Protein stabilization with retained function of monellin using a split GFP system

Sweet proteins are an unexploited resource in the search for non-carbohydrate sweeteners mainly due to their low stability towards heating. Variants of the sweet protein monellin, with increased stability, were derived by an in vivo screening method based on the thermodynamic linkage between fragment complementation and protein stability. This approach depends on the correlation between mutational effects on the affinity between protein fragments and the stability of the intact protein. By linking the two fragments of monellin to the split GFP (green fluorescent protein) system, reconstitution of GFP was promoted and moderately fluorescent colonies were obtained. Two separate random libraries were produced for the monellin chains and the mutant clones were ranked based on fluorescence intensity. Mutants with increased affinity between the fragments, and subsequently increased stability, caused increased fluorescence intensity of split GFP. Single chain monellin variants of the top-ranked mutants for each chain, S76Y in the A-chain and W3C + R39G in the B-chain and all combinations thereof, were expressed and the increase in stability was verified by temperature denaturation studies using circular dichroism spectroscopy. Functionality studies showed that mutant S76Y has retained sweetness and has potential use within the food industry.

produced by error-prone polymerase chain reaction (PCR) and to rationally designed libraries. Lindman et al. 7 used the split GFP system to screen for B1 domain of protein G (PGB1) mutants with increased stability from a designed library as proof of concept of the approach. Here we report the first unbiased fragment libraries screened with this method. Furthermore, we couple the screen for stabilized variants to a functional readout and demonstrate the importance of such validation of functionality of the stabilized protein.
The two fragments of split GFP do not reconstitute spontaneously and consequently do not emit any fluorescence when co-expressed. By fusing the fragments of GFP to other molecules that interact, reconstitution can be promoted and fluorescence obtained. The fused molecules facilitate complex formation of the fluorescent fragments by bringing them in proximity to each other. Regan and co-workers showed that an affinity above ca. 10 3 M −1 between the fused molecules is sufficient for detectable green fluorescence to emerge 8 , whereas high affinity above 10 9 M −1 promotes very bright fluorescence 9 . The main fields of use of split GFP and other bimolecular fluorescence complementation (BiFC) assays are to study protein interactions and to visualize the subcellular localization of molecular interactions 10 . Fragments of several different proteins have been used to detect protein interactions, e.g. ubiquitin, β-galactosidase, luciferases and different versions of GFP [11][12][13][14] , though fluorescent proteins remain the most common approach. GFP emits bright green fluorescence when folded correctly and it obtains its native fold slowly through several intermediates 15 . The protein has been mutated (F64L, S65C, Q80R, Y151L, I167T, and K238N) to get one excitation peak at 475 nm and emission at 505 nm 16 . The chromophore consists of residue Cys65-Tyr66-Gly67 and is in the interior of a β-barrel structure. Split GFP linked to monellin fragments results in moderate fluorescence which enables detection of monellin variants with increased stability.
Sweet proteins acquire attention as there is a zeal for non-carbohydrate sweeteners from food industry. Sweeteners within food industry must maintain their sweetness at high temperature, high pressure and at varying pH. Monellin has an intense and persistent sweetness and is approximately 100 000 times sweeter than sucrose on a molar basis 17 . The protein belongs to a family of seven structurally heterogeneous, sweet and sweet taste-modifying proteins: monellin 18 , brazzein 19 , thaumatin 20 , mabinlin 21 , lysozyme 22 , neoculin 23 and miraculin 24 . Monellin was originally isolated from the tropical serendipity berry (Dioscoreophyllum cumminsii) 18 and the sweetness of monellin can be sensed by old world monkeys and humans only 25 . The protein interacts with the human sweet taste receptor T1R2-T1R3 (a heterodimeric G-protein coupled receptor (GPCR)) on the tongue [26][27][28] . The exact mode of interaction between sweet proteins and the receptor is unknown but the "wedge model" has been proposed to explain how sweet proteins stabilizes the active form of the receptor by binding to an external cavity spanning the T1R2-T1R3 dimer 29,30 . The sweetness of proteins depends on their net charge, surface charge distribution and the three-dimensional structure of the protein and thereby its surface [31][32][33] . Monellin is a heterodimer of two subdomain fragments interacting tightly in a single domain with a β-grasp fold, which consists of an α-helix packed perpendicular to a five-stranded anti-parallel β-sheet 17,34 . A single-chain derivative of monellin (scMN) is used in this study. The sweetness is retained in scMN and it has increased stability compared to wild type monellin heterodimer [35][36][37] . However, after heating or boiling, scMN may not refold reversibly depending on the time spent at high temperature, and on the protein concentration, due to sample aggregation and precipitation 38 . Here we therefore utilize the split GFP assay with the monellin subdomain fragments to obtain an even more stable variant of scMN, with retained sweetness, that can meet the stability requirements from the food industry.

Results
Nomenclature. Wild type monellin consists of two non-covalently attached chains, MNA and MNB 17 . The split GFP system used here relies on non-covalent association of the NGFP-MNA and MNB-CGFP gene products. NGFP-MNA denotes a fusion construct where the C-terminus of the N-terminal GFP fragment (residues 1 to 157) has been linked through an eight-residue linker to the N-terminus of MNA. MNB-CGFP is a fusion construct where the C-terminus of MNB has been linked to the C-terminal fragment of GFP (residues 158 to 238) through a seven-residue linker (Fig. 1). Parent MN (pMN) denotes a construct where cysteine 41 in the B-chain has been replaced by serine to avoid disulfide bridges. ScMN denotes a single-chain construct of monellin where Figure 1. A cartoon model of split GFP fused to monellin. MNA was fused to NGFP (green) and MNB was fused to CGFP (blue). The model is produced from pdb files 1EMA 54 and 2O9U 55 using Pymol 56 .
the C-terminus of the B-chain (MNB) has been linked to the N-terminus of the A-chain (MNA) by a Gly-Phe linker.
Expression of NGFP-MNA and MNB-CGFP constructs. The genes for NGFP-MNA and MNB-CGFP were fused to the same plasmid by ligation independent cloning (LIC) to ensure similar expression levels of the two constructs 39 . Transformation of this plasmid into Escherichia coli (E. coli) and expression of the parent NGFP-MNA and MNB-CGFP, under inducing conditions, resulted in green fluorescent colonies after approximately 2-4 days at room temperature (Figs 2 and 3a). Culturing on non-inducing plates showed no green fluorescence; neither did a negative control on inducing plates expressing NGFP and CGFP without MNA and MNB.
Ranking of fluorescent monellin mutants with increased stability. We set out to search for split GFP library colonies with increased green fluorescence compared to colonies containing the combined parent fragments, NGFP-MNA and MNB-CGFP. This was motivated by the correlation between the stability of the singe-chain protein and the affinity between protein fragments; the MN fragment association serving as a driver of GFP fragment association, folding and maturation of its chromophore. Two separate random libraries were produced for the A and B chain of monellin, respectively (Fig. 2). The MNA library was co-expressed with parent MNB and the MNB library was co-expressed with parent MNA. The Mutazyme ® II DNA polymerase was chosen to obtain similar mutation frequency for all nucleotides. The desired mutation level of 1-2 mutations per monellin chain was verified by DNA sequencing, though some triple mutations were also found. The libraries where spread on inducing plates and plasmids were purified from colonies identified with increased fluorescence intensity compared to the split GFP reconstituted with parent monellin chains. Selected mutant clones were expressed on separate inducing plates and ranked based on fluorescence intensity and compared to parent MN. The ranking was performed by six volunteers and the top 30 ranked mutants were sequenced. The two selected top ranked mutations were S25Y in the A chain of monellin (which corresponds to S76Y in single chain monellin) and W3C + R39G in the B chain of monellin (which corresponds to W3C and R39G in single chain monellin) (Fig. 3). The brightest fluorescence was emitted by split GFP fused to the S76Y mutant. The S76Y mutation was also found in a double mutant (MNA L11R + MNA S25Y) as well as in a triple mutant (MNA S25Y + MNA P43S + MNA P45S). Arginine at position 39 in the B chain of monellin was found to be mutated to five different amino acids (G, C, S, H, L) within the 30 top ranked colonies. The W3C mutation was not found as a single mutation but the tryptophan was replaced by a glycine in one of the top ranked mutants. Other mutants from the top ranked colonies with increased fluorescence were F38S, D27N, F1S, V42I, N39Y, P43L, K18I + A22V, G32D, F1L, F1T + G32D, Y12stop for MNA and E2D, N14K, G1S, E22D, F18Y, R39H + K43I, E48V for MNB.
In an attempt to increase the stability of MN even further, a random library of MNB was expressed together with the MNA S25Y mutant but no mutants with increased fluorescence intensity compared to MNA S25Y mutant with parent MNB were found.
Protein expression and purification of single chain monellin mutants. Single chain monellin variants (devoid of GFP) were produced to enable verification of increased stability of the mutants by thermal denaturation experiments. Parent scMN, single mutants (W3C, R39G and S76Y), double mutants (W3C + S76Y, W3C + R39G and R39G + S76Y) and the triple mutant (W3C + R39G + S76Y) were expressed in E. coli and purified using weak cation exchange chromatography and gel filtration. Dithiothreitol (DTT) was added to variants containing the W3C mutation to retain monomeric samples. Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) analysis revealed that highly pure protein was obtained for all mutants (Fig. 4a).

Mass spectrometry (MS).
Length and identity of the proteins were verified by mass spectrometry. Parent scMN and mutants were cleaved by trypsin and investigated by MS and MSMS. For parent monellin and mutants W3C, S76Y, W3C + S76Y, W3C + R39G and W3C + R39G + S76Y, both full length protein and protein where the N-terminal methionine was cleaved off were found. For example, both peptide MGEWEIIDIGPFTQNLGK and GEWEIIDIGPFTQNLGK were found from parent scMN.
Secondary structure by farUV circular dichroism (CD) spectroscopy. Secondary structure of parent scMN and S76Y was evaluated by CD spectroscopy in the far-UV region (Fig. 4b). The spectra suggest that both proteins are folded and that the global fold is similar. The minima observed at approximately 200-210 nm in spectrum for the S76Y mutant can most likely be ascribed to the addition of an aromatic residue 40 . The changes in intensity are likely due to concentration differences.  Nuclear magnetic resonance (NMR) spectroscopy. NMR spectra revealed highly pure protein with well-folded structures for parent scMN and S76Y, (Fig. 4c). Spectra for both proteins have similar line width and similar distribution of peaks. There are only minor differences in chemical shifts. This indicates that the structures of parent scMN and S76Y are very similar. The peak observed at 8.5 ppm originates from the NMR tube.
Thermal stability of single chain monellin mutants. The stabilities of top ranked mutants and parent scMN were examined in 5 mM sodium phosphate buffer, pH 5.5 (with addition of 1 mM DTT for cysteine-containing mutants) by thermal denaturation monitored by the CD signal at 215 nm (Fig. 5). For all protein variants, a loss in signal intensity was recorded as the temperature increased. The denaturation data appear to follow a sigmoidal curve, indicating cooperative denaturation. Previous studies have shown that temperature denaturation of parent scMN is reversible when the protein is briefly heated to 95 °C, while prolonged heating causes aggregation and precipitation 38 . Since monellin has been shown to be aggregation prone, the protein concentration was kept low (6 µM) during temperature denaturation experiments. Reverse thermal scans indicate that the denaturation process is reversible for parent scMN, S76Y, W3C and W3C + S76Y. The sigmoidal part of the curve is reversible for all samples, though mutants R39G, R39G + S76Y, W3C + R39G and the triple mutant W3C + R39G + S76Y experience a small gain of signal during the reverse scan (Fig. S1). The most stable monellin variants did not reach the final plateau of the sigmoidal curve, indicating that the plateau lies beyond 95 °C. The midpoint of thermal denaturation (T m ) (which is the primary variable of interest), enthalpies of unfolding (ΔH°T m ) and the free energies of unfolding (ΔG°) ( Table 1), were acquired by fitting equations (1) and (3) to the data. The mid-point of thermal denaturation of parent scMN is 83 °C, which correlates well with previously published data 38 . Top-ranked mutations, S76Y and W3C + R39G, obtained from the ranking of the fluorescent GFP-MN variants, contributed to increased stability compared to parent scMN. The stability order of the mutants is the same as the order of the green fluorescence intensity, the top ranked S76Y being the most stable. The single mutation R39G may slightly increase the stability; however, the single W3C mutant have decreased stability and the triple mutant (W3C + R39G + S76Y) have unchanged stability. The most stable variant is the double mutant R39G + S76Y, where the midpoint of thermal denaturation has increased with 13 °C compared to parent scMN. The double mutant W3C + S76Y also experience higher stability than parent scMN. Due to the homogeneity caused by the presence/lack of the N-terminal Met in single-chain monellin the reported values of ΔH°T m , T m and ΔG° are averages for protein with different termini. Sweetness detection. The sweetness of parent scMN, R39G, S76Y and R39G + S76Y were evaluated by six volunteers. Monellin concentrations ranging from 19 to 5607 nM were tested. Sweetness of sucrose was detected by all testers. One tester judged MQ water (negative control) as sweet and was therefore discarded from the study. Parent scMN and S76Y were recorded as sweet by all testers (Table S1). The average sweetness threshold for parent scMN and S76Y were 317 nM (219-492 nM) and 263 nM (219-328 nM), respectively, which is comparable to previously published data for wild type single chain monellin, natural monellin and wild type synthetic monellin 31,37,41,42 . In accordance to earlier studies, monellin was approximately 100 000 times sweeter than sucrose on a molar basis. Two test people did not detect any sweetness for R39G + S76Y and three judged the samples with the second highest or the highest concentration as sweet. Sweetness of R39G was sensed at high concentrations (1108-5607 nM) for all testers. The sweetness of R39G was confirmed by one volunteer at a concentration of 12 µM.

Discussion
Proteins are marginally stable, the free energy of the folded state of a protein lying only 20-60 kJ mol −1 below that of the unfolded state. This may reduce the usefulness of proteins for industrial purposes, e.g. as pharmaceuticals or food. Hence it is of great interest to stabilize proteins. Since many proteins in nature are optimised for function rather than stability, there is often room for increased stability; however, it is critical that the stabilizing mutations do not impede the function of the protein. Protein stability is defined as the free energy difference between the  folded and unfolded state. Mutations increasing the energy of the unfolded state more than the folded state, as well as mutations decreasing the free energy of the folded state more than the unfolded state, are thus stabilizing.
To design more stable mutants remains hard despite that key factors influencing protein stability, such as packing of the hydrophobic core, hydrogen bond formation, electrostatic interactions, conformational entropy and bond strain are rather well understood [43][44][45] . Here we utilize the thermodynamic linkage between fragment complementation and protein stability to find more stable variants of monellin. Mutations affecting the affinity between fragments also affect the stability of the protein. The correlation between fluorescence intensity of split GFP and fragment affinity of the attached protein fragments thus enables in vivo selection of proteins with enhanced stability. Improved stability of monellin may allow heating to higher temperatures which aids industrial usage of the protein. Monellin was chosen for this study since it naturally consists of two chains that reconstitute the protein spontaneously. The complementation rate is slow with an association rate constant, k on of 8.8 × 10 3 M −1 s −1 , and a dissociation rate constant, k off of 3.1 × 10 −4 s −1 , resulting in an equilibrium dissociation constant (K D ) of 35 nM 46 . Fragment complementation of NGFP-MNA and MNB-CGFP results in moderately fluorescent colonies which allows for detection of colonies with increased fluorescence intensities. Detection of colonies with increased fluorescence intensity was performed by human eye. The human eye can capture differences in fluorescence intensity between colonies better than a fluorescence plate reader. However, this is a time-consuming step which limits the number of screened colonies. To further improve the efficiency of this method, fluorescence-activated cell sorting (FACS) might be used as a ranking method. According to the in vivo fluorescence ranking the most stabilizing mutation was S76Y, which was confirmed by temperature denaturation studies. S76Y retained its functionality and was considered as sweet as parent monellin. The increased stability is likely due to the additional van der Waals interactions formed by the tyrosine and its surrounding amino acids. Tyrosine residues are more often buried within proteins than serine residues and generally have more favourable van der Waals interactions of the aromatic ring compared to serine, whereas both can participate in hydrogen bonds via the -OH group 47 . S76 is situated in the middle of the β4 strand and tyrosine has higher β-sheet forming propensity than serine which may also increase the stability of the mutant 48 . The serine residue is placed in a pocket on the protein surface and tyrosine might fill the pocket better. Tyrosine in this position may thus stabilise the folded state relative to the unfolded state in which residue 76 is likely fully exposed to solvent. The double mutations W3C + R39G were the most stabilizing mutations found in the B-chain of monellin according to the in vivo fluorescence ranking. Increased stability of the mutant was verified by temperature denaturation where the W3C + R39G mutant had an increased T m by approximately 8 °C compared to parent scMN. The mutations were also examined separately and the single mutations did not improve stability to any great extent. The single W3C mutation reduced the stability and the R39G mutation had minor, if any effect on the stability. The positively charged arginine 39 is exposed on the surface of the protein and mutation to neutral glycine may be stabilizing by decreasing local electrostatic repulsion. In five out of the top thirty variants with increased fluorescence intensities R39 was mutated, implying that R39 has negative effect on the stability. Introduction of W3C mutation removes a large hydrophobic residue from the surface of the protein which may influence the stability. Addition of reducing agents during temperature denaturation studies minimize the risk of dimerization (of mutants containing W3C) affecting the stability. The likelihood of dimerization in the in vivo assay is low due to the linkage of the monellin fragments to the bulky GFP molecule, though it cannot be ruled out. Mutational effects on protein stability can be additive or non-additive. The stability of monellin with all combinations of mutations were investigated by temperature denaturation which showed that the W3C + R39G and W3C + R39G + S76Y mutations have non-additive effects on the stability while the effects of W3C + S76Y and R39G + S76Y were additive (Fig. 6). S76Y is situated on the opposite side of the protein compared to W3C and R39G, (Fig. 3), resulting in the additive effects of S76Y together with either W3C or R39G. Non-additivity may occur when the mutated residues are in contact with each other. The residues can be in direct contact or through electrostatic interactions. Structural rearrangements propagated by intervening residues can also cause non-additivity effects. By using unbiased libraries we were able to find mutations with non-additive effects that would not be present in a rationally designed library.
As expected, the mutation of arginine 39 resulted in a dramatic decrease of sweetness. The importance of a charged surface to retain sweetness of monellin has been shown repeatedly 31 . Sweet assessment studies have shown that the substitution of arginine 39 with glutamic acid or aspartic acid produced mutants with detection thresholds at 13 000 nM and 89 000 nM, respectively 49 . According to the "wedge model" R39 together with D7, R88 and R65 are crucial to retain sweetness 50 . Our findings correlates well with the wedge model in which arginine 39 interacts with the sweet taste receptor while serine 76 is situated far away from the interaction site 32 . A meta-analysis of the sweetness of published mutants of monellin by Xue et al. 31 showed a correlation between reduced net positive charge and reduced sweetness. Variants with added negative charge always showed reduced sweetness (R39D, R39E, K43E, R72E, R88E, F34D, M42E, Y63E, Y65E). Replacement of a negative residue with a neutral residue showed no change in sweetness detection levels (E2N, MNA D16N (in two-chain monellin), MNA D22N, MNA E25Q, MNA D26N, E59C) except for D7N that experience reduced sweetness. No variation of sweetness was detected when a positively charged residue was replaced by another positively charged residue (R39K) neither did the C41S mutation effect the sweetness of monellin. Other mutations that decrease the sweetness of monellin were D7E (charge-neutral), M42R and D68R (addition of positive charge). Two mutants with increased positive charged (Y63R, Y65R) showed no variation in sweetness. Most mutational studies have aimed to shed light on the interaction of monellin and the sweet taste receptor though some also intend to increase the stability of the protein. Mutation E23Q increased the sweet taste of monellin 50 . A variant of the E23Q mutant with increased stability was constructed by Leone et al. 50 , by rational design, which contained Q28K, C41S and Y65R mutations. Removal of the buried ionizable glutamic acid 23 is expected to improve protein stability and a study showed that by mutating it into alanine the stability became pH independent between pH 4 and 8. An additional mutation, C41A, resulted in the stability being pH independent from pH 4 to 10 51 . Arginine 39 has been extensively mutated and although focus has been on its effect on sweet taste, one study showed that the double mutant D7E + R39K had decreased stability towards temperature denaturation 49 .
The present study corroborates the usefulness of an in vivo protein stabilization method based on fragment complementation and a split GFP system for protein stabilization and adds a functional assay to the work-flow. Several novel mutations with increased stability were found, one of which have retained sweetness. However, the top hit for the B-chain of monellin, W3C + R39G, is not suitable as a sweetener. The removal of the positively charged arginine results in decreased sweetness and for industrial purposes cysteine residues are not applicable due to the need of reducing agents to maintain monomeric protein. The functional mutant S76Y, which was the top hit among the A-chain substitutions of monellin in the in vivo screen, has similar structure to parent scMN and has potential use as a low-calorie sweetener. Improving the stability of monellin increases its utility in industrial processes where heating is a key procedure. Stabilized variants of monellin can hopefully meet the demand of a non-carbohydrate sweetener for the food industry in the future.

Materials and Methods
Construction of a tandem NGFP-MNA-CGFP-MNB pQLinkN plasmid. The original pQLinkN plasmid was a kind gift from Konrad Buessow (Addgene plasmid # 13670) 52 . The cloning procedure is described in Supplementary Information.

Construction of a library of mutant MNA and MNB.
Two libraries of mutant MNA and MNB, respectively, were produced by error prone PCR using GeneMorph II Random mutagenesis kit (Agilent Technologies). By varying the template amount in the PCR reaction, the mutation frequency can be controlled. Mutation levels of 1-2 mutations/gene were desired and the primers used were epNGFPpQLinkNfor, epNGFPpQLinkNrev, epCG-FPpQLinkNfor and epCGFPpQLinkNrev, with sequences as listed in Supplementary Information. To remove parent MNA from the NGFP-MNA-CGFP-MNB pQLinkN plasmid and to enable ligation of mutated MNA, the plasmid and the PCR product were separately digested at 37 °C with KpnI (Thermo Scientific) over night, purified with GeneJET PCR purification kit (Thermo Scientific) and digested with SacI (Thermo Scientific) at 37 °C for 1 h. For the last 15 min of the digestion, FastAP Thermosensitive alkaline phosphatase (Thermo Scientific) was added. To enable insertion of mutant MNB, the PCR product and the plasmid were separately digested at 37 °C with NdeI and PstI overnight and FastAP Thermosensitive alkaline phosphatase was added for the last 15 min. The digestion products were purified using GeneJET PCR purification kit (Thermo Scientific). The mutated MNA och MNB fragments were ligated into the appropriately digested plasmid by using T4 DNA ligase (NEB) for 16 h at 16 °C. digital CCD camera (Sony Cybershot DSC-W5). The fluorescence intensity was judged by human eye. Colonies with higher fluorescence intensity than parent MN were picked and cultivated overnight in 2 mL LB with 100 µg/ mL ampicillin, followed by plasmid preparation using Illustra plasmid preparation kit (GE Healthcare). Forty colonies with increased fluorescence compared to the parent were selected. Finally, the plasmids prepared from the selected colonies were retransformed into E. coli ER2566 and plated on individual inducing plates and the colonies were ranked independently by 6 human volunteers based on fluorescence intensity. The 30 top ranked mutants were sequenced using BigDye Terminator v1.1 Cycle sequencing Kit (Applied Biosystems). The primer 5′-C AAC ATT CTG GGA CAC AAA TTG G-3′ was used to sequence MNA and 5′-TGC TAG TTG AAC GCT TCC ATC TTC-3′ was used to sequence MNB (purchased by BM Unit). The top ranked MNA mutant was co-expressed with the MNB library to find mutants with even higher stability.

Ranking
Expression and purification of single chain monellin mutants. Pet3a plasmids containing genes for single chain monellin with single (R39G, W3C, S76Y), double (R39G + S76Y, W3C + S76Y, W3C + R39G) or triple (W3C + R39G + S76Y) mutations were purchased from Genscript (Piscataway, New Jersey). Plasmids containing parent scMN and mutants were transformed into Ca 2+ competent BL21(DE3) Ply Star E. coli cells and cultivated on LB agar plates containing 50 µg/ml ampicillin and 30 µg/ml chloramphenicol. Single colonies of each mutant were used to inoculated 5 ml LB containing 50 µg/ml ampicillin and 30 µg/ml chloramphenicol at 37 °C overnight. The overnight cultures were used to inoculate 500 ml LB containing 50 µg/ml ampicillin and 30 µg/ml chloramphenicol. Protein expression was induced by addition of 0.4 mM IPTG at OD600 = 0.6. The cultures were harvested after 3-4 hours by centrifugation and the pellet was resuspended in 10 ml of water or 2-(N-morpholino) ethanesulfonic acid (MES) buffer pH 6 (with addition of 1 mM DTT for plasmids containing the W3C mutation) and frozen at −20 °C overnight. The resuspension was thawed on ice, diluted with approximately 30 ml of MES buffer pH 6 (with addition of 1 mM DTT for plasmids containing the W3C mutation) and sonicated for 5 min, 30 s on/30 s off, followed by centrifugation at 16 000 rpm for 20 min. The supernatants were pooled, DNAse was added and the supernatant was left on ice for 15 min, then centrifuged at 16 000 rpm for 20 min. The supernatant was concentrated on a Vivaspin 20 filter, MWCO 3000 (Sartorius). The protein was added to a 5 ml HiTrap CM FF column (GE Healthcare) and washed with 10 mM MES buffer pH 6. The sample was eluted using a 0-0.25 M linear NaCl gradient. Fractions containing monellin mutants were pooled, concentrated and separated on a Superdex 75 (10/300) gel filtration column (GE Healthcare). The running buffer was 5 mM sodium phosphate, pH 5.5, with addition of 1 mM DTT for proteins containing the W3C mutation. The residual salt was removed by gel filtration using a Nap10 column (GE Healthcare) in 5 mM sodium phosphate buffer, pH 5.5, with addition of 1 mM DTT for proteins containing the W3C mutation. The purity of the expressed proteins was analyzed by SDS-PAGE (15% gel) with 1x running buffer (25 mM Tris, 192 mM glycine, 0.5% SDS pH 8.3). PageRuler™ Prestained protein ladder, 10-180 kDa, (Thermo Scientific) was used as protein standard.
Stability measurements using CD spectroscopy. CD spectra were recorded using a JASCO J-815 CD spectrometer with a JASCO PTC-423S/15 Peltier type thermostated cell holder. The thermal denaturation of parent scMN and mutants was monitored by recording the CD signal at 215 nm, in a 2 mm quartz cuvette, from 4 °C to 96 °C (scan rate 1 °C min −1 , signal recorded every 1 °C, response time 16 s). The protein concentrations were 6 µM in 5 mM sodium phosphate buffer, pH 5.5, with 1 mM DTT added to samples containing proteins with the W3C mutations. Reversibility was monitored by recording a thermal scan from 96 °C to 4 °C directly after the forward scan. The secondary structures of the proteins were evaluated by far-UV CD spectroscopy before and after each thermal scan. Far-UV CD spectra were recorded between 190 and 250 nm, 50 nm/min, response 8 s, band with 1 nm, accumulations 3. A baseline recorded for only buffer was subtracted from each spectrum. Buffers were filtered through 0.2 µm filter and samples were centrifuged for 10 min at 13 000 rpm before measurements to remove possible aggregates.
The stability of parent scMN and mutants were modeled assuming a two-state reversible unfolding process. The temperature at the midpoint of unfolding (T m ) and the enthalpy of unfolding (ΔH°T m ) were obtained by fitting equation (1) to CD data, (1) H  T  T  C T T T  T  T  RT   H  T  T  C T T T  T  T  where ɛ obs is the observed ellipticity at 215 nm, k N , b N , k D and b D define the baselines before and after the transition, T is the temperature in Kelvin, ΔH°T m is the enthalpy of unfolding, R is the gas constant and ΔC°p is the difference in heat capacity between the denatured and native states. ΔC p was assumed to be constant, 5858 J mol −1 , which is reasonable for a protein of this size 53 .
Since the final plateaus were not reached for mutants S76Y, R39G + S76Y and W3C + R39G during temperature denaturation studies, the final plateau, X FP , was estimated by equation (2) where Y IP is the initial plateau for S76Y, R39G + S76Y or W3C + R39G before denaturation, parent MN FP is the final plateau for parent scMN and MN IP is the initial plateau for parent scMN. The standard free energy of unfolding, ΔG°, was obtained by extrapolation from T m to 25 °C by equation (3). Sweet taste assessment. The sweetness of monellin mutants was assessed by six volunteers, three males and three females, ranging from 29 to 63 years of age with ethical approval via Lund University. The study was conducted in accordance with the ethics regulations of Lund University and informed consent was obtained from all participants. The study was one-way blinded (for the test persons) and the person preparing the samples was not present during the taste assay. The samples included parent scMN, R39G, S76Y, R39G + S76Y, MilliQ water (negative control) and sucrose (positive control). Samples were dissolved in MilliQ water. Mutants containing W3C were excluded due to requirement of 1 mM DTT in the samples. Dilution series of samples with concentrations 19, 29, 43, 65, 97, 146, 219, 328, 492, 738, 1108, 2492 and 5607 nM were tested starting from the lowest concentration. 1 mL of each sample was taken into the mouth and kept in the mouth for approximately 10 s then spit out. Before each new sample set testers rinsed their mouth with tap water until no sweet taste remained. A sample was recorded as sweet when the test person judged two consecutive samples as sweet. The detection threshold is reported as the lowest concentration where sweetness was detected. All datasets generated and analysed during the current study are available from the corresponding author on reasonable request.