ThermoFRET: A novel nanoscale G protein coupled receptor thermostability assay functional in crude solubilised membrane preparations

Sensitive protein stability assays for membrane proteins are crucial for developing purification protocols, for structural and biophysical characterisation and drug discovery. Here, we describe a novel high-throughput 384-well FRET-based thermostability methodology, ThermoFRET, allowing for the ultrasensitive determination of G protein coupled receptor (GPCR) stability. This method measures FRET between a terbium-cryptate labelled GPCR and BODIPY-FL-Cystine, a thiolreactive dye that reacts with cysteine residues exposed upon protein unfolding in response to thermal denaturation. ThermoFRET is functional in crude solubilised membrane preparations, without protein purification and can detect receptor stabilising ligands, making it ideally suited for orphan receptor screening.


Introduction
A detailed knowledge of GPCR structure is essential for effective drug discovery (Congreve, de Graaf et al. 2020) and contributes to successful in silico ligand screening and ligand design (Lyu, Wang et al. 2019). The process of obtaining a structure starts with purification of the target protein (Tate and Schertler 2009, Tate 2012, Milic and Veprintsev 2015. Membrane proteins are inherently unstable outside the native lipid environment and often require stabilisation under specific solubilisation conditions. Optimising the receptor purification conditions, in terms of choice of detergent, selection of a stabilising ligand or the introduction of stabilising point mutations or fusion of soluble proteins (eg. T4 lysozyme) all help to improve receptor stability, reduce conformational flexibility and facilitate crystallisation and structure determination (Rosenbaum, Cherezov et al. 2007, Serrano-Vega, Magnani et al. 2008, Warne, Serrano-Vega et al. 2008, Scott, Kummer et al. 2014, Milic and Veprintsev 2015. Such optimisation processes are routinely assessed by thermal denaturation of the protein of interest. The optimisation of thermal stability of proteins is a widely used technique in structural biology to maximise the chances of successful crystallisation for all classes of proteins studied (Ericsson, Hallberg et al. 2006, Vedadi, Niesen et al. 2006).
Thermal-shift assays are also used to discover novel ligands for a target in the absence of a known tracer making them particularly attractive for orphan ligandreceptor profiling and as secondary biophysical screens in drug discovery projects.
Thermostability assays involve thermal denaturation of detergent solubilized receptors at increasing temperatures, the idea being to determine how much correctly folded protein remains. While a large variety of protein stability assays are available, specific challenges in their application are posed by membrane proteins and especially GPCRs due to their low expression levels and low thermostability in detergents commonly used for their purification. Often, their thermostability has to be optimised before any purification. Traditionally GPCR protein stability assays have relied on the availability of a high-affinity radioligand to act as a tracer for receptor functionality (Galvez, Parmentier et al. 1999, Serrano-Vega, Magnani et al. 2008, Robertson, Jazayeri et al. 2011, Magnani, Serrano-Vega et al. 2016. In the absence of the radioactive tracer, temperature-induced aggregation based techniques such as temperature shift fluorescence size exclusion chromatography TS-FSEC (Hattori, Hibbs et al. 2012) can be used, although throughput is limited. Alternative fluorescence-based techniques with higher throughput exist, such as the N-[4-(7diethylamino-4-methyl-3-coumarinyl)phenyl]maleimide (CPM) assay, which utilises a thiol-reactive micro-fluorescent fluorochrome. This dye reacts with exposed cysteines, acting as a sensor of protein stability in the temperature-dependent unfolding process (Alexandrov, Mileni et al. 2008). Other thiol-reactive dyes such as BODIPY-FL-Cystine (BLC) or 4-(aminosulfonyl)-7-fluoro-2,1,3-benzoxadiazole (ABD) are also available for stability measurements (Isom, Marguet et al. 2011, Bergsdorf, Fiez-Vandal et al. 2016. However, both these techniques currently require purified protein in microgram quantities which is a considerable drawback.
The novel ThermoFRET method we describe here has distinct advantages compared with the current methods routinely used to detect protein stability.
Importantly, this method has no requirement for purified protein and is easily miniaturised improving overall assay throughput. It is very sensitive and applicable to characterise membrane proteins with low levels of expression. Target specificity is provided by a SNAP-tagged receptor covalently labelled with the substrate SNAP-Lumi4-Tb (Lumi4-Tb), the fluorescent donor for our ThermoFRET-based stability assay. Here, BLC acts as a FRET acceptor with exposed reactive sulfhydryl groups on cysteine residues acting as a sensor of protein unfolding on heating. Consequent measurement of donor and acceptor emissions following energy transfer between the excited donor (Lumi4-Tb) and acceptor (BLC) provides us with a temperaturesensitive time-resolved fluorescence (TR-FRET) signal corresponding to protein stability.
This system is applicable to other membrane proteins with buried cysteine residues for high-throughput determination of protein stability using crude solubilised membrane preparations. Additionally, we can multiplex ThermoFRET with an assay based on the binding of a fluorescent ligand. This allows the simultaneous detection of the loss of the receptor ligand binding activity in addition to unfolding. Due to the nature of the homogeneous assay format, negating the need to separate bound and unbound ligand, the usual requirement for a high affinity radioligand is avoided.
Since we employ TR-FRET detection, these assays are safer than radiometric alternatives and can be readily performed in a more convenient 384-well assay format.

Thermostability of the β 2 adrenoceptor and A 2A adenosine receptor
To validate this novel assay system, we utilised two structurally and pharmacologically well-characterised receptors, the β 2 adrenoceptor (β 2 AR) (Cherezov, Rosenbaum et al. 2007, Rasmussen, Choi et al. 2007) and the adenosine A 2A receptor (A 2A R) (Rasmussen, Choi et al. 2007, Lebon, Bennett et al. 2011). The stability of the detergent stabilised β 2 AR and A 2A R preparations were characterised by two independent methods which make use of the terbium cryptate labelled receptor, to create FRET-based thermostability assays. The ThermoFRET assay uses the chemical reactivity of the native cysteines embedded in the protein interior as a sensor for the integrity of the native folded state with these cysteines being exposed upon protein unfolding. The exposed cysteines are modified by the thiol-reactive fluorochrome BODIPY-FL-Cystine ligand (BLC), that acts as the acceptor for the terbium cryptate donor (see Fig. 1). Additionally, we used a time resolved FRET (TR-FRET) fluorescent ligand binding assay to measure the amount of the native receptor left in a sample following the incubation of receptor and ligand at elevated temperature. In all cases receptor thermostability measurements were assessed over a standard incubation period of 30 min on a dual block PCR thermocycler capable of producing a combined temperature gradient of 60 o C (eg, 10 to 70 o C). The samples were cooled to room temperature prior to measurement.

ThermoFRET assay optimisation
The melting temperature (T m ) of the n-Dodecyl β -D-maltoside (DDM) solubilised β 2 AR was measured in the presence of increasing concentrations of BLC. While the concentration of the protein of interest in the crude solubilised membrane fraction or crude lysate could be very low (sub-nanomolar), the total concentration of reactive cysteines in the sample could be much higher as it contains many different proteins.
The concentration of the BLC should therefore be adjusted accordingly. For example, 1 mg/ml concentration of total protein corresponds to 20 μM of a hypothetical protein of 50 kDa, giving a starting point for empirical determination of an optimal concentration of the BLC. The effect of BLC concentration on assay performance is shown in Supplementary Fig. 1A. A final assay concentration of 10 μM BLC substrate at total protein concentration of 0.015 mg/ml was deemed sufficient to produce a reliable thermal stability signal with a reduced background signal and a low signal to noise ratio and was routinely employed in all subsequent experiments. ThermoFRET assay sensitivity was assessed by titrating β 2 AR receptor concentration and assessing receptor stabilisation by monitoring the change in FRET ratio as a function of increasing temperature, see Supplementary Fig. 1B. This experiment confirms that the ThermoFRET assay has sensitivity in the subnanomolar region, in line with that of a traditional ligand binding assay. Finally, we assessed the effect of DMSO, a commonly used co-solvent for ligand screening, on assay performance ( Supplementary Fig. 1C). DMSO concentration up to 20% did not affect our ability to detect protein unfolding.

ThermoFRET thermostability measurement
The T m of the DDM solubilised β 2 AR measured using ThermoFRET was 36 ± 1 o C, consistent with the literature data (Serrano-Vega and Tate  ThermoFRET data for the A 2A R and β 2 AR in the absence and presence of stabilising ligand are summarised in Fig. 2E and F, respectively. The effect of a shortened incubation time is shown in Supplementary Fig. 2A, with a reduced incubation period (5 min) producing a higher apparent melting temperature than seen at the standard incubation time of 30 min. Perhaps not surprisingly given that one of the primary functions of the cell membrane is to provide support, the melting temperature of the β 2 AR in the native membrane measured using ThermoFRET was much higher than observed in any of the detergents tested ( Supplementary Fig. 2B).

ThermoFRET versus TR-FRET-based fluorescent ligand thermostability measurements
The stability of the detergent solubilized β 2 AR and A 2A R was also characterized in TR-FRET-based fluorescent ligand binding assays. These measurements are equivalent to radioligand binding assays which measure T m as a function of radioligand bound, but importantly can also be performed in a non-purified assay setup due to the signal specificity afforded by TR-FRET. One important distinction between tracer formulated thermostability assays and the ThermoFRET assay stems from the ability of the tracer to further stabilise the ligand-free (or apo) form of the receptor.
A comparison of β 2 AR stability measurements using ThermoFRET and fluorescent ligand binding with fluorescent propranolol (F-propranolol) revealed significant differences between the two methods when performing a detergent/buffer screen P>0.05) were equivalent to the value measured by ThermoFRET without fluorescent ligand supplementation, suggesting that F-Sch-442,416 is sufficient to observe unfolding of the receptor, but that the ligand itself is unable to significantly stabilise the receptor. These data are summarised in see Fig. 3D.

TR-FRET-based saturation binding assays
A more detailed study of receptor functionality was undertaken by employing the technique of saturation binding. Functionality was maintained following the solubilisation process as shown by binding parameters derived from TR-FRET-  Fig. 3B) highlighting a potential caveat of using only a single tracer to assess receptor functionality. If a particular detergent changes a tracer's apparent affinity one might conclude that the protein is not folded correctly.
However, the findings with a different tracer, F-carazolol ( Supplementary Fig. 3A) shows that the receptor is correctly folded, and possibly exists in a conformation more favourable to inverse agonist binding. Interestingly, we can rescue the binding of F-propranolol by adding CHS (Supplementary Fig. 3D) suggesting that CHS is acting either as a reservoir for the enhanced association of the F-propranolol, possibly by increasing its local concentration proximal to the receptor, or that the CHS is creating a different receptor conformation, an idea consistent with bicelle formation (Thompson, Liu et al. 2011).    μ M CA-XAC and 100 nM F-Sch-442,416 (F-Sch) were read separately using either the 337/620/520 HTRF module for BLC or the 337/665/620 HTRF module for F-Propranolol. Samples that were co-incubated were measured on the PHERAstar FSX using dual-read protocol with simultaneous 337/620/520 and 337/665/620 HTRF readouts of the same sample.

Discussion
Our data on thermostability of both the β 2 AR and A 2A R obtained using the ThermoFRET technique are consistent with the literature values that were obtained using radioligand binding experiments (Serrano-Vega and Tate 2009). The increased thermal stability of the β 2 AR in the presence of a series β 2 AR antagonists and of the A 2A receptor in the presence of XAC is a good indicator that the ThermoFRET assay is sensitive to detect the unfolding of the apo (ligand free) receptor and its stabilisation in response to ligand binding. Equally, it can be used to screen for ligands that bind and stabilise the receptor, possibly via locking it in a particular conformation (Congreve, Oswald et al. 2017). Perhaps the biggest advantage of the ThermoFRET over the existing radioligand assays is its compatibility with the presence of other ligands and its ability to report stabilisation of the receptor by the "cold" ligand.
Additionally, we were able to demonstrate clear detergent-dependent differences in β 2 AR thermostability demonstrating that ThermoFRET has great potential as a high throughput detergent screening tool. The stability of the A 2A R was assessed using the mild detergent LMNG (Strege, Carpenter et al. 2017). Comparing our ThermoFRET A 2A R thermostability assay data with previously documented stability measurements, we see T m values (25.4 ± 1.0 o C) corresponding to those obtained with purified A 2A R, solubilised from HEK293 cells by lauryldimethylamine-oxide (LDAO) (30 o C) (Robertson, Jazayeri et al. 2011). Our results also confirm that the A 2A R is more stable in LMNGthan in other detergents, such as decyl-β-maltoside (22 o C) (Robertson, Jazayeri et al. 2011). We observed an increase in receptor stability in the presence of the A 2A R antagonist XAC and its fluorescent derivative, indicating that the solubilised receptor is functional and capable of ligand binding.
The value of this assay is in reporting relative increases in protein stability in the presence of ligands or following an optimisation of detergent or buffer conditions. As the unfolding of GPCRs and covalent modification of the cysteine residues is an irreversible process, it does not necessarily report a true thermodynamic stability of the protein. Some additional possible limitations are associated with using cysteine as a sensor for protein stability. It has been previously documented that 66% of membrane proteins on known structure contain one or more thiol groups, with 91% of cysteine residues found buried within the internal structure of the protein (Alexandrov, Mileni et al. 2008). It is therefore possible that our thiol-reactive dye is binding non-specifically to both exposed thiol-reactive groups or other hydrophobic residues prior to melting of the protein (Wang, Ye et al. 2015), a situation that could complicate the measurement of thermal unfolding due to a raised background signal.
However, the reactivity of the exposed cysteines will not be affected by receptor unfolding. Our finding that T m values obtained in the ThermoFRET assay agree closely with those measured as a function of fluorescent ligand binding (see Fig. 3) provides good evidence that what we observe with the BLC reagent is in fact a direct measurement of protein unfolding.
In summary, we have developed a novel nanoscale FRET-based thermostability assay that allows for the ultrasensitive determination of membrane protein stability.
Importantly, the ThermoFRET assay is functional in crude preparations, with no need for a protein purification step, and requiring minimal amounts of active protein, making it ideal for hight throughput screening. ThermoFRET can be readily applied to other GPCRs and membrane proteins which contain buried cysteine residues, without the need for a fluorescent or radioactive ligand. In addition, functionality of the solubilised receptor can be readily tested using receptor specific fluorescent ligands by monitoring FRET between the Lumi4-Tb-labelled receptor and bound fluorescent ligand. In the absence of fluorescent ligands for the target receptor, stabilisation by non-fluorescent ligands can used as an indication of receptor activity.
In conclusion, the ThermoFRET assay as described provides a rapid screening platform for the optimisation of membrane protein solubilisation conditions and has the potential to discover novel molecular entities targeting orphan receptors for which there are no known ligands. We are grateful to Thomas Roux and Eric Trinquet from Cisbio for providing the fluorescent ligand Sch-442,416-red.

Methods
The Tag Following the labelling step, SNAP-labelling medium was carefully removed, and cells were washed twice with PBS (GIBCO, Carlsbad, USA). Terbium labelled cells were detached using 5 ml of cell-dissociation buffer (GIBCO, Carlsbad, USA) and collected using 5 ml of DMEM containing 10% FCS. Cells were pelleted by centrifugation (400 g for 5 minutes) and pellets were stored at -20 o C prior to membrane preparation.

Membrane Preparation
All steps are performed on ice or at 4 o C using ice-cold buffers. Cell pellets were

Membrane Solubilisation
All steps are done on ice or at 4 o C using ice-cold buffers. Membranes expressing β 2 AR and A 2A R were diluted in buffer C to total protein concentrations of 0.15 mg/ml and 0.5 mg/ml respectively, pelleted by centrifugation (16,873 g, 4 o C, 30 min) and resuspended in solubilisation buffer (20 mM HEPES, 150 mM NaCl, 10% glycerol, 0.5% BSA, pH 7.5) containing either 1% DDM, 1% DDM & 0.2% CHS or 0.5% LMNG for β 2 AR and 0.5% LMNG for A 2A R. Membranes were solubilised for 60 min under rotation at 4 o C, followed by ultra-centrifugation using the Beckman Coulter Optima MAX-CP ultracentrifuge and TLA 100.3 rotor (Fullerton, USA) at 55,000 RPM or 100,000 g at 4 o C for 30 min to remove non-solubilised material. Solubilised samples were used immediately, the supernatant containing the solubilsed receptor being diluted ten-fold in solubilisation buffer without detergent to a final detergent concentration of 0.1% DDM for β 2 AR and 0.05% LMNG for A 2A R.

ThermoFRET Thermostability Assay
Our TR-FRET-based detection method utilises a SNAP-tagged receptor system previously described (Cole 2013). Briefly, covalent labelling of a SNAP-tagged receptor with the Lumi4-Tb s ubstrate creates a FRET donor. The energy transfer between the donor (Lumi4-Tb) and acceptor (BLC) provides the basis for the T m determination (Zwier, Roux et al. 2010, Cole 2013, through the measurement of donor and acceptor emission and calculation of the HTRF ratio (acceptor emission/donor emission*10,000), a homogeneous time-resolved fluorescence (HTRF) signal corresponding to protein instability and unfolding. BLC is prepared in DMSO to a stock concentration of 10mM and stored at -20 o C, protected from light. Samples were incubated at increasing temperatures as described above for the ThermoFRET assay. Following incubation, samples were transferred in duplicates to a 384-well Proxiplate before performing a dual-read on the PHERAstar FSX at RT using the HTRF 337/520/620 and 337/665/620 modules or for samples containing fluorescent ligand or BLC alone by taking separate measurements using individual modules.

Fluorescent Ligand Saturation Binding Assays
Saturation binding assays were performed using increasing concentrations of Fcarazolol and F-propranolol with non-specific binding (NSB) determined by 1 µM cyanopindolol (CYP). For CA-XAC saturation binding assays and F-Sch-442,416, NSB was determined by 10 µM Sch-442,416 and 10 µM XAC respectively. Assays were performed in the same solubilisation buffers as described above but in a 384well Optiplate, a total assay volume of 40 µl was used for both detergent-solubilised β 2 AR (0.01 μ g/ml) and A 2A R (0.05 μ g/ml, final protein concentration) (Miljus, Sykes et al. 2020). Measurement of equilibrium saturation was performed at RT on the PHERAstar using the HTRF 337/665/620 module following a 120-minute incubation at RT for β 2 AR and at 4 o C for A 2A R.

Data Analysis
One-way ANOVA tests with Sidak's multiple comparison test and Student's T-tests were performed to determine statistical difference between individual data sets.
ThermoFRET and fluorescent ligand thermal stability assays T m values were obtained by fitting the data into the Boltzmann sigmoidal equation using GraphPad Prism 7.0 (San Diego, USA) : corresponding non-fluorescent ligand to obtain NSB and total binding, respectively.
Data are expressed as a mean ± SEM of 3 individual experiments performed in singlets. N.D. = Not determined.