known pharmacophores and related chemical equity for CRAF. This data provides the first evidence that high-throughput Alpha CETSA can be used successfully as a primary screening method in HTS, allowing it to be usefully applied to more novel targets, potentially intractable to other assay methodologies.
Identifying relevant hits for novel targets is essential to start chemistry and drive drug design. Traditionally, affinity-based screening applications have been limited to isolated proteins in biophysical and biochemical assay formats, such as differential scanning fluorimetry (DSF) or probe displacement assays. These non-native experiments may provide the opportunity for rapid tool finding but typically require large scale protein supply which is not always feasible for novel targets and therefore removes these approaches from consideration for truly large-scale screening. Cell-based screens have often bridged the gap between physiological relevance and throughput, but traditional cell-based assays often require dedicated specific assay design and read-outs proximal to the target, to ensure the assay can identify relevant mechanisms. Additionally, cell-based assays, such as reporter gene assays or assays that rely on a luminescent/fluorescent read outs, typically require modification of the cellular genome, meaning any ‘hits’ found in these assays may not directly demonstrate activity in non-modified cells. Thus, the application of CETSA has some advantages over traditional cell-based screening.
We have successfully shown that CETSA can be applied at scale and that the approach can identify genuine hits, including previously known pharmacophores and novel compounds. The vast majority of hits identified showed pharmacophores consistent with orthosteric active site inhibitors, and while a small number of more novel hits were identified, activity against recombinant CRAF observed for these compounds could also indicate an orthosteric mechanism. The precise mechanism of these hits will require further study. We believe that having demonstrated the validity of HTS screening by CETSA, that it will have utility in screening against unprecedented targets with a reasonable likelihood of success for identifying genuine hits. A key potential weakness of this approach is the reduced apparent potency observed in CETSA assays that may limit the propensity to find hits with weaker affinity. However, we observed a low level of interference in the screen, and this may be a particular strength of the approach when applied to intractable targets. We also found a very low hit rate from testing the uMOA collection in CETSA compared to other previously used common assay formats. Taken together, the detection of known active compound sets and the minimal effect of nuisance compounds demonstrates that the CETSA assay was a robust and reliable method to identify CRAF binders. It gives a first indication that this approach could be expected to provide a beneficial way for screening more difficult targets.
We also have demonstrated that CETSA can reliably be run in isothermal mode, rather than requiring full melt curve generation, resulting in the heat shock stage being much simpler and faster. Additionally, the application of an antibody-based detection system allowed cells requiring no genetic modification to be used, facilitating procurement and ensuring physiological relevance. This format assay allowed us to screen approximately 500,000 compounds in just 18 screening runs. The ability to miniaturise into a 384 well format was essential to allow this scale of screen.
Initial experiments investigating the miniaturisation of the detection reagents proved critical to enable this scale, facilitated by the high signal produced by the ALSU kit compared to the toolbox assay originally developed. Automation of the liquid transfer step from heated PCR plates to detection plates also was essential to allow the high throughput achieved.
The large number of compounds tested necessitated significant numbers of cells. Frozen cells proved to be unsuitable for this assay with continuous culture from 10-layer stacks required to deliver the required cell numbers. This produced an increased demand on resource and consumable costs, as well as a potentially increased risk due to cell contamination or slow cell growth. We suggest that this is an area for potential improvement.
We have mentioned the benefit from automation of the liquid transfer step above, but there are many other stages of the assay that we believe are limited due to the manual nature of these steps, which resulted in increased screening time, the resource required and the probability for errors. We screened 1320 plates, of which 103 were rendered unusable at some point during the assay process, equating to a failure rate of almost 8%. Process improvement to reduce this error rate could be achieved by addressing the steps that were responsible for repeat faults, decreasing the number of steps, as well as reducing manual steps where possible. For example, the day one processes of cell seeding, heating and lysing were performed manually, preparing up to 100 plates per day. To attempt to ensure timings were kept as consistent as possible, a delay of 1 min was introduced between each cell plate dispense to match the downstream LC480 throughput in applying the 47 °C heat-shock. Cells were then incubated with compound at room temperature for one hour before being placed on the LC480. The availability of four LC480 instruments increased throughput, but the manual protocol set-up on each machine sometimes introduced delays so that tick times were not always maintained. Following heating, cells were lysed as soon as possible, introducing another potential source of error, with four plates requiring lysis all at the same time, but with this not always being possible. Additionally, both the MultiDrop Combi and Certus dispensers sometimes became blocked or mis-dispensed reagent, leading to missed wells or delays to timings, as the issues were resolved.
To mitigate the time-consuming process of preparing the plates, a freeze thaw cycle was included to allow the assay to be split across multiple days while keeping the lysate stable. This step introduced other, unexpected issues. The freezing and subsequent thawing of plates caused the plastic to warp and produce a concave shape, resulting in plates not lying flat on the automation stages. This caused the robot arm to ineffectively grip the plates, resulting in them being dropped. Plates therefore required manually straightening to be reliably placed on the automation.
Together, the manual plating and heating, freezing, liquid transfer and overnight detection resulted in the assay taking 3 days run time for around 100 plates. These areas are potential points for improvement.
Despite these potential issues, the CETSA HT assay performed reliably and reproducibly between runs, with a low hit rate. The assay was found to have a low susceptibility to known nuisance compounds. These aspects might suggest that the CETSA assay has lower sensitivity than other assays and that compounds may appear to have lower potency in this assay format. From simulations in isolated protein systems, the degree of thermal stabilisation has been shown to increase with increasing ligand concentration, even beyond saturation of the protein [
]. This influenced our decision to screen compounds at an elevated concentration of 50 µM to increase the chances of identifying weaker hits. The assay was tolerant to increases in DMSO concentration and permitted this increased screening concentration.
The observed reduced potency in this assay also suggests that EC50 values obtained from concentration response testing, whilst internally quantitative and comparable to inform compound ranking, may not accurately reflect the potency and affinity of the compound in other assay formats. This lower sensitivity may explain why some CRAF inhibitors did not show activity in the CETSA assay. As this is an indirect measurement of binding affinity, relying on the linkage to thermal stabilisation, it also is possible that the CETSA approach may be biased towards compound classes with particular physicochemical properties, influencing the enthalpic and entropic contributions to binding, and so less sensitive for lower potency, enthalpically driven compounds. We realise that this may mean that the CETSA approach may introduced a risk for false negatives. However, when positioned at an early stage, sometimes as the only approach for hit finding for novel or intractable targets, it provides an opportunity to identify tool compounds that may facilitate the development of alternative assays capable of picking up a wider range of chemical equity.
Considerations for Automating CETSA Assays:
It is essential to use hard-shell PCR microtiter plates when automating CETSA assays, as soft-shell PCR plates bend & deform after even gentle heating steps (thermal shock & sealing) resulting in robot handling errors. Limitations in automation availability required the assay to be run over several days, with a requirement to seal plates mid-assay for overnight storage. We encountered issues when sealing the soft-shell PCR plates, as use of an aluminium thermal seal also induced plate bending, especially during the freeze/thaw step, which again resulted in robot handling errors. This issue could be resolved by manually re-straightening plates after thawing. We found that using a clear thermal seal reduced the plate bending although these seals sometimes detached when plates were frozen. Improvements in automation have now allowed us to perform the whole automated assay (apart from the final read step) in one day, eliminating the requirement for sealing, freezing and de-sealing plates mid-assay. We also found that it was important to mix, centrifuge and then re-mix the contents of the PCR plate post thawing to ensure efficient cell lysis, prior to transferring lysate into the AlphaPlates. We explored using the Multidrop Combi shake functionality to mix the plates after they had thawed, but a BioShake was found to offer superior mixing. However, care must be taken to optimize the speed and time of this BioShake shaking.
Overall, this initial proof-of-concept study demonstrates that CETSA can be applied at scale and is able to reliably identify small molecule binders for targets directly in live cells. The study also highlights valuable learning that can be applied to future CETSA screens.
Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:
Geoffrey A. Holdgate, Hannah Rowlands, Kirsten Tschapalda, Carolyn Blackett, Delyan Ivanov, Darren Plant, Joseph Shaw, Andrew Thomas, Martin Packer reports financial support was provided by AstraZeneca UK Ltd Macclesfield. Laurence Arnold reports financial support was provided by Pelago Bioscience AB. Geoffrey A. Holdgate reports a relationship with Society for Laboratory Automation and Screening that includes: speaking and lecture fees and travel reimbursement. Geoffrey A. Holdgate is an Editorial Board member of SLAS Discovery
Acknowledgments
We thank Mark Wigglesworth and Thomas Lundbäck (AstraZeneca) and Michael Dabrowski (Pelago Biosciences) for input into study design. We thank Volker Eckelt and Vincent Dupriez (PerkinElmer) and Antony Sheehan (TGR BioSciences) for their support in enabling Alpha and ALSU detection.
A high-throughput dose-response cellular thermal shift assay for rapid screening of drug target engagement in living cells, exemplified using SMYD3 and IDO1.
Iglewicz B, Hoaglin D. 1993. How to detect and handle outliers. Volume 16. In The ASQC Basic References in Quality Control: Statistical Techniques, E.F. Mykytka, Ph.D., Editor, ASQC Quality Press.
00012-6&id=gr1.jpg){kind=link}
00012-6&id=gr2.jpg){kind=link}
00012-6&id=gr3.jpg){kind=link}
00012-6&id=gr4.jpg){kind=link}
00012-6&id=gr5.jpg){kind=link}
00012-6&id=gr6.jpg){kind=link}
00012-6&id=gr7.jpg){kind=link}
00012-6&id=gr8.jpg){kind=link}
00012-6&id=gr9.jpg){kind=link}
00012-6&id=gr10.jpg){kind=link}
00012-6&id=gr11.jpg){kind=link}