Ca MalDH overexpression was done accordingly to Dalhus et al. (2002). The cells were lysed by sonication in a 50 mM Tris-HCl buffered at pH 7. The crude extract was incubated for half an hour at 70°C and centrifugated for 15 min at 17,000 g. The soluble portion of the extract was loaded on a Q sepharose column equilibrated in 50 mM Tris-HCl buffer at pH 7. The protein was eluted using a linear gradient of 0–1 M NaCl. Fractions containing Ca MalDH were extensively dialyzed against 50 mM potassium phosphate buffer (pH 7) and deposited on a hydroxyapatite column equilibrated with the same buffer. The enzyme was eluted with a linear gradient of 50–1000 mM ammonium phosphate. The active fractions were pooled and concentrated by centrifugation using an Amicon PM30. They were deposited on a Sephacryl S300 gel filtration column (1 × 100 cm) and then eluted using an isocratic buffer of 50 mM Tris-HCl buffered at pH 7. The purified fractions were concentrated at 20 mg/ml and stored at 4°C.

Crystallization was performed by vapor diffusion using the hanging-drop method at 293 K. Native Ca MalDH crystals (≈500 × 400 × 400 μm³) were grown within 2 days by mixing 1.5 μL of 20 mg· mL⁻¹ protein solution and 1.5 μL of 4–14% PEG 400, 100 mM sodium acetate buffer at pH 4.6 and 40 mM cadmium acetate reservoir solution. Ca MalDH derivative crystals were obtained by a 10 s soaking of a native crystal in a 2.0 μL solution equivalent to the mother liquor containing 100 mM of GdHPDO3A lanthanide complex (Girard et al., 2003). Then the crystal was quickly back-soaked in 2.0 μL of the corresponding reservoir solution without the lanthanide complex.

Prior to data collection, native and derivative crystals were cryo-cooled in liquid nitrogen using mother liquor containing 25% PEG 400 as cryo-protectant.

Gd-derivative data were collected on a Nonius FR591 X-Ray home source (1.541 Å). Native data were collected on the FIP-BM30A beamline at the ESRF (Grenoble, France) with the X-ray beam wavelength set to 0.979 Å. Diffraction frames were integrated using the program XDS (Kabsch, 2010) and the integrated intensities were scaled and merged using the CCP4 programs SCALA and TRUNCATE (Winn et al., 2011) respectively. A summary of the processing statistics is given in Table 1.

Ca MalDH crystals belong to the P3₁21 space group with one A-D dimer per asymmetric unit leading to a solvent content of 49.5%.

Ca MalDH structure was determined de novo by the SIRAS (Single Isomorphous Replacement with Anomalous Scattering) method. As shown in Table 1, the high value of Rano clearly indicated the presence of GdHPDO3A complex binding sites, which was then confirmed by inspection of the anomalous Patterson map. Gadolinium positions were determined within the asymmetric unit using the program SHELXD (Sheldrick, 2010). Heavy-atom refinement and initial phasing were performed using the program SHARP (Bricogne et al., 2003). Phases from SHARP were improved by density modification using the CCP4 program DM (Cowtan and Main, 1996) leading to figures of merit of 0.235 and 0.793 after SHARP and density modification respectively. Automatic model building was performed with the program BUCCANEER (Cowtan, 2006) leading to an initial model consisting in 552 over the expected 618 A-D dimer residues.

The model was manually completed and improved in COOT (Emsley et al., 2010) prior to refinement with PHENIX (Adams et al., 2010). This model was then optimized through iterative rounds of refinement and model building. At the end stages of the refinement, TLS was used with TLS-groups determined with the TLSMD server (Brünger, 1992; Painter and Merritt, 2006a,b). The 1.7 Å resolution Ca MalDH final model consists in the complete (N-terminus, C-terminus and catalytic loop) residues sequence for each monomer of the Ca MalDH A-D dimer. The analysis of this final model (Table 2) showed no residues in disallowed regions of the Ramachandran plot (99.7% in preferred regions).

In order to precisely assign the 945 water molecules in the model, we allowed the PHENIX program to automatically build solvent molecule up to 5.0 Å above the protein surface, with a distance of 1.7–3.0 Å between two water molecules or between a water molecules and the coordinated residue and only if the 1.0 σ contored 2F_o−F_c electron density map was interpretable. At the end, each water molecule was manually verified in COOT.

All the figures were made by using the Pymol program: The PyMOL Molecular Graphics System, Version 1.5.0.4 Schrödinger, LLC. All electrostatic calculations were performed using the Pymol plugin for APBS (Baker et al., 2001).

The structure of Ca MalDH enzyme was determined at 1.7 Å resolution using SIRAS phasing. The asymmetric unit contains a dimer, the physiological tetramer being generated by the crystal symmetry operators of the P3₁21 space group (Figure 1). This dimer delineated A-D will serve as the reference for all comparisons through this study. Our Ca MalDH model (4BGT) does not present major fold difference compared to the previously deposited (PDB accession code: 1GUY) structure (Dalhus et al., 2002), as confirmed by a root-mean-square deviation (RMSD) value of 0.42 Å for 594 A-D dimer superimposed residues. Moreover, the mobile loop (residues 83–89, following the linear numbering of 4BGT) covering the catalytic site, as well as the residues of the N- and C- termini have been modeled in each monomer of this new Ca MalDH structure. The detailed analysis of Ca MalDH fold and stabilization mechanism based on the 1GUY model has previously been published (Dalhus et al., 2002), and thus will not be further described in this study. The striking new feature in our model is the incredibly large number of modeled water molecules, i.e., 945 for the dimer A-D, which allows a detailed analysis of water organization.

Sr MalDH shares more than 72% of sequence similarity with its non-halophilic counterpart Ca MalDH. The Sr MalDH model was obtained at a resolution of 1.55 Å, and also contains a large number of water molecules: 680 for the equivalent Ca MalDH A-D dimer (Coquelle et al., 2010). The overall structural similarity between one monomer of Sr and Ca MalDHs led to a RMSD of about 0.6 Å for 258 superimposed Cα.

Therefore, these two structures of excellent resolution, with a large number of water molecules in their solvent layers, provide a unique combination to finely compare the water organization at their surface.

A detailed analysis of the geometry of the 945 water molecules surrounding the dimeric Ca MalDH model (distance and angle) was performed and is presented in Table 3. It is outside the scope of this study to describe in great details both the geometry and interactions with the protein of all these water molecules. The role of water molecules in the folding process and stabilization of proteins has been well described in a work based on a larger set of proteins (Matsuoka and Nakasako, 2009). The most interesting feature of the water molecules in Ca MalDH structure is that 28% of them are organized in polygons (pentagons or hexagons), which can form extended clusters (Figure 2). These polygons are only observed at the surface of apolar residues. Geometrical properties of these polygons (Table 3) are in good agreement with those determined from a large statistical study using high-resolution structures (Lee and Kim, 2009).

Based on Ca MalDH water analysis, a careful inspection of the halophilic Sr MalDH hydration layer at the surface of the protein was performed to detect any water polygon. Even though 43% of Sr MalDH water molecules were considered to be superimposable with those from Ca MalDH (using a cut off distance of 1.5 Å), no polygons were observed at the surface of the halophilic MalDH. However, 14 water molecules lie in the catalytic pocket of Sr MalDH, all of which are conserved in Ca MalDH. Five are organized as a pentagon, the only one observed in Sr MalDH (Figure 3). In Ca MalDH, the same water pentagon is present, but the catalytic pocket of Ca MalDH contains an extra water molecule, which closes a second pentagon in the catalytic pocket, adjacent to the first one (Figure 3A). A black arrow indicates the missing water molecule in Sr MalDH (Figure 3B).

We therefore decided to have a closer look at surface regions where polygons are present in Ca MalDH to figure out the reasons why none are observed in Sr MalDH.

As mentioned, large networks of connected water polygons are present in Ca MalDH (Figure 2A). An example of such network is shown in Figure 4A. This network is anchored between helices α1G-α1G and αH and is made up of five pentagons and one hexagon. In the same protein region, no water polygon is observed in Sr MalDH (Figure 4B), which possesses four extra negative charges compared to Ca MalDH, due to substitutions at positions 199, 203, 283, and 285. These substitutions led to important electrostatic surface changes, with a highly negative one for Sr MalDH compared to the apolar surface of Ca MalDH (Figures 4C,D). The lateral chain of acidic residues D287 in Sr MalDH is orientated in such a conformation that the Sr MalDH hydration pattern is modified when compared to that of Ca MalDH. The data suggest that the replacement of non-polar amino acid residues by acidic amino acid in a halophilic protein modifies properties of the hydration shell. Around apolar surfaces of the non-halophilic MalDH, water molecules cannot form direct hydrogen bonds with the protein, and thus organize themselves as polygons with their nearest stable water neighbors. Acidic amino acids enrichment in these regions of Sr MalDH surface favors direct hydrogen bonding with water and therefore prevents polygons formation.

We also observe that water polygons formation is hampered in halophilic Sr MalDH, not only by direct acidic amino acid substitution but also by the side chain reorganization of conserved residues, as illustrated in Figure 5. In Sr MalDH compared to Ca MalDH, two acidic amino acids are observed at position 158 and 204. Glutamate at position 158 induces a direct perturbation of water pentagon P1, as previously observed. But Glutamate 204 promotes an interaction with R201 side chain, which moved to a new position that hinders appropriate hydrogen bonding geometries requested for the formation of water polygon P2 (Figure 5).

These two examples clearly illustrate the key influence of acidic amino acid enrichment in halophilic protein on the water organization at their surface; either through direct impacts or via conformational rearrangements of surrounding residues. This leads to the destabilization of almost all water polygons observed in the non-halophilic protein structure.

This is an important issue that should be discussed. Because of the chemical properties of the protein surface, the solvent composition at the vicinity of a given protein surface is different from the bulk. In a simple binary system containing water and protein without any cosolvents, such as salt or other macromolecular solutes, a hydration shell surrounds the protein. In the presence of high concentration of additional compounds such as salts, sugars, precipitating agents etc., the protein solution should be described as a ternary system in which the protein is enveloped by a solvation shell. The thermodynamics of proteins in the three-component system is well understood in terms of preferential binding parameters (Von Hippel and Schleich, 1969; Inoue and Timasheff, 1972; Arakawa and Timasheff, 1982; Zaccai and Eisenberg, 1990; Timasheff, 1991; reviewed in Zaccai, 2013). In conditions that maintain protein solubility, the chemical potential of the solvation shell and the bulk are equilibrated (Figure 6). In salting-out conditions that favor protein aggregation and crystallization, the equilibrium is strongly perturbed because the small solutes are excluded from the solvation shell (Tardieu et al., 2002). In this case the solvation shell looks like a hydration shell.

Cytoplasmic protein isolated from extreme halophilic prokaryotes that use the KCl-in adaptive strategy, such as S. ruber or the Halobacteriaceae, maintain a high solubility at molar concentration of various salt (Coquelle et al., 2010). In the case of the tetrameric MalDH from Haloarcula marismortui, the measurements of the preferential binding parameters have shown that the enzyme obey the general thermodynamics rules of the three components system (Costenaro et al., 2002; Ebel et al., 2002): In salting out conditions, the solvation envelope of Hm MalDH is strongly depleted in salt and it looks like a hydration shell; such behaviors is equivalent to the situation encountered with a non-halophilic protein. However, in high concentration of various physiological salts, it has been measured that Hm MalDH preferential binding parameters depend on salt type, demonstrating that the composition of its solvation shell varies (Costenaro et al., 2002; Ebel et al., 2002). In these physiological salts, Hm MalDH solvation envelope is enriched in salt, reflecting its halophilic adaptation. Consequently, as the chemical potential of the solvation layer and the bulk solvent are close, Hm MalDH remains highly soluble at high salt concentration. We determined that Sr MalDH remains highly soluble in high concentration of physiological salts (Coquelle et al., 2010). Based on the observation made on Hm MalDH, this suggests that Sr MalDH solvation layer should also be enriched with salts.

The relationship between an increase in protein solubility and the shift toward negatively charged protein surfaces is not restricted to halophilic protein (Trevino et al., 2007). The favorable effect of acidic residues on protein solubility has been highlighted by an elegant thermodynamical work based on seven non-halophilic proteins, which displayed pIs ranging from 3.5 to 8 (Kramer et al., 2012). In the case of halophilic proteins, it has been demonstrated that their high negative charge density maintains a weak repulsive protein-protein interactions in high salt concentration (Costenaro et al., 2002; Ebel et al., 2002). Theoretically, this repulsive effect between macromolecules of same net charge could also be induced by positively charged amino acids. However, calculation of the solvent-accessible areas of the side-chain components between negatively and positively charged residues unravel their relative efficiency on solubility. Compared to positively charged residues, the favorable effect of acidic residues is due to the lower hydrophobic solvent exposed surface of their side chain (Britton et al., 1998).

Our comparative study of the hydration shell of Ca and Sr MalDHs sheds light on the close relation between solubility, acidic residue enrichment and solvation.

First, conclusions from the solvent properties analysis using X-ray crystallography should be drawn with cautiousness. Our observation confirms that salting-out conditions deplete the solvation layer in salts. Indeed, the additives used for crystal growth shift the conditions toward salt depletion in the solvation envelope. Consequently, even if several halophilic proteins have been crystallized in the presence of high concentrations of physiological salts, one should take precaution when discussing the role of the solvent layer as obtained from X-ray structures.

Second, the favorable change in solubility of halophilic proteins is driven by their protein surface enrichment in acidic residues, which plays a dual role. Indeed, our study shows that acidic residues, through their carboxyl groups that are known to form strong hydrogen bonds, can organize the solvation shell by direct as well as indirect interactions. They are therefore good candidates for interactions with hydrated salt ions as proposed by Zaccai (2013). Moreover, they promote slightly repulsive inter-particular interactions between each protein molecule, favoring solubility.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.