Type I ‘antifreeze’ proteins

Structure–activity studies and mechanisms of ice growth inhibition


M. M. Harding, School of Chemistry, University of Sydney, NSW 2006, Australia. Fax: + 61 29351 3329, E-mail:harding@chem.usyd.edu.au


The type I ‘antifreeze’ proteins, found in the body fluids of fish inhabiting polar oceans, are alanine-rich α-helical proteins that are able to inhibit the growth of ice. Within this class there are two distinct subclasses of proteins: those related to the winter flounder sequence HPLC6 and which contain 11-residue repeat units commencing with threonine; and those from the sculpins that are unique in the N-terminal region that contains established helix breakers and lacks the 11-residue repeat structure present in the rest of the protein. Although 14 type I proteins have been isolated, almost all research has focused on HPLC6, the 37-residue protein from the winter flounder Pseudopleuronectes americanus. This protein modifies both the rate and shape (or ‘habit’) of ice crystal growth, displays hysteresis and accumulates specifically at the {2 0 2¯ 1} ice plane. Until very recently, all models to explain the mechanism for this specific interaction have relied on the interaction of the four threonine hydroxyls, which are spaced equally apart on one face of the helix, with the ice lattice. In contrast, proteins belonging to the sculpin family accumulate specifically at the {2 1¯ 1¯ 0} plane. The molecular origin of this difference in specificity between the flounder and sculpin proteins is not understood. This review will summarize the structure–activity and molecular modelling and dynamics studies on HPLC6, with an emphasis on recent studies in which the threonine residues have been mutated. These studies have identified important hydrophobic contributions to the ice growth inhibition mechanism. Some 50 mutants of HPLC6 have been reported and the data is consistent with the following requirements for ice growth inhibition: (a) a minimum length of approx. 25 residues; (b) an alanine-rich sequence in order to induce a highly helical conformation; (c) a hydrophobic face; (d) a number of charged/polar residues which are involved in solubility and/or interaction with the ice surface. The emerging picture, that requires further dynamics studies including accurate modelling of the ice/water interface, suggests that a hydrophobic interaction between the surface of the protein and ice is the key to explaining accumulation at specific ice planes, and thus the molecular level mechanism for ice growth inhibition.


ice growth inhibition compound


ice growth modifiers.

Freezing is almost always lethal to cellular organisms as it deprives biological reactions of the aqueous medium they require, causes concentration of ions and other solutes in the plasma, denaturation of biomolecules, and ruptures cell membranes [1]. Despite this, fish are found thriving in the polar and subpolar oceans which are as much as 1.2 °C below the equilibrium freezing point of their body fluids [2–5]. Although these fish contain elevated levels of sodium chloride relative to temperate forms, along with comparable levels of other small ions, the concentrations of these ions account for only about half of the observed freezing point depression of water in the blood plasma. These fish have evolved with molecules which function by depressing the freezing point in a noncolligative manner, and thus allow them to survive in conditions where they should freeze and die [6]. In three decades of research into these ‘antifreeze’ molecules (defined below), the general details of the mechanism by which they function has been uncovered [7], but the exact chemistry at the molecular level has remained elusive.

‘Antifreeze’ glycoproteins and ‘antifreeze’ proteins comprise several structurally diverse classes of molecules that have in common the ability to inhibit the growth of ice. The general structural features of each class have been summarized in a number of reviews [5,6,8–11]. The antifreeze glycoproteins [12–18] are carbohydrate rich 2.6–34 kDa proteins containing an (Ala-Ala-Thr)n repeat with a disaccharide attached to threonine. At least four classes of structurally independent proteins have been identified: type I, alanine-rich, α-helical 3.3 to 4.5-kDa proteins [19–21]; type II, cysteine rich globular proteins that contain five disulfide bonds [22–24]; type III, approximately 6 kDa globular proteins [25–27]; and very recently type IV, glutamate- and glutamine-rich proteins that contain α-helices but appear to be unrelated to other proteins [28].

This review will focus on the structure and activity of type I proteins (Table 1) and is organized as follows. The general structural features of proteins in this class will be presented, including solution conformations and X-ray data. The definition of ‘antifreeze’ protein is discussed and an unambiguous definition introduced for this article. Structure–activity studies on the winter flounder protein HPLC6 are summarized, followed by molecular dynamics and modelling studies that have led to a number of proposed mechanisms for ice growth inhibition. We conclude with an outline of a plausible molecular mechanism for ice growth inhibition which has emerged very recently.

Table 1. Sequences and hysteresis values of type I proteins.
CodeSourceSequenceHysteresis (°C)
10 mg·mL−1

Structural features of type I proteins

Table 1 summarizes the sequences of type I proteins reported to date. These proteins are characterized by being rich in alanine (> 60%), possess a high helical content, and contain 11-residue repeat sequences that commence with threonine. Proteins in this class were first isolated from the blood serum of the winter flounder, Pseudopleuronectes americanus, inhabiting the near shore waters off the coast of Nova Scotia [19,20]. Since then type I proteins have been found in the skin of the winter flounder [29], the yellowtail flounder, Limanda ferruginea[30], Alaskan plaice [31], the grubby sculpin, Myoxocephalus aenaeus[32], the shorthorn sculpin, M. scorpius[21] and the Arctic sculpin [33].

Two major proteins have been isolated from the winter flounder [19,20]. These are denoted in the literature HPLC6 and HPLC8 [34]. The sequences both show a high degree of homology, differing only in residues 22 and 26. AFP9 is a 52-residue protein that contains an additional repeat unit [35], and a fourth sequence SAPF1, isolated from the skin of the winter flounder [29] contains significant differences in both the N- and C-terminal residues, compared with HPLC6, HPLC8 and APF9.

HPLC6 is by far the most extensively studied of all known type I proteins and is one of the few type I ‘antifreeze’ proteins for which both a solid state structure [36] and detailed NMR studies [37,38] have been reported. The 37 residue protein contains three 11-amino acid repeats of the sequence ThrX2AsxX7, where X is usually alanine or another amino acid that favours α-helix formation. Extensive conformational studies on HPLC6 using CD, NMR and X-ray diffraction have been carried out and have shown that this protein is completely α-helical in conformation with the exception of the last unit which adopts a 310-helix conformation. The X-ray structure (Fig. 1) showed an elaborate terminal cap structure, which is also likely to contribute to the stability of the helix [36]. The N-terminal cap structure consists of an ordered network of eight hydrogen bonds involving the side chains of residues Asp1, Thr2, Ser4, Asp5 and two tightly held water molecules. The C-terminal cap structure makes use of the side chain of Arg37 and the amidated C terminus to form three hydrogen bonds. The protein has also been studied by NMR spectroscopy [37,38]. Due to the high number of alanine residues in the sequence, which led to significant spectral overlap, assignment of all residues was not possible, and a full solution structure at low temperature has not been published. However, full characterization of the rotamer conformations of individual threonine side chains was possible, and showed that these side chains are mobile and dynamic at low temperatures. Although all four threonines are free to occupy different rotamer configurations, the middle two (Thr13, Thr24) have a preference for −60 degrees 55% of the time.

Figure 1.

Figure 1.

X-ray crystal structure of HPLC6.

Although detailed NMR or X-ray structures of the related proteins HPLC8, AFP9, YTF and AP are unavailable, the high degree of sequence homology with HPLC6 supports similar fully α-helical proteins. CD studies of AFP9 [35], which contains four 11-amino acid repeats, are consistent with a fully α-helical conformation at 0 °C.

The six type I proteins isolated from the shorthorn [21,39], grubby [32] and Arctic sculpins [33] as well as SAFP1 from the winter flounder [29] are distinct from the other sequences in Table 1 in the composition of the N-terminal region of each protein, which commences with methionine and contains helix destabilizing amino acids including proline and glycine. While these proteins do contain the 11-residue internal repeat structure ThrX2AsxX7 present in the other type I proteins, the repeat structure is less stringent with some variation of the Asx residues, and additional lysine residues. The shorthorn sculpin proteins have been proposed to contain two different structural and possibly functional domains [21]. Domain 1 of SS8 is the 9-residue N-terminal region which was cleaved and shown to have little secondary structure. The second domain is helical and contains two contiguous 11-amino acid residue repeats, followed by another somewhat similar repeat. The helical content of SS8 is high (73%), but SS3 is only moderate (about 45%). Molecular dynamics studies on SS8 concluded that the protein conformation compares favourably with an idealized α-helix, except for a short section of the N-terminus [40]. Despite the presence of proline, three hydrogen bonds were calculated to stabilize the helical conformation at this end of the protein. However, this calculated conformation has not been confirmed experimentally.

Whereas a α-helical conformation was predicted for both GS5 and GS8 [32], no detailed conformational studies have been reported for either the grubby or the Arctic sculpin proteins.

Ice growth inhibition compounds

Before discussing the properties of the type I proteins, in this section we define the terms antifreeze activity, ice growth inhibition and hysteresis and review briefly the experimental evidence that is typical of type I proteins. An overview of some of these experiments have been included in previous general reviews on protein interaction with ice [10,11].

The freezing point of a solution at fixed pressure is the temperature at which the liquid and solid phases have the same Gibbs free energy. If the solution is in thermal equilibrium, then by definition the freezing point of the solution is exactly the same as the melting point of the solid phase. Freezing point depression is a thermodynamic phenomenon, lowering the freezing temperature by an amount proportional to the molal concentration of the solute molecule or of the segments of a polymer in solution [41]. Freezing point depression (in addition to boiling point elevation, osmotic pressure, and solvent vapour pressure) is an example of a colligative property of the solution, namely a property that depends on the mole fraction of solute particles present but not on the molecular mass or chemical properties of those particles. As a thermodynamic phenomenon, this effect will depress the melting point of the solution by the same amount. Common antifreezes, such as ethylene glycol, also act thermodynamically. In contrast, the proteins and glycoproteins discussed in this article kinetically depress the freezing point of a solution on a molal basis 300–500 times more than equilibrium colligative effects predict. The term ‘antifreeze’ is perhaps misleading from the physical chemistry perspective. Unfortunately, of late various authors have generalized its use to many, physically different circumstances, so we take the opportunity in this review to define separate terms for the different classes of interaction with ice.

We define first ‘ice growth inhibition compound’ (IGIC), a term used in this review to refer to molecules that act kinetically not thermodynamically, and possess a nonzero thermal hysteresis effect. IGICs which have a proven biological function will be called ‘antifreeze’ molecules in the biological context (a definition due to DeVries [5]). Molecules which merely modify the growth of ice, but do not arrest it, will be called ‘ice growth modifiers’ (IGM). A good example is the mutant 15EKlac [42] discussed below.

An important fingerprint of ice growth inhibition is thermal hysteresis, defined to be the difference between: (a) the equilibrium melting point and (b) the ice growth temperature, the temperature at which seed ice crystals will grow in the solution. In pure water, this difference is zero: any ice crystal seed inserted into the solution at any temperature below 0°C will grow; the rate of growth is determined by the temperature. Thus, an ice growth inhibitor is defined as a molecule for which the thermal hysteresis is nonzero, and much greater than any colligative effect. Despite the marked differences in amino acid composition and protein structure between the type I–IV fish proteins and glycoproteins, they demonstrate very similar thermal hysteresis effects. Table 1 includes the hysteresis values for the naturally occurring type I proteins at the concentration 10 mg·mL−1.

Other phenomena associated with ice growth inhibition include accumulation of protein at specific faces of the ice crystal, detected by hemisphere etching [31], and modification of the crystal habit [43] when ice is grown in a thermal gradient. Observations of ice crystal growth under the microscope shows that the presence of IGIC not only lowers the freezing point of the solution but also alters the growth habits and growth rates of ice.

Ice may exist in many polymorphic forms. Ice 1h is stable at 1 atmosphere below zero °C. The hexagonal ice 1h lattice unit may be characterized by four axes, a1, a2, a3 and c (one of which is redundant). The surface of the hexagonal unit, shown in Fig. 2, comprises eight faces, two basal faces normal to the c-axis and six prism faces. As it is normal to the c-axis, the basal face is known as the c-face or (0001). Directions and vectors within the ice lattice are also described in terms of the four axes and are distinguished by the types of brackets that encloses the coordinates. For example {2 0 2¯ 1} designates the group of 12 equivalent surfaces of a hexagonal bipyramid of which (2 0 2¯ 1) is one specific surface; 〈1¯ 1 0 2〉 designates the group of 12 equivalent directions of these surfaces of which [1¯ 1 0 2]is the direction appropriate to the specific surface (2 0 2¯ 1). Under atmospheric pressure, at 0 °C, ice grows most rapidly along the a-axis to give hexagonal shaped crystals. This growth along the a-axis is markedly inhibited by type I proteins. When the temperature of the solution is lowered, ice crystal growth eventually recommences, but at an accelerated rate and primarily along the crystallographic c-axis to give bipyramidal crystal forms.

Figure 2.

Figure 2.

Schematic representation of the hexagonal unit of ice crystal showing definition of axes and prism faces and planes at which type I proteins accumulate. (A) {2 0 2¯ 1} is the winter flounder plane (B) {2 1¯ 1¯ 0} is the sculpin plane.

Historically in the literature, many authors have stated that molecules ‘bind to’ ice surfaces. Although this terminology may be appropriate for ice/vacuum studies, in this article we write that molecules ‘accumulate at’ surfaces, in recognition of the fact that a variety of mechanisms are possible. The type I proteins inhibit ice growth by accumulating at specific interfaces of ice, summarized in Fig. 2. The winter flounder sequence HPLC6 accumulates on the 12 equivalent {2 0 2¯ 1} bipyramidal planes of the ice 1h crystal; one of these 12 planes is shown in Fig. 2A and is formed from one edge of the lower surface of the hexagon to the mid-point of the top-face of the hexagon. In contrast, the shorthorn sculpin accumulates on the six equivalent {2 1¯ 1¯ 0} planes; these six planes are formed between four vertices of the top and bottom of the hexagon, with one shaded plane shown in Fig. 2B. Experimentally, these patterns are determined by the ice hemisphere test [31], a simple test to determine which crystal planes (if any) are recognized by the protein. A single ice crystal in a dilute solution of the IGIC is grown into a large hemispherical single crystal, such that all interfacial orientations are present during growth. The solution is so dilute as to allow essentially unretarded ice growth. The crystallographic orientation of the IGIC is subsequently determined from measurement of the interface orientations at which the IGIC is incorporated into the growing crystal by evaporation etching of the grown crystal. Fig. 3 shows the patterns obtained with HPLC6 and the sculpin protein SS8.

Figure 3.

Figure 3.

Top view of oriented single crystal ice hemispheres grown from (A) winter flounder HPLC6 [53] and (B) shorthorn sculpin SS8 taken from Wierzbicki et al. [40].

Truncated analogues: effect of molecular mass

The effect of molecular mass on the ice growth inhibition properties of HPLC6 has been examined by sequential removal of N-terminal residues in HPLC6 (Table 2, entries 1–6) [44,45]. Elimination of up to four residues can be tolerated without loss of hysteresis but further elimination of 10 residues or more resulted in complete loss of hysteresis and the ability to modify ice growth. The loss of ice growth inhibition properties in the analogues containing less than 25 residues was attributed to removal of polar amino acid residues, but a critical helix length required for activity could not be ruled out.

Table 2. Effect of molecular mass on activity of HPLC6.
 ProteinSequence   Hysteresis
% 1 mMa
Helicity %Reference
  1. a  100% = 0.68 °C (3 mm); 0.5 °C (2 mm); 0.3 °C (1 mm) for HPLC6 [19].

5Glu22-Arg37 ELTAANAAAAAAATAR082[44,45]
6Leu23-Arg37 LTAANAAAAAAATAR073[44,45]
7TTTB   NH (CH2)2 NH    
1115EKlac  ||    

Similar investigations have been carried out in an attempt to differentiate between the properties of a number of proteins of varying length and composition. Using protein engineering, a series of proteins were formed, containing two, three, four, five and six repeat units of the sequence (TAXXAAAAAAX)n, respectively [46]. Although no actual data was provided, all proteins, except that containing only two repeats of the 11-amino acid unit, were reported to show hysteresis behaviour. These results support the hypothesis that a minimum number of ice interaction residues or minimum helix length is required for ice growth inhibition.

In order to establish whether the lack of interaction with ice of shorter HPLC6 analogues was a result of reduced protein helicity, two truncated proteins, stabilized by covalent bridges (TTTB, 15EKlac) have been studied [42,47]. Protein TTTB is the native protein less one repeat unit, and was stabilized by introduction of an amide bridge via Glu15 and Glu22 (Table 2, entry 7) [47]. The design was based on the successful synthesis of short constrained highly helical peptides via introduction of a covalent lactam bridge via glutamic acid residues [48]. While CD and NMR studies confirmed that the bridge does indeed stabilize the helical conformation, no ice growth inhibition properties were observed. However, it should be noted that the bridge stabilizes one turn of the helix between residues 15 and 22, and NMR studies suggested that residues outside this region were considerably less structured consistent with equilibration between random coil and folded conformations.

Houston et al. [42] prepared a series of related truncated analogues in which the native protein was shortened to delete two of the three 11-amino acid ice ‘binding’ repeats, but retained the N- and C- capping residues (Table 2, entries 8–11). The resulting 15 residue protein (15KE) and its variants (Ac15KE, 15EK) were less helical and showed no ice growth inhibition. However, when the helicity of the protein was reinforced by an internal lactam bridge between Glu7 and Lys11, the minimized protein (15EKlac) was able to stabilize the pyramidal plane {2 0 2¯ 1 } on the surface of growing crystals, but no hysteresis behaviour was detected.

Chirality: variation of amino acid and side chain stereochemistry

There are two levels of chirality that may have an influence on the ice growth inhibition properties of type I proteins: (a) the l chirality of the natural α-amino acids, which gives rise to a right handed α-helix, and (b) the chirality present in some of the side chains of individual residues, notably the threonine residues, which have been implicated in the mechanism of ice growth inhibition discussed in the next section.

The all d-isomer of HPLC6 (Table 3, entry 2) has been synthesized and exhibits identical hysteresis and rates of ice crystal growth to the l-enantiomer as does a 50 : 50 mixture of the d and l enantiomers [49]. However, the enantiomeric forms were predicted to ‘bind’ to the ice surface with different orientations. This hypothesis was supported by molecular modelling and energy minimization calculations of the interaction of the d- and l-enantiomers with ice [50], and was confirmed experimentally in an independent study [51].

Table 3.  Hysteresis, and relative helicity of HPLC6 mutants.
        Hyseresis %aHelicity % 
 ProteinSequence     2 mm1 mmat 3 °CReference
  1. a  100% = 0.5 °C for HPLC6. b  HPLC6 referred to as TTTT in later publications to highlight mutations. c  AFP(KE)2 is equivalent to TTTT2KE.

 TTTT2KE      100 100[58]
         31 [52]

The importance of the side chain chirality, particularly of threonine, has been addressed only very recently by two groups independently [47,52]. On the basis of activity of analogues lacking chirality at this position, it was suggested [53] that the natural proteins may have evolved with threonine residues to provide an additional directional bonding mechanism in the interaction with the ice surface. The allo-threonine substituted analogue (Table 3, entry 34) was studied independently [52], and showed reduced thermal hysteresis compared with the threonine substituted parent protein, and did not arrest ice crystal growth in an identical manner to the native protein.

Mutations of polar and nonpolar residues in HPLC6

In a series of papers, Wen and Laursen [54–56] describe systematic mutations of HPLC6 to yield 22 mutants (Table 3, entries 3–23), designed to assess the role of neutral, polar amino acids, the role of charged amino acids, and the effect of added bulky groups on activity.

From rearrangement of the neutral polar amino acids [56], it was concluded that a specific arrangement of both threonine and asparagine (or aspartic acid) residues is critical for maximal ice growth inhibition, and that the proteins probably ‘bind’ to the pyramidal faces of ice with a specific orientation. Rearrangement of Thr13/Leu12 (S04) or Asn27/Ala30 (S11) resulted in reduction in hysteresis by about one-third whereas incorporation of additional mutations to S11 (S22,23) also reduced hysteresis by the same amount. Analogues S01,14,32-36 involved rearrangement, deletion and replacement of the charged amino acids in HPLC6 [55]. While in some cases these changes resulted in partial reduction of helicity and hysteresis, the effects were not dramatic. Furthermore it was concluded that the salt bridge between Lys18 and Glu22, which helps to stabilize the helix, is not essential for activity, but the molecule was sensitive to modification of the C-terminal Arg37 residue. In contrast, modification of the N-terminal Asp1 was tolerated and the presence of a charged residue at this position does not appear to be crucial. Based on these mutants, a two-step mechanism was proposed whereby at low concentrations activity depends on hydrogen bonding of both threonine and Asx residues and above some critical concentration, hydrophobic inter-protein interactions become important [57].

Mutants S40-43 and S50-52 contained up to four alanine residues mutated to either glutamine or leucine in order to assess the importance of a hydrophobic surface on the outward facing portion of the helix [54]. As glutamine replacements were generally well tolerated, it was concluded that bulk hydrophobicity of the non ice ‘binding’ face is probably not a major factor in inhibition of ice growth. However, introduction of either glutamine or leucine at position 17 completely abolished activity, which was attributed to steric hindrance that prevents effective association or packing of the protein molecules on the ice surface.

Two extra salt bridges (Lys7/Glu11 and Lys29/Glu33) have been incorporated into HPLC6 (which contains a single salt bridge Lys18/Glu22) as a mechanism to stabilize further the helical conformation of the protein [58]. The residues were positioned at residues (i, i + 4) such that when the protein adopts a α-helical conformation, the salt bridges form on the opposite face to that containing threonines 2, 13, 24, 35 (Table 3, entry 35). While AFP(KE)2 was reported to be more helical than the native protein from CD studies, the hysteresis of both proteins was identical. However, it should be noted that the helicity of HPLC6 (75%) which was reported to increase to 96% upon incorporation of two salt bridges is at variance with a number of independent studies of both proteins [53,55,56]. Variations in pH, which induced changes in helical content, paralleled trends in hysteresis values (i.e. lowering helicity lowered the hysteresis value) but at pH 8.5 the proteins displayed identical concentration dependence. Although both HPLC6 and AFP(KE)2 affected the growth of ice along the a- and c-axes, the AFP(KE)2 acted at 7–8 fold lower concentrations than HPLC6.

Recently, Haymet et al. [53,59] have incorporated two salt bridges into the design of a series of mutants (Table 3, XXXX2KE, entries 36–39), as a mechanism to improve the solid-phase synthesis and aqueous solubility of the mutants rather than to increase helicity (although the design was based on the previous report of Chakrabartty and Hew [58]). Identical hysteresis values over a range of concentration, as well as similar hemisphere growth patterns, ice hemisphere and crystal habit tests, were observed for both TTTT and TTTT2KE. In the mutants studied, the aqueous solubility of the proteins was improved greatly by the salt bridges, which effectively disrupt the long alanine stretches in the sequence, and minimize potential aggregation behaviour, but no enhanced effects on helicity were observed with all proteins being 100% helical at temperatures close to 0 °C.

Mutations of threonine residues in HPLC6

A number of sequences discussed in the previous section have included mutation of the threonine residues (e.g. S21, S14); however, limited studies have been carried out on systematic mutation of the threonine residues while retaining the rest of the sequence unchanged.

Two groups have replaced systematically the two central threonine residues, Thr13 and Thr24 with serine and valine (Table 3, entries 24, 25), respectively [53,60]. While the serine analogue was virtually inactive (11% of hysteresis of native protein), the valine analogue suffered only a minor loss in ice growth inhibition [60]. CD, ultracentrifugation and NMR studies indicated no significant structural changes or aggregation of the mutants compared to HPLC6. Thus a reduced role for hydrogen bonding of the threonine hydroxyls to the growing ice crystals was proposed and a more significant role attributed to entropic effects and van der Waals interactions in ‘binding’ of the protein to ice [60]. In an independent study, similar results were observed with TSST as well as the fully serine-substituted analogue SSSS [53].

Our group has studied the proteins XXXX2KE (X = T, S, V, A, G, Table 3, entries 35–39) [53,59]. Compared to HPLC6 (TTTT), these sequences contain two additional salt bridges (Lys7, Glu11 and Lys29, Glu33, indicated by the 2KE suffix), and all four threonine residues have been mutated simultaneously to each of serine, valine, alanine and glycine, respectively. These mutations were made in order to assess the effect of relative size, hydrophobicity and hydrogen bonding characteristics of these side chains on the interactions of the proteins with ice. CD studies showed that all mutants are 100% helical at low temperature, except for the glycine derivative which was estimated to be 70% α-helical. SSSS2KE and GGGG2KE displayed unfaceted growth and showed no hysteresis. Hysteresis values, ice growth patterns and the helicity measurements on TTTT2KE showed that the additional salt bridges present in series 2 proteins do not alter significantly the properties of the protein (Fig. 4). The valine substituted mutant VVVV2KE gave a distinct etching pattern in which the protein accumulates on the {2 0 2¯ 1} plane of ice 1h (Fig. 4), and exhibited thermal hysteresis comparable to that of the native protein. In the case of the alanine substituted mutant AAAA2KE, reduced hysteresis behaviour was measured, together with a distinct etch pattern in the ice hemisphere test, in which ‘binding’ to the {2 1¯ 1¯ 0} sculpin plane was observed (Fig. 4). In addition, both VVVV2KE and AAAA2KE also modified the ice growth habit, determined by the crystal habit test [43](Fig. 5). Sharp faceting anywhere except on the basal plane is diagnostic of orientation-specific surface interactions, which are a characteristic fingerprint of IGICs. This is an extremely sensitive test, which can detect IGICs above 10−7 weight percentage concentration. Distinct faceting and stepped ledges were seen in ice grown from a 0.1 mg·mL−1 solution of VVVV2KE (Fig. 5A) compared with pure water which shows no detectable faceting (Fig. 5B). AAAA2KE also showed distinct facets (Fig. 5C).

Figure 4.

Figure 4.

Summary of properties of key ice growth inhibitors (A) XXXX2KE and modification of crystal growth and ice hemispheres for (B) X = T, (C) X = V and (D) X = A.

Figure 5.

Figure 5.

Crystal habit test [43]. (A) VVVV2KE, (B) water, (C) AAAA2KE.

These combined results (Figs 4 and 5) show that there is a significant contribution to the mechanism of ice growth inhibition from the hydrophobic methyl group in threonine and valine which both contain a common hydrophobic face on the surface of the protein (in the case of threonine, this is achieved with the methyl group on the surface of the protein). However, whereas the native protein TTTT, and TTTT2KE and VVVV2KE were shown to accumulate on the {2 0 2¯ 1} face, the closely related AAAA2KE accumulated on the {2 1¯ 1¯ 0} face, giving a similar etching patter to the shorthorn sculpin protein. The origin of this accumulation is not understood.

The relative importance of the residues at positions 2, 13, 24 and 35 to ice growth inhibition has been tested recently by single and double point mutations of threonine for serine (Table 3, entries 27–33) [52]. While replacement of the central two threonine residues for serine caused a drastic reduction in hysteresis (90%), mutation of the terminal two residues resulted in only about a 30% loss. These studies concluded that there are two central hydrophobic clusters (Ala9-Leu12-Thr13; Ala20-Leu23-Thr24) that provide the driving force for initial ‘binding’ and overall stability. It was proposed that the methyl groups of Thr13 and Thr24 participate in hydrophobic interactions with ice, which provide the driving force for ‘binding’ and stability, whereas the threonine hydroxyls and other polar residues control ‘binding’ specificity and impart additional stability through hydrogen bonding. VVVV2KE was also studied [52], but gave reduced hysteresis and different modification to ice growth than an independent study of VVVV2KE [53]. The apparent discrepancy between these results may lie in the highly hydrophobic nature of this mutant; full characterization of the exact species present at high and low concentrations is required as self-aggregation is possible. Our studies [53] were restricted to low sample concentrations due to gelling of the samples at > 1 mm and hence errors are possible in measurements above this concentration, which were reported in the later study [52].

Molecular modelling and dynamics studies

In 1977, Raymond and DeVries [61] proposed that HPLC6 ‘binds’ preferentially to the prism faces of the ice that are parallel to the c-axis, and subsequently inhibit ice growth of the basal plane by raising the curvatures of surfaces on the ice crystal. This model however, did not explain why the protein molecules adsorb preferentially to prism faces. Later studies from the same group [62,63] attributed this directional preference to matching of the distance (4.5 Å) between threonine and aspartate groups of the protein to the distance between the water molecules on the prism faces. However, the 4.5 Å separation is present in a large family of ice crystal planes including the basal plane and hence the lattice match model predicts that type I proteins ‘bind’ to different ice planes with a similar affinity.

In 1991, Knight et al. [31] demonstrated that the structurally similar proteins of the winter flounder and the Alaskan plaice adsorb onto the {2 0 2¯ 1} pyramidal planes of ice, whereas the sculpin proteins adsorb on {2 1¯ 1¯ 0}, the secondary prism planes. A common vector was identified, 〈0 1 1¯ 2〉 on the [2 0 2¯ 1] and [2 1¯ 1¯ 0] planes containing the repeating distance of 1.67 nm which matches with that of the α-helical protein (1.65 nm for 11 residues) and the origin of ‘binding’ preferences was attributed to matching of lattice distances with the four threonine hydroxyls. Based on this seminal paper [31], a number of molecular modelling and dynamics studies of HPLC6 were reported, in an effort to understand the mechanism of ice ‘binding’ by type I proteins to specific planes.

Monte Carlo simulated annealing and energy minimization studies of the conformation of HPLC6 in a vacuum spawned a model in which the protein ‘binds’ to an ice nucleation structure in a zipper-like fashion by hydrogen bonding of the hydroxyl groups of the four threonine residues to the oxygen atoms along the [0 1¯ 1 2] direction in ice lattice, subsequently stopping or retarding the growth of ice pyramidal planes so as to depress the temperature of ice growth [64]. Although solvent was not included in the study, severely limiting the conclusions, the study did highlight that the alignment and equal spacing of the side chain hydroxyl atoms of the four successive threonine residues are very sensitively dependent on their side chain angles and can vary from 13.4 to 18.4 Å.

The first molecular dynamics studies were reported by two groups independently at CRYO ′92. Kay and Haymet [65] simulated HPLC6 in a periodically replicated box of water using biosym software and the cvff force field. In contrast, Brady et al.[66] reported a simulation using the AMBER force field. This work was later extended to a simulation of ‘binding’ at one idealized ice/water interface rather than the ice/vacuum interface [67]. The protein was positioned such that the Threonine residues were pointing away from the interfacial region throughout the simulation and a mixture of contacts from other polar groups and improved steric matching of the protein shape to the ice/water interface was noted.

Additional molecular dynamics studies using the AMBER forcefield [68], which included solvent water at 0 °C supported the positioning of the threonine hydroxyls close to 16.6 Å and also indicated a number of other hydrogen bonding features involving aspartic acid and arginine that assist in stabilization of the helical structure. An independent modelling study [69]in vacuo (using commet software) proposed that the high affinity of the winter flounder protein for the the {2 0 2¯ 1} plane derives mainly from the steric compatibility between this plane and the protein molecule, giving rise to a many-fold increase in the Van der Waals’ component of the surface/molecule ‘binding’ energy.

Molecular dynamics simulations [70] examined the behaviour of the water molecules surrounding HPLC6, at room temperature and at 0 °C. The structure of the water around the threonine residues appeared to be much better defined than on the opposite face of the helix (which is predominantly hydrophobic) and it was proposed that the protein inhibits ice crystal growth by ‘binding’ into the surface of the crystal via the polar residues, while the apolar residues restrict the ‘binding’ of further water molecules and keep them in flux, thus preventing further growth.

A detailed molecular dynamics study on HPLC6 and two mutants [71] in the gas phase, solvated by water, adsorbed on the (2 0 2¯ 1) crystal plane in the gas phase and in aqueous solution, identified an important role for other potential hydrogen bonding groups specifically the Asx residues. Four distinct ice ‘binding’ regions: Asp1/Thr2/Asp5; Thr13/Asn16; Thr24/Asn27; Thr35/Arg37 were noted with an overall ‘binding’ surface of the protein complementing the ice surface topology, using both hydrogen bonding and van der Waals’ interactions. This work also noted the potential importance of hydrophobic interactions and suggested that deletion of hydrophobic residues such as leucine may well provide insight into the role of hydrophobic effects.

The high resolution crystal structure of TTTT [36,72] provided important molecular level detail about the protein structure that allowed the validity of earlier molecular modelling results to be assessed. The crystal structure did not reveal the expected close spatial match between the protein and ice. Furthermore, the threonine hydroxyls groups did not protrude sufficiently from the surface of the protein to clear sterically hindering groups and a less stringent hydrogen-bonding criterion was proposed to explain interaction of the protein to ice. The topology of the [2 0 2¯ 1]plane was noted to be complementary to the protein aligned along the ‘binding’ axis and as the hydrogen bonding groups extend minimally from the protein’s flat ice ‘binding’ surface, water molecules on the ridges of the [20,21] ice plane were proposed to be the most probable ‘binding’ sites. The 4.5 Å spacing of water molecules along the ridges provides accessible hydrogen bonding targets for both ice ‘binding’ groups within an ice ‘binding’ motif. It was concluded that the relative flatness of the surface of the protein and the rigidity of the side chains are critical in the mechanism of action and that a flat and rigid ice ‘binding’ surface, albeit with different spatial arrangements of ice ‘binding’ groups, may be a general feature of all type I proteins.

In contrast to the flounder and plaice sequences, the polar and charged residues of the sculpin sequences do not contain an obvious repetitive motif that can match with the ice lattice periodicity. Ice etching measurements, molecular modelling and dynamics studies [40] have been used to rationalize the preferential ‘binding’ of the 42 residue shorthorn sculpin SS8 to the {2 1¯ 1¯ 0}plane. Specific interaction of the protein with the ice surface was suggested to be based on the spacing of Lys9 and Lys31 (33.8 Å) as well as Arg12 and Lys23 (16.9 Å). Two models were proposed, the first requiring accommodation of ‘binding’ surface residues within ice cages, and the second involving inclusion of lysine side chain tetrahedral groups into the ice lattice. This proposed mechanism is significantly different from all proposals for other type I proteins and examination of the effect of systematic mutations of the key charged residues is required in order to confirm this proposed mechanism. Furthermore, two structural domains have been identified in the protein and cleavage of the N terminus of SS3 showed reduction in hysteresis consistent with this proposal.

Proposed mechanism of ice growth inhibition by type I proteins: the importance of hydrophobic interactions

In view of the very recent structure–activity studies that have clearly identified hydrophobicity as a key element required for ice growth inhibitor proteins, all previously proposed mechanisms for the interaction of HPLC6 with ice must be reassessed. As outlined above, most of these studies assumed threonine hydroxyls ‘binding’ to the ice surface as a starting point in simulations, or in the design of mutants in structure–activity relationships. Furthermore, the absence of solvent in most molecular modelling and dynamics studies limits the conclusions in terms of how these proteins inhibit ice growth in the serum of fish. The nature of the pure water/ice interface is clearly essential in discussion and modelling of possible ice ‘binding’ mechanisms by IGICs. The limitations of models which approximated the interface as an ice/vacuum interface are seen in Fig. 6, which displays the best available molecular view of the real ice/water interface, in particular the ice {2 0 2¯ 1} interface [73] (J. A. Hayward and A. D. J. Haymet, unpublished results). Since 1987 [74], the molecular details of the ice/water interface have been known [75–77], in remarkable agreement with available microscopic experimental evidence [78]. To explain the action of all kinetic ice growth inhibitors, and indeed growth inhibition in related systems [79], the three most essential features are: (a) the interfacial region is broad, and is described by overlapping ‘order parameters’ which describe the smooth evolution from bulk ice to bulk water, in terms of the separate quantities average density, translational order, and orientational order; (b) these order parameters are different for each face of ice, and hence provide an explanation for the selectivity of one interface over another; and (c) the diffusion constant is reduced below its bulk value well outside the interfacial region.

Figure 6.

Figure 6.

Snapshot from molecular dynamics simulation of Hayward and Haymet [73] (J. A. Hayward and A. D. J. Haymet, unpublished results) showing two statistically independent water/ice {2 0 2¯ 1} faces. These are periodically replicated in all three dimensions to fill space. A total of 568 molecules are shown. The long dimension of the simulation box is 66.61 Å (average of whole simulation). The interfacial region is seen to be highly disordered, and many tens of Angstroms wide.

Central to understanding the mechanism of ice growth inhibition is the understanding of the reason HPLC6 and other type I proteins recognize and accumulate at the {2 0 2¯ 1} plane of ice 1h. While hydrogen bonding involving the threonine residues has provided a model consistent with this experimental data, there is a simple general argument against the role of hydrogen bonding as the dominant interaction in accumulation of HPLC6 at a specific ice surface [53]. Previous mechanisms have included a favourable enthalpy of interaction between the solute and the ice surface formed by four or more threonine residues (plus in addition possibly other residues), denoted ΔHsoluteice. The traditional argument implies simply that this number is negative and hence provides a mechanism for interaction between the solute and ice surface. It has been argued recently [53] that the traditional argument neglects the interactions which these same residues would have with water, denoted ΔHsolutewater. The crucial quantity is the difference (ΔHsoluteice−ΔHsolutewater) estimated to be at most a few tenths of a kcal·mol−1, and hence it probably plays little or no role in the mechanism of action of these proteins. Further general considerations of the ease of hydrogen bonding between the solute and molecules in the water phase, which can adjust their position with minimal entropy penalty [81] to make a more optimal hydrogen bond (as opposed to molecules in the ice phase constrained to oscillate about lattice positions), lead to an argument that the difference (ΔHsoluteice−ΔHsolutewater) is in fact positive and further argues against hydrogen bonding as a possible explanation.

Fig. 7A shows a helical wheel representation of HPLC6 and the key mutants that have shown ice growth inhibition properties, highlighting a common hydrophobic face (for X = T this is achieved with the γ-methyl group oriented towards the surface). The hydrophilic region that includes the Asx residues is also shown; systematic mutations of the Asx residues are required to unequivocally establish whether these residues are essential to ice growth inhibition properties. We have proposed that this hydrophobic face is involved in interaction with the ice surface, resulting in inhibition of ice growth [53]. In support of this mechanism, perturbation of the hydrophobic face (encompassing Ala10/Ala25 to Ala9/X24) in a number of synthetic analogues (Table 3, S22, S41, S42, S50, S52) has always been accompanied by reduction in hysteresis.

Figure 7.

Figure 7.

(A) Helical wheel representation of type I mutants showing hydrophobic and hydrophilic regions and (B) comparison of relative size and hydrophobicity of X side chains.

Fig. 7B shows the relative sizes, hydrophobicity and hydrogen bonding characteristics of the X side chains introduced at positions 2, 13, 23 and 35 which act as IGICs. The CPK representations highlight the similarity in both the size and orientation of the methyl group in threonine and valine and the smaller size of alanine which is probably unable to interact as closely with the ice surface due to steric effects of the neighbouring (more bulky) amino acid side chains, leading to reduced hysteresis. The key question, which is not understood at present, is how this hydrophobic interaction leads to recognition of the {2 0 2¯ 1}plane, and the unusual ice hemisphere of the alanine derivative that indicates accumulation at a different ice plane. The reduction in hysteresis of the allo-Threonine analogue [52] provides an important clue: the reduced hysteresis shows that the orientation of the methyl group is important and may explain why the natural protein has evolved with threonine at this position. Detailed molecular dynamics studies of the interaction of both TTTT2KE and VVVV2KE with the ice/water interface, incorporating the γ-methyl group of the threonines (plus other surrounding side chains) should provide insight into these interactions and allow an increased understanding of the exact mechanism of ice growth inhibition.

Is a similar hydrophobic face present in all of the type I proteins in Table 1? The yellowtail flounder and Alaskan plaice proteins are structurally similar to HPLC6 in the presence of regularly spaced threonine residues but there is some variation in the hydrophilic residues. However, all proteins have an obvious hydrophobic face (Fig. 8). The sculpin sequences also contain a α-helical section with a hydrophobic face, but full structural characterization of these proteins, in particular the N-terminal sequence is required. However, interaction of the hydrophobic face of the α-helical regions of the sculpin proteins (Fig. 8) with ice, in a similar manner to that proposed for the flounder proteins is possible. The role of the N-terminal sequence is however, not clear. Determination of the solution conformation of this sequence in the presence and absence of the rest of the protein is required as a starting point to hypotheses for how these proteins may interact with ice.

Figure 8.

Figure 8.

Helical wheel representation of type I proteins. The hydrophobic face is indicated by bold residues; in the case of threonine this is achieved with γ-methyl group oriented on the surface.

In summary, although the exact molecular chemistry is currently unknown, and awaits sophisticated, nonlinear spectroscopy, further experimental data on carefully designed mutants, and further computational studies including full solvation and free energy calculations, it is clear that a number of subtle effects are involved in the ice growth inhibition mechanism by type I proteins, and that hydrophobicity is important. More detailed computer simulations of the ice/water interface [73] (J. A. Hayward and A. D. J. Haymet, unpublished results) including hydrophobicity [80], and molecular dynamics studies of type I proteins that focus on hydrophobic interactions rather than hydrogen bonding, may provide further insight into the molecular basis for these interactions, and assist in the design of new generation IGICs.

A very recent paper [81], published after this review was accepted for publication, has reported a study of seven mutants of HPLC6 in which the role of the leucine and asparagine residues have been investigated. The results were consistent with a primary role for leucine in preventing protein aggregation at high concentrations, and a role for the asparagine residues in enhancing solubility, rather than being involved in hydrogen bonding and/or steric complementarity.


We thank the Australian Research Council (A.D.J.H. and M.M.H.), the Australian Antarctic Division (ASAC Grant 2050 to ADJH) and the American Chemical Society (PRF Grant 33707-AC9 to A.D.J.H.) for financial support. A.D.J.H. thanks C. Knight and A. DeVries for many helpful discussions.