The GFP thermal shift assay for screening ligand and lipid interactions to solute carrier transporters

Solute carrier (SLC) transporters represent the second-largest fraction of the membrane proteome after G-protein-coupled receptors, but have been underutilized as drug targets and the function of many members of this family is still unknown. They are technically challenging to work with as they are difficult to express and highly dynamic, making them unstable in detergent solution. Many SLCs lack known inhibitors that could be utilized for stabilization. Furthermore, as they bind their physiological substrates with high micromolar to low millimolar affinities, binding and transport assays have proven to be particularly challenging to implement. Previously, we reported a GFP-based method for the overexpression and purification of membrane proteins in Saccharomyces cerevisiae. Here, we extend this expression platform with the GFP thermal shift (GFP-TS) assay, which is a simplified version of fluorescence-detection size-exclusion chromatography that combines the sample versatility of fluorescence-detection size-exclusion chromatography with the high-throughput capability of dye-based thermal shift assays. We demonstrate how GFP-TS can be used for detecting specific ligand interactions of SLC transporter fusions and measuring their affinities in crude detergent-solubilized membranes. We further show how GFP-TS can be employed on purified SLC transporter fusions to screen for specific lipid–protein interactions, which is an important complement to native mass spectrometry approaches that cannot cope easily with crude lipid-mixture preparations. This protocol is simple to perform and can be followed by researchers with a basic background in protein chemistry. Starting with an SLC transporter construct that can be expressed and purified from S. cerevisiae in a well-folded state, this protocol extension can be completed in ~4–5 d. This protocol extension describes the GFP thermal shift assay for monitoring ligand interactions of solute carrier transporters using either crude detergent-solubilized membranes or purified samples.

with genome-wide RNA interference screens 4,5 and CRISPR knockout and knockin experiments can provide useful insights into potential substrates 6,7 , it still remains necessary in most cases to validate such interactions using in vitro-based binding and transporter assays. A requirement for in vitrobased assays is overexpression and purification of the SLC transporter protein. However, like many membrane proteins, SLC transporters are often difficult to produce and are typically unstable in detergent solution, making them difficult to isolate 13 . GFP-based tags have become a mainstream approach that facilitates homolog screening and construct modification to enable isolation of wellexpressed and detergent stable membrane proteins [8][9][10][11][12][13] . Previously, we outlined a GFP-based screening platform to optimize the expression, stability and purification of eukaryotic membrane proteins in Saccharomyces cerevisiae 14,15 . This platform has since been used to carry out functional and structural investigation for a number of SLC transporters and other types of membrane proteins [15][16][17][18][19] .
Here, we extend the S. cerevisiae GFP-based protein production platform with the GFP thermal shift (GFP-TS) assay for monitoring ligand interactions of SLC transporters using either crude detergent-solubilized membranes or purified samples 20 . The GFP-TS assay offers an attractive complementary approach to characterize SLC transporters by bridging the gap between the relative ambiguity of large-scale phenotypic screening and challenging transport assays in proteoliposomes using purified components.

Development of the GFP-TS assay for monitoring ligand-protein interactions
An attractive high-throughput strategy to functionally characterize SLC transporters is to screen for ligand binding rather than transport. Specific binders could turn out to be bona fide substrates or at least inhibitors, which are nevertheless useful for, e.g., functional characterization of transporters in cell-based assays. Moreover, knowledge that a ligand binds serves as a useful positive control to facilitate characterization by other methods, such as the solid-supported-membrane-based electrophysiology 16 and proteoliposome transport assays 17 . Arguably, the most successful highthroughput label-free binding technologies are thermal shift assays (TSAs), which are based on measuring a ligand-induced change in the thermostability of the target 18,19 . The most common TSA is differential scanning fluorimetry, otherwise commonly referred to as thermofluor assay, which monitors heat-induced protein unfolding by including an environmentally sensitive fluorescent dye that binds to the protein and increases fluorescence signal as the protein unfolds [18][19][20] . For membrane proteins, the sulfhydryl-binding dye N-[4-(7-diethylamino-4-methyl-3-coumarinyl)phenyl]maleimide (CPM) has been successfully used for monitoring unfolding and ligand binding to many different types of membrane proteins 11,13,21 . Recently, the CPM assay was successfully applied to detect specific substrate binding in a library of different compounds to mitochondrial SLC transporters 11 .
While our GFP-based overexpression and purification pipeline in S. cerevisiae facilitates the isolation of SLC transporters for the CPM assay 13 , the intrinsic fluorescence of the GFP-fusion partner also includes the possibility to monitor thermal denaturation by fluorescence-detection size-exclusion chromatography (FSEC-TS) 22 . Encouragingly, FSEC-TS is also compatible with unpurified samples for detecting drug or ligand binding 22 . This is an advantage over dye-based unfolding assays, which cannot cope with unpurified samples, require large amounts of purified protein and sometimes give uninterpretable results 23 . There are also mixed reports for the suitability of the CPM dye for monitoring lipid-protein interactions 24,25 , as the interactions detected in principle could be a result of competition between the hydrophobic dye and lipids. Furthermore, in an unpurified sample, the target resides in a more native-like lipid environment, which presents an attractive advantage for cases when it is difficult to purify the protein, or when lipid-driven modulation of drug binding could be an important factor 26 .
Despite these advantages, FSEC-TS is a relatively low-to mid-throughput method, as it relies on a size-exclusion step for separating the protein peak from heat-induced aggregates at each temperature, which remain fluorescent. Here, we show that the SEC step can be replaced by centrifugation, if the short-chain detergent n-octyl-β-D-glucoside (OG) is added prior to heating to force precipitation of heat-induced aggregates to accelerate and simplify the overall procedure, resulting in GFP-TS ( Fig. 1) 27 . The melting temperatures (T m ) measured by GFP-TS in crude detergent-solubilized membranes are comparable to the T m values calculated by FSEC-TS and the unfolding rates measured by the CPM assay with purified transporters (Fig. 2a-d) 27 . Importantly, as outlined here, the GFP-TS assay is transferrable to a 96-well format and, given its simplicity, should be transferrable to  Fig. 2 | Development and validation of the GFP-TS assay. a, GFP fluorescence of DDM-solubilized membranes containing a bacterial homolog of the apical sodium-dependent bile acid transporter from Neisseria meningitis (ASBT NM )-GFP remaining in solution after heating and centrifugation (cyan circles), measured using a 96-well plate spectrofluorometer. The fluorescence from the respective pellets resuspended in identical volumes of buffer to the supernatant is also shown (black squares). Apparent T m values were calculated using data from the nine different temperatures and fitted to a sigmoidal dose-response equation (bars show the range of two technical replicates, and values reported are mean ± s.e.m. of the fit). b, FSEC-TS traces of the ASBT NM -GFP supernatants from a; the aggregates are highlighted by the dotted ellipse, and the corresponding FSEC-TS melting curve is shown as an inset. c, As in a except that the detergent OG was added to the DDM-solubilized ASBT NM -GFP samples prior to heating (bars show the range of two technical replicates, and values reported are mean ± s.e.m. of the fit; T m values are tabulated in Table 1. d, As in b except that the detergent OG was added to the DDM-solubilized ASBT NM -GFP samples prior to heating. The position of the aggregates removed by addition OG is indicated by a dotted ellipse and the corresponding FSEC-TS melting curve as an inset. e, Sensitivity of the GFP-TS assay to free GFP. FSEC profiles of detergent-solubilized ASBT NM -GFP were measured in the presence of increasing concentrations of free GFP, and the apparent T m values were determined by GFP-TS. The percentage of free GFP fluorescence is expressed as a fraction of the total protein sample fluorescence measured on a plate reader. f, Apparent T m values determined for detergent-solubilized membrane fractions of different membrane proteins expressed as GFP fusions (x-axis labels) without OG (open bars), and FSEC-TS values from the same samples (filled bars). FI, fluorescence intensity.

Comparison with other methods
There are other TSAs that are compatible with unpurified samples, in particular, the cellular thermal shift assay (CETSA), which can monitor a change in protein thermal stability upon ligand binding in intact cells or lysates 18,29 . The combination of CETSA principle with high-resolution MS (MS-CETSA), also known as thermal proteome profiling 'TPP', has proven a powerful method 30 , which uses high-resolution mass spectrometry (MS) for calculating melting curves and subsequent ligand-protein induced thermostabilization 31 . Nevertheless, without recombinant protein expression, lowly abundant membrane proteins have proven challenging for CETSA and thermal proteome profiling, with few de novo targets identified 32,33 . Those ligands that have been paired to a membrane protein target have had very high (nM) binding affinities 32,33 . The major drawback of CETSA for deorphanization of a single target is that it requires the availability of highly specific antibodies, which predominantly recognize the folded state of the membrane protein 32 . Given that many membrane proteins and SLC transporters lack specific and high-quality antibodies, overexpression of the target in a well-folded state becomes essential, which is facilitated by GFP-based tagging. Moreover, the GFP-TS assay has fewer technical steps than CETSA and facilitates downstream purification, which is a requirement for studying lipid-protein interactions.
In addition to ligands and substrates, specific lipids are often found to regulate the function of SLC transporters and membrane proteins in general [34][35][36][37] . Recently, MS has gained traction as a tool for probing the intricate connections between membrane proteins and lipids 38 , shedding light on different lipid binding modes 39 , lipid-mediated protein stabilization and even thermodynamics and allosteric effects of protein-lipid interactions 40,41 . Despite its proven usefulness for studying individual protein-lipid contacts and the development towards analyzing more complex mixtures 42 , currently MS, like other spectroscopic and computational approaches, is not well suited for identifying specific interactions in native membranes or in high-throughput formats, impeding large-scale investigations.
Here, we further demonstrate that FSEC-TS and the GFP-TS assay can be applied to screen for and detect specific lipid interactions of purified SLC transporters. Although the simpler GFP-TS method increases sample capacity and makes it possible to screen and assess hundreds of ligands simultaneously, FSEC remains an important initial quality check for the GFP-TS assay. Furthermore, FSEC-TS can be used to further assess if ligands or lipids alter oligomerization of the target 27 . We have found these indirect GFP-based methods for assessing ligand interactions to be a particularly useful tool in combination with native MS for confirming and mapping specific lipid interactions to SLC transporters, which could otherwise be perturbed when assessed by thermofluor dyes 27,43 .

Overview
The current protocol (a schematic overview is shown in Fig. 1) begins with a yeast membrane fraction harboring the heterologous overexpressed GFP fusion, which was prepared according to our previous protocol 14 . Initially, the target-containing membranes are solubilized in the mild detergent n-dodecylβ-D-maltopyranoside (DDM), and the sample is assessed for its suitability for GFP-TS by FSEC (Steps 1-5). Once protein quality has been established, the next stage is to assess target thermostability in membranes (Steps 6-12). Briefly, the target is solubilized in the mild detergent DDM and aliquoted. The harsh detergent OG is added, and the samples are heated over a range of different temperatures from 10°C to 90°C, and then pelleted to remove heat-induced aggregates. The fluorescence remaining in the supernatant from the non-aggregated GFP fusion is then measured in a plate reader, and the apparent melting temperature T m is calculated. Downstream, we outline the optimization steps that might be required on a case-by-case basis, prior to large-scale ligand screening, for example, the addition of the solvent dimethyl sulfoxide (DMSO) to assess potential interference with detecting interactions with ligands that have been stored in DMSO (Steps 13-21) and the optimal temperature for screening for ligand interactions (Steps [22][23][24][25][26][27][28][29]. Once these parameters have been adjusted, ligand screening can be conducted by GFP-TS using detergent-solubilized membranes (Steps 30-38). Moreover, affinity estimates can be obtained simply by heating the GFP fusion with increasing ligand concentrations (Steps 39-48). GFP-TS can also be used tο rapidly assess how mutations in the protein affect ligand binding, which is most optimal when structural information is available (Steps 49-54).
Overall, the GFP-TS is applicable for GFP fusions in either crude membranes or as purified fusions, with minimal adaptations between the two sample types. In the end of this protocol, we outline the similar steps for obtaining an apparent T m using purified material (Steps 55-57) and how the detergent purified protein can be used to screen for specific lipid-protein interactions (Steps 58-66).

Experimental design
Starting material: target protein in membranes As outlined in the preceding protocol 14 , the targets and constructs are first optimized for monodispersity in the mild detergent DDM by FSEC 44 . To omit the FSEC for analysis by GFP-TS, the amount of free GFP should not exceed 30% of the protein-GFP peak to ensure the melting temperature stays within a 10% error (Fig. 2e); this is because free GFP is stable and will not be precipitated by heating, masking the signal reduction of the protein-GFP fusion. In addition, aggregates formed during target overexpression should not be more than 30% of the folded proteinfusion peak as they could also potentially mask unfolding, since the GFP-TS assay is optimized to remove heat-induced aggregates, and not aggregates present prior to heating.
Addition of OG to DDM-solubilized protein-GFP fusions OG is considered a 'harsh' detergent, as its addition tends to cause protein aggregates to precipitate rather than stay in solution 13 . As OG is added to all samples, it is recommended for experiments to proceed swiftly after addition of OG to decrease unwanted time-cumulative protein aggregation. On average, melting temperatures with DDM + OG are~7.5°C lower than with DDM only 28 . Although the addition of OG might not be strictly necessary in all cases, on average, the apparent T m values obtained with GFP-TS are more comparable to FSEC-TS when OG is included (Fig. 2b,c,f and Table 1). For this reason, the GFP-TS assay has been optimized with inclusion of OG (Fig. 3a). We observe a strong correlation between GFP-TS and the CPM assay when OG is included (R 2 = 0.78, Fig. 3b). Furthermore, an average T m of 47°C was estimated by thermal proteome profiling of 371 membrane proteins in a human cell line when proteins were solubilized in a mild detergent 31 , suggesting that many targets should be stable enough in DDM + OG detergents. However, in cases when the T m drops to <30°C with OG present, adjustments to this protocol might be necessary, such as reducing the final concentration of OG from 1.0% to 0.5% wt/vol (Tables 1, 2).

Heating block versus thermal cycler for heating
Initial experiments were carried out using a heating block. However, to further develop our assay and render it suitable for high-throughput applications, we adapted the protocol to use a thermal cycler, which allows for processing of up to 96 samples simultaneously. Moreover, gradient thermal cyclers allow generation of full melting curves faster and with the same accuracy as a standard heating block

PROTOCOL EXTENSION
NATURE PROTOCOLS (Fig. 3c). Therefore, our procedure instructs for the use of a gradient or conventional thermal cycler. However, all experiments can be performed using a heating block yielding the same quality if a thermal cycler is unavailable.
Increasing temperature Increasing fluorescence intensity c, Apparent T m measured by GFP-TS for detergent-solubilized human CMP-sialic acid transporter (hCST)-GFP membranes in heating block (black squares) and thermal cycler (magenta circles) format. d, Effect of increasing DMSO concentration on signal from detergentsolubilized hCST-GFP membranes. Values of the heated samples (filled bars) are normalized against the nonheated control sample with 0% (vol/vol) DMSO (empty bar). Data were obtained from two independent experiments. e, Optimization of assay temperature for optimal signal intensity from detergent-solubilized hCST-GFP membranes. The intensity values obtained from two independent experiments were normalized against the nonheated control sample kept at 4°C (empty bar). The red box indicates the suitable temperature range to assay, resulting in 20-40% of the original nonheated control intensity. f, In some cases, due to low expression levels, it might be necessary to concentrate the sample and thus increase the total protein-GFP fluorescence by increasing the membrane to detergent ratio. The apparent T m values for detergent-solubilized ASBT NM -GFP using GFP-TS only differ by 1.5°C over a total membrane protein concentration range of 3.0-7.5 mg/ml. b reproduced from ref. 27 .
Assay check-points for ligand screening In addition to measuring the melting temperature of the target protein, GFP-TS can be applied to screen for interacting ligands. To ensure the success of such a screen, it is critical that appropriate control experiments are carried out. Typically, ligands are dissolved in organic solvents such as DMSO or ethanol. Consequently, prior to performing a ligand screen, the possibility of solventinduced effects on the target protein stability must be determined. By monitoring the protein stability over a range of solvent concentrations, potential solvent effects can be identified and eliminated (Fig. 3d). As already established, GFP-TS is a thermostability-based assay, which means a temperature-induced destabilization of the system is necessary when screening for ligands. For GFP-TS, potential stabilizing effects of ligands are more prominent when the remaining protein signal is~20-50% compared with that of the unheated and untreated control. To identify this window of reduced signal intensity, it is recommended that a range of temperatures are investigated prior to large-scale ligand screening (Fig. 3e).

Limitations
The use of GFP as a protein marker has become a mainstream approach in protein production and purification systems 14,45-47 , making GFP-TS an attractive and easily applicable option for many laboratories. This offers a great amount of adaptability considering the intrinsic stability of GFP as well as the many varieties of GFP that have been engineered overtime. At the same time, one should carefully consider the experimental setup, which might require optimization, to avoid uninterpretable or inconsistent results. For example, a possible issue arising from protein overexpression is increased protein aggregation and, in the case of GFP-based protein production pipelines, cleavage of GFP during target preparation. Since there is no possibility to distinguish between signals from target-GFP aggregates, target-GFP and 'free' GFP in the sample, it is important that one first ensures sample quality by FSEC 22 prior to this assay to avoid false signal generation or signal masking caused by free GFP, i.e., as also outlined in the proceeding protocol 14 The GFP-TS assay relies on the precipitation of heat-induced aggregates through the addition of a relatively harsh detergent β-OG to the mild detergent DDM. Consequently, a degree of protein destabilization due to interaction with OG can be expected. In our experience, if the target remains monodisperse (well folded) once purified in DDM as previously outlined 14 , then the target will be stable enough for using the GFP-TS assay. Nevertheless, we anticipate that, in some cases, the degree of destabilization caused by OG addition might make the target no longer stable enough for GFP-TS (T m < 30°C). Alternatives to overcome this are offered in the troubleshooting section. Since GFP is no longer fluorescent >76°C, the GFP-TS assay cannot be used to assess membrane proteins with higher melting temperatures. Nevertheless, we have not come across eukaryotic SLC transporters with an apparent T m > 76°C in the presence of DDM and β-OG detergents.
On a general level, TSAs rely on changes of the target thermal stability resulting from interaction with a binder. This approach has been widely explored for structural studies as well as in drug discovery 18 . Although particularly useful for large-scale screening, thermal stabilization is not always observed upon ligand binding, and thus, the GFP-TS assay, like any other TSA, is prone to falsenegative results 30,48 . Furthermore, ligand screens conducted on crude material might fail to detect protein-ligand pairing, due to off-target interactions, or ligand modifications caused by other proteins in the system 49 . Consequently, the amount of ligand available for interaction with the target is diminished or no longer recognized. In such cases, purification of the GFP target might be necessary before application of the GFP-TS assay.

Biological materials
Membrane resuspension containing GFP-fused protein target. See Steps 27-32 in ref. 14 .     c CRITICAL STEP If the solubilization efficiency is >70% but the fluorescent signal is weak (less than tenfold higher than background), the amount of membranes used can be increased to boost the signal-to-noise ratio (Fig. 3f). ? TROUBLESHOOTING 4 Transfer 100 µL of supernatant into an FSEC vial, and inject the sample into a Biorad Enrich 650 SEC column (flow rate 1 mL/min) equilibrated in FSEC buffer. 5 Analyze the FSEC trace to confirm the target-GFP fusion is the predominant species and any aggregation or free GFP is <30% of the total peak height.

? TROUBLESHOOTING
Determining melting temperature (T m ) in crude detergent-solubilized membranes • Timing 3 h 6 Adjust total membrane protein concentration to at least 3.5 mg/mL in 6.5 mL by addition of GFP-TS reaction buffer. Solubilize under mild agitation at 4°C for 1 h. c CRITICAL STEP When preparing the protein-solubilization mix, subtract the volume of OG that will be added in the following step (10% of the total volume, i.e., solubilization is carried out in 6.5 − 0.65 = 5.85 mL). ? TROUBLESHOOTING 7 Add 650 µL OG stock (for a final concentration of 1% wt/vol), and mix sample carefully by inverting the tube. 8 Aliquot 120 µL of the solubilized membrane solution with OG (from Step 7) into triplicate 0.2 mL PCR tubes. 9 Incubate each triplicate in a set temperature gradient (10-90°C, incrementing steps of 5°C) in a gradient thermal cycler, for 10 min. Cool immediately after incubation by returning the tubes to ice. c CRITICAL STEP The selection of temperatures can be adjusted depending on the available amount of sample. 10 Centrifuge samples at 5,000g using a Beckman Coulter table top centrifuge at 4°C for 45 min. 11 Carefully transfer 80 µL of the supernatant into a 96-well black plate, and measure GFP fluorescence emission at 512 nm by excitation at 488 nm in a microplate spectrofluorometer. c CRITICAL STEP Aspirating 80 µL gives a good margin to avoid aspirating precipitated protein aggregates; however, if necessary, a smaller volume of~50 µL can be aspirated. 12 Plot the values of GFP fluorescence against the respective temperature, and calculate the melting temperature of the target-GFP fusion, using a sigmoidal four-parameter logistic function in GraphPad Prism or an equivalent graphing program. For data normalization, set the highest and lowest data points of the set as 100% and 0%, respectively. c CRITICAL STEP GFP has a melting temperature of 76°C, above which it no longer fluoresces. Therefore, only proteins with a melting temperature <76°C can be routinely analyzed using the GFP-TS assay. ? TROUBLESHOOTING

Screening of ligand libraries using detergent-solubilized membranes
Assessing solvent-induced effects on target protein stability • Timing 2 h 30 min 13 Prepare a series of stock solvent concentrations, ensuring that you have enough stock for triplicate reactions (10 µL of each stock will be used in the reaction mix). The span of solvent concentrations to be tested depends on the solvent chosen in the intended ligand screen. 14 Aliquot 10 µL of each solvent stock into triplicate 0.2 mL PCR tubes. In parallel, set up solvent-free controls in triplicate, and add 10 µL water to them. 15 Adjust total membrane concentration to 3.5 mg/mL by addition of GFP-TS reaction buffer.
Solubilize under mild agitation for 1 h at 4°C. c CRITICAL STEP The required amount of detergent-solubilized membranes is 98 µL per sample. 16 Aliquot 98 µL of the solubilized protein sample into the tubes containing the solvent. 17 Add 12 µL OG stock (for a final concentration of 1% wt/vol) into each sample. 18 Heat the tubes at the determined T m of the untreated protein from Step 12 (plus a triplicate set at 4°C) for 10 min using a thermal cycler. Cool immediately by returning the tubes to ice. 19 Centrifuge samples at 5,000g using a Beckman Coulter c CRITICAL STEP Using crude detergent-solubilized membranes allows for screening in more native-like conditions, without the need of purified material. However, certain ligands may interact with proteins present in the membranes and may even be modified, thus reducing the concentration of ligand available for interaction with the target. If this is a concern, the ligand screen could instead be performed using purified protein.

Determination of ligand binding affinity • Timing 3 h 15 min
39 Generate a series of stock ligand concentrations, ensuring that enough of each stock is available to use 10 µL in the reaction mix (for triplicate reactions). If solvent tolerance is a problem, smaller volumes of more concentrated ligand stocks can be used. The respective ligand concentration to be tested varies on an individual target protein basis. Generally, a span of 0-10 mM is recommended, but lower concentrations may be used when ligand solubility and/or availability limits exist, e.g., when compound availability is scarce, when using compound libraries with a uniform concentration and/or when restricted by solvent tolerance of the target.
? TROUBLESHOOTING 40 Aliquot 10 µL of each ligand stock solution into triplicate sets of 0.2 mL PCR tubes, with increasing ligand concentrations. In parallel, prepare a ligand-free control by adding 10 µL of solvent stock in place of the ligand stock for an identical final solvent concentration. 41 Adjust total membrane concentration to 3.5 mg/mL by addition of GFP-TS reaction buffer, and solubilize under mild agitation at 4°C for 1 h. 42 Aliquot 98 µL of the solubilized protein sample in the tubes containing the ligands, and mix carefully. 43 Incubate the protein-ligand samples at 4°C for 1 h. 44 Add 12 µL OG stock (for a final concentration of 1% wt/vol), and mix samples carefully. 45 Heat the tubes at the optimal heating temperature determined at Step 29 for 10 min in a thermal cycler. Cool immediately by returning the tubes to ice. 46 Centrifuge samples at 5,000g using a Beckman Coulter table top centrifuge at 4°C for 45 min. 47 Remove 80 µL of the supernatant, and measure GFP fluorescence as described in Step 11. 48 Plot the values of GFP fluorescence of each sample against the respective ligand concentration, and fit the data points using an appropriate nonlinear regression binding function in GraphPad Prism, e.g., a one-site total binding equation.

Mutation analysis of ligand-binding site residues • Timing 3 h 40 min
49 Adjust total membrane concentration to 3.5 mg/mL in 13 mL (subtract ligand and OG volume that will be added in later steps) by addition of GFP-TS reaction buffer. Solubilize under mild agitation for 1 h at 4°C. 50 Divide the solubilized material into two aliquots of equal volume. 51 Add 65 µL of ligand stock to one aliquot, and an equal volume of solvent to the other. 52 Incubate both aliquots at 4°C for 1 h. 53 Repeat the T m measurement as described in Steps 6-12. 54 Calculate the ΔT m of each mutant construct by comparing the individual T m values in absence and presence of ligand. c CRITICAL STEP When analyzing amino acid variants in a given target, it is possible that the T m of the mutant(s) is different from the T m of the initial construct. Therefore, calculating ΔT m (by comparing the values of ligand-free and ligand-present states) is more accurate for assessing the effects of mutations on ligand binding.
Measuring melting temperature of DDM-purified GFP fusions • Timing 2 h c CRITICAL The following steps describe procedures for performing melting T m measurements on purified target protein, which are very similar to those outlined for detergent-solubilized membranes. Using purified protein is required when screening for specific lipid-protein interactions, or if the fluorescent signal needs to be higher to accurately monitor ligand interactions. The GFP fluorescence intensity should result in a signal-to-noise ratio greater than 10 (recommended target protein concentration~0.01-0.03 mg/mL). 55 Adjust target protein to~0.01-0.03 mg/mL in 6.5 mL (subtract OG volume) SEC buffer. 56 Add 650 µL OG stock (final 1% wt/vol), and mix sample carefully by inverting the tube. 57 Repeat Steps 8-12.
Screening of stabilizing lipids with purified GFP fusion • Timing 1 h 30 min 58 Follow Steps 22-29 (with addition of DDM instead of solvent) to determine the optimal heating temperature of the protein-lipid mix. c CRITICAL STEP Lipid stocks are prepared by solubilization in 10% (wt/vol) DDM overnight under mild agitation at 4°C, and are diluted tenfold in the protein solution (final concentration of DDM 1% wt/vol). Consequently, the control samples must contain 1% (wt/vol) DDM. 59 Aliquot 12 µL of each lipid stock into triplicate 0.2 mL PCR tubes. Aliquot an equal volume of 10% (wt/vol) DDM stock into the control tubes. 60 Based on the GFP fluorescence intensity determined in Step 55, adjust target protein tõ 0.01-0.03 mg/mL in SEC buffer. 61 Aliquot 96 µL of protein solution from Step 59 into the tubes, and mix carefully by pipetting. 62 Add 12 µL of OG stock (final concentration 1% (wt/vol)), and mix carefully. 63 Incubate each triplicate set at the optimal heating temperature (determined at Step 29 for detergentsolubilized membranes) for 10 min using a thermal cycler. Cool immediately after incubation by returning the tubes to ice. 64 Centrifuge samples at 5,000g using a Beckman Coulter table top centrifuge at 4°C for 45 min. 65 Carefully transfer 80 µL of the supernatant into a 96-well black plate, and measure GFP fluorescence as described at Step 11. 66 Plot the values of GFP fluorescence from the lipid-containing samples against the DDM-only control. For data normalization, set the fluorescence value of the DDM-only sample as 100%.

Troubleshooting
Troubleshooting advice can be found in Table 2.

Timing
Steps

Anticipated results
We have performed the GFP-TS assay using crude membrane material and purified sample in detergent for several SLC transporters 27,28 . Using crude detergent-solubilized membranes, we demonstrate here that the GFP-TS assay can discriminate the binding of the substrate nucleotide cytidine monophosphate (CMP) to human CMP-sialic acid transporter (CST) SLC35A1 from the closely related nonsubstrate nucleotide uridine monophosphate (UMP), and 192 nucleoside analogs in a 96-well format (Fig. 4a,b). We confirm that 0.5 mM of CMP stabilizes human SLC35A1 by ΔT m of 5°C (Fig. 4c), as visually apparent when one compares the FSEC traces of sample heated with a temperature of 40°C with and without the nucleotide present (Fig. 4d). Using human SLC35A1 as a c, Apparent T m determined using GFP-TS for hCST-GFP membranes without (black circles) and after (green squares) addition of 0.5 mM CMP. d, FSEC profiles of hCST-GFP heated at 40°C in absence (black line) and in presence (green line) of 0.5 mM CMP. The samples were collected from a 96-well plate after GFP-TS. e,f, Binding of CMP to Zea mays CMP-sialic acid transporter (CST ZM )-GFP (black squares) and hCST-GFP (cyan circles) in membranes (e) and CMP-sialic acid to purified CST ZM -GFP (black squares) and hCST-GFP (cyan circles) (f) as monitored by GFP-TS. Binding affinities (K d ) were calculated from data points recorded over a range of CMP and CMP-sialic acid concentrations, and these were fitted by nonlinear regression using a one-site total binding function. Respective K d values obtained by ITC are shown in parentheses. Data show mean ± s.e.m from three independent experiments. e adapted from ref. 27 ; f adapted from ref. 28 . control, we have further shown that the plant homolog from Zea mays is specifically stabilized by the nucleotide monophosphate CMP, and is therefore almost certainly a transporter for CMP-sialic acid 27,28 . It was later confirmed that the plant homolog could transport CMP-sialic acid (in exchange for CMP) by proteoliposome-based transport assays using radiolabeled substrates 28 Owing to the challenges associated with setting up proteoliposome-based assays 11,17 , and the limited availability of labeled compounds, the GFP-TS assay is an important stepping stone to in-depth functional characterization.
The GFP-TS was further utilized with crude detergent-solubilized membranes to determine the CMP binding affinities (K d ) for plant and human CST, which were found to be consistent with each other and with isothermal titration calorimetry (ITC) measurements from the purified proteins 27 (Fig. 4e). Interestingly, CMP-sialic acid was found to bind with much lower affinity than CMP, and we could therefore only obtain a K d value using the GFP-TS as the signal was too weak for ITC (Fig. 4f); the differences between CMP and CMP-sialic acid binding affinities were further consistent with IC 50 value differences 28 .
Membrane protein crystals grown in meso or using the lipidic cubic phase method generally produce higher-resolution structures, as they have a lower solvent content (type I crystals) than those grown by traditional vapor-diffusion crystallization (type II crystals) 50 . To grow lipidic cubic phase crystals of membrane proteins, the purified membrane protein solution is mixed with the synthetic lipid monoolein 51 . Using GFP-TS assay, we added monoolein to purified transporters, and found that the monoolein lipid was in fact very destabilizing for most of the eukaryotic membrane proteins tested, including plant CST 27 . To compensate, we found that if CMP was added during in meso based crystallization of plant CST we could improve its stability and enable the formation of lipidic cubic phase crystals, which were used to determine the structure by X-ray crystallography 28 . We next employed the GFP-TS assay to carry out extensive mutagenesis-based analysis of the CMP binding site of plant CST (Fig. 5a-c) in parallel with human CST (Fig. 5d-e). Importantly, all CMP and CMP-sialic acid binding measurements could be carried using small amounts of crude detergentsolubilized membranes, and would have otherwise been more expensive and time-consuming if each mutant needed to be purified in sufficient amounts for binding analysis using most other methods, such as ITC.
Lipids are of critical importance to membrane protein function 38 , and yet molecular details are often difficult to obtain due to their hydrophobic nature and weak interactions with membrane proteins. Although FSEC-TS was shown to detect the binding of ligands and drugs to channels and transporters, e.g. refs. 22,52 , it was unclear if it could detect specific lipid interactions. We recently confirmed that FSEC-TS could be used to detect such specific interactions, demonstrating that the lipid cardiolipin binds to the bacterial Na + /H + antiporter NhaA from Escherichia coli and not to its close homolog NapA from Thermus thermophilus 27 ; both of these observations were originally reported by native MS measurements 53,54 (Fig. 6a-d). We were able to support this finding by generating a monomeric mutant of NhaA, which no longer binds to cardiolipin and thus does not show the cardiolipin-driven dimerization of wild-type NhaA 27 . More recently, we reported the structure of the mammalian Na + /H + exchanger SLC9A9 (NHE9), and studied its interactions with lipids using native MS (Fig. 6e) 43 . However, native MS could only estimate potential lipid interactions, whereas with FSEC-TS we confirmed the specific lipid-binding interactions of PIP 2(3) lipids that were consistent with the masses detected by native MS (Fig. 6e-g) 43 . The PIP 2(3) binding could be abolished by mutagenesis of several NHE9 residues as confirmed by combining thermostability and native MS measurements 43 . Our results suggest that the negatively charged lipids cardiolipin and PIP 2 bind to NhaA and NHE9, respectively, to stabilize the functional homodimer 43 . Here, we show that GFP-TS gives comparable results to FSEC-TS for cardiolipin binding to NhaA or PIP 2 binding to SLC9A9 (Fig. 6c,g,h). We have also used GFP-TS to screen for the lipid binding preference of SLC transporters and several bacterial homologs in high throughput 27 . Surprisingly, we found that, on average, the melting temperatures of SLC transporters and the bacterial homologs were similar prior Fig. 5 | GFP-TS assay for ligand interaction studies to human SLC35A1 and a plant homolog. a, The substrate-binding site of CST ZM in the outwardfacing conformation (PDB id: 6I1R). Transmembrane segments are shown in cartoon representation colored brown. Water molecules are shown as brown spheres; CMP and coordinating residues are shown as sticks. Putative hydrogen bonds are depicted as dashed lines, as is the π-cation interaction between W209 and cytosine nucleobase. b, Apparent T m shift after addition of 1 mM CMP to WT (open bars) and CMP sialic-acid (CMPsia) binding residue mutants (filled bars) of CST ZM . The mutated residues were selected based on the substrate-binding site presented in a. UMP was added to measure nonspecific binding (red bars). c, As b but with 1 mM CMP-sia as ligand, and sialic acid added to measure nonspecific binding (red bars) instead. d, As b but with hCST. e, As c but with hCST. Data show mean ± s.e.m from three independent experiments. Figure adapted from ref. 28 . to purification 27 . Destabilization of the SLC transporters was more apparent upon purification, but could be recovered again by the addition of specific lipids 27 . Our results suggest that the GFP-TS assay can identify lipids that recover lipid-specific stabilization of SLC transporters, which is lost during the de-lipidation process of protein purification.
Taken together, the GFP-TS assay is a valuable tool in the effort to functionally characterize SLC transporters and their interaction with inhibitors and lipids for establishing important mechanistic  details. While our focus has been on the use of the GFP-TS assay for transporters produced in S. cerevisiae, we see no reason why the described methodology cannot be utilized for screening ligand and lipid interactions in other membrane protein families using protein produced from other GFP-based expression and screening platforms, e.g., as outlined in refs. 45,55 .

Data availability
All data generated or analyzed during this study are included in this published article.