Chapter 22.2. Hydrogen bonding in biological macromolecules

The hydrogen-bonding capacities of the various hydrogen-bonding groups in proteins are shown in Fig. 22.2.3.1. All, with the exception of the peptide NH and Trp side-chain NH groups, can participate in more than one hydrogen-bond interaction. Peptide and side-chain C=O groups, for example, can act as acceptors for two hydrogen bonds by using both lone pairs of electrons on the sp²-hybridized oxygen. Likewise, the —OH groups of Ser or Thr can act as donors through their single H atom, and acceptors through their two lone pairs. In Tyr side chains, the C—O bond has some double-bond character, and the phenolic —OH is thus likely to prefer only two hydrogen bonds, both in the ring plane. The carboxylate groups of Asp and Glu are normally ionized above pH 4 and their C—O bonds also have partial double-bond character; each carboxylate oxygen should then be able to accept two hydrogen bonds, although the restriction to two may be less severe than for C=O.

Figure 22.2.3.1| top | pdf | Hydrogen-bonding potential of protein functional groups. Potential hydrogen bonds are shown with broken lines. Arg, Lys, Asp and Glu side chains are shown in their ionized forms.

Several uncertainties exist. Crystallographically, it is not usually possible to distinguish the amide oxygen and nitrogen atoms of Asn and Gln, and the decision as to which is which has to be made on environmental grounds by considering what hydrogen bonds would be made in each of the two possible arrangements. Likewise, two possibilities exist for His side chains by rotating 180° about C^β—C^γ. This problem has been analysed by McDonald & Thornton (1994b), and corrections can be made with HBPLUS .

For some side chains, the ionization state is uncertain. Arg and Lys are assumed to be fully protonated, as in Fig. 22.2.3.1, and Asp and Glu are assumed to be fully ionized. Nevertheless, a survey by Flocco & Mowbray (1995) has shown that a small but significant number of short O···O distances between Asp and Glu side chains must represent O—H···O hydrogen bonds, with one carboxyl group protonated. His side chains, in addition to the orientational uncertainty, have a pK_a (∼6.5) that implies that they may be in either their neutral or their protonated form, depending on pH and environment. In the neutral form, only one N atom is protonated (more often , but sometimes ), but in the protonated form both N atoms carry protons; again, the actual state has to be deduced from their environment.

The three components of nucleic acids, i.e. phosphate groups, sugars and bases, all participate in hydrogen bonding to greater or lesser extent. The phosphate oxygen atoms can potentially act as acceptors of two or more hydrogen bonds and are frequently the recipients of hydrogen bonds from protein side chains in protein–DNA complexes. The sugar residues of RNA have a 2′-OH which can act as both hydrogen-bond donor and acceptor, and the 4′-O of both ribose and deoxyribose can potentially accept two hydrogen bonds.

It is the bases of DNA and RNA that have the greatest hydrogen-bonding potential, however, with a variety of hydrogen-bond donor or acceptor sites. Although each of the bases could theoretically occur in several tautomeric forms, only the canonical forms shown in Fig. 22.2.3.2 are actually observed in nucleic acids. This leads to clearly defined hydrogen-bonding patterns which are critical to both base pairing and protein–nucleic acid recognition. The —NH₂ and >NH groups act only as hydrogen-bond donors, and C=O only as acceptors, whereas the >N— centres are normally acceptors but at low pH can be protonated and act as hydrogen-bond donors.

Figure 22.2.3.2| top | pdf | Hydrogen-bonding potential of nucleic acid bases guanine (G), adenine (A), cytosine (C) and thymine (T) in their normal canonical forms.

The net contribution of hydrogen bonding to protein folding and stability has been the subject of much debate over the years. The current view is that although the hydrophobic effect provides the driving force for protein folding (Kauzmann, 1959), many polar groups, notably peptide NH and C=O groups, inevitably become buried during this process, and failure of these groups to find hydrogen-bonding partners in the folded protein would be strongly destabilizing. This, therefore, favours the formation of secondary structures and other structures that permit effective hydrogen bonding in the folded molecule. Not surprisingly, the contribution of specific hydrogen bonds to stability depends on their location in the structure (Fersht & Serrano, 1993). Mutagenesis studies have shown that even the loss of a single hydrogen bond can be significantly destabilizing (Alber et al., 1987) and that the energetic contribution can vary depending on whether or not the groups involved are charged (Fersht et al., 1985).

A consistent conclusion from analyses of protein structures is that virtually all polar atoms either form explicit hydrogen bonds or are at least in contact with external water. The extent to which their full hydrogen-bond potential is fulfilled in a folded protein (for example, the potential of an Arg side chain to make five hydrogen bonds) has been examined in several studies. Baker & Hubbard (1984) considered the explicit hydrogen bonds made by main-chain and side-chain atoms in a number of refined protein structures and established general patterns for both, but did not differentiate buried and solvent-exposed atoms or allow for unmodelled solvent. Savage et al. (1993) used the solvent accessibilities of polar groups to estimate their assumed numbers of hydrogen bonds to external water. This supplemented the explicit hydrogen bonds that could be derived from the atomic coordinates and allowed an estimate of the extent to which potential hydrogen bonds are lost during protein folding. McDonald & Thornton (1994a) focused specifically on buried hydrogen-bond donors and acceptors in order to determine the extent to which the hydrogen-bond potential of these is utilized.

The results of these analyses can be summarized as follows. Almost all polar groups do in fact make at least one hydrogen bond. Hydrogen-bond donors are almost always hydrogen bonded; only 4% of NH groups `lose' hydrogen bonds as a result of protein folding (Savage et al., 1993). On the other hand, hydrogen-bond acceptors often do not exert their full hydrogen-bonding potential. For example, for main-chain C=O groups, which are expected to accept two hydrogen bonds, 24% of possible hydrogen bonds are estimated to be lost during folding (Savage et al., 1993). Among buried C=O groups, although very few make no hydrogen bonds (as little as 2% if hydrogen-bonding criteria are relaxed), the majority fail to form a second hydrogen bond (McDonald & Thornton, 1994a). Steric factors, particularly in β-sheets or where Pro residues are adjacent, restrict hydrogen-bonding possibilities, although some of the `lost' interactions may be recovered through C—H···O interactions (see Section 22.2.7.1). McDonald & Thornton also point out that failure to form a second hydrogen bond is less energetically expensive than failure to form the first. Among polar side chains, the ionizable side chains (Asp, Glu, Arg, Lys, His) show a very strong tendency to be fully hydrogen bonded or solvent exposed. Buried Arg side chains, for example, frequently form all five possible hydrogen bonds. The side chains that most often fail to fulfil their full hydrogen-bond potential are Ser, Thr and Tyr; these almost always donate one hydrogen bond but frequently fail to accept one.

Secondary structures provide the means whereby the polar C=O and NH groups of the polypeptide chain can remain effectively hydrogen bonded when they are buried within a folded globular protein. In doing so, they provide the framework of folding patterns and account for the majority of hydrogen bonds within protein structures. The three secondary-structure classes (helices, β-sheets and turns) are each characterized by specific hydrogen-bonding patterns, which can be used for objective identification of these structures (Stickle et al., 1992).

An important concept in understanding the patterns of side-chain hydrogen bonding in proteins is that of local versus non-local interactions; local means that a side chain hydrogen bonds to another residue that is relatively close to it in the linear amino-acid sequence. Baker & Hubbard (1984) were first to introduce this distinction, with local defined as ±4 residues. Bordo & Argos (1994) define local as ±6 residues and Stickle et al. (1992) as ±10 residues. The distinction is not important, but the distributions in all three analyses show that ±5 would encompass all the significant populations of local hydrogen bonds. Local hydrogen bonds, in which side chains interact with nearby main-chain atoms or other side chains, are evidently critical for protein folding. Non-local hydrogen bonds, although fewer in number (see below), in turn can be very important for stabilization of the folded protein.

If hydrogen bonds with water are excluded, a rule of thirds applies. Approximately one-third of the hydrogen bonds made by side chains (sch's) are with main-chain (mch) C=O groups, one-third are with main-chain NH groups, and one-third with other side chains. Within these populations, however, there are significant differences. For sch–mch(C=O) hydrogen bonds, approximately 45% are local; for sch–mch(NH) hydrogen bonds, a much higher proportion is local (69%), and for sch–sch hydrogen bonds, the proportion is much less (35%) (Bordo & Argos, 1994).

The distribution of local sch–mch(NH) hydrogen bonds shows a marked positional preference (Fig. 22.2.5.1) that highlights consistent hydrogen-bonding motifs found in all proteins (Fig. 22.2.5.2). The major peak involves side chains that interact with an NH group two residues further on in the polypeptide, an n–NH(n + 2) hydrogen bond. This motif primarily involves Asp, Asn, Ser and Thr side chains and is most often found (i) in turns, where a side chain from position 1 hydrogen bonds to the NH of residue 3, (ii) in loop regions where it stabilizes the local structure but is not necessarily associated with chain reversal, and (iii) at helix N-termini.

Figure 22.2.5.1| top | pdf | Distribution of side-chain–main-chain hydrogen bonds as a function of the separation (Δ a.a.) along the polypeptide between the side-chain (sch) and main-chain (mch) groups involved, i.e. Δ a.a. = −n means that a side chain interacts with a main-chain group n residues earlier in the polypeptide (towards the N-terminus). Reproduced with permission from Bordo & Argos (1994). Copyright (1994) Academic Press.

Figure 22.2.5.2| top | pdf | Schematic representations of common classes of side-chain–main-chain hydrogen bonds (a) in turns and (b) at helix N-termini. Arrows represent side chains that hydrogen bond to main-chain CO or NH groups (NH identified by the small circle for H).

Helix N-termini are also the site of other characteristic local side-chain–NH hydrogen-bonding motifs (Baker & Hubbard, 1984; Presta & Rose, 1988; Richardson & Richardson, 1988; Harper & Rose, 1993; Bordo & Argos, 1994). Prominent among these are sch–NH(n + 3) hydrogen bonds involving Ser, Thr, Asp and Asn side chains, but sch–NH(n − 3) interactions, in which Glu or Gln side chains hydrogen bond back to a main-chain NH, form an important lesser category. Other motifs, such as that in which a Glu or Gln side chain bends round to hydrogen bond to its own NH group, are also found. Collectively, these contribute to helix capping motifs (Fig. 22.2.5.2b) that help satisfy the hydrogen bonding of the `free' NH groups of the helix N-terminus and in effect extend the helix; the sch–mch(NH) hydrogen bond mimics the mch–mch C=O···HN hydrogen bonds of the helix. Helix N-capping by side chains is probably a very important influence in protein folding, acting as a stereochemical code for helix initiation (Presta & Rose, 1988; Harper & Rose, 1993).

The distribution of sch–mch(CO) hydrogen bonds also shows a striking preference, this time for positions −3 and −4. These sch–CO(n − 3) or sch–CO(n − 4) hydrogen bonds account for the vast majority of local hydrogen bonds between side chains and main-chain C=O groups. Almost all (∼85%) are in helices, with most of the remainder in turns. They involve predominantly (∼80%) Ser and Thr side chains but other side chains (Asn, His, Arg) can also participate. These local hydrogen bonds can occur at any point along a helix, where they are often associated with helix bending or kinking (Baker & Hubbard, 1984). However, they are most frequently found at helix C-termini (Bordo & Argos, 1994) and may constitute a termination motif.

Local side-chain–side-chain hydrogen bonds, although common, do not seem to fit into any obvious patterns; the only recurring interaction identified so far is between side chains on succeeding turns of helices, i.e. separated by approximately four residues. These frequently involve charged side chains, which can form hydrogen-bonded ion pairs. In sections of extended chain, side chains that are two residues apart may similarly interact.

Non-local hydrogen bonding by side chains is less easy to categorize but is no less significant; more than 50% of side-chain–main-chain(C=O) hydrogen bonds are non-local, as are ∼65% of side-chain–side-chain hydrogen bonds. In most proteins, a small number of polar side chains with multiple hydrogen-bonding capability act as the centre for networks of hydrogen bonds; these appear to be particularly important for stabilizing non-repetitive polypeptide chain structures (coil, loops). Examples are given in Baker & Hubbard (1984). Most often these involve larger side chains with more than one hydrogen-bonding centre (Asn, Asp, Gln, Glu, Arg, His) which cross-link different sections of the polypeptide. Arg side chains interacting with main-chain C=O groups seem to be particularly effective; Ser and Thr, on the other hand, are seldom used, even though both have the potential to form three hydrogen bonds.

The geometry of side-chain hydrogen bonding has been analysed by Baker & Hubbard (1984) and, more extensively, by Ippolito et al. (1990). The former concentrate on hydrogen-bond lengths and angles and show that the preferred angles fit well with stereochemical expectations. Ippolito et al. examine the preferences for the various hydrogen-bonding sites around each side-chain type by means of scatter plots (Fig. 22.2.5.3) from which probability densities are computed. These show that well defined preferences exist, determined by both steric and electronic effects.

Figure 22.2.5.3| top | pdf | Typical scatter plots showing the distribution of hydrogen-bonding partners around protein side chains, shown for (a) Asn or Gln and (b) Tyr. Reproduced with permission from Ippolito et al. (1990). Copyright (1990) Academic Press.

Water molecules, with their small size and double-donor, double-acceptor hydrogen-bonding capability, are ideal for completing intramolecular hydrogen-bonding networks, e.g. by linking two proton acceptor atoms, or two protein donor atoms, that cannot otherwise interact. Thus, buried water molecules, making multiple hydrogen bonds, help satisfy the hydrogen-bond potential of internal polar atoms and contribute to protein stability; internal waters average about three hydrogen bonds each (Baker & Hubbard, 1984; Williams et al., 1994). From the survey of Williams et al. (1994), most (58%) occupy discrete cavities, while 22% are in clusters housing two waters and 20% are in larger clusters; some examples of larger clusters are given in Baker & Hubbard (1984). Buried waters are often conserved between homologous proteins (Baker, 1995), and each buried water–protein hydrogen bond is estimated to stabilize a folded protein by, on average, 0.6 kcal mol⁻¹ (1 kcal mol⁻¹ = 4.184 kJ mol⁻¹) (Williams et al., 1994). More loosely bound external waters exchange much more rapidly and presumably contribute less energetically.

Several patterns of hydrogen bonding are consistently observed. Water molecules are most often seen interacting with oxygen atoms rather than nitrogen atoms and acting as hydrogen-bond donors rather than acceptors. Possible reasons include the greater number of acceptor sites in proteins and the fewer geometrical restrictions imposed by acceptors (Baker & Hubbard, 1984; Baker, 1995). There is also a predominance of interactions with main-chain atoms rather than side-chain atoms: on average ∼40% with main-chain C=O groups, 15% with main-chain NH and 45% with side-chain groups (Baker & Hubbard, 1984; Thanki et al., 1988). Favoured main-chain binding sites include the N- and C-termini of helices, C=O groups on the solvent-exposed sides of helices, the edge strands of β-sheets, and the ends of strands where they add extra inter-strand hydrogen bonds at the position where the strands diverge (Thanki et al., 1991). Among side chains, the most highly hydrated appear to be Asp and Glu, whose COO⁻ groups bind, on average, two water molecules each (Baker & Hubbard, 1984; Thanki et al., 1988). On the other hand, the best-ordered water sites are created by residues whose side chains simultaneously make hydrogen bonds to other protein atoms (His, Asp, Asn, Arg) or may be sterically restricted (Tyr, Trp).

The distributions of water molecules around protein groups follow the geometrical patterns expected from simple bonding ideas (Baker & Hubbard, 1984; Thanki et al., 1988). Interactions with NH groups are linear, and those with C=O groups show a preferred angle of ∼130° at the oxygen-atom acceptor, consistent with interaction with an oxygen-atom lone pair; restriction to the peptide plane is not very strong, however. Although the distributions around polar side chains generally follow the expected patterns (Thanki et al., 1988), there is little evidence of ordered water clusters around non-polar groups. This may be because water clusters need to be `anchored' by hydrogen bonding to polar groups to be seen crystallographically.

DNA structure is dominated by the prevalence of duplex structures and hence by the classic Watson–Crick hydrogen-bonding pattern of A–T and G–C base pairs. This hydrogen-bonding pattern is not affected by whether the double helix has A-form, B-form, or Z-form geometry. Other hydrogen-bonding modes in DNA are probably very rare, arising only as a result of mutations (which produce mismatches), chemical modifications, such as methylation, or other disturbances, such as the binding of drugs or proteins so as to alter DNA conformation. Mismatches can give stable hydrogen bonding but at the expense of local perturbations of the DNA structure.

In contrast to DNA, RNA molecules generally form single-stranded structures, which are correspondingly much more complex and less regular. This means that catalytic and other activities can be generated in addition to their information-carrying roles. Current knowledge of detailed RNA three-dimensional structure is limited to transfer RNAs and several ribozymes, including a large ribosomal RNA domain (Cate et al., 1996). Even from this small sample, however, it is clear that a great diversity of hydrogen-bonding interactions exists; RNA molecules contain regions of double-helical structure, often with classical Watson–Crick A–U and G–C base pairing, but these regions are interspersed with loops and bulges and tertiary interactions between the various secondary-structural (double-helical) elements. These interactions include many unconventional base pairings (e.g. see Fig. 22.2.6.1).

Figure 22.2.6.1| top | pdf | Hydrogen-bonding interactions in RNA tertiary structure. In (a), a triple base interaction is shown. In (b), G150 and A153 of a GAAA tetraloop participate in multiple hydrogen-bond interactions involving bases, riboses and phosphate. Reprinted with permission from Cate et al. (1996). Copyright (1996) American Association for the Advancement of Science.

Some RNA structural motifs may prove to be of widespread general importance in RNA molecules. One example is a sharp turn with sequence CUGA in the hammerhead ribozyme that exactly matches turns in tRNAs (Pley et al., 1994). Another is the GNRA tetraloop structure (N = any base, R = purine). This loop has a well defined structure, stabilized by hydrogen bonding and stacking involving its own bases, and it also presents further hydrogen-bonding groups that can dock into `receptor' structures in other parts of the RNA molecule. This results in triple or quadruple base interactions (Fig. 22.2.6.1) that tie different parts of the RNA structure together; the parallel with hydrogen-bonding side chains in proteins is very strong. The 2′-hydroxyls of ribose groups are also used in some of these interactions (Fig. 22.2.6.1). Further ribose interactions involve interdigitated ribose groups that line the interfaces between adjacent helices such that pairs of riboses interact by hydrogen bonding through their 2′-hydroxyl groups, forming `ribose zippers' As many more RNA structures are determined experimentally, it is likely that more hydrogen-bonding motifs will be recognized, and their full role in RNA structure can be better assessed than at our present, imperfect state of knowledge.

Sutor (1962) first summarized evidence for C—H···O hydrogen bonds following earlier suggestions by Pauling (1960), and current evidence has been nicely summarized in several recent articles (Derewenda et al., 1995; Wahl & Sundaralingam, 1997). The energy of C—H···O hydrogen bonds has been generally estimated as ∼0.5 kcal mol⁻¹ (about 10% of an N—H···O interaction) but may be higher, especially in hydrophobic environments. It also depends on the acidity of the C—H proton, with methylene (CH₂) and methyne (CH) groups being most favourable.

A number of examples of C—H···O hydrogen bonds can be found in nucleic acid structures (Wahl & Sundaralingam, 1997). The best known is that between the backbone O5′ oxygen and a purine C(8)—H or pyrimidine C(6)—H, when the bases are in the anti conformation. Another example is given by a U–U base pair, in which the two bases form a conventional N(3)—H···O(4) hydrogen bond and a C(5)—H···O hydrogen bond.

In proteins, two groups are regarded as being particularly significant (Derewenda et al., 1995). These are the C^ɛH of His side chains and the methylene H atoms of the main-chain α-carbon atoms. C—H···O hydrogen bonds involving His side chains have been found for the active-site His residues of proteins of the lipase/esterase family and in other proteins (Derewenda et al., 1994). The C^αH atoms appear to provide much more widespread C—H···O hydrogen bonding, however, especially in β-sheets, where they are directed towards the `free' lone pairs of the main-chain C=O groups. C—H···O hydrogen bonds may thus play a previously unrecognised role in satisfying the hydrogen-bond potential of C=O groups. In general, Derewenda et al. (1995) find a significant number of C···O contacts that meet the criteria for C—H···O hydrogen bonds; the H···O distance peaks at 2.45 Å (C···O 3.5 Å), which is less than the van der Waals distance of 2.7 Å, and the angles indicate that the H atoms are directed at the acceptor lone-pair orbitals.

Sulfur atoms are larger and have a more diffuse electron cloud than oxygen or nitrogen, but are nevertheless capable of participating in hydrogen bonds. Given that the radius of sulfur is ∼0.4 Å greater than that of oxygen, hydrogen bonds can be assumed if the distance H···S is less than ∼2.9 Å, or S···O(N) is less than ∼3.9 Å, providing the angular geometry is right. In proteins, the SH group of cysteine can be a hydrogen-bond acceptor or donor, whereas the sulfur atoms in disulfide bonds and in Met side chains can act only as acceptors.

The clearest example of hydrogen bonding involving Cys residues is given by the NH···S hydrogen bonds in Fe-S proteins (Adman et al., 1975); here, peptide NH groups are oriented to point directly at the S atoms of metal-bound Cys residues, with H···S distances of 2.4–2.9 Å. Similar NH···S hydrogen bonds are found in blue copper proteins, involving the Cys ligands. In these cases, the cysteine sulfur is deprotonated and therefore more negative, making it a stronger hydrogen-bond acceptor, and it is likely that hydrogen bonding to cysteine S⁻ atoms is common. A large survey of Cys and Met side chains in proteins has given evidence of both N—H···S and S—H···O hydrogen bonds involving the SH groups of Cys side chains (Gregoret et al., 1991). In particular, Cys residues in helices frequently hydrogen bond to the main-chain C=O group four residues back in the helix in interactions analogous to those seen for Ser and Thr residues in helices. On the other hand, O—H···S or N—H···S hydrogen bonds to the S atoms of Met or half-cystine side chains, although they do exist, are rare (Gregoret et al., 1991; Ippolito et al., 1990).

Surveys of protein structures have shown that aromatic rings (of Trp, Tyr, or Phe) are frequently in close association with side-chain NH groups of Lys, Arg, Asn, Gln, or His (Burley & Petsko, 1986). Energy calculations further suggest that where an N—H group, as donor, is directed towards the centre of an aromatic ring, as acceptor, a hydrogen-bonded interaction with an energy of ∼3 kcal mol⁻¹ (about half that of a normal N—H···O or O—H···O hydrogen bond) can result (Levitt & Perutz, 1988). Whether the close associations observed by Burley & Petsko can truly be regarded as hydrogen bonds has been controversial, however. Mitchell et al. (1994) have analysed amino–aromatic interactions and shown that by far the most common form of association between sp² nitrogen atoms and aromatic rings involves approximately plane-to-plane stacking, which cannot represent hydrogen bonding. There is still, however, a significant number of cases where the H atoms of N—H groups are directed towards aromatic rings, and these represent genuine hydrogen bonds (Mitchell et al., 1994). It is clearly essential to consider the donor–acceptor geometry, both distances and angles, before assuming an amino–aromatic hydrogen bond; the N···ring distance should be less than ∼3.8 Å, and N—H··· C angle greater than 120°, where C is the ring centre (Mitchell et al., 1994).