Combined approaches of EPR and NMR illustrate only one transmembrane helix in the human IFITM3.

Interferon-inducible transmembrane protein IFITM3 was known to restrict the entry of a wide spectrum of viruses to the cytosol of the host. The mechanism used by the protein to restrict viral entry is unclear given the unavailability of the membrane topology and structures of the IFITM family proteins. Systematic site-directed spin labeling (SDSL) and electron paramagnetic resonance (EPR) studies of IFITM3 in detergent micelles identified a single, long transmembrane helix in the C-terminus and an intramembrane segment in the N-terminal hydrophobic region. Solution NMR studies of the same sample verified the secondary structure distribution and demonstrated two rigid regions interacting with the micellar surface. The resulting membrane topology of IFITM3 supports the mechanism of an enhanced restricted membrane hemi-fusion.

Scientific RepoRts | 6:24029 | DOI: 10.1038/srep24029 biophysical studies are urgently required to illustrate the three-dimensional structures, or at least the membrane topologies of IFITMs.
In this report, a combination of electron paramagnetic resonance (EPR) and nuclear magnetic resonance (NMR) was applied to investigate the structure and membrane topology of the IFITM3 protein in detergent micelles. Systematic site scanning of spin labeling, EPR dynamic and accessibility analysis identified a C-terminal transmembrane α -helix and an N-terminal IFITM3 segment (composed of two short α -helices) lying on the surface of micelles. Further triple resonance solution NMR studies verified the secondary structures of IFITM3 and also illustrated the backbone flexibility through NMR relaxation analysis. Collectively, a tentative IFITM3 model was proposed. This model adopts a topology similar to model I (Fig. 1a), which is consistent with recent antiviral mechanism studies.

Results
EPR analysis revealed the single transmembrane topology of IFITM3. With site-directed spin labeling (SDSL), EPR spectroscopy is a powerful tool to analyze the mobility and secondary structure of a membrane protein [24][25][26][27] . Before implementing SDSL-EPR studies, three endogenous cysteine residues (C71, C72 and C105) on IFITM3 were first substituted with serine to construct the Cys-less "WT". Then, a total of 72 sequential residues (from W60 to Y132, except T65, which was not expressed after cysteine mutation) covering the two predicted hydrophobic regions were mutated to cysteine residues (supplementary Figure 1). All purified IFITM3 mutants in dodecyl-phosphocholine(DPC) micelles were identified in a monomeric state using SDS-PAGE, same as the wild-type (WT) IFITM3 protein (Fig. 1c). The spin labeling reaction with methanethiosulfonate (MTSL) was subsequently performed to attach the side chain R1 to the IFITM3 mutants, showed as Fig. 1b 27 . Then, the systematic site-specific continuous wave (CW)-EPR spectra of spin-labeled IFITM3 were acquired in DPC micelles at ambient temperature(298K).
The EPR spectra line shape could reflect the mobility of the nitroxide probe R1, which also contains structural information of proteins 28 . Total of 71 CW-EPR spectra are presented in Fig. 2, and the typical nitroxide three-line CW-EPR spectra of spin-labeled IFITM3 variants were observed. Although most of the spectra exhibited only one motional state, several residue sites (S61R1, L73R1, G74R1, A79R1, S81R1 and D92R1) had multiple components. For a clear presentation, spectra of spin labels at these residue sites were plotted in Fig. 3a, indicating the presence of both the immobilized (i) and mobilized (m) components. Interestingly, all these residues with multiple motional states were identified in the first half of the putative hydrophobic region(from W60 to G95 ), which strongly suggests that this segment adopts two different motional or conformational states. (a) Three different topology models proposed recently for IFITM3. The hydrophobic region of IFITM3 from W60 to Y132 was analyzed using EPR methods. (b) The spin labeling reaction for cysteine substituted IFITM3 mutants to introduce the nitroxide side chain, which is denoted R1. (c) SDS-PAGE analysis of IFITM3 in detergent micelles. Both wt-IFITM3 and IFITM3-W60C were purified as monomers in detergent micelles. Figure 2. CW-EPR spectra of 72 spin-labeled Cysteine mutants, covering the hydrophobic region of human IFITM3 (form W60 to Y132, except T65). Spectra composed of multiple components were highlighted with . All spectra were normalized by the height of the central peak. The CW-EPR spectra exhibit multi-components in detergent micelles. "i" and "m" represent the "immobile" and "mobile" components, respectively. (b,c) represent the oxygen accessibility (Π O2 ) and NiEDDA accessibility (Π NiEDDA ) of the hydrophobic region of IFITM3 in detergent micelles. Sine waves with a periodicity of 3.6 were drawn to indicate the α -helix secondary structure. (d) The membrane immersion depth (Φ ) of sequential residues span the hydrophobic region of IFITM3 in detergent micelles. The grey region represents the putative transmembrane region and the dashed line indicates the suggested membrane interface. Then, EPR power saturation studies were conducted as shown in supplementary Figure 2. These experiments illustrate the accessibility of spin radical side chains, which can be used to derive secondary structure and the membrane topology information of proteins 24,29,30 . The systematic accessibility parameters (Π ) were investigated for the sequential residues (form W60 to Y132, except T65), covering the two putative hydrophobic regions of IFITM3. Oxygen in the air was used as the hydrophobic paramagnetic reagent to evaluate membrane exposure, while NiEDDA served as the hydrophilic paramagnetic reagent to study the extent of aqueous exposure. Π O2 and Π NiEDDA were plotted against the residue numbers in Fig. 3b,c. As we can see, the Π NiEDDA values progressively decreased from A98 to G114, followed by a lower flat region (from I115 to I121). Then, the Π NiEDDA values increased from residue I122 to A131. On the contrary, the value of Π O2 first increased starting from residue A98 and then slowly decreased. These two typical changing patterns of Π O2 and Π NiEDDA demonstrated that a highly hydrophobic transmembrane region exists in IFITM3 starting from residue A98 to A131 in detergent micelles. However, no similar patterns in the accessibility parameters (both Π O2 and Π NiEDDA ) were observed in the first half of the putative hydrophobic region (from W60 to G95), indicating that no transmembrane property exist in this segment of IFITM3. Moreover, a close examination of the accessibility parameter values (Π O2 and Π NiEDDA ) of the spin labeled residues revealed some periodic behaviors. The periodicity covered the residues from A96 to A131 in Fig. 3b,c, indicating that this segment adopts an α -helical secondary structure in detergent micelles. In addition, the accessibility parameter values of two short segments (L62-F67 and I76-R85) in the N-terminal region were also observed with same periodicity, suggesting that two short α -helices potentially exist in this segments. Moreover, the immersion depth parameter (Φ ) was derived from the conjugated accessibility parameters of Π O2 and Π NiEDDA 29 . As shown in Fig. 3d, the Φ values of the residues covering the entire putative hydrophobic region of IFITM3 can be separated into two apparently different patterns. In the first pattern starting from W60 to G95, the Φ value fluctuated around a relatively low value, indicating that the residues in this segment were partially buried in the micelles. However, the second pattern highlighted in gray ( Fig. 3d) was totally different ( the Φ values progressively increased from residue A98 to T118, followed by a decrease until residue Y132), indicating that a transmembrane segment was embedded in this region of IFITM3. Collectively, the analysis of the accessibility parameter (Π O2 and Π NiEDDA ) and the immersion depth parameter (Φ ) derived from the power saturation experiments demonstrated that the IFITM3 protein contains a long transmembrane α -helix covering the residue sequence from A96 to A131 in the putative hydrophobic region. This coincides with the previously predicted topology I and topology III model, but totally different from the topology II model (Fig. 1a). At the same time, two short α -helices without transmembrane properties were observed in the first half of the putative hydrophobic region, which was similar to the predicted topology model I or model II (Fig. 1a). Moreover, as shown in Fig. 3a, several residues in the first half of the putative hydrophobic region were revealed to contain multi-motional states by the CW-EPR spectra. Thus, our EPR analysis of IFITM3 in detergent micelles suggested that IFITM3 might adopt a membrane topology structure similar to the topology model I in Fig. 1a.
Solution NMR structure determination and backbone relaxation analysis of IFITM3. With information on the secondary structure and membrane topology of IFITM3 as determined by the EPR studies, solution NMR studies were also conducted to determine the atomic resolution structure of IFITM3 in DPC micelles. The backbone chemical shift assignments ( 1 H, 15 N, 13 CO, 13 C α , and 13 C β ) of full-length IFITM3 protein in DPC micelles were achieved using series of two-and three-dimensional solution NMR spectra (Fig. S3). Heteronuclear single quantum coherence (HSQC) spectra of the specifically 15 N labeled amino acid ( 15 N-Leucine, 15 N-Isoleucine, 15 N-Valine, 15 N-Methionine or 15 N-Phenylalanine) in the IFITM3 samples were acquired to assist the residue assignments (Fig. S4). Based on the backbone chemical shift values of assigned residues, the program TALOS + was used to estimate the secondary structure of IFITM3 31 . As shown in Fig. 4a, three segments with relatively large negative TALOS + index values were obtained, representing three α -helix segment. The remainder of the IFITM3 protein exhibited TALOS + index values close to zero, indicating a random coil. Compared with the previously predicted topology model that contains two long hydrophobic regions, the first two short α -helixes of IFITM3 resided in the first putative hydrophobic region while the third long α -helix correspond to the second putative hydrophobic region. To further verify the secondary structure, backbone relaxation analysis of IFITM3 was performed. As shown in Fig. 4c, residues with small T2 values were mainly observed in three regions, suggesting high backbone rigidity. This finding is consistent with the α -helix structure predicted by program TALOS + in the same region. While most residues in the N-terminus have large T2 values, indicating high flexibility property of the backbone. Thus, the overall results were consistent with the derived secondary structure of IFITM3 as determined by TALOS + analysis (Fig. 4). However, the final atomic resolution structure of full-length IFITM3 could not be obtained, mainly due to the lack of an adequate number of NOE restraints and the intrinsic random coil structure in the N-terminal region. Instead, low resolution structural topology of full-length IFITM3, especially the putative hydrophobic region, can still be derived from the limited NMR data. As shown in Supplementary Figure 5, an ensembles of 10 conformers with the lowest energy and best convergence were selected from 100 calculated structures of the N-terminally truncated IFITM3 (from W60 to A131). A kink or unusual dihedral angles could be found in the C-terminal of the transmembrane helix, which were probably derived from the unassigned residues nearby and the existence of Pro125 ( Figure S5). When superimposing the relatively long C-terminal α -helix (from A96 to A131), the root-mean-square deviation (RMSD) values of the backbone atoms and heavy atoms were 3.139 Å and 2.985 Å, respectively. For a clear presentation of the structure and predicted membrane topology of the putative hydrophobic region of IFITM3, one conformer was selected out (Fig. 5a). The structural model of IFITM3 shows that a long C-terminal transmembrane helix and two short intramembrane α -helices in the N-terminal of the hydrophobic region was connected by a small flexible loop between them, which is perfectly consistent with the structural topology derived from the EPR analysis. Taken together with the EPR and solution NMR results, we conclude that the hydrophobic region of IFITM3 adopts a topology containing two short intramembrane α -helices followed by a long transmembrane α -helix (Fig. 5b).

Discussion
Three different membrane topology models for IFITM proteins have been previously proposed 14 . The first model suggested that IFITM3 adopted an intramembrane topology with both the N-and C-termini facing cytoplasm, based on the observation that the inserted N-linked glycosylation site within the N-and C-terminal region cannot be modified (Model II of Fig. 1a) 18 . The results from flow cytometry experiments and cell surface immune-staining experiments argued that IFITM3 is a dual-transmembrane protein with both its N-and C-termini exposed to the endoplasmic reticulum (ER) lumen or extracellular space (Model III of Fig. 1a) 3,19 . The third topology model was recently proposed demonstrating that IFITM3 is a type II transmembrane protein (model I of Fig. 1a) 22,23 .
Through the combined approaches of EPR and solution NMR, the structural model and the membrane topology of human IFITM3 protein were determined. As illustrated in Fig. 5b, IFITM3 adopts a transmembrane topology, supporting the proposed type II transmembrane protein topology (model I in Fig. 1a). At the C-terminus of human IFITM3, a long α -helix covering residues from A96 to A131 spans through the micelles. While two short discrete intramembrane α -helices were discovered in the first predicted hydrophobic segment of IFITM3. As shown in Fig. 3d, the EPR immersion depth (Φ ) data demonstrated that these two short α -helices possessed no transmembrane properties, but only buried partially in the micelles. And the relatively larger Φ values of the residues in the second short helix indicated that it might be more deeply buried in micelles than the first short helix. Moreover, the random coil property of the N-terminus of IFITM3 was revealed by NMR analysis.
Recently, functional studies by Li et al. have suggested that IFITMs could hamper viral membrane hemi-fusion through reducing membrane fluidity and conferring a spontaneous curvature 16,32,33 . Here in our topology structure of IFITM3 in micelles, two short intramembrane α -helices were discovered in the first predicted hydrophobic region. These two short helices were likely to induce membrane curvature when they inserted into only one  leaflet of the lipid bilayers (Fig. 5b), although lipid bilayers adopt different curvature and membrane thickness from micelles. Hence, our membrane topology of IFITM3 could provide direct structural evidence to support the antiviral mechanism proposed by Li et al. Moreover, three types of post-translational modifications have been reported to regulate the antiviral activity through influencing the subcellular localization, trafficking, clustering or interaction with potential receptors. The S-palmitoylation occurs at Cysteines 71, 72 and 105 17 , and ubiquitination takes place at Lysines 24, 83, 88 and 104 18 , whereas Tyrosine 20 is phosphorylated 8,34 . These post-translational modifications can only occur when these residues are exposed to cytoplasmic side of the membrane, which is consistent with our IFITM3 structure and proposed membrane topology model (Fig. 5b), with an intracellular N-terminal domain and an intramembrane segment facing the cytoplasm. Moreover, CW-EPR spectra analysis indicated that residues with multi-motional or multi-conformational states all resided in the intramembrane hydrophobic segments (W60-G95). We suggested that the multi-conformational property of the intramembrane hydrophobic segments might closely relate to its multiple roles in post-translational modifications, such as different structural states might interact with different enzymes.
Therefore, the combined EPR and solution NMR studies clearly supported the type II transmembrane protein topology model for IFITM3, which is consistent with the previously proposed mechanism of its antiviral function. It remains unclear why the other contradictory membrane topology models for IFITM3 are derived from functional studies. We cannot exclude the possibility that IFITM3 may be able to adopt multiple topologies in different stages of the viral infection period, or in different host cell types. Further structural and functional correlation studies of IFITM3 in different cell types and studies focusing on other IFITM proteins are still required to completely unveil the antiviral mechanism of IFITM proteins.

Constructions of Cysteine mutants ofIFITM3.
Three endogenous cysteines (C71, C72 and C105) were substituted by Serine through site-directed mutagenesis to construct Cysless "WT" IFITM3. Then, 72 sequential residues spanning from W60 to Y132 (except T65) on full-length IFITM3 were mutated to Cysteine, respectively. The target IFITM3 gene sequences were subsequently inserted into the vector pET21b (Novagen) using the restriction sites Nde I and Xho I. All constructs were verified through gene sequencing.
Protein expression and purification. IFITM3 protein was over-expressed in BL21-GOLD (DE3) competent cells (Novagen) transformed with vector pET21b containing IFITM3 coding sequence in M9 media at 25 °C. A final concentration of 0.8 mM IPTG (isopropyl-D-thio-galactoside) was added at cell density around OD 600 = 0.8 to induce protein over-expression in the following 20 hours. The cell pellets harvested by high-speed centrifugation (Allegrax-15R, Beckman) were resuspended in lysis buffer(70 mM Tris, 300 mM NaCl, pH 8.0). Then, cell membrane was broken using ultrasonication, followed by separation procedure using high-speed centrifugation. After that, the pallets fraction containing IFITM3 protein was dissolved using the binding buffer (20 mM Tris, 200 mM NaCl, pH 8.0) containing 1% (w/v)SDS. Finally, IIFITM3 protein was purified using Ni-NTA affinity chromatography (QIAgen, Germany) and eluted in binding buffer containing 0.5% DPC. SDS-PAGE (Sodium dodecyl sulfate polyacrylamide gel electrophoresis) was applied to analyze the purified IFITM3 protein in DPC micelles.
For power saturation studies and accessibility analysis, the experiments were performed using the same spectrometer equipped with an ER4123D CW resonator (Bruker BioSpin, Germany). Samples with a total volume of ~3 μL were loaded into gas permeable TPX capillary tubes. During the power saturation analysis, the range of incident microwave power was from 0.7 mW (25dB attenuation) to180 mW(1dB attenuation) with the step of 2dB.To obtain the EPR spectra of IFITM3 sample in the N 2 atmosphere, O 2 atmosphere of the sample was purged through N 2 blowing or equilibration. At the same time, EPR spectra were acquired in air (O 2 as the hydrophobic paramagnetic reagent), or in N 2 with the presence of 50 mM Ni 2+ -EDDA complex (NiEDDA) as the hydrophilic paramagnetic reagent. Power saturation curves were measured as the vertical peak-to-peak amplitude(A) of the first derivative M i = 0 line as a function of incident microwave power (P). To determine the value of P 1/2 , the data were fitted using an R software script according to the equation (1): 1/ 1/2 I is a scaling factor, P 1/2 is the incident power where the first derivative amplitude is reduced to half of its unsaturated value, and ε is a measure of the homogeneity of saturation of the resonance line. With the obtained P 1/2 , the accessibility parameter Π O2 and Π NiEDDA were calculated according to the equation (2): where ΔP 1/2 is the value difference between two P 1/2 values in the presence and absence of relaxing agent, ΔHpp is the peak-to-peak line-width of the first derivative spectrum, P 1/2 (DPPH) and ΔHpp (DPPH) were the P 1/2 and peak-to-peak line-width values of a standard sample of crystalline 2,2-diphenyl-1-picrylhydrazyl (DPPH) in KCl. Then, the immersion depth parameter Φ could be calculated using the power saturation parameter:  15 N-Isoleucine, 15 N-Valine, 15 N-Methionine and 15 N-Phenylalanine selectively labeled IFITM3 samples. Details of the experimental parameters were described previously. All the spectra were analyzed using NMRPipe and NMRView. Backbone resonance assignment was then performed. Dihedral angle constraints were predicted using Talos + program, based on the obtained chemical shifts of 13 Cα , 13 Cβ , 13