A combined computational-biophysical approach to understanding fatty acid binding to FABP7

Abstract

Members of the fatty acid binding protein (FABP) family function as intracellular transporters of long-chain fatty acids and other hydrophobic molecules to different cellular compartments. Brain FABP (FABP7) exhibits ligand-directed differences in cellular transport. For example, when FABP7 binds to docosahexaenoic acid (DHA), the complex relocates to the nucleus and influences transcriptional activity, whereas FABP7 bound with monosaturated fatty acids remains in the cytosol. Preferential binding of FABP7 to polyunsaturated fatty acids like DHA has been previously observed and is thought to play a role in differential localization. However, we find that at 37°C, FABP7 does not display strong selectivity, suggesting that the conformational ensemble of FABP7 and its perturbation upon binding may be important. We use molecular dynamics simulations, NMR, and a variety of biophysical techniques to better understand the conformational ensemble of FABP7, how it is perturbed by fatty acid binding, and how this may be related to ligand-directed transport. We find that FABP7 has high degree of conformational heterogeneity that is substantially reduced upon ligand binding. We also observe substantial heterogeneity in ligand binding poses, which is consistent with our finding that ligand binding is resistant to mutations in key polar residues in the binding pocket. Our NMR experiments show that DHA binding leads to chemical shift perturbations in residues near the nuclear localization signal, which may point toward a mechanism of differential transport.

Significance

This work studies FABP7 at physiological temperature and shows that nuclear localization of FABP7 cannot be controlled solely by differences in fatty acid binding affinity, which are largely absent at physiological temperatures. Instead, through biophysical experiments and simulations, we demonstrate evidence of ligand-dependent conformational changes that may be associated with biological outcomes. Further, extensive simulations reveal heterogeneity in available ligand-binding conformations, supported by the observation of ligand binding resistant to mutations in binding site residues. Our work suggests subtle ligand and protein conformation differences may be important for differential cellular transport.

Introduction

Fatty acid binding proteins (FABPs) belong to the intracellular lipid-binding protein family and function as chaperones that transport hydrophobic cargo within the cell (1,2,3). Depending on the bound cargo, this transport can influence a host of cellular functions, including cell growth and mobility, gene expression, energy storage, and lipid metabolism (1,2,4).

Here, we focus on brain FABP (FABP7), which is found in various tissues but is particularly important in the brain. High expression levels of FABP7 are associated with increased cell proliferation, which is critical for neuron migration during brain development and has been linked to more invasive tumors and poor prognosis in glioblastomas, melanomas, and kidney and bladder cancers (5,6). Beyond its role in cell proliferation, genome-wide association studies have linked FABP7 to a variety of diseases, including schizophrenia (7) and autism (8).

FABPs bind to a diverse range of hydrophobic cargo, including long-chain fatty acids (9), endocannabinoids (10,11), phytocannabinoids (12), and organic dyes (13) (Fig. 1). Binding to specific ligands can lead to nuclear localization and changes in gene expression (14). For example, nuclear localization occurs when FABP7 binds to polyunsaturated docosahexaenoic acid (DHA) but not when FABP7 binds to monounsaturated or saturated fatty acids despite their similar structures (5). This differential behavior is thought to arise due to DHA binding leading to unique structural changes to the protein (5). Unfortunately, an atomistic-scale understanding of the precise ligand-dependent structural changes that lead to nuclear localization of FABP7 are not known.

Figure 1.

Overview of FABP7 sequence, structure, and known ligands. (a) The structure of FABP7, showing β-strands (blue), ⍺-helices (teal), and loops (orange). (b) Key residues in the portal region including nuclear localization signal (K21, R30, and Q31). (c) The sequence and secondary structure motifs of FABP7. NLS residues are shown in bold and other significant residues to this study in red. (d) Common ligands that bind to FABP7.

NMR and x-ray crystallographic structural studies reveal that FABPs share a common tertiary structure with a beta-barrel and flexible lid domain composed of two α-helices (Fig. 1, A–C). A common feature among FABPs is a gap between the fourth and fifth strands (denoted βD and βE in Fig. 1 A). Contacts across the gap are primarily between side chains rather than interstrand backbone H-bonds.

Members of the FABP family share a dynamic portal region (βCD, βEF, and H3, highlighted in red; Fig. 1, B and C) responsible for mediating ligand entry and exit. Located along the βCD, residue F57 is responsible for opening and closing the portal region and interacts with ligands in the binding pocket (4). F57 also interacts with H3, which contains three residues (K21, R30, and Q31) that form a nuclear localization signal (NLS) (5). Binding to specific ligands may change the conformation of the NLS and promote nuclear localization. X-ray crystal structures of FABPs bound to fatty acids reveal that the fatty acids maintain similar poses in the FABP7 beta-barrel with H-bonds to R126 and Y128 (15). Based on these structures, it is unclear why DHA-induced nuclear localization of FABP7 occurs and why the other fatty acids bound to FABP7 do not localize to the nucleus.

Herein, we use biophysical experiments and simulations to examine FABP7 binding to three fatty acids: oleic acid (OA), stearic acid (SA), and DHA. Our goals are to understand the interactions essential for fatty acid binding and shed light on the ligand-induced conformational changes that contribute to different biological outcomes. Our main findings are as follows:

• •Both NMR and computer simulations show that the structures of the βCD, βEF, and H3 regions of apo FABP7 are heterogeneous and dynamic, facilitating ligand entry. This heterogeneity is significantly reduced upon binding.
• •Previous studies have suggested a robust selectivity for polyunsaturated fatty acids, like DHA, over other fatty acids (5,14). In contrast, at physiological temperatures, we find only modest differences in binding affinity between polyunsaturated DHA, monounsaturated OA, and saturated SA.
• •By performing experiments at multiple temperatures, we show that fatty acid binding is strongly temperature dependent. The differences between fatty acids are bigger at lower temperatures, which is consistent with previous experimental findings (16).
• •Previous literature has suggested that R126 and Y128 are key fatty acid-binding residues. We find that some mutations to these residues (R126L, Y128F, and R126L/Y128F) make only modest changes to ligand affinity.
• •Extensive computer simulations reveal that multiple sites within the binding pocket can bind the fatty acid carboxylate group. Even wild-type FABP7 displays substantial heterogeneity in fatty acid binding poses. Upon mutation, the population of binding modes shifts to other poses that can better coordinate the carboxylate group.
• •We observe that DHA binding leads to substantial chemical shift changes in residues within the portal region, most notably in H3, including the side-chain chemical shift of NLS residue Q31. Similar changes are not observed for OA or SA.

Materials and methods

Computational

Molecular dynamics

The crystal structure of FABP7 bound to antinociceptive SBFI-26 (PDB: 5URA) (17) was used as the starting structure for our models. To simulate apo FABP7, the inhibitor was removed. Hydrogen atoms were initially assigned with H++ (18) and adjusted to maintain H-bonding networks within the protein. The systems were solvated with a 10 Å octahedral water box and neutralized with sodium ions using the tLEaP utility from AmberTools (19).

The OpenMM (20) python library was utilized to run all molecular dynamics simulations. The Amber ff14SB force-field (21) was used for all protein atoms. Each simulation was minimized using limited memory Broyden-Fletcher-Goldfarb-Shanno optimization, heated, and equilibrated. Production simulations were performed under NPT conditions controlled using a Monte Carlo barostat (1 bar) and Langevin thermostat (2 fs timestep and 1 ps⁻¹ collision frequency). Constraints were applied to bonds involving hydrogen and the system equilibrated for 500 ns using a 2 fs timestep. For all simulations, particle mesh Ewald was applied to electrostatic interactions, and all nonbonded interactions were cut off at 10 Å.

Ligand parametrization

Each ligand was optimized at the B3LYP/6-31G(d) level of theory using Gaussian 16 (revision C.01) (22). Following optimization, each ligand was assigned GAFF (23) atom types and AM1-BCC charges using the Antechamber utility from AmberTools (19).

Docking

AutoDock Vina (24) was used to generate docked poses of each ligand docked to structures obtained from a 1 μs simulation of apo FABP7. Poses within 1.0 Autodock Vina scoring unit of the minimum were then clustered according to the position of the binding pocket residues and the bound ligand to 30 clusters using Ward’s hierarchical agglomerative clustering algorithm, producing several unique binding poses (Fig. S1) using the cpptraj (25) from AmberTools (19).

Simulation of docked poses

The medoids of each cluster were used as starting structures for molecular dynamics simulations. At least 50 μs simulation time was obtained per system for a total of 800 μs sampling time.

Hidden Markov modeling

As it is difficult to run individual simulations that reach the long timescales needed to study biological phenomena, it is common to run many shorter, but still relatively long, simulations and extract a kinetic model using tools like hidden Markov models (HMMs) or Markov state models (MSMs). Briefly, these tools (1) identify slowly relaxing degrees of freedom, (2) discretize the system along these slow degrees of freedom, and (3) extract a kinetic model based on the transition statistics between the discrete states (26). HMMs and MSMs make different trade-offs. MSMs are more detailed and can provide a richer kinetic picture, but they can be more challenging to fit when sampling is limited. HMMs provide a more coarse-grained view of kinetics, but they are often better able to approximate the dynamics of high-dimensional and difficult-to-discretize biological systems when data are scarce (27). Despite having at least 50 μs simulation for each system, we found that it was difficult to reliably build MSMs, so the bulk of our analysis uses HMMs.

HMMs were generated from simulations of docked poses using the PyEMMA v.2.5.7 (28). The input molecular dynamics coordinates were first analyzed with time-independent coordinate analysis (tICA) (29) to reduce dimensionality and isolate slowly evolving modes. These components were clustered using k-means to 100 clusters using PyEMMA default settings. The clusters were used as input to generate HMMs. Implied timescale plots and Chapman-Kolmogorov tests were used to select the number of hidden macrostates and the shortest lag time that generated stable models (Fig. S2). For each generated HMM, this resulted in ∼2–4 states with a lag time of 10–20 ns. From the HMMs, transition times and stationary distributions of each macrostate were calculated.

Experimental

Cloning

FABP7 gene sequences for wild-type FABP7 and R126L, Y128F, and R126L/Y128F mutants were ordered from Life Technologies (Carlsbad, CA, USA) in the pENTR plasmid that is compatible for GATEWAY cloning. Each sequence included a TEV cleavage site followed by a 4xGly spacer at the N-terminus. The LR Clonase reaction was used to clone all sequences into the pHGWA expression vector (30) with an N-terminal 6xHis tag.

Protein expression, purification, and delipidation

Proteins were expressed in Escherichia coli BL21 (DE3) cells and induced using 1 mM IPTG. Cell pellets were resuspended in lysis buffer (25 mM imidazole, 3 mM DTT, 50 mM sodium phosphate, 250 mM NaCl [pH 7.8]) and lysed with sonication. The soluble fraction was separated from cellular debris by centrifugation at 40,000 × g. 6xHis-tagged FABP7 was then immobilized with an Ni-NTA IMAC column (AKTA Pure, GE Healthcare, Chicago, IL, USA) and eluted in high imidazole solution (250 mM imidazole, 50 mM sodium phosphate, 250 mM NaCl [pH 7.8]). The protein was incubated overnight with TEV (VWR, Radnor, PA, USA) at 4°C, followed by the removal of the cleaved His tag and TEV protease using Ni-NTA beads (GE Healthcare). Because FABP7 binds to hydrophobic molecules, the recombinant protein must be delipidated from endogenous bacterial lipids. This is performed by two successive incubations of FABP7 with Lipidex-5000 beads (PerkinElmer, Waltham, MA, USA) for 1 h each at 37°C. Purified protein was then dialyzed into a final solution (20 mM sodium phosphate, 120 mM NaCl [pH 7.4]). Delipidation was validated by measuring the apparent binding affinity to 1-anilino-8-naphthalene sulfonate (ANS) in samples spiked with a known amount of OA (Table S1; supporting material).

¹⁵N isotopically labeled protein used for NMR experiments followed the same protocol except cells were grown in M9 minimal media with ¹⁵N-ammonium chloride. ¹⁵N-FABP7 was dialyzed into NMR buffer (20 mM sodium phosphate [pH 7.4]) for all NMR experiments.

NMR

Spectra were collected using a 600 MHz Bruker Avance spectrometer equipped with a ¹H, ¹⁵N, ¹³C TXI probe with a single axis z-gradient. Data processing was completed using NMR Pipe and NMR Draw and analyzed with NMR View. All spectra were collected at 25°C and referenced to sodium trimethylsilylpropanesulfonate.

¹H-¹⁵N heteronuclear single quantum coherence (HSQC) spectra were collected with a 50 μM FABP7 concentration. Fatty acids were titrated into a 50 μM FABP7 sample from a DMSO stock to a maximum of 2.5% DMSO volume content. Control spectra of FABP7 with 2.5% v/v DMSO showed no changes from DMSO-free spectra. Assignment of amide backbones was taken from the literature for apo-, DHA-, and OA-bound protein complexes (31,32). SA-bound FABP7 assignment of ¹H-¹⁵N HSQC spectra was completed by following peaks over the course of the titration experiment. The following peaks could not be unambiguously assigned were excluded from this study: S13, N15, F27, A28. K37, P38, G46, T56, D89, and D110.

Weighted chemical shift perturbations (CSPs) between apo- and fatty acid-bound spectra were calculated as follows:

$$CSP=\sqrt{{\Delta {\delta }_{HN}}^2+{0.15{\Delta \delta }_N}^2,}$$

where $\Delta {\delta }_{HN}$ and ${\Delta \delta }_N$ are the chemical shift differences between apo and holo spectra for the proton and nitrogen dimensions, respectively. For residues that have multiple peaks in the apo spectra, the most intense peak was used for the CSP calculation.

Fatty acid serial dilutions

OA, DHA, and ANS was purchased from Sigma (Burlington, MA, USA) and SA from Larodan (Solna, Sweden). Stock solutions of fatty acids were made up in DMSO. Titrations consist of a series of 1:1 serial dilutions in DMSO. The serial dilutions were mixed with the protein to solution to a maximum 2% DMSO content. All dilution series were completed in low-protein-binding Eppendorf tubes.

Microscale thermophoresis

Microscale thermophoresis (MST) measurements were carried out using a Monolith NT.115Pico instrument. Measurements were carried out in MST buffer (20 mM phosphate, 120 mM NaCl [pH 7.4], in 0.005% Tween-20) using premium capillaries (Nanotemper, Cambridge, MA, USA). Proteins were labeled with RED-tris-NHS second-generation dye (Nanotemper) following the supplier protocol. Measurements were taken with the instrument set to high MST power and ∼30% LED power. A solution of 20 nM FABP7-NHS-RED-Tris-labeled FABP7 was combined with fatty acid dilution series. Samples were incubated for 30 min at measurement temperature before loading samples into MST capillaries. Analysis of MST traces was completed using MO.Affinity Analysis v.2.3 (Nanotemper) (33). All measurements are completed in triplicate.

A Bayesian Van’t Hoff analysis of MST data was performed as described in the supporting material.

ANS displacement

All assays were completed at 25°C using a Cytation 5 plate reader (BioTek, Winooski, VT, USA). Binding interaction measurements were performed in the same manner for FABP7 wild-type and mutant proteins. ANS signal intensity increases 40-fold when in the hydrophobic environment of the protein binding pocket. The binding affinity was determined as previously described (13).

Full experimental and computational methods are available in the supporting material.

Results and discussion

• 1.FABP7 does not display a binding-affinity preference for polyunsaturated fatty acids at physiological temperature.

Previous literature reports a preferential interaction between FABP7 and polyunsaturated fatty acids (e.g., DHA) compared with monounsaturated (e.g., OA) and saturated fatty acids (e.g., SA) (5,16). To characterize the affinity of these interactions, we performed three different binding assays: MST, fluorescence anisotropy, and ANS displacement. Further, we performed binding-affinity experiments at multiple temperatures to understand the thermodynamic driving forces.

Fig. 2 shows that at 25°C, we measured tighter binding affinities for DHA and OA over SA, which is consistent with the proposals that binding favors unsaturated over saturated fatty acids (5,16,34). However, FABP7’s preference for OA and DHA over SA unexpectedly disappears at 37°C.

Figure 2.

FABP7 preferentially binds OA and DHA at 25°C but not at 37°C. Binding curves measured for SA (green), OA (blue), and DHA (orange) from microscale thermophoresis (MST) measurements at (a) 25°C and (b) 37°C show preferential binding that is not seen at physiological temperatures. Bars show standard deviation from triplicate measurements. (c) Van’t Hoff analysis of MST binding data (circles) suggests temperature-dependent binding for OA and DHA but not SA. Literature data using the Lipidex assay (triangles) obtained at 4°C are shown for comparison, with error bars indicating the spread of the literature data (16).

Van’t Hoff analysis of MST binding assays at 25°C, 29°C, 33°C, and 37°C (Figs. S3–S7) reveals a strongly temperature-dependent binding affinity for OA and DHA, while there is little change in binding affinity for SA at different temperatures (Fig. 2). The Van’t Hoff analysis reveals that the binding of fatty acids is driven by enthalpy, which is consistent with previous isothermal titration calorimetry studies (16). The binding of OA and DHA weakens with increasing temperature indicating that binding is entropically unfavorable.

Previous work suggest FABP7 exhibits a strong binding preference for unsaturated (OA, DHA) over saturated fatty acids (SA) (5,16,34), whereas our data indicate that FABP7 exhibits little selectivity at 37°C (Fig. 2). This discrepancy may arise because previous studies were based on Lipidex assays (16,34), which include a long incubation step at 4°C. When we extrapolate our data to 4°C, we also observe a preference for unsaturated fatty acids in agreement with previous data (Fig. 2).

• 2.FABP7 affinity for fatty acids is resistant to mutations of binding residues.

X-ray crystal structures of FABP7 bound to fatty acids show that R126 and Y128 form H-bonds with the bound fatty acid, and previous work indicated that a F104A/R126A/Y128A triple mutant showed no discernible binding for DHA at 37°C in vivo (5), suggesting that these amino acids are essential for fatty acid binding.

We produced and purified FABP7 with R126L, Y128F, or R126L/Y128F mutations. The motivation behind these specific mutations is to remove the ligand H-bonding ability of these residues while minimally perturbing the structure. Our aim was to determine the importance of these residues for binding and to observe any ligand-dependent differences, which could provide insight into binding and transport. We expected these mutations to dramatically alter the binding affinity of FABP7 toward fatty acids in vitro, as previous work suggests that these residues may be critical for fatty acid binding (15).

We used ANS displacement assays to measure the binding affinity for all three mutants and MST for R126L and R126L/Y128. The thermophoretic signal was too small to reliably determine an accurate binding affinity for Y128F using MST. We note that the absence of a thermophoretic signal does not indicate an absence of binding, as it could arise from the bound and unbound states having similar thermophoretic mobility.

Our binding assay results indicate that the R126L, Y128F, and R126L/Y128F mutations have only modest effects on fatty acid binding. Specifically, for a given fatty acid, the binding affinity is within an order of magnitude regardless of the mutation (Figs. 3 and S4–S6). This result is surprising because the fatty acid carboxylate consistently coordinates to R126 and Y128 in x-ray crystal structures (16), leading us to expect much larger changes in affinity.

Figure 3.

FABP7 binding affinity for fatty acid is resistant to R126L and Y128F mutations. Each panel shows the binding affinities for a particular variant of FABP7 binding to fatty acids at 25°C: (a) ANS displacement assays and (b) MST. Error bars indicate standard deviation across three replicates. MST experiments with Y128F resulted in minimal thermophoresis, precluding analysis.

Combining our temperature series and mutagenesis binding-affinity data, we conclude that 1) FABP7 does not have a strong preference between the three fatty acids tested at physiological temperature and 2) fatty acid binding is partially resistant to mutation of critical binding residues. The lack of specificity toward DHA is perhaps not surprising given that FABP7 can accommodate many structurally diverse ligands (Fig. 1).

• 3.Apo FABP7 is highly dynamic with H3- and gap-unfolded conformational states.

To investigate how binding to different ligands changes the structure and dynamics of FABP7 and how this might be linked to differential transport, we used a combination of ligand-titration protein NMR experiments and molecular dynamics simulations.

We first establish a point of reference by examining the dynamics of apo FABP7. Our ¹H-¹⁵N HSQC NMR spectra reveal that several residues (G24, V25, G33, T36, F57, K58, N59, and T60) produce multiple peaks, suggesting the presence of multiple metastable states that exchange slowly relative to NMR timescales (Figs. 4, A and B, and S8). Previous studies have reported that several residues within the portal region have multiple peaks for both FABP7 (31,35) and FABP3 (36).

Figure 4.

The portal region of apo FABP7 is highly dynamic, and ligand binding leads to reduced dynamics. (a) Location of residues that are assigned multiple HSQC peaks. (b) Example showing multiple peaks for T60. (c) Hidden Markov model-generated kinetic network of metastable states for apo FABP7. Red color of protein indicates unfolding in that region.

Using molecular dynamics simulations, we sought to understand the protein dynamics that give rise to the multiple HSQC peaks. Using HMMs, we could identify alternative conformational states characterized by the partial unfolding of either H3 or βD (Fig. 4 C). The H3-unfolded state is associated with less persistent backbone H-bonds between residue pairs V32/T36 and T29/G33 compared with the native fold. Similarly, for the partially unfolded βD state, T54 and T60 have diminished H-bonding (Figs. S9 and S10). These partially unfolded states are associated with flexible βCD/βEF turns and α-helices (Fig. S11). The multiple ¹H-¹⁵N HSQC peaks observed for T36 and T60 may arise due to states with partial unfolding to H3 or βD.

Interestingly, multiple peaks are observed in the turn between H2 and H3, yet the simulations do not predict significant structural changes in this region and only predict unfolding in the last turn of H3 (residues 29–33). The differences may be caused by insufficient sampling of a fully unfolded state that is captured by the NMR experiments. Our coarse-grained three-state HMMs underestimate the transition times between the states (low μs) compared with the existence of multiple 2D ¹H-¹⁵N HSQC peaks for a residue, which instead suggest a high μs to ms timescale. Unfortunately, due to the long lifetimes of the states, we have insufficient sampling to estimate an MSM that captures the folded → H3 unfolded process, which would allow a better estimation of the state lifetimes compared to our HMM. Nevertheless, we can build an MSM that captures the folded → gap unfolded transition, revealing that we underestimate this timescale in our HMM (∼10 μs in MSM compared with ∼1 μs in HMM; Fig. S12). It is likely that we similarly underestimate the folded → H3 unfolded transition timescale in our HMM.

Unfolding of H3 and the gap region has also been observed for other apo FABPs. Extensive NMR and simulation studies performed on intestinal-FABP (FABP2) (35,37,38,39,40) and epidermal-FABP (FABP5) reveal that (36,41,42,43), like FABP7, these proteins also undergo H3 and gap region unfolding. For FABP2, ¹⁵N relaxation dispersion and chemical shift saturation transfer NMR experiments reveal minor populations with structural changes to H3 and gap region residues (37). X-ray crystal structures of FABP5 suggest higher H3 flexibility (41), while data from hydrogen-deuterium exchange experiments (37,38,39) on both proteins show lower protection factors and increased flexibility for residues in H3.

Simulations of FABP2 and isolated secondary structure motifs reveal states with partially unfolded H3 (38), which is consistent with reduced FABP2 hydrogen-deuterium exchange protection factor data and our predicted dynamics of FABP7. Equilibrium simulations on apo FABP2 revealed an intermediate along the unfolding pathway, while full unfolding of H3 was not observed due to insufficient sampling (38). Nevertheless, metadynamics simulations captured minima corresponding to the native fold, partially unfolded H3, and a fully unfolded H3 (38).

Combining our data with the apo structural dynamics of FABP2 and FABP5, we propose that the unfolding of H3 or βD leads to the multiple peaks in HSQC experiments. Our predicted partially unfolded states are likely intermediates along an unfolding pathway. We suspect that our simulations are too short to capture the entire H3- or βD-unfolding pathway, which likely occurs on longer timescales.

Taken together, we provide support for the presence of an H3- and βD-unfolded states in FABP7 that appear to be common across the FABP family. Previous literature reports that for FABP2 (35,37,38,39,40) and FABP5 (41,42,43), the unfolding of H3 occurs on a similar timescale to ligand binding, which may indicate that H3 unfolding is an essential step along the binding pathway (37). It is also possible that minor states are involved other biological activities of FABP7, such as driving membrane interactions or nuclear localization.

• 4.Ligand binding stabilizes the portal region of FABP7.

Our 2D ¹H-¹⁵N HSQC spectra reveal several differences between apo and holo FABP7 (Figs. S13–S16), especially in the portal region (βCD loop ends and H3; Fig. 1). Regardless of the fatty acid bound, the largest CSPs between apo and holo FABP7 are located at the βCD loop ends (T54, N59, and T60) and H3 (G33 and T36; Figs. 5 and S11). Notably, the residues with the largest CSPs are identical to or located near residues with multiple peaks in the apo state, including G33 and T60. Furthermore, the ligand-saturated HSQC spectra no longer show multiple backbone peaks for these portal region residues, indicating that ligand binding has a large effect on the protein heterogeneity or dynamics within these regions (Fig. S8). This supports the idea that ligand binding modulates these residues’ conformational state.

Figure 5.

Fatty acid binding leads to perturbations in portal region residues. (A) Backbone chemical shift perturbations between apo- and DHA-, OA-, and SA-bound spectra are mapped onto the structure of FABP7. The largest perturbation is observed with portal region residues for all FAs. Binding of DHA shows more residues in H2-H3 and βCD/βEF turns with greater CSPs. Absolute CSP bar plots for each fatty acid can be found in Fig. S16.

To analyze how ligand binding might change FABP7 protein dynamics, we analyzed molecular dynamics simulations using a dimensionality reduction method, TICA. For wild-type/apo FABP7, we inputted key protein-protein distances as features for TICA and plotted the first two components (Tables S2 and S3; top left panel in Fig. S17). These components are orthogonal coordinates corresponding to the slowest-evolving protein dynamics, which qualitatively map to gap unfolding (primarily βD; y axis) and H3 unfolding (x axis) for our system (Fig. S9). The wild-type/apo simulations (top left panel of Fig. S17) show a distinct basin corresponding to the native fold. The native fold is separated by a small barrier from gap-unfolded structures, while structurally diverse conformations with low populations that involve the unfolding of H3 are also present. These alternative minor states are predicted to be associated with the portal region residues with multiple HSQC peaks in the apo spectra. When the simulations of holo FABP7 are projected onto the slowest-evolving apo FABP7 TICA components, we observe a disappearance of the partially unfolded gap and H3 structures.

The increased stability of the native fold is driven by tighter intraprotein interactions. Contact map occupancies show that contacts between βCD-H3 and βCD-βEF become more extensive upon ligand binding (Fig. 6). F57 tightly associates with T29, V32, and G33, while several contacts are more persistent between K58/T60 and the βEF turn. Interestingly, we also observe a decrease in protein fluorescence anisotropy upon ligand binding (Fig. S7). We interpret this as being associated with the more compact closed conformation tumbling faster in solution. The reduced conformational heterogeneity in the portal region upon ligand binding can explain why residues (G33, T36, T54, F57, N59, and T60) with multiple peaks in the apo HSQC spectra converge to single peaks in the holo HSQC spectra.

Figure 6.

Ligand binding tightens the contacts between portal region residues. (a) Color scheme. (b and c) Shaded squares indicate contact occupancy (<4.5 Å separating closest two heavy atoms) between two residues in (B) βD and H3 or (c) βD and H3.

Our combined experimental and theoretical data suggest that ligand binding stabilizes the portal region and prevents the unfolding of H3 and the gap region. Similar decreased protein conformational plasticity upon ligand binding has been identified for other FABP family members, including I-BABP (42) and H-FABP (36).

• 5.Changes to the dynamics of H2-H3 may help explain fatty acid-dependent biological outcomes.

Despite only small differences in binding affinities between the ligands at 37°C, NMR data and molecular dynamics simulations reveal that each fatty acid has a unique effect on the conformation of FABP7.

Compared with apo FABP7, OA-bound FABP7 has the smallest CSPs (Figs. 5 and S16). SA binding results in similar perturbations to portal region residues as OA but with a larger perturbation to N59.

Intriguingly, DHA results in more significant perturbations to residues in H3 and the βCD/βEF turns (Fig. 5). This includes notable side-chain perturbations to NLS residue Q31 and nearby residue N15 that are not observed when the other fatty acids are bound (Fig. 7). Therefore, we speculate that DHA binding could uniquely affect the structure and dynamics of H2/H3, including the NLS, which could possibly explain the differential localization of FABP7 bound to DHA.

Figure 7.

Some peaks are differently affected by DHA binding compared with OA and SA. ¹H-¹⁵N HSQC of the apo and holo wild-type FABP7 shows that DHA binding leads to chemical shift perturbations that are not observed for OA or SA. (a) Q31 backbone displays a modest chemical shift perturbation upon DHA binding. (b) Q31 and N15 side chains display large chemical shift perturbations upon DHA binding.

However, significant caution is warranted, as the magnitude of CSP depends on a variety of factors (44). Furthermore, whereas we observe a large change in the side-chain chemical shift for NLS residue Q31, the backbone chemical shifts for K21, R30, and Q31 are small. (The sidechains of K21 and R30 are not visible.)

Our simulation data correlate with the significant perturbations observed by NMR for DHA binding to FABP7 compared with the other ligands. Specifically, DHA binding leads to increased flexibility and weaker contacts among gap-region residues compared with OA or SA (Fig. 6). Analysis of the conformational heterogeneity shows that DHA-bound FABP7 is more flexible than apo FABP7 in the βCD, βEF, and H3 regions (Fig. S11). However, the differences for H3 are modest, and the simulations do not provide a clear explanation for the side-chain shifts of Q31 and N15. Among several possible explanations for this discrepancy, it is possible that DHA binding perturbs the dynamics of H3 on timescales that we have not sampled with our simulations, as was the case for apo FABP7.

Our data reveal that DHA binding leads to changes to H2, H3, and βCD/βEF turns of FABP7 that do not occur upon OA or SA binding. While there are significant caveats to our NMR and simulation results, we speculate that DHA-induced changes to the H2-H3 conformational dynamics could contribute to the nuclear localization of FABP7.

• 6.Binding pose heterogeneity explains changes to protein dynamics and mutation-resistant binding.

It is unclear why mutation of key binding residues has little effect on binding of the three fatty acids studied (Fig. 3). Indeed, previous studies indicate that the F104A/R126A/Y128A mutation dramatically reduces FABP7’s binding affinity for fatty acids (5). Even though our mutations are roughly isosteric (R126L and Y128F) and may have less of an impact than mutation to alanine, we expected that the effect of binding affinity would be larger than what we observed due to the removal of charged and H-bonding groups. Using molecular dynamics and HMMs, we analyze how the ligands bind to wild-type and mutant FABP7 to understand why R126L and/or Y128F mutants maintain tight binding to fatty acids.

We observe three unique binding conformations. The two most populated conformations have an H-bond between the ligand and both R126 and Y128 (Fig. 8, top and bottom left). These two poses differ in their tail conformations. The most populated conformation (Fig. 8, top) orients the lipid tail toward βD and is most similar to the bound conformation adopted by OA and DHA in x-ray crystal structures (Fig. S18) (16). Specifically, OA adopts a U-shape conformation, while the DHA tail adopts a helix-like conformation, and interactions with F16 and F104 stabilize both tail conformations. The binding pose of DHA does not precisely match the crystal structure with the tail not looping across the gap region, which may be due to the flexibility and heterogeneity of the lipid tail. The flexible lipid tail is consistent with NMR data that suggest the C12-C18 carbons of the FABP3-bound fatty acids become disordered at room temperature (45).

Figure 8.

Each fatty acid exhibits a dynamic lipid tail when bound, and an alternative carboxylate binding site exists for DHA and SA. Hidden Markov model-generated kinetic network of ligand binding poses for DHA, OA, and SA. For clarity, only the pose of DHA is shown in orange and key binding residues in gray.

A third binding pose has the ligand carboxylate interacting with R78 and K100, with the tail extending toward H3 (Fig. 8, bottom right). This conformation is more probable with DHA (11% occupancy) than SA (2% occupancy) and is not observed at all for OA. DHA’s higher occupancy may be due to its tail length and flexibility and stabilized by hydrophobic interactions with T29, L23, V25, F16, and F57. The location of these interactions is near H3, and they help explain why larger significant HSQC CSPs occur upon binding DHA compared with the other fatty acids.

To understand why R126L and/or Y128F mutants had little impact on the binding affinity of FABP7 toward OA, DHA, or SA (Fig. 3), we examined the ligand-protein dynamics using HMMs. We summarized the data by analyzing 1000 HMM-generated structures from each unique system with principal-component analysis (PCA) (Fig. 9). Subsequently, we combined the structures from all systems, calculated nine distances that correspond to different ligand-binding sites, and performed PCA on the results to project each system onto a consistent coordinate system. The structures are colored according to their varying conformations (Fig. 9).

Figure 9.

R126L and/or Y128F mutations lead to fatty acids sampling alternative conformations more frequently. PCA was used to combine coordinates corresponding to different ligand binding sites for 1000 HMM-generated structures from each system. (a) Representative ligand binding poses. (b) The structures from each system are projected onto the PCA coordinate system and coloured according to the ligand pose, as in panel (a).

For the wild-type systems, PCA identifies a fourth binding conformation with the carboxylate H-bonding to T36 and T55. Based on our HMMs, this pose is not kinetically distinct from other poses and rapidly transitions to structures where the ligand interacts with R126 and Y128. Despite this new binding conformer, the observed ligand binding poses usually interact with R126/Y128.

In general, the R126L and Y128F mutations increase the heterogeneity of the ligand binding poses adopted in the FABP7 binding pocket (Fig. 9). Each mutant increases the frequency of the T36/T55- and R78/K100-interacting binding poses being adopted, with the Y128F mutation having a smaller effect compared with R126L. The diversity within each major binding pose increases depending on the mutation. Specifically, with R126L mutation, there is a more significant overlap between the R126/Y128-interacting binding pose and more heterogeneity within the other poses.

For both apo and holo FABP7, the R126L mutation has a larger effect on the diversity of protein conformations than the Y128F mutation (Fig. S17). This is because R126 forms water-mediated H-bonds with G33 and N34 backbones (Fig. S19), while these interactions are not present for the R126L mutant. Losing these H-bonds destabilizes the secondary structure of H3, leading to a higher proportion of H3-unfolded structures and suggesting that R126 may also regulate the conformation of H3 in addition to its ligand-binding role. The water-mediated H-bond between R126 and H3 is also present in FABP3 (46).

The mutation-resistant binding we observe via MST (Fig. 3) is explained by the ability of FABP7 to accommodate fatty acids in several distinct conformations, which permits binding in the absence of H-bond donors at the R126 or Y128 positions (Fig. 9). Even for wild-type FABP7, there is still considerable structural diversity in the binding poses observed. More broadly, the ability for FABP7 to bind fatty acids in several conformations may be a factor in its promiscuous binding to structurally diverse ligands, including endocannabinoids (10,11), phytocannabinoids (12), and dyes (13) that may locate to different regions of the binding pocket.

Conclusions

By using simulations and a variety of biophysical techniques, we provide fundamental insight into how FABP7 binds fatty acids and how binding affects the protein structure. We establish that FABP7 has no preference between binding SA, OA, and DHA at physiological temperature. Our simulations and NMR data reveal that the portal region of apo FABP7 is highly dynamic, and ligand binding leads to unique changes to these dynamics depending on the fatty acid bound. Notably, when DHA is bound, the DHA structure of NLS-containing H3 is uniquely perturbed, which may help explain why the FABP7-DHA complex is preferentially relocated to the nucleus. We also establish that FABP7’s fatty acid binding affinity is resistant to mutation of key binding-site residues and that this resistance is conferred by FABP7’s ability to bind fatty acids in several unique binding poses. Overall, we provide comprehensive insights into how FABP7 interacts with key ligands and hints toward how these interactions may govern FABP7’s behavior within the cell.

Author contributions

I.B. conceptualized the study, performed all NMR experiments, analyzed NMR data, and wrote and edited the article. S.L. conceptualized the study, performed all computational modeling and analysis, and wrote and edited the article. M.R.-Y. conceptualized the study, secured funding, performed all ANS, MST, and melting temperature experiments, analyzed the corresponding data, and wrote and edited the article. T.S. analyzed experimental data and edited the article. H.I. analyzed NMR data and edited the article. H.J.V. analyzed NMR data and edited the article. J.L.M. conceptualized the study, secured funding, supervised the study, analyzed experimental and computational data, and wrote and edited the article.

Editor: Michael F. Brown.