Conformational exchange of fatty acid binding protein induced by protein-nanodisc interactions

Abstract

Fatty acid binding proteins (FABPs) can facilitate the transfer of long-chain fatty acids between intracellular membranes across considerable distances. The transfer process involves fatty acids, their donor membrane and acceptor membrane, and FABPs, implying that potential protein-membrane interactions exist. Despite intensive studies on FABP-membrane interactions, the interaction mode remains elusive, and the protein-membrane association and dissociation rates are inconsistent. In this study, we used nanodiscs (NDs) as mimetic membranes to investigate FABP-membrane interactions. Our NMR experiments showed that human intestinal FABP interacts weakly with both negatively charged and neutral membranes, but it prefers the negatively charged one. Through simultaneous analysis of NMR relaxation in the rotating-frame (R_1ρ), relaxation dispersion, chemical exchange saturation transfer, and dark-state exchange saturation transfer data, we estimated the affinity of the protein to negatively charged NDs, the dissociation rate, and apparent association rate. We further showed that the protein in the ND-bound state adopts a conformation different from the native structure and the second helix is very likely involved in interactions with NDs. We also found a membrane-induced FABP conformational state that exists only in the presence of NDs. This state is native-like, different from other conformational states in structure, unbound to NDs, and in dynamic equilibrium with the ND-bound state.

Significance

Transport of fatty acids and other lipophilic molecules between intracellular membranes involves interactions between fatty acids and fatty acid binding proteins as well as between the proteins and membranes. Despite many studies, the protein-membrane interaction mode is still unclear. Through novel simultaneous analysis of multiple sets of NMR relaxation-based data, we demonstrated that human intestinal fatty acid binding protein interacts weakly with negatively charged membranes mainly through two positively charged regions and the transient interactions induce changes of a minor portion of protein molecules in structure and dynamics. Such changes may be necessary for the uptake of ligands from membranes to the protein and the release of ligands from the protein to membranes.

Introduction

Fatty acid binding proteins (FABPs) are a family of lipophilic intracellular proteins expressed in many animals with different isoforms (1). They have low sequence identity but adopt a similar overall structure with a helical cap containing two α-helices and a β-barrel consisting of 10 β-strands (2,3) (Fig. 1). They likely function as carriers of water-insoluble long-chain fatty acids and other lipophilic molecules for facilitating their intracellular transport between different organelles (2). In the transport process, protein-membrane interactions are indispensable for some FABP isoforms (4). A number of studies were conducted to elucidate the mechanisms of how FABPs transfer their ligands to and from mimetic membranes (5, 6, 7, 8, 9, 10). Two distinct transfer modes have been revealed: aqueous diffusion mechanism and collision mechanism. The diffusion mechanism involves release of ligands from the protein to the aqueous milieu or uptake of ligands from the aqueous milieu to the protein, and the diffusion of ligands to/from membranes (8,9). This mechanism requires no direct protein-membrane contacts. The diffusion mechanism is considered as a suitable description of fatty acid transfer by liver FABP (LFABP). However, direct interactions between LFABP and lipid vesicles were observed under conditions of low ionic strength by different techniques (11, 12, 13). The collision mechanism requires initial protein-membrane contacts and then transfer of ligands between the protein and membrane in the protein-membrane complex. This mechanism has been proposed for several types of FABPs including intestinal FABP (IFABP) (8), adipocyte FABP (AFABP) (5), heart FABP (7), and brain FABP (6) isoforms.

Figure 1.

Ribbon diagram of hIFABP structure. Residues are colored according to backbone ¹⁵N chemical shift differences between native state N and ND-bound state (a) and between state N and ND-induced state I₄ (b): blue (|Δδ| ≥ 1.0 ppm), purple (0.5 ≤ |Δδ| < 1.0 ppm), cyan (|Δδ| < 0.5 ppm), and gray (unavailable data). The chemical shift differences were derived from analysis of relaxation data. To see this figure in color, go online.

FABP-membrane interactions have been shown to be mediated mainly through electrostatic-charge-charge force because the interactions are dependent on membrane net charge, protein surface charge, and buffer ionic strength. Site-directed mutagenesis studies on IFABPs (10), AFABPs (14), and heart FABPs (15) show that the positively charged residues in the helical cap play a critical role in the protein-membrane interactions and regulate the transfer rates of anthroyloxy-labeled fatty acids in vitro. Computational studies suggest that the conserved positively charged residues in α-helix II of all FABPs except LFABP are involved in the direct protein-membrane contacts (16). In silico mutations indicate that the diffusional and collisional transfer modes can be switched by reversing two conserved positively charged residues in α-helix II and one negatively charged residue on the opposite side of the helical cap, suggesting the importance of other residues outside the helical cap (16). In addition, residues in regions other than the cap such as βA strand and βC-D turn have also been shown to be involved in FABP-membrane interactions by mutagenesis (14,15), indicating the complexity of the protein-membrane interactions.

Because lipid composition of the membranes and hydrophobicity of the protein surface have effects on the anthroyloxy-labeled fatty acid transfer rates (8,17), hydrophobic interactions also contribute to FABP-membrane interactions despite less importance than charge-charge interactions. The involvement of hydrophobic interactions is further supported by mutagenesis studies and hydrophobic photo-labeling experiments on IFABPs (10). A recent fluorescence quenching study on IFABP mutants suggested that the protein penetrates partially into the membrane bilayer and the portal region (including the cap and its surrounding β-turns) is in close interactions with the hydrophobic core of model membranes (18). However, another similar study on LFABPs explained the fluorescence quenching effects by the insertion of one lipid acyl chain into the protein cavity rather than by the protein penetration into the membrane (12). A recent study on LFABP also suggested that the protein interacts peripherally with liposomes (13). Despite these studies, the exact FABP-membrane interaction mode is still unclear.

Although the protein structure in the FABP-membrane complex is unavailable, the membrane-bound protein conformation is suggested to be dependent on lipid composition (18) and differ from the closed conformation observed in aqueous solution (10). However, recent NMR studies on LFABP (13) and drosophila FABP (19) indicate no significant structural changes upon binding to phospholipid membranes. Our studies on human IFABPs show that local unfolding of α-helix II is necessary for fatty acids to enter the protein cavity for binding in the absence of membranes (20, 21, 22), as no obvious openings exist on the protein surface. To investigate the effects of phospholipid membranes on FABP structure and conformational exchanges, here we used nanodiscs (NDs) as mimetic membranes. The interactions of NDs with human IFABP (hIFABP) were examined by NMR. We found that hIFABP interacts much more strongly with negatively charged NDs than with neutral NDs. The affinity of hIFABP to negatively charged NDs and the binding kinetics parameters were determined from simultaneous fitting of multiple sets of NMR relaxation data. We also revealed that NDs alter the conformational exchanges of hIFABP. Our data suggested that a small portion of the ND-bound conformation may adopt a structure different from the native closed form.

Materials and methods

ND preparation and purification

1,2-ditetradecanoyl-sn-glycero-3-phosphocholine (DMPC; 14:0) and 1,2-ditetradecanoyl-sn-glycero-3- phospho-(1′-rac-glycerol) (DMPG; 14:0) were purchased from Avanti Polar Lipids (Alabaster, AL). Phospholipid NDs were prepared in a way similar to that described previously (23). Briefly, stock solutions of 20 mg/mL DMPC and DMPG were prepared in 10 mM potassium phosphate buffer (pH 7.4), respectively. Peptide 22A (PVLDLFRELLNELLEALKQKLK, Bio Basic Asia Pacific Pte Ltd., (Singapore)) was dissolved in deionized water at a concentration of 10 mg/mL as a stock solution. For DMPC NDs, the lipid and peptide stock solutions were mixed in equal volumes. For DMPC/DMPG ND, the DMPC and DPMG stock solutions were first mixed by a volume ratio of 7:3, and then the mixture was mixed with the peptide stock solution in equal volumes. The mixture was incubated at 50°C for 10 min, then cooled down to room temperature and kept for 10 min. This process was repeated for 10 times. Subsequently, the mixture was incubated at 70°C for 10 min, then cooled down to room temperature and kept for 10 min. This process was repeated three times. Finally, the mixture was frozen at −80°C refrigerator for 10 min, then thawed to room temperature and kept for 10 min. This was repeated three times. Well-assembled NDs were further purified by a size-exclusion column (Superdex 200 increase 10/300GL; GE Healthcare (Chicago, IL)) to remove unassembled lipid and peptide. Purified NDs was concentrated and kept in a buffer containing 20 mM sodium phosphate (pH 7.1) and 50 mM NaCl for later experiments.

Protein sample preparation and NMR spectroscopy

The wild-type hIFABP was expressed, purified, and delipidated using the protocols described previously (24). For the samples with NDs, the protein solution was added to the purified ND solution in a protein/lipid molar ratio of 1:5 or protein/ND molar ratio of 1:∼0.04. The final samples used for NMR experiments contained 0.5 mM ¹⁵N-labeled hIFABP, 20 mM sodium phosphate (pH 7.1), 50 mM NaCl, 1 mM EDTA, and 0.05% NaN₃. All NMR experiments were performed at 25°C on a Bruker (Hanau, Germany) 800 MHz instrument equipped with a cryoprobe.

¹⁵N relaxation dispersion (RD) data were acquired with a continuous wave (CW) decoupling and phase-cycled Car-Purcell-Meiboom-Gill (CPMG) method (25, 26, 27) using a constant time relaxation delay of 40 ms and interscan delay of 2 s. RD data at 16 different CPMG fields from 25 to 1000 Hz were collected by varying the separation of CMPG pulses. To estimate uncertainties in the apparent relaxation rates, the measurements at a CPMG field of 50 Hz were repeated three times.

Chemical exchange saturation transfer (CEST) profiles were obtained at two different weak saturation fields with a saturation time of 0.5 s and interscan delay of 1.5 s (28,29). For the sample containing NDs, the saturation field strengths were 12.5 and 6.25 Hz. For the sample without NDs, the saturation field strengths were 9.4 and 4.2 Hz. For the higher saturation field strength, 55 HSQC-based spectra were acquired using a series of ¹⁵N carrier frequencies ranging from 106 to 133 ppm at a spacing of 0.5 ppm. For the lower saturation field strength, 91 HSQC-based spectra were recorded from 106 to 133 ppm at a spacing of 0.3 ppm. Reference spectra were also recorded with similar parameters except a saturation time of zero. The uncertainties of the data points were estimated from the standard deviation of the points over a region far away from the CEST dips.

Dark-state exchange saturation transfer (DEST) profiles were obtained at two different weak saturation fields (150 and 300 Hz) with a saturation time of 0.5 s and interscan delay of 1.5 s (30). For each saturation field, 14 HSQC-based spectra were acquired at a series of ¹⁵N carrier frequencies: −229, 0, 56, 76, 91, 106, 121, 136, 151, 166, 186, 246, 346, and 471 ppm. The uncertainties of the data points were estimated from the standard deviation of the points at 246, 346, and 471 ppm.

¹⁵N relaxation rates in the rotating-frame (R_1ρ) and longitudinal relaxation rates (R₁) were measured as described previously (31,32). Six data points with relaxation delays of 10, 150, 300, 450, 700, and 1000 ms were collected for determination of R₁-values. R_1ρ-values were determined by collecting six points with delays of 1, 15, 30, 45, 60, and 81 ms using a spin-lock field strength of 1600 Hz. Transverse relaxation R₂-values were calculated from R₁ and R_1ρ as described previously (33). The R₂-values obtained by this method contained nearly no contributions from exchanges between two states which are distinct in chemical shifts (R_ex).

Dynamic light scattering measurement

The hydrodynamic diameters of hIFABP and NDs were examined by Dynamic light scattering (DLS) (DynaPro NanoStar; Wyatt Technology(Santa Barbara, CA)). The samples used for DLS were the same as those for the NMR experiments. Before DLS measurements, the samples were centrifuged at 14,500 rpm for 15 min. The data were analyzed to obtain size distribution by using DYNAMICS 5.0 software.

Simultaneous analysis of DEST and ΔR₂ data

The exchange process between the ND-free and –bound protein in the binding equilibrium can be simply described as:

$$\text{N }+\text{ ND}\leftarrow \to \text{N}/\text{ND},$$ (1)

where N, ND, and N/ND represent hIFABP, ND, and ND-bound hIFABP (or protein/ND complex), respectively. The apparent forward association and backward dissociation rates are given by k_on′ and k_off, respectively. k_on′ = k_on × [ND], where k_on is the association rate, and [ND] is the concentration of protein-free NDs. For such an equilibrium binding model, k_on′ and k_off can be extracted from ¹⁵N DEST profiles and ΔR₂-values (¹⁵N transverse relaxation rate difference between the samples in the presence and absence of NDs) using the method described previously (30,34,35).

Combined analysis of RD, CEST, DEST, and R_1ρ data

In the absence of NDs, our previous studies showed that the residues outside and inside the region F47-T67 (βC and βD region) of hIFABP exist in three and four conformational states, respectively (21). So, the exchange model described in Eq. 1 is oversimplified. Assuming that one new conformational state is induced by NDs, the three exchange models shown in Fig. 2 are more reasonable for the description of RD, CEST, DEST, and R_1ρ profiles of hIFABP in the presence of NDs.

Figure 2.

Conformational exchange models of hIFABP in the presence of NDs. Model I: three-state, Model II: five-state, and model III: six-state. N represents native state, whereas I_j represents the j^th minor state. N/ND represents the ND-bound state or ND-protein complex. k′_on is the apparent association rate of N to ND, whereas k_off is the dissociation rate of N from N/ND complex. k_ex1 (k_ex2, k_ex3) is the total exchange rate between state N and state I₁ (I₂, I₃), k_ex4 is the total exchange rate between states N/ND and I₄. The rates, which were obtained from fitting relaxation data of selected residues to each model, are indicated. The populations of minor states (in percentage) are also indicated. To see this figure in color, go online.

A standard χ²-based minimization procedure was employed for the combined analysis of RD, CEST, DEST, and R_1ρ profiles. The χ² is given by the following equations:

$${\chi }^2={\chi }_{CEST}^2+{\chi }_{RD}^2+{\chi }_{DEST}^2+{\chi }_{R1\rho }^2,$$ (2)

$${\chi }_{CEST}^2=\sum _{i,j}\left\{\frac{{\left[I{C}_{i,j}^{exp}\left({\omega }_1\right)-I{C}_{i,j}^{cal}\left({\omega }_1\right)\right]}^2}{N_{C1}\times \delta C_i^2\left({\omega }_1\right)}+\frac{{\left[I{C}_{i,j}^{exp}\left({\omega }_2\right)-I{C}_{i,j}^{cal}\left({\omega }_2\right)\right]}^2}{N_{C2}\times \delta C_i^2\left({\omega }_2\right)}\right\},$$ (3)

$${\chi }_{RD}^2=\sum _{i,j}\frac{{\left[R{D}_i^{exp}\left({\nu }_{CPMGj}\right)-R{D}_i^{cal}\left({\nu }_{CPMGj}\right)\right]}^2}{N_{RD}\times \delta R{D}_i^2},$$ (4)

$${\chi }_{DEST}^2=\sum _{i,j}\left\{\frac{{\left[I{D}_{i,j}^{exp}\left({\omega }_1\right)-I{D}_{i,j}^{cal}\left({\omega }_1\right)\right]}^2}{N_{D1}\times \delta D_i^2\left({\omega }_1\right)}+\frac{{\left[I{D}_{i,j}^{exp}\left({\omega }_2\right)-I{D}_{i,j}^{cal}\left({\omega }_2\right)\right]}^2}{N_{D2}\times \delta D_i^2\left({\omega }_2\right)}\right\},$$ (5)

$${\chi }_{R1\rho }^2=\sum _{i,j}\frac{{\left[I{R}_i^{exp}\left(t_j\right)-I{R}_i^{cal}\left(t_j\right)\right]}^2}{N_R\times \delta R_i^2},$$ (6)

where χ²_CEST, χ²_RD, χ²_DEST, and χ²_R1ρ are the χ²-values for CEST, RD, DEST, and R_1ρ data, respectively; IC_i,j^exp(ω_m) and IC_i,j^cal(ω_m) are the experimental and calculated intensities of the j^th CEST data point for the i^th residue at a weak radiofrequency (rf) field strength of ω_m (m = 1, 2), respectively, N_C1 and N_C2 are the total numbers of CEST data points at ω₁ and ω₂, respectively; RD_i^exp(ν_CPMGj) and RD_i^cal(ν_CPMGj) are the experimental and calculated relaxation rates of the j^th RD data point at a CPMG field ν_CPMGj for the i^th residue, N_RD is the total number of RD points; ID_i,j^exp(ω_m) and ID_i,j^cal(ω_m) are the experimental and calculated intensities of the j^th DEST data point for the i^th residue at a weak rf field of ω_m (m = 1, 2), respectively, N_D1 and N_D2 are the total numbers of DEST data points at ω₁ and ω₂, respectively; IR_i^exp(t_j) and IR_i^cal(t_j) are the experimental and calculated intensities of the j^th R_1ρ data point for the i^th residue; δC_i(ω_m), δRD_i, δD_i(ω_m), and δR_i are the errors of CEST, RD, DEST, and R_1ρ data points for the ith residue, respectively. The summation extends over all the data points for a given residue for individual fitting, whereas it extends over all the selected residues for global fitting. IC_i,j^cal(ω_m), ID_i,j^cal(ω_m), RD_i^cal(ν_j), and IR_i^cal(t_j) were calculated based on the McConnell equations (36), which are detailed in Supporting materials and methods.

Error estimation of the fitting parameters was done using the jack-knife method with random deletions of 20% residues.

Results and discussion

Interaction of hIFABPs with mimetic membrane

The NDs formed by wrapping DMPC and DMPC/DMPG bilayers with the 22A peptide had similar size distributions when the molar ratio of the lipid to the peptide was fixed to 7.2 (Fig. S1, a and b). The average diameters were ∼10.1 and 10.5 nm for DMPC and DMPC/DMPG NDs as estimated by DLS, which are similar to that obtained conventionally from use of DMPC and membrane-scaffolding protein (MSP) (37). The size of the NDs was ∼3 times as large as that of hIFABP (3.6 nm) (Fig. S1 c). The NDs were used to mimic membranes to examine the interactions of FABPs with cell membranes.

To examine protein-membrane interactions, we measured NMR peak intensities of hIFABP in the absence and presence of DMPC and DMPC/DMPG NDs, respectively. The intensities of most ¹H-¹⁵N correlation peaks were slightly smaller in the presence of DMPC NDs than in the absence of DMPC NDs (Fig. 3 a). The average of the intensity ratios (intensity for the sample with DMPC NDs to that without DMPC NDs) was ∼0.95. Moreover, the ratio varies from one residue to another and the variation was significantly larger than intensity uncertainties (<1%). The result suggests the presence of extremely weak interactions between hIFABP and DMPC NDs. On addition of DMPC/DMPG NDs to hIFABP, peak intensities decreased by ∼40% on average (Fig. 3 b), but HSQC peak positions (chemical shifts) remained nearly unaffected (Fig. S2). The intensity ratios ranged from 0.4 to 0.9. The differential effects on different residues were also observed in a previous study on interactions of Drosophila FABP with POPC/POPG NDs (19). Although the intensity reduction by DMPC NDs is significantly less than that by DMPC/DMPG NDs, the intensity ratios in the two types of NDs correlate well (Fig. S3). Our results demonstrate that hIFABP interacts much more strongly with DMPC/DMPG NDs than with DMPC NDs. Therefore electrostatic force plays a dominant role in hIFABP-membrane interactions because one DMPG molecule carries the net charge of −1, whereas DMPC has no net charge. This agrees with previous studies showing that binding of FABP to membranes is mediated primarily through charge-charge interactions (4,9,10).

Figure 3.

Intensity ratios of HSQC peaks in the presence of NDs to those in the absence of NDs for hIFABP in DMPC NDs (a) and in DMPC/DMPG NDs (b). To see this figure in color, go online.

If hIFABP bound tightly to DMPC/DMPG NDs with high affinity, the intensity ratios would be similar for different residues. This is because all the residues in the ND-bound form are nearly invisible in the HSQC spectrum. Moreover, the ND-bound and -free forms would be comparable in population despite a low ND/protein ratio (0.04:1) as one ND can bind multiple protein molecules in the case of strong binding. However, our experimental data showed that the intensity ratio varies significantly from one residue to another (Fig. 3 b), suggesting that the protein-membrane interactions are weak with low affinity and the dominant species is the ND-free form which is in exchange or in dynamic equilibrium with the minor species, ND-bound form (Eq. 1).

Interaction kinetics

To confirm the protein-ND interactions, we measured transverse relaxation rates (R₂) of backbone ¹⁵N spins in the protein. In both the absence and presence of NDs, ¹⁵N R₂-values were relatively uniform except for those spins located in the loops and terminal regions (Fig. 4, a and b), consistent with the well-folded and rigid structure of hIFABP. The relaxation rates were only ∼2 s⁻¹ larger for most residues in the presence of DMPC/DMPG NDs than in the absence of the NDs (Fig. 4 b), further supporting weak protein-ND interactions and presence of a minor population of ND-bound state. A similar trend was also observed for LFABP in the presence and absence of POPC/POPG vesicles (13). The relaxation rate differences (ΔR₂) should result from chemical exchange between the ND-free and ND-bound states (forms) because of chemical shift differences or/and significant R₂ difference between the ND-free state (R_2f, ∼13 s⁻¹) and -bound sate (R_2b, ∼130 s⁻¹, which was estimated based on the ND size and the published overall tumbling time of the NDs with a diameter of 9.4 nm (37)). Because the former contribution was largely suppressed in our R₂ measurements and the chemical shifts were shown to be similar in the presence and absence of NDs (Fig. S2), the latter contribution should be dominant.

Figure 4.

Transverse relaxation rates of backbone ¹⁵N in the presence of DMPC/DMPG NDs, R₂(+ND) (a); Relaxation rate differences (b), ΔR₂ = R₂(+ND) − R₂(−ND), where R₂(−ND) is the relaxation rate in the absence of DMPC/DMPG NDs. To see this figure in color, go online.

A previous study found that nanoparticle-bound LFABP undergoes chemical exchange with nanoparticle-free LFABP in the presence of POPC/POPG vesicles (13). In that study, the apparent association rate (k_on′) was estimated to be ∼1.9 s⁻¹ from DEST and ΔR₂ data by assuming a slow exchange model. In addition, the dissociation rate of LFABP from the vesicles (k_off) was found to be ∼0.7 s⁻¹ from fluorescence intensity decays. To examine if the interconversion of the ND-bound and ND-free hIFABP is also slow with respect to the relaxation rate difference between the two states (R_2b − R_2f = 120 s⁻¹), we acquired DEST profiles of backbone ¹⁵N spins (Fig. S4). Following the previously established method (30), we obtained the total exchange rate (k_ex = k_on′ + k_off), population of the minor ND-bound form (P_b), and individual residue-specific R_2b values by fitting globally the DEST and ΔR₂ data of 114 well-resolved residues to the binding model shown in Eq. 1. Assuming a slow exchange process (k_ex < 0.2(R_2b − R_2f)) by forcing k_ex ≤ 24 s⁻¹, the results from the best fitting were k_ex = 23.4 s⁻¹ and P_b = 10.2% (k_on′ = 2.4 s⁻¹, k_off = 21.0 s⁻¹). Comparing the calculated and experimental DEST profiles (Fig. S4, a–c) and as well ΔR₂ values (Fig. S4 d), we found that the experimental data fitted to the slow exchange model very badly (χ² = 707). Assuming a fast exchange process (k_ex > 4(R_2b − R_2f)), we obtained k_ex = 5991 s⁻¹ and P_b = 1.8% (k_on′ = 108 s⁻¹, k_off = 5883 s⁻¹). The data fitted to the fast exchange model well (χ² = 161, Fig. S5, a–d). In fact, χ²-values were relatively insensitive to k_ex from 550 to 6000 s⁻¹ for the fast exchange model, e.g., χ² = 162 when k_ex = 2992 s⁻¹ and P_b = 1.9%, whereas χ² = 170 when k_ex = 590 s⁻¹ and P_b = 2.1%. Thus, the exact exchange rate could not be determined from the DEST and ΔR₂ data. Nevertheless, we can conclude that the ND-bound hIFABP undergoes fast chemical exchange with the ND-free hIFABP, different from the nanoparticle-LFABP system. The significant difference between our work and the previous study may result from differential properties of LFABP and IFABP. It may also result from the use of different exchange models and experimental techniques for estimation of k_off and k_on′.

To further investigate the exchange timescale between the ND-bound and ND-free states, we performed CEST and RD experiments as they are sensitive to exchanges on millisecond timescale. In comparison with the protein in the absence of NDs, hIFABP is more dynamic in the presence of DMPC/DMPG NDs, and many residues displayed significantly enhanced RD (Figs. S6 and S7). In addition, several residues each exhibited an additional large CEST dip (Figs. S6 and S7). For instance, L102 had one CEST dip and no RD in the absence of NDs, but it displayed large RD and two CEST dips in the presence of NDs (Fig. S6, e–h). The data suggest that the NDs induce a new conformational state, which exists only in the presence of the DMPC/DMPG NDs and is invisible directly in the HSQC spectrum.

Our previous studies showed that two and three minor native-like states coexist with the major native state for the residues outside and inside the βC and βD region of hIFABPs in the absence of NDs, respectively (21). According to CEST data, the minor dips corresponding to the minor states I₁ and I₂ in the absence of NDs remained nearly unchanged in location and had small changes in depth upon addition of DMPC/DMPG NDs (Figs. S6 and S7). The data indicate that NDs do not significantly affect the original minor states observed in the absence of NDs. On the basis of this point, the observation of a new state (denoted as I₄), the presence of a ND-bound state, and the previously established exchange models (21), we proposed exchange models II (five-states) and III (six-state) for the residues outside and inside the βC and βD region in the presence of NDs, respectively (Fig. 2).

To extract global exchange parameters from CEST, RD, DEST, and R_1ρ data, we firstly focused on the residues which each displayed no RD nor CEST minor dip in the absence of NDs, but had large RD and one CEST minor dip in the presence of NDs (Fig. S6, e–h). For these residues, the five-state model can be reduced to a three-state model (model I) shown in Fig. 2. In total, eight residues (V23, L78, G80, F93, L102, N103, T104, and V118) met the requirements. We fitted the data of these residues to model I globally. From the best fit (Fig. 5, a–d), we obtained k_ex = 4876 ± 287 s⁻¹ (k_ex = k_on′+k_off), P_N/ND = P_b = 1.74 ± 0.06%, k_ex4 = 661 ± 36 s⁻¹, and P_I4 = 17.4 ± 0.6%. The chemical shifts of ¹⁵N spins in the ND-bound state and state I₄ were obtained too and are listed in Table S1. The result shows that the interconversion between the ND-bound and ND-free states is a fast exchange process (k_ex ≫ Δν, R_2b−R_2f, where Δν is the frequency difference of a ¹⁵N spin between the two states), which is consistent with the result derived from the analysis of only the DEST and ΔR₂ data. In the data analysis of our system, the population of the ND-bound state was mainly determined by DEST and R_1ρ data. So, for such a fast exchange, the chemical shift difference between the ND-bound and -free states could be obtained from our analysis although it could not be determined from RD data alone.

Figure 5.

Representative CEST (a and e), RD (b and f), DEST (c and g), and R_1ρ (d and h) profiles for the residues outside the F47-T67 region (βC and βD region). The experimental CEST data at lower (6.25 Hz) and higher (12.5 Hz) rf fields are indicated by “o” and “Δ,” respectively, so are the DEST data at lower (150 Hz) and higher (300 Hz) rf field strengths. The red and blue solid lines in (a, c, e, and g) are best fits obtained with model II at higher and lower rf field strengths, respectively. The locations (or chemical shifts) of states N, I₁, and I₄ in the CEST profiles are indicated by arrows. (a–d) are for residues T104, whereas (e–h) are for H33. To see this figure in color, go online.

Secondly we fitted the residues outside the βC and βD region, which had R_ex > 5 s⁻¹ and two or three CEST dips, to model II (Fig. 2) in a global way by fixing k_ex, k_ex4, P_b, and P_I4 at the values derived from model I. From the fitting of 28 residues (Fig. 5, e–h), we extracted the populations of minor states I₁ and I₂ (P_I1 = 2.84 ± 0.20%, P_I2 = 1.73 ± 0.14%) and their exchange rates with the major state N (k_ex1 = 31 ± 2 s⁻¹, k_ex2 = 571 ± 39 s⁻¹). Lastly, we fitted the residues inside the βC and βD region with R_ex > 5 s⁻¹ and two or three CEST dips to model III. Fixing k_ex, k_ex1, k_ex2, k_ex4, P_b, P_I1, P_I2, and P_I4 at the values derived from models I and II, we obtained the global exchange rate (k_ex3 = 2825 ± 284 s⁻¹) and population of state I₃ (P_I3 = 0.60 ± 0.02%) through global fitting of 12 residues (Fig. S8). For other residues, which were not used in the global fitting, their ¹⁵N chemical shifts in the minor states were extracted by individual fitting by fixing global exchange rates and populations. All the chemical shifts are listed in Table S1 and the chemical shift differences between states N and N/ND and between states N and I₄ are visualized in Fig. 1.

According to our results obtained from the combined analysis of CEST, RD, DEST, and R_1ρ, hIFABP associates to and dissociates from the DMPC/DMPG NDs in a fast manner. The apparent association rate (k_on′ = k_ex × P_b/(P_b + P_N)) was ∼103 s⁻¹, whereas the dissociation rate (k_off = k_ex × P_N/(P_b + P_N)) was ∼4773 s⁻¹. The association rate or on-rate (k_on = k_on′/[ND], where [ND] is the concentration of protein-free NDs) was calculated to be 9.8 × 10⁶ M⁻¹s⁻¹. In this calculation, we estimated that each ND is comprised of 18 peptide molecules and 130 lipid molecules and that the total ND concentration was ∼19.2 μM in our sample. This estimation was based on that 1) two copies of MSP1D1 (membrane scaffold protein variant 1) wrap around a patch of phospholipid bilayer to form one ND (37), 2) one MSP1D1 with 189 residues is equivalent to ∼9 22A peptides in length, and 3) peptide/lipid = 1:7.2 (molar ratio). In addition, we assumed that one bilayer ND binds to at most one hIFABP due to the weak binding feature. The on-rate obtained here is about two orders of magnitude smaller than the diffusion-limited rate (10⁹–10¹⁰ M⁻¹s⁻¹) for protein-protein associations. However, it is close to the upper limit of the association rates observed for many protein-protein complexes (10⁵–10⁶ M⁻¹s⁻¹) (38). This suggests that hIFABP-membrane association requires preferred protein orientations rather than random collision of protein onto membrane.

On the basis of k_off- and k_on-values from the fast exchange model, the binding affinity (K_d = k_off/k_on) was estimated to be 487 μM. This affinity agrees very well with those of AFABP to cardiolipin (CL) containing vesicles (egg phosphatidylcholine (EPC)/CL/egg phosphatidylethanolamine (EPE) = 65:25:10) (417 μM) and to phosphatidic acid (PA) containing vesicles (EPC/PA/EPE = 65:25:10) (588 μM) estimated by the sucrose-loaded vesicle assay (39). The affinity measured here also agrees qualitatively with those of IFABP (23.5 μM) and LFABP (50 μM) to CL/EPC (molar ratio 1:3) vesicles estimated by the sucrose-loaded vesicle assay (40) and liver bile acid binding protein to DMPC/DMPG (1:1) vesicles (31 μM) estimated by the agarose gel method (35). The result further supports the weak FABP-membrane interaction feature.

Structural information about ND-bound state (N/ND) and state I₄

To obtain structural information about the ND-bound state, we compared its chemical shifts with those of the native and unfolded (U) states. In terms of ¹⁵N chemical shifts, the ND-bound state is very similar to state N, but very different from state U (Fig. S9, a and b; Table S1), indicating that the ND-bound state is native-like and distinct from the native state in conformation. Using electron spin resonance, a previous study on BFABP showed that the residues located at the interface of two helices become more flexible and some other residues in the helices become more ordered upon binding to phospholipid micelles (41), supporting conformational changes caused by protein-membrane interactions. Another study proposed deep immersion of the helical region into the hydrophobic core between the headgroups and the middle of the lipid bilayer and as well potential conformational changes of the immersed part (18).

In comparison with state N, the residues with relatively large chemical shift differences (Δδ = |δ(N/ND) − δ(N)| ≥ 1.00 ppm) are located mainly in two clusters: one in αI-αII loop (V23, N24), αII (V26-H33), and αII-βB loop (D34-L36); the other in βD (V61-E63), βE(V66, T67, Y70-L72), βF(L78-W82), and βG (N87, L89, G91, R95, T96) (Fig. 1 a). This suggests that these two clusters may correspond to the two protein-ND interfaces in the ND-bound state. There are three positively charged residues in αII (K27, R28, and K29), which likely mediate contacts of the helical region with NDs through charge-charge interactions. Previous mutagenesis and computational studies have also demonstrated these residues are important for the protein-membrane interactions (10,16). There are three solvent-exposed positively charged residues (R79, K92, and K94) in the second cluster. So protein-ND interactions may also be mediated through the charge residues in this cluster. A previous study showed that K92I mutant had a similar anthroyloxy-labeled fatty acid transfer rate from the protein to vesicles and K92E mutant had a larger transfer rate compared with wild-type IFABP, indicating that this charged residue plays a certain role in the protein-membrane interactions. Further mutagenesis studies are needed to examine the importance of R79 and K94 to the protein-membrane interactions. Besides these two positively charged regions, another positively charged region exists on the surface of hIFABP, including K7, K37, K129, and K130 (Fig. S10). This region may not be involved in direct contacts with NDs because this region is surrounded by four negatively charged residues (D9, D111, E112, and D130, Fig. S10) and displays no significant chemical shift differences (Fig. 1 a).

Although deep insertion of the helical region was proposed previously, our data cannot tell whether the helical region is inside the membrane or just on the membrane surface. The two helices of FABPs are stable in aqueous solution or in a hydrophilic environment. When they are placed in a hydrophobic environment (inside the membrane), the helices will likely unfold. Even in aqueous solution, local unfolding of the second helix has been demonstrated in the absence of NDs (20). However, the ¹⁵N chemical shifts of αII in the ND-bound state (Table S1) shows that αII is mainly folded. Nevertheless, we could not rule out the presence of another ND-bound state (N′/ND) in which the helix is mainly unfolded. This is because the locally unfolded ND-bound state (N′/ND) was undetectable by our experiments, provided that it exchanges with the native-like ND-bound state (N/ND) on microsecond timescale. As the unfolding can promote ligand release to or uptake from the membranes, further studies on the ND-bound states are necessary.

Similarly, we analyzed the chemical shifts of state I₄. From comparisons of ¹⁵N chemical shifts (Fig. S9, c and d), we found that state I₄ is also native-like and distinct from state N. Only some of the residues that have obvious chemical shift differences between states I₄ and N (|δ(I₄)–δ(N)| ≥ 0.5 ppm) (Fig. 1 b) also have relatively large shift differences between the ND-bound state and state N (Fig. 1 a). Our data can be explained well by assuming the presence of a single ND-bound state (N/ND) as shown in Fig. 2. However, another intermediate ND-bound state (e.g., N′/ND, the locally unfolded ND-bound state mentioned above) likely exists between states I₄ and N/ND (N←→N/ND←→N′/ND←→I₄) because states N and I₄ are different in conformation. In principle, inclusion of N′/ND in model I, we can obtain structural information about the protein in state N′/ND from the relaxation data. Because of limited data and insensitivity of our data to fast exchanges on microsecond timescale, we did not use this model to analyze the data. In addition, the presence of another pathway from I₄ to N without visiting the ND could not be excluded because I₄ is less stable than N.

Effect of NDs on minor states I₁, I₂, and I₃

To examine the effect of NDs on conformational exchanges between state N and its minor native-like states I₁, I₂, and I₃, we performed ¹⁵N RD and CEST experiments on hIFABP in the absence of NDs at the same temperature as that for the sample in the presence of NDs (25°C). Following the same analysis as that described previously for the data recorded at 30°C, we obtained populations of minor states I₁, I₂, and I₃ (P_I1 = 4.2 ± 0.5%, P_I2 = 1.95 ± 0.04%, and P_I3 = 1.1 ± 0.1%) and their respective exchange rates with state N (k_ex1 = 27 ± 3 s⁻¹, k_ex2 = 589 ± 8 s⁻¹, and k_ex3 = 2803 ± 106 s⁻¹). Lowering temperature from 30 to 25°C reduced P_I1, P_I2, and P_I3 from 4.3%, 2.09% and 1.2% to 4.2%, 1.95% and 1.1%, respectively, but the changes were within the uncertainties. From the populations at the two temperatures, the enthalpy changes (ΔH) for the N to I₁, I₂, and I₃ conversions were estimated as described previously (42), and were found to be slightly positive. In other words, the physical interactions in the native state are just slightly stronger than those in the minor states, suggesting the minor states are native-like. This is consistent with the conclusion drawn previously from the chemical shifts (21). In addition, the decrease of temperature also slowed down k_ex1 by ∼25%, k_ex2 by ∼30%, and k_ex3 by ∼10%, indicating presence of the transition barriers from both N to I and I to N state.

The chemical shifts of states I₁, I₂, and I₃ in the presence of NDs are similar to those in the absence of NDs (Table S1), showing that the conformations of the minor states are nearly not affected by NDs. The DMPC/DMPG NDs reduced the populations of states I₁, I₂, and I₃ from 4.2%, 1.95% and 1.1% to 2.8%, 1.73% and 0.6%, respectively. Because of experimental data uncertainties and unrealized exchange pathways, it is difficult for us to conclude if NDs affect the minor states more than the major state in stability. The total exchange rates between state N and minor states (k_ex1, k_ex2, and k_ex3) were nearly not affected by the NDs, suggesting that NDs do not significantly influence the original conformational exchanges in the absence of NDs.

Conclusions

Our results show that hIFABP interacts weakly with both neutral (DMPC) and negatively charged (DMPC/DMPG) NDs, but it has a much higher affinity to the negatively charged NDs. Because of weak protein-membrane interactions, hIFABP exists mainly in membrane-free forms and slightly in membrane-bound forms in in vitro environments, and dissociates away rapidly from the membranes after their transient association. The protein may interact with the membranes primarily through solvent-exposed positive residues located in the helical region and βD-βG region. Although the interaction is transient, a membrane-free native-like state can be induced by the interaction. This membrane-induced state is distinct from the native state and other minor states of hIFABP in the absence of membranes. This membrane-induced state is also in slow conformational exchange with a protein-membrane complex or membrane-bound state. Our results suggest that membranes can cause additional conformational changes of FABPs and the membrane-bound FABP adopts a structure different from the membrane-free FABP. The altered structure may be favorable to the release and uptake of fatty acids.

Author contributions

D.Y. conceived the project and performed data analysis. Y.L. performed experiments and data analysis. D.Y. wrote the manuscript with input from Y.L.

Editor: Michael Brown.