Heating-mediated purification of active FGF21 and structure-based design of its variant with enhanced potency

Fibroblast growth factor 21 (FGF21) has pharmaceutical potential against obesity-related metabolic disorders, including non-alcoholic fatty liver disease. Since thermal stability is a desirable factor for therapeutic proteins, we investigated the thermal behavior of human FGF21. FGF21 remained soluble after heating; thus, we examined its temperature-induced structural changes using circular dichroism (CD). FGF21 showed inter-convertible temperature-specific CD spectra. The CD spectrum at 100 °C returned to that at 20 °C when the heated FGF21 solution was cooled. Through loop swapping, the connecting loop between β10 and β12 in FGF21 was revealed to be associated with the unique thermal behavior of FGF21. According to surface plasmon resonance (SPR) experiments, in vitro cell-based assays, and model high-fat diet (HFD)-induced obesity studies, heated FGF21 maintained biological activities that were comparable to those of non-heated and commercial FGF21s. Based on sequence comparison and structural analysis, five point-mutations were introduced into FGF21. Compared with the wild type, the heated FGF21 variant displayed improved therapeutic potential in terms of body weight loss, the levels of hepatic triglycerides and lipids, and the degree of vacuolization of liver in HFD-fed mice.

FGFs are divided into FGF1, FGF4, FGF7, FGF8, FGF9, FGF11, and FGF19 subfamilies depending on their sequence similarities. Except for the intracellular FGF11 subfamily, FGFs form complexes with FGF receptors (FGFRs) on the cell membrane, which are generally composed of three immunoglobulin (Ig)-like extracellular domains, a transmembrane domain, and a cytoplasmic tyrosine kinase domain 2,3,5 . FGFs bind to the interface between the second and third Ig-like ectodomains to induce receptor dimerization, leading to the phosphorylation of tyrosine residues in the cytoplasmic domain to stimulate signaling pathways 1,2,6 .
The canonical FGFs (FGF1, FGF4, FGF7, FGF8, and FGF9 subfamilies) have a positively charged sector clustered by lysine and arginine residues and thus display avidity for negatively charged heparin/heparan sulfate

Results
Temperature-responsive structural reversibility of FGF21. To investigate how susceptible FGF21 is to thermal denaturation, FGF21-containing solutions were incubated at high temperatures for 10 min. When proteins are denatured, they form coarse aggregate particles that scatter visible light to make solutions turbid. There was no change in turbidity of FGF21 solutions stored at 20 °C after incubation at 40, 60, 80, and 100 °C, when observed by naked eyes, indicating the absence of large protein aggregates ( Supplementary Fig. S1). In contrast, FGF2 solutions irreversibly became turbid after heating; turbidity was maintained regardless of storing temperatures ( Supplementary Fig. S1). To confirm that FGF21 is resistant to thermal denaturation, the heattreated FGF21 solutions were centrifuged to remove protein aggregates, and then supernatants were analyzed using SDS-PAGE. As shown in Fig. 1A, the amount of FGF21 in the solutions remained at almost 30 mg/mL before and after heat treatment, indicating that no protein aggregates were formed due to thermal denaturation. However, FGF2 was dramatically reduced in supernatants after incubation at 40 °C or more (Fig. 1A).
Thermal denaturation curve is a good indicator for assessing the thermal stability of proteins. To assess the thermal denaturation of FGF21 using CD spectroscopy, spectral changes were monitored at 215 nm at a temperature range of 20-100 °C 32,33 . The FGF21 spectra displayed gradual changes as a function of temperature, which contrasted with the steep transition slope of the FGF2 spectra ( Fig. 1B,C). The magnitude of enthalpy and entropy changes was smaller in FGF21 than FGF2 (Fig. 1B,C), and the T m values were estimated to be 62.26 ± 0.6306 °C and 52.69 ± 0.1991 °C for FGF21 and FGF2, respectively.
For further characterization of the temperature responsiveness of FGF21, we examined the secondary and tertiary structural states of FGF21 as the temperature increased from 20 to 100 °C by measuring far-UV (190-260 nm) and near-UV (240-340 nm) CD spectra, respectively 34 . In both CD spectra, temperature-induced spectral changes are sequential; for example, the spectrum at 60 °C is located between those at 40 °C and 80 °C ( Fig. 2A,B). The spectral feature of FGF21 accompanied by temperature change encouraged us to measure the CD spectra while cooling the heated FGF21 solutions. The CD spectra of FGF21 at 20 °C were virtually identical to that of FGF21 that was cooled to 20 °C after heating at 100 °C (Fig. 2C,D), indicating that structural changes of FGF21 induced by heating can be restored by cooling (Fig. 2E). Consequently, FGF21 is highly likely to have temperature-responsive structural reversibility.
The unique loop region associated with the heat response of FGF21. The N-and C-terminal loops extending out from the β-trefoil core structure of FGF21 (core-FGF21) show large sequential disparities when compared with the corresponding regions of other FGFs ( Supplementary Fig. S2). Therefore, we constructed a deletion mutant without the N-terminal loop (∆29-40) and C-terminal loop (∆174-209) to investigate whether both distinct terminal loops are related to the remarkable response of FGF21 to heating. However, similar to the wild type FGF21, the deletion mutant remained soluble at a concentration of ~ 30 mg/mL after high-temperature incubations (Fig. 1A), indicating that core-FGF21 is responsible for the thermal behavior of FGF21.
FGFs share the β-trefoil core structure with disparities in loop regions connecting β-strands 3,4 . Among connecting loops, the β10-β12 loop (residues His145-Pro161) of FGF21 is outstanding in that it has a unique amino acid composition. First, the sequence identities of the β10-β12 loop in FGF21 to corresponding loops in other FGFs lie in the range of 5-25%, whereas the entire FGF21 shows 14-36% sequence identities to other FGFs. Second, the β10-β12 loop of FGF21 contains 5 proline residues, constituting 29.4% of the loop, which contrasts     Fig. S2). Proline residues in loop regions have been reported to contribute to the thermal stability of proteins 35,36 . Based on these analyses, we tested the contribution of the unique β10-β12 loop to the thermal response of FGF21 though loop swapping. Sequence identities among the β10-β12 loops of FGFs were low ( Supplementary  Fig. S2), which led us to hypothesize that rather than individual residues, the overall feature of the β10-β12 loop is associated with the unique thermal behavior of FGF21. This was the reason why we took the loop swapping strategy instead of point mutations. Our aim was to design a mutant that is sensitive to heat treatment but maintains structural and functional properties of the wild type FGF21. When the β10-β12 loop of FGF21 was replaced by the corresponding region of FGF2, we failed to get soluble form of the mutant, which indicates defects in folding or structural integrity. As an alternative, we searched for another FGF protein with the β10-β12 loop that is similar to that of FGF21. The β10-β12 loop of FGF23 shows a relatively high sequence identity to that of FGF21 and furthermore, the FGF23 loop adopts a random coil conformation like the β10-β12 loop of FGF21 (Supplementary Fig. S2). In terms of overall feature, the β10-β12 loop of FGF23 seemed to be suitable for loop swapping with minimal structural and functional effect (Fig. 3). Therefore, the β10-β12 loop of core-FGF21 was replaced by the corresponding loop (residues 137-154) of FGF23 to make a loop-swapped mutant (core-FGF21-LW).
As shown in Fig. 1A, the amount of core-FGF21 in solutions was nearly identical, regardless of heat treatment. However, core-FGF21-LW in supernatants was reduced by more than 85% after 10 min incubations at 80 °C and 100 °C (Fig. 1A, Supplementary Fig. S1). The formation of protein aggregates was monitored by measuring optical density since aggregates scatter visible light strongly due to their large sizes 37 . A solution containing core-FGF21-LW displayed the initial optical density at 600 nm (OD 600 ) of 0.005 when incubated at 75 °C. However, over the duration of the treatment, the OD 600 value of the solution reached 1.086, indicating that core-FGF21-LW formed aggregates. In contrast, the OD 600 value of a solution containing core-FGF21 remained constant, which led us to assume that the β10-β12 loop is related to the heat response of FGF21 ( Supplementary Fig. S3). Similar to the wild type, the loop-swapped FGF21 mutant with the N-and C-terminal loops (FGF21-LW) activated FGF receptor 1c (FGFR1c) (Fig. 3A), and its CD spectra in the far-UV region (190-260 nm) was similar to that of the wild type (Fig. 3B). Taken together, the loop replacement disturbed only the thermal behavior without affecting the structural and functional properties of FGF21.
Heat treatment purification of active human FGF21. E. coli is a mesophilic organism frequently used to produce recombinant proteins. Since most E. coli proteins are heat-labile, incubation at high temperatures is an efficient strategy for the purification of thermostable proteins produced in E. coli. As described in the "Materials and methods" section, heat treatment at 100 °C was implemented to purify FGF21. In our construct design, FGF21 was expressed with the N-terminal Trx tag. Since the tag is heat stable, the Trx-FGF21 fusion protein was incubated at 100 °C for 10 min 38 , which separated the Trx-FGF21 protein from heat-labile E. coli proteins that precipitated at high temperatures 39 (Supplementary Fig. S4). Subsequently, the Trx tag was removed by the cleavage of the linker between the tag and FGF21 to secure only FGF21 40 . According to DLS analysis, the size of FGF21 purified through heat treatment (htFGF21) was highly similar to that of FGF21 purified without heating (nFGF21) (Supplementary Fig. S5). This strongly suggests that heat treatment has little influence on the quality of FGF21 41 .
Our CD and DLS analyses revealed that heat treatment had little effect on the structure of FGF21 (Fig. 2, Supplementary Fig. S5). Therefore, we investigated whether heat treatment influences the activity of FGF21 by comparing FGFR1c/β-Klotho activation levels of nFGF21, htFGF21, and commercial FGF21 (cFGF21) using FGF21-responsive reporter cells (modified HEK293 cells) that allow for monitoring FGFR1c/β-Klotho activation. www.nature.com/scientificreports/ Both nFGF21 and htFGF21 displayed little difference in FGFR1c/β-Klotho activation levels, which were comparable to the activity of cFGF21 (Fig. 4A). Since adipocytes are one of the major targets of FGF21, we also examined the activities of the three FGF21s by detecting the phosphorylation of FGFR substrate 2 (FRS2) and extracellular signal-regulated kinase (ERK) in 3T3L1 adipocytes. The treatment of nFGF21, htFGF21, and cFGF21 led to similar phosphorylation levels of FRS2 and ERK (Fig. 4B,C). Furthermore, the levels of glucose uptake by FGF21s were similar in 3T3L1 adipocytes (Fig. 4D). These results indicate that FGF21 remains active after heating. Furthermore, we determined the rate constants for association (k a ) and for dissociation (k d ) of the wild type FGF21s for β-Klotho in ligand binding assay by SPR. The results showed that both nFGF21 and htFGF21 bind to β-Klotho with similar equilibrium dissociation constants (K D ) of 12.8 nM and 12.9 nM, respectively ( Table 1, Supplementary Fig. S10). The band density of nFGF21 treated cells was set to 1, and the other values were calculated relative to this value (n = 3, average ± SEM). (D) 3T3L1 adipocytes were treated with FGF21s at the indicated concentration for 6 h, followed by detection of glucose uptake levels. The glucose uptake values of non-treated cells were set to 1, and the other values were calculated relative to this value (n = 3, average ± SEM). The significance was evaluated using the one-way ANOVA, followed by Tukey's post hoc test. n.s. not significant. www.nature.com/scientificreports/ The administration of human or murine FGF21 alleviates the progression of NAFLD in HFD-induced obesity models [42][43][44][45] . Considering that nFGF21 has traditionally been used in those previous studies, we sought to determine whether htFGF21 was also effective in ameliorating fatty liver. To investigate the effects of htFGF21 on NAFLD, HFD-fed mice were intraperitoneally injected with 1 or 10 mg/kg htFGF21, once daily for 14 days. The injections of either 1 or 10 mg/kg htFGF21 markedly reduced body weight in HFD-fed mice compared with vehicle controls (Fig. 5A,B). Significant increases in intrahepatic TG contents were observed in vehicletreated HFD-fed mice compared with vehicle-treated NCD-fed mice; however, the HFD-induced increases in intrahepatic TG dramatically diminished after treatment with 1 or 10 mg/kg htFGF21 (Fig. 5C). Histological analyses of liver tissue stained with hematoxylin and eosin (H&E) showed that there was extensive hepatocyte vacuolation in vehicle-treated HFD-fed mice, suggesting intrahepatic fat accumulation (Fig. 5D). Conversely, few hepatocellular vacuolations were observed in liver sections of either 1 or 10 mg/kg htFGF21-treated HFD-fed mice, which was comparable with that of vehicle-treated NCD-fed mice (Fig. 5D). Moreover, Oil red O staining of liver sections showed that htFGF21 injections dramatically attenuated the formation of intense lipid droplets in HFD-fed mice (Fig. 5D). These results suggest that htFGF21 treatment can prevent HFD-induced NAFLD in mice. Collectively, cell-based functional assays and in vivo mouse model studies strongly suggest that heating allows for an easy and effective purification of active FGF21. Data are represented as mean ± SEM, calculated using one-way ANOVA followed by Tukey's multiple comparisons test, *p < 0.05, **p < 0.01, *** and † † † p < 0.001, **** and † † † † p < 0.0001. n = 5-6 mice/group. † † † and † † † † indicate vehicle-injected NCD-fed mice versus vehicle-injected HFD-fed mice. *, **, ***, and **** indicate vehicle-injected HFD-fed mice versus htFGF21-injected HFD-fed mice. www.nature.com/scientificreports/ Structure-based design of a FGF21 mutant with enhanced potency. According to the crystal structure of the C-terminal loop of human FGF21 complexed with β-Klotho 11,12 , residues at positions 201 and 208 are directly involved in interactions with β-Klotho; the side chains of Gln201 and Ala208 fit into a small pocket and canyon, respectively, on the surface of β-Klotho ( Supplementary Fig. S7). Sequence comparison shows that Gln201 and Ala208 of human FGF21 are replaced by His201 and Thr208, respectively, in some mammalian FGF21s (Supplementary Fig. S8). It is notable that regions of β-Klotho interacting with the residues at positions 201 and 208 of FGF21 (Gln201/His201 and Ala208/Thr208) were highly conserved in mammalian β-Klotho ( Supplementary Fig. S9). To assess the effects of the two amino acids replacements, we introduced Q201H and A208T mutations into the C-terminal loop of FGF21. The htFGF21 variant with Q201H and A208T mutations (htM2) displayed 2-fold increased potency for FGFR1c/β-Klotho activation than the wild type htFGF21 (Fig. 6A).
In terms of the structural aspect, the C-terminal loop protrudes from the main body of FGF21 and thus is vulnerable to proteolytic degradation. Fibroblast activation protein (FAP) in serum inactivates FGF21 by cleaving the peptide bond between Pro199 and Ser200 in the C-terminal loop, and the P199G mutation is effective to prevent the FAP-mediated cleavage of the loop 31 . With a rationale that the enhanced flexibility of the C-terminal loop can have a positive effect on β-Klotho binding 46 , we introduced two more glycines to the C-terminal loop (S195G and S200G) in addition to the P199G mutation. In the crystal structure of the C-terminal loop complexed with β-Klotho 11 , the side chains of both serine residues are exposed to the solvent without interacting with β-Klotho ( Supplementary Fig. S7). Therefore, the Ser → Gly mutations are likely to have no effect on β-Klotho interactions but could increase the loop flexibility. The htFGF21 variant with S195G, P199G, and S200G mutations (htM3), which has three glycine residues in the C-terminal loop, showed 3.65-fold higher potency for FGFR1c/β-Klotho activation than the wild type htFGF21 (Fig. 6A).
Encouraged by the two mutational studies that produced mutant FGF21s with higher FGFR1c/β-Klothoactivating functions, we combined the five point-mutations in the C-terminal loop. The htFGF21 variant with S195G, P199G, S200G, Q201H, and A208T mutations (htM5) showed the highest potency for FGFR1c activation; Figure 6. Cell-based activity tests of htM2, htM3, and htM5. (A) Modified HEK293 cells used for reporting FGFR1c/β-Klotho activity was treated with the wild type htFGF21 (htWT), htM2, htM3, and htM5 at the indicated concentration for 6 h. Cell lysates were prepared for detection of luciferase activity. The levels of FGFR1c/β-Klotho activity in non-treated cells were set to 1, and the other values were calculated relative to this value (n = 4, average ± SEM). Non-linear regression was performed using GraphPad Prism 8, and the best-fit non-linear regression curve is depicted. EC 50 s of htWT, htM2, htM3, and htM5 were calculated as 2.74 ± 0.24, 1.37 ± 0.06, 0.75 ± 0.17, and 0.52 ± 0.01 ng/mL (average ± SEM), respectively. The statistical differences at each concentration were evaluated using one-way ANOVA followed by Tukey's post hoc test and are represented as a, b, c, and d. p < 0.05. (B) 3T3L1 adipocytes were treated with htWT, htM2, htM3, and htM5 30 ng/mL for 15 min. Protein samples were prepared for western blotting. The original blots are presented in Supplementary Fig. S15. (C) The band density of htWT treated cells was set to 1, and the other values were calculated relative to this value (n = 3, average ± SEM). Statistical significance was evaluated using one-way ANOVA followed by Tukey's post hoc test and was marked as a, b, c, d, and e. p < 0.05. www.nature.com/scientificreports/ this htM5 mutant showed 5.27-fold higher potency than the wild type htFGF21 (Fig. 6A). The activities of htM2, htM3, and htM5 were also confirmed in 3T3L1 adipocytes. Consistent with FGFR1c/β-Klotho activation in Fig. 6A, the three mutants increased the levels of phosphorylated FRS2 and ERK more than the wild type htFGF21 (Fig. 6B,C). We performed SPR assays to examine the influence of the five point-mutations in the C-terminal loop on β-Klotho binding. The K D value of htM2 (10.9 nM) was slightly higher than that of the wild type nFGF21 (12.8 nM) and htFGF21 (12.9 nM) ( Table 1, Supplementary Fig. S10). In the case of htM3 and htM5, they exhibited 2.47-and 2.96-fold higher affinities to β-Klotho than wild type FGF21s; their K D values were 5.19 nM and 4.32 nM, respectively (Table 1, Supplementary Fig. S10). Compared to wild type FGF21s, both mutants displayed similar k a and lower k d , indicating that slow dissociation is the cause of their tight binding to β-Klotho (Table 1, Supplementary Fig. S10).

Scientific
To compare effects on NAFLD between the wild type and htM5, HFD-fed mice were intraperitoneally injected with 0.1 or 0.5 mg/kg of htFGF21 or htM5, once daily for 14 days. Although the injections of either the 0.1 mg/ kg htFGF21 or htM5 decreased body weight compared with vehicle-treated mice, the trend for body weight loss was greater in the htM5 group than in the htFGF21 group (Fig. 7A,B). Mice injected with either the 0.5 mg/kg htFGF21 or htM5 treatment showed significant body weight loss compared with vehicle-treated mice, and there was no difference in the extent of weight loss between the htFGF21 and htM5 groups (Fig. 7A,B). Either the htFGF21 or htM5 injections substantially reduced hepatic TG content in a dose-dependent manner but htM5 significantly decreased hepatic TG content more than htFGF21 at 0.1 mg/kg (Fig. 7C).
Although the vacuolization was still observed in the both livers of mice injected with 0.1 mg/kg of the htFGF21 or htM5, the degree of vacuolization was reduced compared with vehicle-injected mice and the effect of htM5 was greater than that of htFGF21 (Fig. 7D). Similarly, Oil red O staining revealed that either the htFGF21 or htM5 administration suppressed HFD-induced hepatic lipid accumulation. htM5 inhibited lipid accumulation www.nature.com/scientificreports/ at a higher degree than htFGF21 at the dose of 0.1 mg/kg (Fig. 7D). Taken together, these results suggest that the introduction of five point-mutations gave rise to the htM5 mutant that is likely to exert better effects for ameliorating NAFLD than the wild type.

Discussion
Characterization of protein stability is an integral step in biopharmaceutical development 47 and the monitoring of thermal behavior is a classical probe to investigate protein stability 48 . To examine temperature-dependent structural changes of FGF21, we performed CD experiments. The CD spectra of FGF21 can be roughly grouped into three distinct states depending on temperatures; CD spectra between 20 and 40 °C, at 60 °C, and between 80 and 100 °C ( Fig. 2A,C). This observation is compatible with the previous report about three thermal unfolding states of FGF21 detected by differential scanning calorimetry 49 . Remarkably, we discovered the temperatureresponsive structural reversibility of FGF21 (Fig. 2E). The β10-β12 loop was identified to be associated with this remarkable thermal behavior; the loop replacement only perturbed thermal response of FGF21 without affecting structural and functional properties (Fig. 3, Supplementary Fig. S3). These ever-first findings led us to employ heating as an efficient purification step to separate recombinant FGF21 from heat-labile proteins (Supplementary Figs. S4, S6). It should be noted that heat treatment has no influence on the structure and function of FGF21 (Fig. 4, Supplementary Fig. S5). Heating has another advantage to prevent the proteolytic cleavage of the essential but fragile C-terminal loop of FGF21 during purification, since heat-labile host proteases lose their activities due to thermal denaturation. On SDS-PAGE, FGF21 purified with heating displayed no degradation whereas there was degradation pattern in the case of FGF21 purified without heating (Supplementary Fig. S6). Consequently, heating-based strategy can be widely applied for the purification of FGF21 analogs that are under therapeutic development. FGF21 is targeted to cells and tissues harboring the tissue-specific transmembrane co-receptor β-Klotho and then forms the ternary signaling complex with β-Klotho and receptors 15,16 . The C-terminal loop (residues 186-209) of FGF21 is responsible for interacting with β-Klotho. We introduced five point-mutations into the C-terminal loop to generate the M5 variant with high affinity to β-Klotho (Table 1) and enhanced potency in activation of FGF21 signaling (Figs. 6, 7). Consistently, this mutant displayed therapeutic potential in terms of body weight loss, the levels of hepatic triglycerides and lipids, and the degree of vacuolization of liver in highfat diet-induced obesity mice (Fig. 7). Among the five mutations, S195G, S200G, Q201H, and A208T are firstly reported in this study whereas P199G is a well-known mutation to escape the proteolytic attack by a serum protease, FAP [28][29][30][31] .
Q201H and A208T were designed based on sequence comparison ( Supplementary Fig. S8), indicating that amino acid substitutions at non-conserved sequence positions among FGF21 orthologs can be used to design FGF21 variants. According to the crystal structure of the complex between β-Klotho and the C-terminal loop bearing the A208E mutation 11 (the FGF21-C/β-Klotho structure), the side chain of Gln201 is accommodated in a pocket surrounded by Arg696, His743, Phe764, Glu768, Arg829, Arg845, and Phe849 ( Supplementary Fig. S7). Histidine and glutamine are similar amino acids in terms of volume and hydropathy. The volume of histidine and glutamine are 153.2 Å 3 and 143.8 Å 3 , respectively, while the hydropathy scale of histidine and glutamine are − 3.2 and − 3.5, respectively 50 . Consequently, the Q201H mutation, which locates a bit larger and a bit more neutral amino acid into the pocket, seems to allow for better contact between the imidazole side chain of histidine and the residues lining the pocket.
In the FGF21-C/β-Klotho structure 11 , Glu208 is located in a canyon and its side chain is sandwiched between His646 and Pro940 of β-Klotho. The hydrophobic hydrocarbon chain of Glu208 contacts with Pro940 and its terminal carboxylate group interacts with the imidazole side chain of His646 ( Supplementary Fig. S7). The side chain of threonine has both a hydrophobic methyl group and a polar hydroxyl group whereas alanine only has a hydrophobic methyl side chain. Therefore, threonine seems to be more suitable for position 208 than alanine since its side chain is able to form favorable interactions with His646 and Pro940. In the case of the side chains of Ser195 and Ser200, they have no interactions with β-Klotho, an indication of little contribution of their side chains to β-Klotho binding. Therefore, the glycine introduction instead of serine seems to be beneficial for coreceptor binding due to the increased flexibility of the C-terminal loop.
The functional consequences of each position in the C-terminal sequences have been elaborately examined through peptide-based alanine scanning at a single amino acid resolution 51 , in which the alanine substitutions at the five mutation positions presented here were suggested to have little effect on β-Klotho binding. The enhanced potency of the M5 variant harboring the five point-mutations (S195G, P199G, S200G, Q201H, and A208T) shows that the introduction of non-alanine amino acids at the inert positions and their synergistic combinations could lead to the development of FGF21 variants with improved therapeutic potential.
To estimate whether the five point-mutations affect immunogenicity, we analyzed the binding affinity of peptides derived from the C-terminal loop (residues 174-209) in the wild type FGF21 and M5 towards major histocompatibility complex (MHC) molecules of T cell by using TepiTool 52 in Immune Epitope Database (IEDB). According to this analysis where binding affinities were represented by IC 50 values, most peptides from both proteins were predicted to have high IC 50 values, suggesting little binding to MHC molecules [53][54][55] (Supplementary Figs. S11, S12). Furthermore, the peptide "GMVGGGHGR" of M5 harboring four mutations (S195G, P199G, S200G, and Q201H in thick letters) has significantly higher IC 50 values (~ 353 nM) towards some MHC class I alleles than the corresponding "SMVGPSQGR" peptide of the wild type (IC 50 of 91-200 nM) ( Supplementary  Fig. S11). We also performed B cell epitope predictions for the C-terminal loop of the wild type FGF21 and M5 using 'Kolaskar and Tongaonkar antigenicity' 56 and 'Bepipred linear epitope prediction 2.0′ 57,58 tools available at IEDB. In these analyses, residues with higher scores than the threshold can be considered to be immunogenic. As shown in Supplementary Fig. S13,  www.nature.com/scientificreports/ the corresponding region of the wild-type, which together with the T cell epitope prediction, indicates that the M5 variant is highly likely to be at most as immunogenic as the wild type.

Materials and methods
Cloning, expression, and heat-treatment-mediated purification of human FGF21. The coding genes of the wild type FGF21 (residues 33-209), the β-trefoil region of FGF21 (residues 41-173; core-FGF21), and all FGF21 variants (FGF21-LW, core-FGF21-LW, M2, M3, and M5; see the "Results and Discussion" sections) were synthesized using a TEV protease cleavage site (ENLYFQ/G) at the N-terminus (COSMOGE-NETECH, Republic of Korea). The synthesized genes were inserted at the BamHI and SalI sites of the expression vector pET32a (+) (Novagen, Darmstadt, Germany) to fuse thioredoxin (Trx) to the N-terminus of the target proteins. They were transformed into Escherichia coli strain Rosetta (DE3). The transformants were grown to an OD 600 of 0.6-0.7 in Luria-Bertani (LB) media containing 50 μg/mL ampicillin and chloramphenicol at 37 °C, and 0.5 mM isopropyl-d-thiogalactoside (IPTG) was added for the expression of the Trx-fused target proteins. Circular dichroism spectroscopy. CD experiments were performed using a Jasco J-1500 equipped with Jasco PTC-517 Peltier cell holder, a thermostat (Jasco Corporation, Tokyo, Japan). CD spectra were collected with a scanning speed of 50 nm/min, a digital integration time of 4 s, and a bandwidth of 1 nm. Quartz cuvettes with a path length of 1 mm were employed. The signal of a protein-free buffer containing 10 mM NaH 2 PO 4 (pH 8.0) and 20 mM NaCl was subtracted from all CD spectra. Each spectrum, an average of 5 scans, was normalized to molar ellipticity (θ) using the mean weight residue and concentration by using a program provided by the manufacturer.
To monitor the thermal denaturation, CD spectra of 0.2 mg/mL proteins were collected every 1 °C at 215 nm as the temperature was changed from 20 to 100 °C with a heating rate of 1 °C/min. Spectral curves were fitted using the OriginPro 2021 program (Origin Lab Corporation, Northampton, MA, USA) through a non-linear adjustment by the Boltzmann method. When temperature reached 20, 40, 60, 80, and 100 °C during the thermal denaturation experiments, the far-UV CD spectra were obtained in the range of 190-260 nm. The near-UV CD spectra were obtained in the range of 240-340 nm using 5 mg/mL of proteins. Heating rate was set at 5 °C/min and the temperature was maintained for 10 min at 20, 40, 60, 80, and 100 °C. The far-and near-UV CD spectra were collected by continuous scanning at 0.5 nm intervals. The structural reversibility was checked by stepwise cooling of the protein solution from 100 to 20 °C. Turbidity measurement. The optical density at a wavelength of 600 nm was measured at 75 °C using the V-730 UV-Vis Spectrophotometer (Jasco Corporation) equipped with a bath circulator. The concentration of proteins in a buffer consisting of 10 mM NaH 2 PO 4 (pH 8.0) and 200 mM NaCl was 25 mg/mL. The value of optical density was determined by subtracting the signal of the protein-free buffer. Measurements began 15 s after putting samples into the spectrophotometer.
Dynamic light scattering. DLS experiments were conducted at 20 °C using a Zetasizer Nano ZS apparatus (Malvern Instruments Ltd., Malvern, UK) by setting an automatic attenuator. Proteins having a concentration of 25 mg/mL in a buffer consisting of 10 mM NaH 2 PO 4 (pH 8.0) and 200 mM NaCl were centrifuged at 13,000 rpm for 15 min at 4 °C. The supernatant was transferred to a glass cuvette having a path length of 1 cm that was located at a position of 4.65 mm. The light scattering intensities were collected 14 times at an angle of 173° using  ). Briefly, cells were incubated with FGF21 proteins for 6 h in growth media, followed by incubation in glucose-and serum-free DMEM for 30 min. Cells were treated with 2-deoxyglucose for 10 min. Next, we measured the level of accumulated 2-deoxyglucose-6-phosphate (2DG6P) using the assay kit according to the manufacturer's protocol using a GloMax®96 microplate luminometer (Promega).

Western blot analysis.
To determine FGFR signaling, 3T3-L1 cells were incubated in serum-free DMEM.

Animals and treatments.
All animal experiments were conducted in accordance with the guidelines on animal care including the ARRIVE guidelines and use as approved by the Institutional Animal Care and Use Committee of Daegu Gyeongbuk Institute of Science and Technology (DGIST-IACUC-19121001-0001). Male C57BL/6 mice at 7 weeks of age were purchased from Koatech. Mice were housed under a 12 h light/12 h dark cycle (lights on from 7:00 am to 7:00 pm) in individually ventilated cages (1 mouse per cage) with chip bedding at 23 °C ± 3 °C and a relative humidity of 50% ± 10%. Before the HFD experiments, mice were acclimated to the housing facility for 1 week and given ad libitum access to water and normal chow diet (NCD) (12% kcal from fat; Lab Supply, 5053). After 10 weeks of a NCD or HFD (60% kcal from fat; Envigo, 06414) treatment, HFD-fed mice were randomly assigned to vehicle or FGF21 treatment groups. The randomization was stratified by body weight. The NCD-fed mice were intraperitoneally injected with vehicle (DPBS, 21-031-CV; Corning, NY, USA) and the HFD-fed mice were intraperitoneally injected with vehicle, wild type FGF21, or mutant FGF21 once daily for 2 weeks. All mice had free access to water and food while remaining on their respective diets during the drug treatment periods. The single injection was performed between 4:00 pm and 6:00 pm, and body weight and food consumption were measured immediately before each injection every day. The following day after the last injection, mice were sacrificed to collect livers after noting final measurements.
Liver TG quantification. Hepatic TG content was measured using Triglyceride Quantification Colorimetric/Fluorometric Kit (K622-100; BioVision, Milpitas, CA, USA). Livers were homogenized in 1 mL of a solution containing 5% NP-40 in water, heated to 80-100 °C, and allowed to cool to room temperature; this step was repeated twice. Insoluble material was removed using centrifugation at 16,000×g for 2 min. Extracted TG was diluted sixfold with distilled water. TG level was measured according to the manufacturer's instructions. Individual TG levels were normalized to liver weight.