Modulating Ligand Dissociation Through Methyl Isomerism in Accessory Sites: Binding of Retinol to Cellular Carriers

Due to the poor aqueous solubility of retinoids, evolution has tuned their binding to cellular proteins to address specialized physiological roles by modulating uptake, storage, and delivery to specific targets. With the aim to disentangle the structure-function relationships in these proteins and disclose clues for engineering selective carriers, the binding mechanism of the two most abundant retinol-binding isoforms was explored by using enhanced sampling molecular dynamics simulations and surface plasmon resonance. The distinctive dynamics of the entry portal site in the holo species was crucial to modulate retinol dissociation. Remarkably, this process is controlled at large extent by the replacement of Ile by Leu in the two isoforms, thus suggesting that a fine control of ligand release can be achieved through a rigorous selection of conservative mutations in accessory sites.

. Representation of the retinol-CRBP complex. The structural fold of CRBP consists of 10 antiparallel β-strands (A-J) and two short α helices (I and II). The regions that define the entry portal site are highlighted in yellow. Retinol (RTL) is shown as blue sticks.
To investigate the binding mechanisms in rat CRBP-I and II and explore their functional implications, a detailed analysis of apo and holo forms was performed by combining extended atomistic molecular dynamics (MD) simulations and parallel-tempering metadynamics (PT-metaD). We characterized the conformational flexibility of the two isoforms as well as the free energy surfaces for the opening/closing of the portal site in both apo and holo forms, and the formation/breaking of interactions between retinol and protein in the holo species. Furthermore, surface plasmon resonance (SPR) was used to examine the thermodynamics and kinetics of retinol binding. Overall, both theoretical and experimental results provide detailed insight into the binding mechanism, disclosing a linkage between retinol binding and the flexibility of the entry portal, particularly regarding the methyl isomerism between Ile and Leu in this accessory site of the two isoforms.
The conformational flexibility of apo and holo forms of CRBP-I and II was examined from three independent MD simulations (5, 3 and 3 s) performed for each system, covering a total of 44 s.
The root-mean square deviation profiles supported the structural stability of the simulated systems along the trajectories (SI Figure S2). In both isoforms, the presence of retinol reduced the structural fluctuations of the protein, as expected from the interactions formed with residues in the binding cavity ( Figure 2; see also SI Figures S3-S6). However, the pattern of residue fluctuations differed in the two isoforms. While apo-I showed increased fluctuations in loops βE-βF, βG-βH and at less extent βC-βD, apo-II exhibited larger fluctuations in helices αI-αII and at less extent in loops βC-βD and βD-βE ( Figure 2A). Remarkably, most of these elements define the entry portal site, suggesting that the two isoforms differ in the dynamics of key structural elements implicated in the entry/release of retinol to/from the binding cavity. 24,25 Essential Dynamics (ED) analysis was employed to gain insight into the distinct flexibility of CRBP-I and CRBP-II. The analysis was performed for the backbone atoms of residues 7-134 to avoid the noise due to the mobile parts at the N-and C-termini. The first essential mode ( Figure   2B) accounted for 15-25% of the entire structural variance, and generally was 2-fold larger than the contribution explained by the second mode. The apo systems exhibited larger structural deformations, especially in the entry portal site, although helices αI-αII were stiffer in apo-I than in apo-II. On the other hand, holo systems were more rigid than their apo forms, as noted in the lower extent of the backbone motions. However, the decrease in conformational flexibility of the protein backbone did not affect similarly the two isoforms. In fact, the rigidification of the entry portal site was more important in holo-I than in holo-II (SI Table S1). At first sight, these results seemed to be in contrast with NMR H/F exchange experiments 15 that suggested a larger flexibility in both apo and holo states of CRBP-II relative to CRBP-I. However, it is worth noting that residues in the βE-βF loop of apo-I could not be assigned, while present results reveal that this structural element has a crucial influence on the dynamics of the portal site. Indeed, upon exclusion of the βE-βF loop in ED analyses, CRBP-II was slightly more flexible than CRBP-I in both apo and holo states (SI Table S1 and Figure S7), thus reconciling the experimental findings about the dynamics of the two isoforms and our results from MD simulations. To estimate the differences in the dynamics of apo and holo systems, the conformational entropy was evaluated for the whole system as well as separately for the entry portal site and the protein core, formed mainly by the -barrel, using the procedure by Harris et al (See Supporting Information for details). 26 As expected, the results (Table 1; see also Table S2 and Figure S8) confirmed that holo systems were less flexible than apo ones, and pointed out that the decrease in entropy was larger for CRBP-I (0.56 Kcal mol -1 K -1 ) than for CRBP-II (0.14 Kcal mol -1 K -1 ).
Furthermore, the conformational entropy (S ∞ ) obtained for apo-I was larger (by 0.35 Kcal mol -1 K -1 ) than for apo-II, whereas the difference between the holo species was reduced to 0.07 Kcal mol -1 K -1 . Noteworthy, the entropy difference between apo-I and apo-II and between holo-I and holo-II was mainly due to the differences in the entry portal site (apo: 0.27 Kcal mol -1 K -1 ; holo: -0.06 Kcal mol -1 K -1 ). Overall, these results confirm that the distinct patterns of conformational flexibility between the two isoforms primarily arise from the entry portal site. Table 1. Entropy (S ∞ ) and entropy difference (ΔS) of the whole protein, and its core and entry portal site, determined from the analysis of the 5s MD trajectory. Values (Kcal mol -1 K -1 ) determined considering only the backbone atoms.

System
Protein [a] Core [b] Portal site [c] S ∞ (apo-I) [ Helices αI-αII and loops βC-βD and βE-βF. [d] The error of the conformational entropy was estimated from the standard deviation of S ∞ obtained in the fitting at increasing simulation windows, with an upper value of 0.008 Kcal mol -1 K -1 for the whole protein and the portal site, and 0.005 Kcal mol -1 K -1 for the protein core.
Since the stiffness of the portal site in holo-I was higher than in holo-II, we hypothesized that the difference in binding affinity between CRBP-I and II might arise from a larger residence time of retinol in the former isoform. To address this question, PT-metaD was used to evaluate the free energy landscape for the opening/closing of the entry portal site in apo and holo systems, and the binding/unbinding of retinol to/from the holo systems. In order to take into account the larger structural flexibility of the open state compared to the closed one, the free energy change for the opening/closing of the entry portal site was estimated by averaging the values determined from three separate calculations, each relying on the use of distinct open structures chosen as reference systems (SI Figures S9-S11). The results pointed out that the opening of the portal site in the apo state of CRBP-I and II was very similar and close to 5.5  0.2 Kcal mol -1 ( Figure 3A). However, the presence of retinol in the -barrel had a marked influence on the opening of the portal site in holo-I, as this process was disfavoured by 3.8 Kcal mol -1 compared to apo-I. Remarkably, the presence of retinol led to a modest increase in the cost of opening the portal site in holo-II (by only 0.7 Kcal mol -1 ) relative to apo-II. These findings agree with the larger decrease in conformational entropy found for CRBP-I relative to CRBP-II upon retinol binding (see above and Table 1).
The analysis of the structures sampled during the opening of the portal site reveals that there is a slight rearrangement of retinol in the binding pocket, although the ligand remains trapped in the interior of the -barrel after opening of the loop in both CRBP-I and II (Figure 4). However, whereas the rearrangement of retinol occurs a fast process during the first 100 ns of the loop opening for CRBP-II, a slower process that involves a gradual rearrangement of retinol is observed for the loop opening in CRBP-I. This suggests the presence of stronger interactions between the ligand with the residues of the portal site in this latter isoform, as will be discussed later.  The snapshots were taken during the last 500 ns of the pT-metaD simulations at 300 K.
Overall, the combination of the free energy estimates obtained for the opening/closing in apo and holo states, and the binding/unbinding of retinol from holo species, indicates that the affinity of CRBP-I for retinol is 2.4 Kcal mol -1 more favourable relative to CRBP-II ( Figure 4).
Noteworthy, this agrees with the experimentally observed greater affinity of retinol for CRBP-I, as the predicted affinity lies between the range of experimental values, which vary from an upper threshold of <100-fold 12 to a lower limit of 3.3-fold greater affinity for CRBP-I. 23 Remarkably, our results also revealed that the difference in binding affinity is mainly determined by the opening/closure of the entry portal site in the holo state. This suggests that the larger cost of opening the holo-I complex cannot be attributed to the interactions formed by the portal site with the rest of the protein, as the free energy changes determined for the opening of the portal site in the apo species are highly similar in the two isoforms (Figure 4; see also SI Figure S12). Therefore, it may be speculated that the interactions formed between retinol and the entry portal site in the holo species are more favourable in CRBP-I than in CRBP-II, thus providing a basis to justify the larger decrease in conformational flexibility observed upon retinol binding to CRBP-I relative to CRBP-II. This assumption was confirmed from the analysis of the interaction energies between retinol and the structural elements that define the portal site in holo-I and holo-II (Table 2). Whereas the interaction energy with helices I-II and loop βC-βD was similar in the two holo systems, the interaction of retinol with loop βE-βF was 2.6 Kcal mol -1 more stabilizing in holo-I. Further decomposition into pairwise ligand-residue contributions revealed that the energy difference was mainly due to the interactions with Gly77 and Ile78 in CRBP-I, which were 2.1 Kcal mol -1 more stabilizing than the interactions with Gly77 and Leu78 in CRBP-II (SI Table S3). In contrast, other residue substitutions located in the loop βE-βF contributed less than 0.2 Kcal mol -1 , even though this can be justified from either the solvent-exposed arrangement of the side chain of these residues or the large distance from the mutated residue to retinol. In contrast, residues at position 78 (Ile in CRBP-I, Leu in CRBP-II) are located at the top of the loop βE-βF, pointing toward the interior of the -barrel, and form van der Waals contacts with the -ionone ring and the unsaturated chain of retinol ( Figure 5). Overall, these results point out that the difference in the interaction energy with retinol can be mainly attributed to the conservative mutation of Ile78 in CRBP-I to Leu78 in CRBP-II, disclosing an unexpected effect related to the methyl isomerism between the side chains of these two residues. [a] Calculations performed for 50 snapshots taken regularly in the last microsecond of the 5μs MD simulations. Helices α-I and α-II comprise residues Glu15-Leu37, and loops βC-βD and βE-βF involve residues Ser55-Asn59 and Glu73-Cys/Val83, respectively.
To confirm the impact of the Ile/Leu mutation at position 78 on the binding of retinol to CRBP-I and II, SPR was used to characterize the kinetic rate constants for the association (kon) and dissociation (koff) of retinol to CRBP-II and its Leu78Ile single-mutated variant (SI Figures S13-S14 and Table S4). The results show that the kon remains essentially unaltered for both CRBP-II and the mutated variant (Table 3). However, the koff of retinol is slowed down by a factor of 2.2 in the mutated protein. The increased residence time originated by the single-point mutation Leu78Ile agrees with the expected strengthening of the interaction of retinol with the mutated residue in CRBP-I (Ile) relative to CRBP-II (Leu), as deduced from the PT-metaD simulations and the decomposition analysis presented above ( Table 2 and SI Table S3). Furthermore, the dissociation constant (KD) is decreased by 2.8-fold in the mutated CRBP-II, which compares with the lower limit of the experimental ratio between CRBP-I and CRPB-II (3.3-fold). 23  While the selectivity of different members of cytosolic binding proteins toward distinct retinoidlike compounds has been related to the presence of specific residues in the -barrel, 27,28 present results point out that a seemingly minor chemical change related to the methyl isomerism between Ile and Leu at the portal site modulates the binding properties of retinol between closely related CRBP isoforms. The net effect is the enhanced free energy penalty associated to the closedopen transition, which would disfavour the release of the ligand and increase the residence time of retinol in the interior of the -barrel. Noteworthy, the affinity for the two isoforms is finely modulated by the differential interaction of the -ionone unit of retinol with the residue (Ile/Leu) at the top of the loop βE-βF, suggesting an unexpected role of the methyl isomerism between the two similar residues.
From a functional point of view, these results unveil a subtle regulation mechanism that underlies the distinct physiological role of the two isoforms. In enterocytes, CRBP-II plays an important, but not essential, role in assisting the transient exchange of retinol from the intestinal lumen to the lymph, making it necessary to have an efficient delivery system. In contrast, CRBP-I is highly expressed in hepatic cells, where it participates in the storage of retinol and controls its mobilization to ensure a steady supply in the blood plasma. Therefore, the strengthened interaction of retinol with the -barrel lid may have evolved as a mechanism to self-regulate long-term retinol mobilization subject to specific requirements of active retinoid metabolites to the target cells without being affected by the fluctuations of the dietary intake.
Finally, these findings demonstrate that conservative changes in specific residues at remote sites distinct from the binding pocket, which should not alter the gross structural and physicochemical features of the protein, may result in a fine-tuning of the ligand's binding properties. Thus, a