The main goal of this study was to evaluate on a quantitative basis the relationship between hydrophobic contacts and proteins adaptation to high temperatures.
An essential prerequisite to carry out such a study is to assemble a large and minimally redundant set of very high resolution crystal structures. Indeed, despite the observation that each protein family seems to adopt different structural strategies to adapt to high temperatures , common trends may be outlined if a large number of structural data is available . At the same time, since computed values of apolar contact area are mostly influenced by the relative position of the interacting residues, their precision is affected by the resolution of the crystal structures analysed. Therefore two datasets were culled from a set of 1563 crystal structures from thermophilic (optimal growth temperature between 50°C and 80°C) and hyperthermophilic (optimal growth temperature above 80°C) organisms, and their mesophilic counterparts. The rationale of this choice was to assure that the obtained results were not biased either by the paucity of data, or by the quality of the collected crystal structures.
As already discussed by Chen et al. , the increase of the apolar contact area in hyperthermophilic and thermophilic proteins may be achieved at least by two different mechanisms: an evenly distributed increase over all residues; a local increase over key residues. The latter mechanism, that has been shown to be a major contribute to the enhanced thermostability of proteins from T. maritima , seems to involve mainly residues already implied in the formation of hydrophobic contacts. This suggests that a better compactness may originate from an even better connectivity in those protein regions that already have a tendency to compactness and not by simply "tightening the loops" . The results obtained in this work on the difference of apolar contact area (ΔACA) agree with this hypothesis: a significant increase of ACA was measured in both datasets only when the analysis was limited to the SCRs of the hyperthermophilic structures. The SCRs were presumably subject to similar constraints during the divergent evolution of a family of proteins from a common ancestor, and therefore they possibly contain most of the determinants necessary to maintain the fold. Considering the role played by hydrophobic contacts in this sense, it is not surprising that the residues composing the SCRs and engaging hydrophobic contacts were mostly involved in the structural modifications necessary to achieve and maintain a proper fold at high temperatures. Moreover, the finding that the measure of the difference of ACA resulted highly significant only when limited to the SCRs, could explain some apparently not significant results previously obtained by measuring accessible surface area  or cavity size .
The statistically significant increase of ~0.75 Å2/residue of apolar contact area was observed only in the SCRs of hyperthermophilic proteins. Therefore, it can be argued that proteins from thermophilic organisms usually adopt different strategies to enhance thermostability. Indeed, it has been demonstrated that moderately and extremely thermostable proteins rely on different mechanisms to achieve greater stability [8, 20]. Ion-pairs interactions represent presumably a predominant force in thermophilic proteins, as well as in many hyperthermophilic proteins [8, 21]. On the other hand, comparisons of mesophilic and hyperthermophilic protein structures indicate that the hydrophobic effect has a contribution to stability only at high temperatures, while only moderately thermophilic proteins show an increase in the polarity of their exposed surface . Two factors could be responsible for this difference: the temperature dependence of the thermodynamic forces involved in protein stabilization, and/or the phylogenetic origin of the extremely thermophilic organisms, that belong to the domain Archaea, and are therefore distinct from moderately thermophilic organisms, which are mostly Bacteria. In any case, the obtained results strongly suggest that packing of hyperthermophilic proteins, in comparison with their mesophilic homologs, has improved significantly, and it is reasonable to deduce that this increased amount of apolar contact area contributes to the stabilization of the native state of the protein.
Our analysis revealed that α-helices were mainly involved in the increased amount of ACA. Surprisingly, no statistically significant trends have been observed in the comparison of the ACA in the β-strands of the SCRs. We cannot provide a clear explanation of this different behaviour between secondary structures. An intriguing possibility is that β-strands are, generally, already almost optimally packed, even in mesophilic proteins, resulting in a small margin of improvement. However, it is also possible that this observation is due to 'sample bias' e.g., the peculiarities of the available protein structures.
Structural stabilization of α-helices in protein cores may therefore represent a component of great importance for the enhanced termostability of hyperthermophilic proteins. A number of studies in the past has stressed the importance of the enhanced stability of α-helices as a general feature of many hyperthermophilic proteins. In order to investigate the role of α-helices in protein thermostability, Petukhov et al.  compared energy characteristics of α-helices from four families of hyperthermophilic and mesophilic proteins, using statistical mechanical theory for describing helix/coil transitions. They found that the magnitude of the observed decrease in intrinsic free energy on α-helix formation of the thermostable proteins was sufficient to explain the experimentally determined increase of their thermostability. Furthermore, protein engineering studies showed that a well-packed α-helix structure is related to large increase in thermostability [23, 24]. It is well known that the flexibility of α-helices is often required to assure protein function, such as conformational transitions in substrate binding or protein-protein interactions . However, an excessive flexibility of this secondary structure element, at high temperatures, could result in an insufficient stability to maintain its native conformation, causing the entire protein to unfold.
According to thermodynamic studies on model peptides in aqueous environments, two main factors appear to play a key role in the structural stability of the α-helices: the presence of amino acids with intrinsic helical propensity, and side chain-side chain interactions [26, 27]. Therefore, we further investigated the nature of the increased stabilization of α-helices composing the SCRs of hyperthermostable proteins, determining the differences in amino acid composition of the residues involved in CHCs. The results of this analysis strongly suggest that isoleucine and, to a lesser extent valine, mostly to the detriment of leucine, are involved in the formation of more hydrophobic contacts in hyperthermophilic proteins, compared to their mesophilic counterparts. Likewise, the importance of isoleucine in the formation of CHCs of hyperthermophilic proteins was confirmed by the analysis of the preferred amino acid interactions in CHCs, where almost all types of interactions scoring at > 3.0 standard deviations involved the amino acid isoleucine, and by the favoured amino acid substitutions between the hyperthermophilic and mesophilic proteins in CHCs. A large amount of theoretical and experimental studies demonstrates the importance of isoleucine in the stabilization of protein structures from thermophilic organisms. Malakauskas and Mayo  reported the computer-aided engineering of a seven-fold mutant of the β1 domain of the Streptococcal protein G, exhibiting a melting temperature above 100°C and an enhancement in thermodynamic stability of 4.3 kcal mol-1 at 50°C over the wild-type protein. Of seven mutations, five were of type XXX→ Ile, and they improved side-chain packing in the interior of the protein. An increased content of isoleucine in thermophilic and hyperthermophilic proteins, to the detriment of leucine, was also noted by Haney et al.  and Kumar et al. . More recently, a structural genomics based study carried out by Chakravarty and Varadarajan  reported that leucine is preferentially substituted by the β-branched residues valine and isoleucine, at buried sites.
Several studies have demonstrated in the past that leucine has a slightly higher α-helix propensity than isoleucine and, generally, β-branched residues [27, 30]. This assumption, which is apparently in contrast with the results obtained by this work, derives from substitution experiments in short polyalanine α-helices-forming peptides in water . This process is mainly associated with the loss of conformational entropy of residues during the folding of α-helices in an aqueous environment: freezing side chain with fewer internal rotational degrees in the α-helix conformation would be entropically less expensive. However, it must be noted that these experiments, and many derived propensity scales, do not take into account solvent entropy effects. As discussed by Creamer and Rose , neglect of solvent entropy appears justified for a peptide side chain because no significant differences in solvation energy are expected in the side chain of a solitary polyalanyl helix during a helix-coil transition. In either case, the side chain is highly solvent-exposed. The same situation would not be appropriate for a protein helix that, upon association with the remainder of the molecule, engages a solvent-shielded interaction surface. In this study, only the α-helices composing the SCRs and therefore mostly found in the protein core were considered for further investigation. Therefore, application of helix propensity scales might be not appropriate in this case. For example, Li and Deber  have shown that α-helices propensity scales are not appropriate for non aqueous environments and that β-branched amino acids, as valine and isoleucine, rank among the best helix promoters in an apolar environment, as a lipid bilayer.
On the other side, hydrophobic contacts deriving by side chain interactions could play a role of great importance in the stabilization of the α-helices composing the SCRs of hyperthermostable proteins. At temperatures above 80°C, the hydrophobic effect, that is considered to be a dominant force in protein folding [32, 33], is mainly enthalpy driven . In fact, while at high temperatures the entropy contribution to the protein stability tends to zero, the loss or gain of van der Waals interactions acquires increased importance. For example, constructing 15 Barnase mutants in which hydrophobic interactions were deleted, Serrano et al.  found a strong correlation between the degree of Barnase destabilization and the number of methyl side chain groups that were lost (r = 0.91). These data agree with the preferred substitutions (RAla→Val = 3.20; RVal→Ile = 6.31) observed in the CHCs of our datasets.