The Atomistic Understanding of the Ice Recrystallization Inhibition Activity of Antifreeze Glycoproteins

Abstract

As the most potent ice recrystallization inhibitors, antifreeze glycoproteins (AFGPs) have been extensively studied since their discovery. However, the molecular mechanism of how they inhibit ice growth remains controversial—notably, which group directly contributes to the binding of AFGPs to ice is hotly debated. Here, we use molecular dynamics simulations to investigate the atomistic details of the binding of AFGP8 to ice. We show that the binding of AFGP8 to ice can be divided into three cases: backbone dominant binding (BDB), disaccharide dominant binding (DDB) and weak binding (WB). Hydrogen-bonding and hydrophobic groups contribute equally to the binding of AFGP8 to ice and synergistically promote the binding. The –CH₃ groups promote the contacting of AFGP8 to ice via hydrophobic effect, and the hydrogen-bonding groups anchor AFGP8 to ice surfaces through direct hydrogen bonding with ice. Specially, we verify that the -CONH- groups anchor the backbone of AFGP8 to ice by forming hydrogen bonds with ice surfaces while the –OH groups not only anchor the disaccharide to ice but also slow down the dynamics of the surrounding water. In addition, we reveal that both the backbone and the disaccharide can bind to ice surfaces while the latter is more flexible, which also perturbs the hydrogen bond network of potential ice-like water molecules by swaying in the solution to further enhance its antifreeze activity. This work provides the atomistic details of the ice growth inhibition mechanism of AFGP8, which is helpful for the design of high-efficacy cryoprotectants.

1. Introduction

Antifreeze proteins (AFPs) and antifreeze glycoproteins (AFGPs) allow organisms to survive in subzero habitats [1,2,3,4,5]. These natural proteins, as well as synthetic mimics, are of enormous interest for their use in biomedicine [6,7,8,9,10,11], the food industry [12,13,14] and engineering deicing [15,16,17,18]. Antifreeze glycoproteins have three macroscopic antifreeze properties resulting from their ice interactions: ice recrystallization inhibition (IRI) [19,20], thermal hysteresis (TH) [1,21] and dynamic ice shaping (DIS) [22,23,24]. The former (IRI) is of particular interest because this property is easier to mimic and central to application in cryopreservation [25,26,27,28,29,30,31]. This property is generally explained by an adsorption-inhibition mechanism in which the proteins recognize and bind to specific ice surfaces, thereby preventing macroscopic ice growth [32,33]. Antifreeze glycoproteins, composed with tripeptide repeats of alanine-alanine-threonine (Ala-Ala-Thr) in which the hydroxyl group of the Thr is glycosylated with β-D-galactosyl-(1,3)-α-N-acetyl-D-galactosamine (Figure 1a) [34,35,36], are the most potent IRI agents [37]. Antifreeze glycoproteins are classified as AFGP1 to AFGP8 based on the number of Ala-Ala-Thr repeat units, and some of Ala in the smaller AFGPs (AFGP7-8) may be substituted by Pro [36]. A series of experiments have indicated that AFGPs are preferentially adsorbed at the prismatic and pyramidal surfaces of ice to inhibit ice growth [37,38,39,40,41]. Despite the previous studies having reached a consensus that poly-L-proline type II (PPII) helix conformations (as shown in Figure 1b) play a crucial role in the binding of AFGPs to ice [42,43,44,45,46,47], the molecular details of how AFGPs bind to ice remain largely elusive due to its high flexibility and complex components [9,48,49,50].

Several scenarios have been proposed for the binding of AFGPs to ice. One is based on the direct interaction of the hydrogen-bonding groups of AFGPs to ice. Some experiments have substituted certain disaccharide moieties and illustrated that the disaccharide plays a significant role in the anti-freezing of AFGPs [37,51,52,53,54]. A 2D infrared spectroscopy study by Giubertoni et al. [47] indicated that AFGPs bind to ice through their disaccharide chains. Knight et al. [40] proposed a model where the hydrogen bonds formed between AFGP8 and the ice is vital for the irreversible binding of AFGP8. Moreover, a delicate experiment by Sun et al. [50] confirmed that all the hydroxyl (–OH) groups on the disaccharides are necessary for the unique antifreeze activity of AFGP8. A recent simulation study by Zhang et al. [55] showed that –OH groups can form hydrogen bonds with ice and promote the binding of AFGP8 to ice. In the study of other IRI agents, such as poly (vinyl)alcohol (PVA) [56,57,58] and nanogold modified with small molecules [59], hydrogen bonding has also been proved to be the dominant driving force for the binding of antifreeze agents to ice. Therefore, it is believed that hydrogen bonding is directly involved in the binding of AFGPs to ice.

This view, however, has been challenged since it is almost impossible to directly observe which part of AFGP binds to ice in experiments. Harding et al. [35] pointed out that the hydrogen bonding may not be as prevalent as hydrophobic effects for AF(G)Ps and the latter may dominate the binding of AFGPs to ice, but they did not prove it directly. Based on MD simulations, Mochizuki and Molinero [45] reported that AFGPs bind to ice via hydrophobic groups rather than the –OH groups of the disaccharides. The substitutional experiments by Tachibana et al. [60] have also shown that the methyl group of threonine is important for the unique antifreeze activity of AFGP8. Therefore, a systematical study of the specific roles of hydrogen-bonding and hydrophobic groups in the IRI activity of AFGP8 is still very necessary.

Here, we performed molecular dynamic (MD) simulations to elucidate the ice growth inhibition mechanism of AFGP8 (the smallest AFGP) by systematically investigating the roles of –CH₃, –CONH– and –OH groups of AFGP8 in the binding of AFGP8 to ice and the water dynamics of AFGP8 solutions. We find that the hydrogen-bonding (–CONH– or –OH) and hydrophobic (–CH₃) groups synergistically promote the binding of AFGP8 to ice. Specifically, the –CH₃ groups promote the contacting of AFGP8 to ice via hydrophobic effect, while the -CONH- groups anchor the backbone of AFGP8 to ice by forming hydrogen bonds with ice surfaces and the –OH groups anchor the disaccharide to ice through hydrogen bonding. In contrast to the backbone, which adsorbs on ice surfaces more easily due to its rigidity, the disaccharides exhibit dual roles due to their hydrophilicity and flexibility. Some of the disaccharides anchor to ice while the others destroy the hydrogen bond network of potential ice-like water molecules by swaying in the solution. Our results provide an important molecular guidance for the design of more efficient antifreeze agents that could be used for tissue cryopreservation and in other fields.

2. Methods

2.1. Molecular Dynamics Simulations

All the MD simulations and free energy calculations are carried out by GROMACS 2019.5 packages [61,62] using the all-atomistic OPLS-AA [63] force field and the TIP4P/Ice water model [64]. The melting temperature of ice Ih in this water model is 270 K [65], in very good agreement with the experimental value of 273.15 K. The Lennard-Jones, bond and angle parameters of C, H and O of CH₃CH₃, CH₃CONH₂ and CH₃OH for the free energy calculations are identical to corresponding groups of AFGP8, without partial charges. The Lennard-Jones parameters for cross-interactions are obtained through the combination rules: ε_ij = (ε_iiε_jj)^1/2 and σ_ij = (σ_iiσ_jj)^1/2. The cutoffs for the van der Waals and coulombic interactions are set to 1.2 nm. The long-range Coulombic interactions are evaluated with the particle-mesh Ewald algorithm. Equations of motion are integrated using leapfrog with a time step of 2 fs. The temperature T and pressure P for production runs are controlled with the Nosé–Hoover thermostat [66,67] and Parrinello–Rahman barostat [68], with time constants of 0.2 and 1.0 ps, respectively. Periodic boundary conditions are applied in three directions. The pressure is set to 1 atm in all NPT-MD simulations.

To investigate the impact of AFGP8 on ice growth when exposed to an advancing ice front on the prism surface of ice Ih, we build the ice slab system, which contains one AFGP8 molecule, and then solvated by ~10000 water molecules, as experiments have reported that AFGP8 exhibits TH activity even at concentrations below 20 mM [69]. The simulation box is set to 6.35 × 6.64 × 10.00 nm³. Two layers of ice Ih (1008 molecules) constrained by a harmonic potential with force constant of 1000 kJ mol-1 nm-2 are generated by the Genice program [70] with the density of 0.909 g cm⁻³. The center of mass of AFGP8 is initially placed approximately 1.1 nm above the ice surfaces. We perform more than a dozen independent simulations with different backbone orientations parallel to the ice surface. First, the energy minimization is conducted using the steepest descent method with the force tolerance of 1000 kJ mol⁻¹ nm⁻¹. Then, a 500 ps NVT-MD simulation is performed for equilibration. Finally, a 1000 ns production NVT-MD run at 268 K is performed to monitor the IRI activity. We employ the CHILL+ algorithm [71] to identify one water molecule in ice or liquid phase. Moreover, the water molecules with the distance less than 3.5 Å away from the ice molecules determined by the CHILL+ algorithm are also considered as ice molecules (~350 ice molecules in each frame), which has a better description for the advancing ice front. To estimate the interaction between the groups and ice, the –CH₃ group is treated as binding to ice surface if there are two ice molecules within 5.5 Å (first solvation shell) of it. The hydrogen bonds are identified with donor-acceptor distance less than 3.5 Å and hydrogen-donor-acceptor angle less than 30° [72].

2.2. Conformation Analysis

We perform the replica exchange molecular dynamic (REMD) simulations [73] to enhance the conformational sampling for AFGP8 in solution. We put an AFGP8 with the arbitrary initial conformation into a box with ~4500 water molecules. The system is energy-minimized and equilibrated for 1 ns in NPT ensembles for each replica. We execute REMD simulation in NPT ensemble for 400 ns in which 48 replicas are prepared by REMD-temperature-generator [74] in the temperature range of 268−398 K (i.e., 268.00, 270.39, 272.80, 275.24, 277.69, 280.15, 282.63, 285.12, 287.63, 290.16, 292.70, 295.26, 297.80, 300.00, 303.00, 305.69, 308.33, 311.00, 313.68, 316.37, 319.09, 321.83, 324.58, 327.35, 330.15, 332.96, 335.78, 338.63, 341.49, 344.38, 347.28, 350.21, 353.16, 356.12, 359.10, 362.10, 365.13, 368.17, 371.23, 374.31, 377.41, 380.53, 383.68, 386.84, 390.03, 393.24, 396.46, 398.00 K). The replicas are exchanged every 10 ps and the average exchange ratio between nearest neighbor temperatures is 26.6 ± 6.0%. The radius of gyration ($R_g$) and the Ramachandran plots are computed from REMD trajectories.

2.3. Free Energies

The umbrella sampling approach [75,76] is adopted to compute the binding free energies (i.e., potential of mean force (PMF)) of CH₃CH₃, CH₃CONH₂ and CH₃OH to ice prism surface, respectively. We first build the ice slab system, which contains one layer of ice molecules (60 molecules) with dimensions 2.27 nm × 2.22 nm × 10.0 nm. Then, about 900 water molecules and 1 solute molecule are added to the systems. The temperature is set to 280 K to avoid freezing. The initial ice molecules are harmonically restrained at their original positions with a force constant of 3000 kJ mol⁻¹ nm⁻² to avoid the melting of ice surface. The center of mass distance along the Z-axis between the gas molecule and ice surface is considered as the reaction coordinate (RC) and the Z distance along the RC (~2.0 nm) is divided into windows every 0.02 nm. A harmonic umbrella potential with a force constant of 1000 kJ mol⁻¹ nm⁻² is applied along the Z-axis. In each umbrella window, a 500 ps equilibration simulation followed by a 20 ns production run is performed under the NVT ensemble. The PMFs are constructed by the weighted histogram analysis method [77] and plotted as a function of the distance from ice surface. We set the average PMF at distances from 1.4 to 1.8 nm as the reference point since the water molecules in this region can be considered as bulk water.

2.4. Adsorption Energies

The adsorption energies of CH₃CH₃, CH₃CONH₂ and CH₃OH on ice prism surface are calculated using the Vienna Ab-initio Simulation Package (VASP) code [78,79], where the core electrons are treated with the projector augmented-wave (PAW) method [80,81]. The valence electrons for all systems are described using the generalized gradient approximation (GGA) with the Perdew–Burke–Ernzerhof (PBE) function [82]. The DFT-D3 correction method of Grimme et al. [83]. is employed to describe the van der Waals interactions. The Brillouin-zone sampling is restricted to the Γ-point [84] because of the large supercell size used in the calculations. The energy cutoff for the plane wave basis sets is set to 600 eV and the electron smearing is described by the Gaussian smearing method [85] with a width of 0.05 eV. Each calculation is considered converged when the total energy changed is less than 10⁻⁵ eV and the forces on each atom are smaller than 0.05 eV/Å. The whole simulation box is set to 18.2 × 14.8 × 23.1 ų containing a vacuum of ∼18 Å and two layers of 2 × 2 ice Ih. The top layer of the ice is allowed to move freely, and the bottom layer of the ice is fixed to describe the bulk ice. Several representative adsorption sites are calculated and we present the most stable one. The adsorption energies of CH₃CH₃, CH₃CONH₂ and CH₃OH molecules on the ice prism surface are calculated as

$$E_{ads}=\left(E_A+E_B\right)-E_{AB}$$

(1)

where E_A and E_B are the total energy of molecules in the gas phase and the relaxed ice prism surface system, respectively. E_AB is the energy of the adsorption system. Larger adsorption energy represents a more stable adsorption.

2.5. Water Dynamics and Flexibility Analyses

In order to compare the dynamics and flexibility of each part of AFGP8 in solution, an AFGP8 is put into a cubic box and then solvated by ~2900 water molecules. Each system is firstly subjected to a steepest-descent energy minimization with a tolerance of 1000 kJ mol⁻¹ nm⁻¹, followed by a 500 ps equilibration NPT-MD simulation at 268 and 300 K. Finally, 60 ns production NPT-MD run is evolved at 268 and 300 K. The intermittent hydrogen bond autocorrelation function [86,87] is calculated as

$$C_{HB}\left(t\right)=\frac{<h\left(0\right)h\left(t\right)>}{<h\left(0\right)h\left(0\right)>}$$

(2)

where the value of $h\left(t\right)$ equals 1 if a tagged pair of different molecules are hydrogen bonded at the time origin and still exists at time t and 0 if the hydrogen bond is absent. The average is taken over different time origins and all possible pairs.

To represent the relative mobility of different groups and the flexibility of AFGP8, we calculate the mean-squared displacement (MSD) and the B factors [88]. The former is calculated as

$$MSD=<{\left|{\stackrel{\rightharpoonup }{r}}_i\left(t\right)-{\stackrel{\rightharpoonup }{r}}_i\left(0\right)\right|}^2{>}_{i\in A}$$

(3)

where the average is taken over different time origins and all particles of type A. The B factor is gained as follows

$$RMS{F}_i=\sqrt{\frac{{{\displaystyle \sum }}_{t_0}^t{\left[r_i\left(t^{\prime }\right)-{\overline{r}}_i\right]}^2}{t-t_0}}$$

(4)

$$RMSF=\frac{{{\displaystyle \sum }}_{i}RMS{F}_i·m_i}{{{\displaystyle \sum }}_i{m}_i}$$

(5)

$$B=RMS{F}^2·\frac{8}{3}{\pi }^2$$

(6)

where $r_i$ and $m_i$ represent the position and the mass of atom $i$ in conceded residue, respectively.

3. Results and Discussion

3.1. PPII Helix Conformations Are the Dominant Structures of AFGP8

To rationalize the conformations of AFGP8 in solution, we performed REMD simulations. Figure 1c shows the distribution of the radius of gyration (R_g) for AFGP8 at 268 and 300 K. Consistent with a previous nuclear magnetic resonance (NMR) spectroscopy study [89], it shows an obvious pick at 1.1–1.2 nm for both temperatures, suggesting that the structure of AFGP8 is adequately extended in solution. Further statistics of dihedral angles between amino acids (the Ramachandran plots) of AFGP8 shown in Figure 1d,e depict that the PPII helix is the dominant conformation of AFGP8 in solution for both temperatures, which is consistent with previous reports [42,45,47,90,91] and corresponds to the peak of the distribution of R_g. In addition, the PPII helix becomes more pronounced at a lower temperature, indicating that AFGP8 mainly adopts the extended conformation with high population of the PPII helix at subzero temperature (268 K).

3.2. Both Hydrogen-Bonding and Hydrophobic Groups Synergistically Contribute to the IRI Activity of AFGP8

In the MD modeling of antifreeze material, such as PVA and AF(G)Ps, the initial structure has a vital influence on its IRI activity [45,56]. Aiming to systematically investigate its IRI activity, we set up a series of MD simulations at 268 K of AFGP8 with the PPII helix, in which different groups are contacting with ice prism surface. All the simulations are performed for at least 1000 ns to monitor the IRI activity. The dependence of IRI activities on the binding strength of AFGP8 to ice is represented in Figure 2. In about 70% of our simulations, we observe that AFGP8 obviously slows down the growth rate of ice compared to the case of pure water but is still pushed along with the growing ice front (Figure 3a), which is denoted as weak binding (WB). In this case, AFGP8 “walks” irregularly on ice surfaces leading to the reversible binding of AFGP8 to ice (Figure 3b). However, we find that for about 20% of our simulations AFGP8 can bind to ice irreversibly through the backbone, which lies flat on the ice surface with most of the methyl (–CH₃) and amide (–CONH–, on the backbone) groups being adsorbed to the ice surface and dramatically prevents the growth of the ice front, as shown in Figure 3c,d. In this case, the –CH₃ groups occur at the cavities of ice through hydrophobic effect and the –CONH– groups anchor to the ice surface by forming hydrogen bonds with ice directly. The disaccharides float in the water phase with almost no hydrogen bond being formed between disaccharides and ice. As a result, we ascribe this irreversible binding to the synergistic effect of the hydrophobic interaction of –CH₃ groups and hydrogen bonding between –CONH– groups and ice and refer to this case as backbone dominant binding (BDB), which has not been reported before. Additionally, we find another case of irreversible binding in our simulations (10%), where the disaccharide of AFGP8 inserts into the ice steps and some –CH₃ groups on the backbone are also adsorbed to the ice front, see Figure 3e,f. In this case, the –OH group on the disaccharides instead of the –CONH– group on the backbone anchors to the ice surface by forming hydrogen bonds with ice directly, which is denoted as disaccharide dominant binding (DDB). Furthermore, it is discovered that the binding strengths of AFGP8 to ice are comparable for both DDB and BDB.

In order to quantitively investigate the cases of WB, BDB and DDB, we elucidate the concrete roles of the –CH₃, –CONH– and –OH groups of AFGP8 played in the IRI activity of the three cases by analyzing the average number of –CH₃ groups (${\mathrm{N}}_{{\mathrm{CH}}_3}$) interacting with ice and the hydrogen bonds formed by the –CONH– (${\mathrm{N}}_{\mathrm{CONH}}$) or –OH (${\mathrm{N}}_{\mathrm{OH}}$) groups with ice, as shown in Figure 4. We find that AFGP8 binds to ice through the co-adsorption of the hydrogen-bonding and hydrophobic groups in all three cases. For the case of WB, we note that ${\mathrm{N}}_{{\mathrm{CH}}_3}$, ${\mathrm{N}}_{\mathrm{CONH}}$ and ${\mathrm{N}}_{\mathrm{OH}}$ are only limitedly verifiable (~4, ~2 and ~2, respectively). This might be the main reason for their reversible binding. In contrast to WB, ${\mathrm{N}}_{{\mathrm{CH}}_3}$(~12) and ${\mathrm{N}}_{\mathrm{CONH}}$ (~8) are significantly increased, but ${\mathrm{N}}_{\mathrm{OH}}$ is even lower (~1) in BDB, which suggests that the –CH₃ and –CONH– groups play a key role in the irreversible binding of AFGP8 to ice. For the case of DDB, ${\mathrm{N}}_{{\mathrm{CH}}_3}$ is similar with WB but its IRI activity is greatly improved compared to WB. This can be attributed to the significant increasing of ${\mathrm{N}}_{\mathrm{OH}}$ (~6), which demonstrates that the –OH group can also play as key for the binding of AFGP8 to ice. Our results directly provide theoretical evidence for previous experiments [50] where –OH groups are indispensable for the significant antifreeze activity of AFGP8.

However, the contributions of the –CH₃ and –CONH– groups to the IRI activity are difficult to distinguish because they are almost adjacent to each other in AFGP8 (Figure 1a,b). To demonstrate the role of the –CONH– groups in IRI activity, we design a model molecule of GLY14 ([Gly-Gly-Gly]₄-Gly-Gly) that consists of –CONH– groups only and perform MD simulations to investigate its IRI activity, see Figure 5. We find that the ice growth inhibition mechanism of fully flexible GLY14 is similar to that of PVA, where the binding mainly occurs from hydrogen bonding and its contact area with ice surface dictates the IRI activity [56]. GLY14 binds to ice sufficiently well and effectively inhibits the growth of ice until it is overgrown or pushed away (~250 ns in Figure 5b) by the advancing ice front of the ice prism surface. This is because the contact area of GLY14 with the ice surface decreases as it is deformed by the growth of the ice front due to its great hydrophilicity and full flexibility. On the other hand, experimental studies have shown that the amphiphilicity is necessary for high IRI activity of AFGPs, and excessive hydrophilicity or hydrophobicity is not conducive to its antifreeze activity [37,46,50,60,67]. Therefore, we speculate that a strong IRI activity needs a synergy of sufficient hydrogen-bonding (–OH or –CONH–) and hydrophobic (–CH₃) groups in AFGP8 to bind to ice.

3.3. Hydrophobic Desolvation and Hydrogen Bonding Drive the Binding of AFGP8 to Ice

We have shown that the –CH₃, –CONH– and –OH groups of AFGP8 contribute to its IRI activity. In order to clarify the driving force behind the binding of AFGP8 to ice, we seek to perform enthalpic and entropic analyses in terms of adsorption energy and free energy calculations. We saturate the three groups with methyl to obtain CH₃CH₃, CH₃CONH₂ and CH₃OH molecules and calculate the adsorption energy of the three molecules on the ice Ih prism surface. By taking the adsorption energy of water molecules (~0.51 eV in Figure 6a) as a reference, we show that the adsorption of the energy of CH₃CONH₂ (~0.85 eV), which forms double hydrogen bonds to the ice surface, is much larger than water molecule adsorption, indicating the less electronegative nature of the nitrogen atom with respect to the oxygen atom. CH₃OH exhibits a similar adsorption energy of ~0.45 eV as water molecules because CH₃OH also adsorbs to the ice surface by forming an O–H–O hydrogen bond. For CH₃CH₃, no hydrogen is formed and the so-called “adsorption energy” of CH₃CH₃ is small (~0.19 eV). We interpret that the –CH₃ group is passively deposited on the ice surface under hydrophobic desolvation. The free energy calculation (Figure 6b) further confirms our explanation of –CH₃ deposited on the ice surface. With the inclusion of the water phase, the free energy of CH₃CH₃ exhibits two minima, corresponding to the molecule in contact with ice and the ice-like water layer. Similarly, the free energies of CH₃CONH₂ and CH₃OH also show two minima near the ice surface, suggesting that all three groups prefer to stay in the region of the ice/water interface. We note that the second minimum of the free energy of CH₃CH₃ is smaller than the first but CH₃OH shows a deeper first minimum. This suggests that the –OH group provides a stronger direct adsorption on the ice surface with respect to the –CH₃ group and prevents AFGP8 from being pushed away by the growing ice front. That is why, in the case of DDB, the AFGP8 hinders ice growth with only one disaccharide anchoring to ice. Additionally, despite the free energy of CH₃CONH₂ being equally conducive for –CONH– to stay on ice surface or ice-like water layer, the enthalpic contribution (adsorption energy) may be stable –CONH– groups on the ice surface by hydrogen bonding. This suggests that when there are enough –CONH– and –CH₃ groups being adsorbed to the ice surface, AFGP8 can also bind to ice tightly (i.e., in the case of BDB). To conclude, the binding affinity to the ice surface is relevant to the contribution of both enthalpy and entropy, associated with the formation of hydrogen bonds and the desolvation of the –CH₃ groups, respectively. Moreover, the hydrophobic desolvation and hydrogen bonding synergistically promote the binding of AFGP8 to ice.

It is also observed that the –CH₃ groups are mainly adsorbed to the S and W sites (Figure 7a) on the ice surface, in agreement with previous work [45]. Instead, the –CONH– and –OH groups tend to be adsorbed at the Hex site (as shown in Figure 7a, the H or O of the surface ice molecule) through hydrogen bonding. Moreover, we note that N and O atoms in the –CONH– group can form hydrogen bonds with ice at the Hex sites. This indicates that the –CONH– group has more ice-binding sites than the –OH group, which may be one of the reasons why the case of BDB is easier to be observed than DDB in simulation. In addition, the “distance matching” between certain groups in antifreeze agents and water molecules at the ice surface is also critical for the binding of antifreeze agents to ice [57,58,92,93,94]. The nearest neighbor distance between the –CH₃ group and the O/N atom in –CONH– is about 4.0/4.5 Å. Moreover, the distance between the S/W and Hex sites of ice is about 5.0 Å and 5.3 Å (Figure 7a), respectively. The proper distances between the –CH₃ groups and –CONH– groups can have a significant “stereoscopic match” with the binding site on ice surfaces due to the flexibility of hydrogen bonds, and the –CH₃ group can be further stabilized on ice surfaces by its adjacent –CONH– group, which binds to ice through directly hydrogen bonding. Thus, they can cooperatively promote the binding of AFGP8 to ice (Figure 7b). This might be one of the reasons why the IRI activity of AFGP8 is orders of magnitude better than PVAs. From the above analysis, we argue that hydrogen-bonding and hydrophobic groups contribute equally to the binding of AFGP8 to ice, in which the hydrophobic groups help AFGP8 to make contact with ice surfaces and the hydrogen-bonding groups anchor AFGP8 on ice surfaces.

3.4. AFGP8 Perturbs the Water Dynamics

Previous works have suggested that AFGPs can also inhibit the growth of ice by perturbing water dynamics [46,95,96,97]. To explore the influence of different groups in AFGP8 on water dynamics, we further analyze the structural relaxation of hydrogen bonds via the intermittent hydrogen bond autocorrelation function C_HB(t), as shown in Figure 8. Obviously, the C_HB(t) between –CONH– groups and water decays more slowly than that between –OH groups and water (Figure 8a), which hints that the –CONH– group of AFGP8 has more stable hydrogen bonding with adjacent water molecules. In this regard, we conjecture that the –CONH– groups of AFGP8 may be more steadily adsorbed on ice surfaces than –OH groups. We also note that the C_HB(t) between hydrogen-bonding (–CONH– or –OH) groups and water decays significantly slower than that of water (Figure 8a,b), suggesting that –CONH– and –OH groups that stay in the ice-like water layer can capture the surrounding water molecules and block the diffusion of water onto the ice surface, thus further reducing the growth rate of ice. Overall, AFGP8 slows down the dynamics of the surrounding water and perturbs the hydrogen bond network of ice-like water (insert of Figure 8b), which is consistent with previous experimental reports [33,95] and simulation studies [46,96,98]. Therefore, it is evident that AFGP8 inhibits ice growth not only by direct adsorption but also by slowing down the dynamics of the surrounding water.

3.5. The Flexible Disaccharides Are as Useful as the Rigid Backbone in Anti-Freezing

In our simulations, we observe that AFGP8 can automatically explore the appropriate binding position on the ice surface. Although it has been explained by the great flexibility of AFGP8 [45], the specific contribution of different residues of AFGP8 to its flexibility is still elusive. Here, we analyze the mobility of different groups of AFGP8 in solution by computing the mean squared displacement (MSD), as shown in Figure 9a. The result indicates that the sequence of the mobility of these three groups is: –OH > –CH₃ > –CONH–, whether at 268 or 300 K. This may explain why the C_HB(t) between –CONH– groups and water decays more slowly than that between –OH and water. To further visualize the flexibility of each residue of AFGP8, we calculate the B factor of AFGP8, see Figure 9b. Clearly, the disaccharide of AFGP8 is more flexible than the peptide. This suggests that the backbone is harder to break off once it binds to the ice surfaces due to its slight rigidity; thereby, it may bind to ice more easily than the disaccharides. Accordingly, the disaccharides are slightly harder to anchor to the ice surface firmly, which requires forming multiple hydrogen bonds with ice simultaneously. This may explain why most of the disaccharide chains of AFGP8 stay in solution and the case of BDB is more common than DDB in simulations. On the other hand, these disaccharides in solution can slow down the dynamics of the surrounding water and prevent water molecules from entering the ice surfaces, thus further enhancing the IRI activity of AFGP8. Combining the idea that high affinity to water is necessary for the effective separation of some IRI agents in the ice-water interface and gaining a high IRI activity proposed by Wang et al. [99], the more flexible and hydrophilic disaccharides may facilitate AFGP8 diffusion through the ice surfaces until finding the suitable adsorption sites and cover a wider area of ice surfaces to gain a better antifreeze activity. Therefore, we argue that both the highly flexible disaccharide motifs and the appropriately rigid backbone play a crucial role in the individual antifreeze effect of AFGP8.

4. Conclusions

We combine all-atom molecular dynamics simulations and density functional theory (DFT) calculations to systematically elucidate the ice growth inhibition mechanism of AFGP8 and the roles of –CH₃, –CONH– and –OH groups in the binding of AFGP8 to ice. On the one hand, AFGP8 can inhibit ice growth by an adsorption mechanism. Our simulations indicate that both the backbone and disaccharides can bind to ice, and the binding of AFGP8 to ice is driven by the synergistic effect of hydrophobic desolvation and hydrogen bonding. The bindings can be divided into three cases: WB, BDB and DDB. The reversible (WB) and irreversible (BDB and DDB) binding between protein and ice surfaces are obviously observed on our finite simulation time scale (1000 ns) and the supercooling condition (268 K), and the binding strengths of AFGP8 to ice are comparable for both DDB and BDB. On the other hand, AFGP8 also enhances IRI activity by slowing down the dynamics of the surrounding water and perturbing the hydrogen bond network. The disaccharides are generally more flexible than the backbone and most of them prefer to sway in solution and perturb the water dynamics, therefore further enhancing the IRI activity of AFGP8.

Furthermore, our results show that the hydrophobic (–CH₃) and hydrogen-bonding (–CONH– or –OH) groups contribute equally to the binding of AFGP8 to ice, and synergistically promote this binding. Specifically, the –CH₃ groups promote AFGP8 contact with ice surfaces through hydrophobic interaction, while the –CONH– and –OH groups anchor the backbone and disaccharides to ice surfaces through direct hydrogen bonding, respectively. The binding affinity of these three groups is relevant to not only the subtle contribution of both enthalpic and entropic interactions but also the local topology of the complicated protein or ice surface. In particular, the ancillary simulations of GLY14 indicate that the –CONH– group also plays an important role in the IRI activity. Additionally, the –OH groups on the disaccharides of AFGP8 more easily perturb the water dynamic due to their high mobility and strong hydrogen bonding. It is also observed in our simulation that the –CONH– groups on the disaccharide moieties can directly bind to ice, which is consistent with the experiment result that the N-acetyl groups are important for the extraordinary antifreeze activity of AFGP8 [60]. Based on our results, we suggest the following for the design of highly effective flexible IRI agents: (1) A separation of the relatively hydrophobic rigid backbone and hydrophilic flexible side chains is necessary, the rigid part is easier to bind to ice and the flexible side chains can bind to ice and perturb the dynamics of the surrounding water; (2) Enough ice binding sites (hydrogen-bonding and hydrophobic groups) are required; thereby, the IRI agents can bind to ice strongly with a variety of configurations and gain a high IRI activity; (3) The hydrogen-bonding groups and hydrophobic groups need to be spatially distributed in a way that fits the ice structure, promoting the binding of IRI agents to ice through synergistic effects. Overall, this work enhances the understanding of the IRI mechanisms of AFGPs at the molecular level and has important implications for guiding the design of more refined materials with high antifreeze activity for medical and industrial, as well as agricultural applications.