Energetic and Molecular Water Permeation Mechanisms of the Human Red Blood Cell Urea Transporter B

Urea transporter B (UT-B) is a passive membrane channel that facilitates highly efficient permeation of urea. In red blood cells (RBC), while the major function of UT-B is to transport urea, it is assumed that this protein is able to conduct water. Here, we have revisited this last issue by studying RBCs and ghosts from human variants with defects of aquaporin 1 (AQP1) or UT-B. We found that UT-B's osmotic water unit permeability (pfunit) is similar to that of AQP1. The determination of diffusional permeability coefficient (Pd) allowed the calculation of the Pf/Pd ratio, which is consistent with a single-file water transport. Molecular dynamic simulations of water conduction through human UT-B confirmed the experimental finding. From these results, we propose an atomistic description of water–protein interactions involved in this permeation. Inside the UT-B pore, five water molecules were found to form a single-file and move rapidly along a channel by hydrogen bond exchange involving two critical threonines. We further show that the energy barrier for water located in the central region coincides with a water dipole reorientation, which can be related to the proton exclusion observed experimentally. In conclusion, our results indicate that UT-B should be considered as a new member of the water channel family.


Construction and Evaluation of structural model for the human sequence
The homology model of the human sequence was based on the bovine structure (pdb: 4EZC), which was solved at 2.36 Å resolution. The sequence alignment was based on HMM profiles available in PFAM database. The resulting alignment showed that both sequences share more than 83% identity on the overall length, the identity being even larger in the pore region defined by Levin et al (92%). We also tested ProbCons that is considered as a very good tool for alignment [1]. The resulting alignment was identical. Thus, as discussed and shown in different papers the identity score is high enough to be confident in the quality of the model [2,3]. In addition, to evaluate the model quality, geometry analyses on model and X-ray structure were performed with Procheck software. The Ramachandran plot gave 95.1% of residues in the most favoured regions, 4.6% for additional allowed region and 0.3% in generously allowed regions for the model, compared to 94.6%, 5.1% and 0.3% for the Xray structure, respectively. Consequently, the quality of the model is at least, as good as the X-ray structure. We also evaluated the quality of our model using ProQM [4] and compared to the score obtained for the X-ray structure. The scores were highly similar and large enough for considering the model as a good one (0.71 for the model and 0.72 for the X-ray structure) Figure S1.

Simulation Convergence:
Convergence of each simulation was characterized by root mean square deviation (RMSD) calculation ( Figure S2) from to the structure at time t=0, i.e after minimization and equilibration with position restraints on the protein. Simulations on human model converged after 75ns of simulation and the two replica showed equivalent deviation for the human model. Regarding simulation of X-ray bovine structure, deviation was slightly smaller. The simulation convergence was also assessed using Principal Component analysis (PCA).
PCA analysis was first performed on each half of simulation to evaluate the convergence of each simulation for the human model. Cα atoms were examined in the analysis that was restricted to the equilibrium part of simulation (75-200ns). N-ter and C-ter of structure were discarded from the analysis due to the high flexibility of these regions ( Figure 5), which would demand larger conformational sampling. Analysis of the contribution of each PC showed that the 10 first PCs characterize more than 90% of fluctuation (93% and 95% for MD1 and MD2, respectively).
Firstly, the overlap between the two halves of each simulation was calculated using the covariance overlap methodology developed by Hess [5] . Covariance overlaps are 0.70 and 0.69 within MD1 and MD2 respectively, which indicates that convergence was reached within each MD.
Secondly, when PCA was performed on the whole set of data, i.e. concatenating the two trajectories, but focusing on the transmembrane region, the overlap equals 0.35. Hence, the two replica explore different conformational spaces. This value is similar to the one observed by Grossfield et al. [6] for other membrane proteins simulated on similar timescale (100ns).
When considering the transmembrane domain of a monomer, the overlap was increased significantly (0.47).
The inner products given in Figure S3 illustrate the similarity in the different cases.
Finally, when focusing on the residues defining the pore, the overlap was 0.60, showing the convergence of pore fluctuations, which is also exemplified by the similarity of permeation parameters.
In conclusion, the mechanism we examined, i.e. water transport seems to occur on a timescale compatible with the present simulation time and we were able to obtain reproducible results.
It seems that large conformational changes are not required for water transport to occur.