ThermoBRET: a ligand-engagement nanoscale thermostability assay applied to GPCRs

Measurements of membrane protein thermostability allows indirect detection of ligand binding. Current thermostability assays require protein purification or rely on pre-existing radiolabelled or fluorescent ligands, limiting their application to established target proteins. Alternative methods detect protein aggregation which requires sufficiently high level of protein expression. Here, we present a ThermoBRET method to quantify the relative thermostability of G protein coupled receptors (GPCRs), using cannabinoid receptors (CB1 and CB2) and the β2-adrenoceptor (β2AR) as model systems. ThermoBRET reports receptor unfolding, does not need labelled ligands and can be used with non-purified proteins. It uses Bioluminescence Resonance Energy Transfer (BRET) between Nanoluciferase (Nluc) and a thiol-reactive fluorescent dye that binds cysteines exposed by unfolding. We demonstrate that the melting point (Tm) of Nluc-fused GPCRs can be determined in non-purified detergent solubilised membrane preparations or solubilised whole cells, revealing differences in thermostability for different solubilising conditions and in the presence of stabilising ligands. We extended the range of the assay by developing the thermostable tsNLuc by incorporating mutations from the fragments of split-Nluc (Tm of 87 ⁰C vs 59 ⁰C). ThermoBRET allows determination of GPCR thermostability, which is useful for protein purification optimisation and as part of drug discovery screening strategies.

Introduction 1 G protein coupled receptors (GPCRs) are a large family of membrane proteins that 2 are important drug discovery targets (Hauser, Attwood et al. 2017). Structural and 3 biophysical studies of GPCRs have significant importance in modern drug discovery 4 (Congreve, de Graaf et al. 2020) but one major hurdle is their successful solubilisation 5 from their native membrane environment and subsequent purification. Optimisation of 6 receptor stability during this process is a key component to success (Tate 2010). 7 Additionally, the ability of a bound ligand to stabilise the receptor structure is a property 8 which can be exploited in screening efforts to find novel drug candidates (Fang 2012, 9 Zhang, Stevens et al. 2015. 10 Existing GPCR protein stability assays rely on the availability of a high-affinity 11 radioligand to act as a tracer for receptor functionality (Galvez, Parmentier et al. 1999, 12 Serrano-Vega, Magnani et al. 2008, Robertson, Jazayeri et al. 2011, Magnani, 13 Serrano-Vega et al. 2016). In the absence of the radioactive tracer, temperature-14 induced aggregation-based techniques such as technology developed by Heptares 15 Therapeutics (now Sosei Heptares) (Marshall, Jazayeri et al. 2013) or temperature 16 shift fluorescence size exclusion chromatography TS-FSEC (Hattori, Hibbs et al. 2012, 17 Vuckovic 2017, Nji, Chatzikyriakidou et al. 2018) can be used. Alternative 18 fluorescence-based techniques with higher throughput exist, such as the N- [4-(7-19 diethylamino-4-methyl-3-coumarinyl)phenyl]maleimide (CPM) assay, which utilises a 20 thiol-reactive fluorescent fluorochrome. This dye reacts with exposed cysteines, acting 21 as a sensor of protein stability in the temperature-dependent unfolding process 22 (Alexandrov, Mileni et al. 2008). Other thiol-reactive dyes such as BODIPY-FL-Cystine 23 (BLC) or 4-(aminosulfonyl)-7-fluoro-2,1,3-benzoxadiazole (ABD) are also available for 24 stability measurements (Isom, Marguet et al. 2011, Bergsdorf, Fiez-Vandal et al. 25 2016). However, both these techniques currently require purified protein in microgram 26 quantities which is a considerable drawback. 27 Low abundance of GPCRs even in over-expressing systems and their inherently low 28 stability in detergents (Milic and Veprintsev 2015) calls for sensitive protein stability 29 assays that can be used without protein purification or pre-existing tracer compounds. 30 Here, we present the ThermoBRET assay based on bioluminescence resonance 31 energy transfer between the bright Nanoluciferase (Nluc,and,correspondingly,32 NanoBRET) (Hall, Unch et al. 2012), acting as a donor of light and a thiol reactive 33 Sulfo-Cyanine3 maleimide (SCM) dye, the acceptor, allowing us to quantify the relative 34 thermostability of non-purified GPCRs solubilised into detergent micelles. As a test 35 case we focus on two GPCRs, the cannabinoid receptor 2 (CB2) as a therapeutically 36 promising (Pertwee 2012) but unstable drug target (Vukoti, Kimura et al. 2012, 37 Beckner, Gawrisch et al. 2019, Beckner, Zoubak et al. 2020) and the previously well 38 characterised b2 adrenergic (b2AR) receptor. This assay detects picomolar 39 concentration, corresponding to nanogram amounts, of target protein. Due to the 40 nature of the homogeneous assay format, negating the need to separate bound and 41 unbound ligand, can be used to detect binding of low-affinity ligands. Since we employ 1 NanoBRET detection, these assays are safer than radiometric alternatives and can be 2 readily performed in 96-and 384-well assay format. 3 4

5
ThermoBRET provides reliable measurements of GPCR stability 6 We fused a Nluc (Hall, Unch et al. 2012) to the receptor N-terminus, preceded by a 7 cleaved signal peptide to ensure its' successful expression and plasma membrane 8 trafficking ( Supplementary Information 1). Detergent solubilised receptor samples 9 containing a thiol reactive Sulfo-Cyanine3 maleimide (SCM) acceptor are incubated at 10 varying temperatures using a gradient forming PCR thermocycler. As the receptor 11 unfolds on heating the SCM covalently binds to exposed cysteine residues ( Figure 1). 12 We chose SCM because of its suitability as a BRET acceptor for Nluc, water solubility, 13 and relatively low cost compared to other thiol-reactive fluorophores. In principle, any 14 maleimide or other thiol-reactive conjugated fluorescent dye with overlapping donor-15 acceptor emission-absorption spectra can be used. The unfolded state of the receptor 16 due to thermal denaturation is measured as NanoBRET between the Nluc tag and the 17 SCM acceptor and is quantified as a ratio of the donor and acceptor light emissions, 18 termed the NanoBRET ratio. The relative thermostability of a receptor in different 19 solubilised non-purified membrane preparations can be easily determined, first by 20 thermal denaturation across a temperature gradient on a thermocycler block, rapid 21 cooling to 4 °C, and then following the addition of the Nluc substrate furimazine and 22 measurement of the NanoBRET ratio in a 384-well luminescence plate reader at room 23 temperature ( Figure 1). The midpoint of the transition curve is found by fitting the data 24 to a Boltzmann sigmoidal equation to obtain a Tm. 25 Detergents affect stability of receptors 26 When solubilised in DDM detergent, CB2 had a Tm of around 33 °C ( Figure 2A)  Differences in the detergent stability of the adrenergic β2-receptor were also found 33 (Supplementary Figure 1). Hence, this assay can be readily used to screen for the 34 best detergent solubilising conditions before attempting a large-scale purification. 35 Ligands stabilise CB 2

36
We also tested a selection of endogenous and synthetic cannabinoid receptor ligands 37 for their ability to increase the thermostability of CB2 ( Figure 2B). The lipophilicity of its 38 ligands has made the CB2 receptor a particularly challenging target for ligand binding 1 experiments due to their high non-specific binding. These ligands were all tested at a 2 concentration of 20 µM, well above their dissociation constant (KD) at room 3 temperature, in order to ensure full occupancy of the solubilised receptors. 4 Interestingly, the endogenous cannabinoid 2-arachidonoylglycerol (2AG) increased 5 the Tm of CB2 by around 6 °C, whereas the other endogenous cannabinoid 6 anandamide (AEA) only increased the Tm by around 2 °C. The most probable reason 7 for these observations are the variable temperature dependence of the affinity of the 8 ligand to the receptor as well as well as the degree of the entropy contribution to the 9 binding (Layton and Hellinga 2010). Other synthetic cannabinoid ligands HU308 and 10 SR144528 also produced appreciable increases in thermostability, and the pattern of 11 ligand stabilisation appeared different for the related CB1 receptor (Supplementary 12 Figure 2). 13 tsNluc extends the range of the ThermoBRET assay 14 One problematic aspect of the ThermoBRET is the thermostability of the Nluc donor 15 itself, which has been reported to unfold at around 55 -60 °C (Hall, Unch et al. 2012). 16 This limits the thermal range for this assay and prevents accurate Tm determination in 17 conditions where the receptor itself is particularly thermostable, for example when CB2 18 is bound to the high affinity non-selective cannabinoid agonist HU210 ( Figure 2C). We 19 therefore combined Nluc mutations which had been developed by Promega as part of 20 their efforts to create a stable split-luciferase system (Dixon, Schwinn et al. 2016) and 21 found that these mutations improved thermostability of the full length luciferase by 22 about 30 °C ( Figure 2D). In line with previous reports (Hall, Unch et al. 2012) we found 23 that purified Nluc had a Tm of 59 °C, and that purified thermostable Nluc (tsNluc) had 24 a Tm of 87 °C ( Figure 2D), making it preferable for thermostability measurements 25 across a wide temperature range. Importantly, tsNluc contains no cysteine residues 26 (Supplementary Information 2) and thus is unaffected by maleimide/thiol chemistry. 27 Further characterisation showed tsNluc to have a similar luminescence emission 28 profile as Nluc with furimazine as a substrate, although with a lower luminescence 29 output (Supplementary Figure 3). Applying this novel tsNluc fusion improved the 30 working temperature range of the ThermoBRET assay and allowed successful Tm 31 determination for CB2 in the presence of HU210 ( Figure 2E). Strikingly, HU210 was 32 able to stabilise CB2 by around 12 °C, the highest level achieved of any of the CB2 33 ligands tested. 34 To further assess the ability of this improved assay format to determine the stability of 35 GPCRs in detergent we created a tsNluc-b2AR expression construct. Previous work 36 employing the CPM assay indicates that this receptor can be stabilised by high affinity 37 antagonists (Wacker, Fenalti et al. 2010). Due to the higher throughput achievable 38 with our BRET-based system we were able to assess the receptor stabilising effects 39 of both b-adrenergic agonists and antagonists ( Figure 3 A and B). Both high affinity 40 antagonist and agonist were able to stabilise the receptor to a degree which was 41 dependent on the affinity of the ligands for the receptor ( Figure 3C). In contrast koff and 42 kon were by themselves more poorly correlated with receptor stabilisation ( Figure 3D  1 and E). 2

3
Having established a robust assay to measure receptor stability, we examined the 4 more subtle effects of lipid-detergent ratio in the solubilised tsNluc-CB2. Firstly, we 5 examined if the stability of CB2 is affected by the receptor concentration, while keeping 6 the membrane fraction/detergent ratio the same by supplementing the fraction of CB2 7 membranes with "empty" non-transfected HEK293 membranes to the same total 8 amount (Supplementary Figure 4A). The reported value was the same, suggesting 9 that CB2 stability is not affected by its concentration, at least within the tested range. 10 Secondly, we examined if an increase in membrane/detergent ratio could affect CB2 11 stability, by supplementing a fixed amount of CB2 membranes with an increasing 12 amount of "empty" membranes (Supplementary Figure 4B). Surprisingly, the receptor 13 stability was decreased by 2-3 °C at higher membrane/detergent ratio, although it 14 plateaued at concentrations above 20 ng/μL, as generally increasing concentration of 15 lipids may lead to stabilisation of receptors (Cecchetti, Strauss et al. 2021). A possible 16 explanation for this is that the increased total protein concentration (HEK293 17 membranes contain a significant amount of protein) may accelerate aggregation 18 process. This observation suggests that for each new target this relationship needs to 19 be explored, and concentration of membrane and detergent should be kept constant 20 and an appropriate point on the plateau region should be chosen. 21

22
Receptor denaturation is a complex process, progressing through a loss of tertiary 23 structure through potential intermediates that may have the overall organisation of the 24 correctly folded receptor (and protecting cysteines from modification) to complete loss 25 of tertiary structure and aggregation. It is important to understand what process is 26 sensed by the ThermoBRET assay. 27 Stabilisation of a receptor can also be monitored through addition of a fluorescent 28 ligand to a detergent solubilized receptor, measured using NanoBRET detection. In 29 this case unfolding of a target protein in response to an increase in temperature will 30 result in a loss in specific binding. Figure 4A shows the saturation binding curve for 31 propranolol-green binding to the tsNluc-b2AR receptor solubilised in DDM. The Kd of 32 propranolol-green for the b2AR was determined to be 5.7 ± 1.1 nM. Figure 4B shows 33 the loss in propranolol-green specific binding signal as the protein unfolds in response 34 to an increase in temperature, following preincubation of a saturating concentration of 35 fluorescent ligand (1 µM) with the receptor. were derived for this smaller test set and correlated with the thermal shift values 2 obtained at fixed concentrations of individual agonist and antagonists. A linear 3 relationship was observed between these two measures ( Figure 5B). Finally, the 4 ThermoBRET IC50 values were correlated with radioligand binding affinity values 5 obtained for the different ligands ( Figure 5C). Again, an excellent correlation was 6 observed between the two data sets apart from for the very slowly dissociating 7 antagonist cyanopindolol. There are two plausible explanations for this: either the very 8 slow off-rate of this ligand means that for this ligand equilibrium is not achieved prior 9 to the determination of ThermoBRET IC50 values, samples being kept on ice prior to 10 melting, or we are observing the phenomenon of ligand depletion due to nM 11 concentrations of receptor present in the reaction mixture. 12 In addition, we assessed the stability of the b2AR solubilised directly from whole cells 26 and the ability of orthosteric ligands to stabilise the receptor. Figure 6 shows the 27 measured stability of the apo b2AR solubilised from whole cells in the absence and 28 presence of a selection of antagonist and agonist ligands and the resulting correlation 29 of Tm values obtained with the same ligands from membrane solubilised receptors. 30 The measurement of receptor thermostability directly from whole cells further reduces 31 the number of steps required for the assay (ie. no need to prepare membranes) and 32 allows target engagement to be assayed when receptors are still in a native membrane 33 environment if it is added prior to solubilisation. 34

35
Here, we aimed to establish sensitive technique that allows to measure receptor 36 stability in crude detergent-solubilised preparations, without a need for a unique tracer 37 compound. ThermoBRET proved to be sensitive and selectively reporting protein 38 stability of solubilised membrane preparations and even directly solubilised cells. It 39 utilises universal cysteine-reactive fluorescent dyes for readout and can be used to 1 detect binding of specific non-fluorescent ligands. 2 The processes of protein unfolding and protein aggregation are related, yet very 3 separate, phenomena. As mentioned in the introduction, a decrease in GFP 4 fluorescence ( to the protein of interest and monitor the decrease in luminescence activity to measure 10 aggregation of the protein of interest after thermal denaturation. In contrast, the 11 ThermoBRET assay described here captures the initial conformational unfolding 12 events which expose maleimide reactive cysteine residues in the protein of interest. 13 In addition, ThermoBRET measurement is buffered from changes in the concentration 14 of the luciferase fused target protein because a ratiometric method is used to calculate 15 resonance energy transfer, contrasting with assays which measure luminescence 16 intensity only. This means the measurement remains robust even at low 17 concentrations of the target protein where the magnitude of the measured signal is at 18 the lower end of detection capabilities. 19 We note that under the test conditions used, the luminescence activity of just 85 pM 20 purified Nluc and tsNluc was measurable in a 96-well plate ( Figure 2D), making the 21 assay extremely sensitive. This makes the assay particularly amenable situations 22 where there are limitations on the amounts of reagent that can be provided, for 23 example protein targets which are poorly expressed in vitro and/or in vivo. This assay 24 principle could even be applied in more physiologically relevant in vivo cellular models 25 whereby tsNluc (or the 11 amino acid HiBiT tag) is fused to an endogenously 26 expressed protein via CRISPR-mediated insertion (White, Johnstone et al. 2019). 27 The novel thermostable tsNluc we describe has clear advantages compared to Nluc. 28 Firstly, due to its improved thermostability it is less likely to unfold before the protein 29 of interest and cause sample aggregation and other possible artefacts. The Tm of 30 tsNluc was 87 ˚C which may put an upper temperature limit on the ThermoBRET 31 method, however such a high thermostability situation would be very unexpected for 32 an integral membrane protein solubilised in detergents. Secondly, whilst previous 33 reports (Hall, Unch et al. 2012) and our own data ( Figure 2D) showed the Tm of Nluc 34 to be 59 ⁰C, more recent use of Nluc to monitor protein aggregation show clear 35 luminescence activity after protein samples had been heated to temperatures >60 ⁰C 36 before cooling (Dart, Machleidt et al. 2018). We speculate that Nluc has propensity to 37 spontaneously refold after thermal denaturation, and that the presence of detergents 38 in the buffers of the latter report either aided Nluc refolding or delayed irreversible 39 protein aggregation. In our ThermoBRET assays, the cysteine exposed upon Nluc 40 unfolding would potentially react with the SCM and prevent its refolding, whereas 41 tsNluc avoids these pitfalls. The cysteine-less sequence of tsNluc also allows easy in 42 vitro chemical tagging of tsNluc-fusions. There is of interest in producing conjugates 1 of Nluc fused to other biomolecules, though usually this involves incorporation of 2 sequence-specific ligation motifs onto Nluc followed by enzyme-mediated ligation to 3 the molecule of interest (Wang, Shao et al. 2017, Mie, Niimi et al. 2019, Wouters, Vugs 4 et al. 2020). By introducing a cysteine at the C-terminus of tsNluc, any molecule could 5 be conjugated to tsNluc by thiol-reactive chemistry and mild reaction conditions, 6 enhancing its potential for protein engineering and enabling a wide scope for future 7 applications. 8 In comparison to our previous ThermoFRET application using a terbium cryptate 9 labelled receptor as a FRET donor (Tippett, Hoare et al. 2020), the ThermoBRET 10 approach offers potential advantages. ThermoFRET requires cell surface labelling of 11 the receptor-fused SNAP tag with the terbium cryptate donor molecule, adding to 12 assay cost, but perhaps more importantly creating an extra labelling step that can be 13 problematic if the tag is not readily exposed at the plasma membrane. In contrast, the 14 use of a genetically encoded bioluminescent donor (ie. tsNluc or Nluc) omits this 15 labelling step. This means that fused proteins which are poorly trafficked to the plasma 16 membrane, are now amenable as they do not require labelling at the cell surface. 17 Additionally, ThermoFRET requires more sophisticated detection by plate readers with 18 time-resolved fluorescence detection capabilities, whereas BRET only requires a 19 luminometer with filtered light detection that are more readily available in many labs. 20 A comparison of biophysical techniques used in drug screening cascades is shown in 21 Figure 7 along with the relative protein requirements and assay throughput potential 22 of each technique. 23 The ability of the ThermoBRET assay to quantify ligand-induced changes in the 24 receptor Tm makes it an ideal tool to study ligand binding to GPCRs. Tm values 25 obtained in the current study for the b2AR specific ligands compare well with those 26 obtained previously (Zhang et al., 2015). In principle, this assay can detect compounds 27 which bind the target at any site, assuming this interaction influences the 28 thermodynamic conformational landscape of the protein. It can be used to screen 29 potential ligands for orphan GPCRs as it does not depend on the availability of tool 30 compounds or known binders to develop a competition assay. Moreover, it could be 31 used to detect the combined stabilisation of several ligands to discover positive and 32 negative allosteric modulators of GPCRs. 33 Despite a number of advantages ThermoBRET offers, and positive results obtained 34 for b2AR, CB2 and CB1 receptors, this technique is not immune to the general 35 limitations of thermal shift assays. The protein needs to be in a native state once 36 solubilised, and some condition optimisation may need to be done for the less stable 37 receptors. Correspondingly, it may not be successful for every target receptor tried as 38 there is a requirement that the receptor contains buried cysteines which become 39 exposed upon thermal denaturation. This limitation is also inherent for the CPM assay, 40 and in situations where no free thiol exists then cysteines could be rationally 41 introduced into the receptor sequence to generate a ThermoBRET signal.Finally, more 42 practical experience with screening larger compound libraries will be needed to 1 establish real life performance and limitations of this technique. 2 Overall, ThermoBRET is an excellent and highly sensitive tool for optimisation of 3 solubilisation conditions and biophysical screening of GPCR compound libraries to 4 support structural biology, aiding the drug discovery efforts. 5

6
Drug compounds and reagents 7 Sulfo-Cy3 maleimide (SCM) (Lumiprobe GmbH, Germany) was obtained in powder 8 form, dissolved in DMSO at a concentration of 10 mM and stored in the dark at -20C. 9 Furimazine, the substrate for Nluc, was obtained from the Nano-Glo Luciferase Assay 10 System kit (Promega, UK) provided at a concentration of 5 mM. with 1 μg/mL tetracycline for 48h to induce receptor expression. Following this, cells 39 were lifted by trituration and centrifuged at 500 g for 10 minutes. Cell pellets were then 1 frozen at -80 ⁰C until membranes were prepared. 2 Membrane preparations 3 HEK293TR cell pellets were resuspended in 20 mL of ice cold buffer (10 mM HEPES 4 pH 7.4, 10 mM EDTA) and homogenised using a Ultra Turrax (Ika Work GmbH, 5 Germany). The homogenised cell suspension was then centrifuged at 4 ⁰C for 5 6 minutes at 500 g to remove whole cells and large debris, and the remaining 7 supernatant was then centrifuged twice at 4 ⁰C and 48,000 g for 30 minutes before the 8 membrane pellet was resuspended in buffer (10 mM HEPES pH 7.4, 0.1 mM EDTA). 9 Protein concentration of resuspended membranes was determined with using Pierce 10 BCA Protein assay kit (ThermoFisher Scientific) and was adjusted to 3 -10 mg/mL 11 before being aliquoted and stored at -80 ⁰C. 12 were performed essentially as described above but receptor solubilisation was 36 performed in the presence of protease inhibitor cocktail (cOmplete TM mini EDTA-free 37 Protease Inhibitor cocktail (Roche)). 38

Nluc and tsNluc expression and purification
1 NiCo21(DE3) chemically competent E. coli were transformed with pJ411 bacterial 2 expression plasmids and plated onto LB/agar plates containing 2% w/v glucose and 3 50 µg/mL kanamycin. After incubation at 37 ºC for 16-24 hours, a single colony was 4 picked to inoculate 20 mL of terrific broth containing 0.2% w/v glucose and 50 µg/mL 5 kanamycin. After 16-24 hours in a shaking incubator set at 37 ⁰C, 15 mL of overnight 6 culture was added to 3 L of terrific broth containing 0.2% w/v glucose and 50 µg/mL 7 kanamycin, grown in a shaking incubator at 37 ⁰C until OD600 of 0.7-1, when 500 μM 8 of isopropyl-β-D-thiogalactopyranoside (IPTG; VWR Chemicals) was added to induce 9 protein expression. Cells were then grown overnight (16-20 hours) at 25 ⁰C in a 10 shaking incubator before being harvested by centrifugation and frozen at -80 ⁰C. Cell 11 pellets were then thawed on ice, and resuspended in 100 mL lysis buffer (100 mM Tris 12 pH 7.5, 300 mM NaCl, 0.25 mg/mL chicken lysozyme, 1 µg/mL bovine DNAse I, 4 mM 13 MgCl2, and 3 cOmplete TM mini EDTA-free Protease Inhibitor cocktail tablets (Roche)). 14 After 1h on ice in lysis buffer, cells were then lysed further by French press. Cell lysates 15 were then clarified by centrifugation at 25,000 rcf for 30 minutes and then by passing 16 through a 0.45 μm syringe filter. The His-tagged proteins from the resulting lysate were 17 then purified using a 5mL HiTrap TALON Crude column on an ÄKTA start protein All data used to support the conclusions are included in the paper. Raw numerical data 7 can be obtained by contacting the corresponding authors. 8 Acknowledgements 9 AK was funded by a COMPARE team science summer studentship (2019) awarded 10 to BH. This research was supported by COMPARE funding to DBV. We would like to 11 thank Uwe Grether for supplying the CB1 antagonist rimonabant. 12 Detergent solubilised non-purified membrane preparations expressing GPCRs fused 3 at the N-terminus with Nluc (or tsNluc) are heated using a PCR thermocycler in the 4 presence of sulfo-Cy3 maleimide (SCM). As the protein unfolds due to thermal 5 denaturation, SCM reacts with newly exposed cysteine residues putting the sulfo-Cy3 6 acceptor fluorophore in proximity with the Nluc donor. At higher temperatures, protein 7 aggregation leads to a decrease in the NanoBRET signal and these points are 8 truncated before fitting to a Boltzmann sigmoidal equation to obtain a melting point 9

Declaration of interests
(Tm). Data is plotted as the mean ± standard error for 3 replicates.