Protein–ice interaction of an antifreeze protein observed with solid-state NMR

Abstract

NMR on frozen solutions is an ideal method to study fundamental questions of macromolecular hydration, because the hydration shell of many biomolecules does not freeze together with bulk solvent. In the present study, we present previously undescribed NMR methods to study the interactions of proteins with their hydration shell and the ice lattice in frozen solution. We applied these methods to compare solvent interaction of an ice-binding type III antifreeze protein (AFP III) and ubiquitin a non-ice-binding protein in frozen solution. We measured ¹H-¹H cross-saturation and cross-relaxation to provide evidence for a molecular contact surface between ice and AFP III at moderate freezing temperatures of -35 °C. This phenomenon is potentially unique for AFPs because ubiquitin shows no such cross relaxation or cross saturation with ice. On the other hand, we detected liquid hydration water and strong water–AFP III and water–ubiquitin cross peaks in frozen solution using relaxation filtered ²H and HETCOR spectra with additional ¹H-¹H mixing. These results are consistent with the idea that ubiquitin is surrounded by a hydration shell, which separates it from the bulk ice. For AFP III, the water cross peaks indicate that only a portion of its hydration shell (i.e., at the ice-binding surface) is in contact with the ice lattice. The rest of AFP III’s hydration shell behaves similarly to the hydration shell of non-ice-interacting proteins such as ubiquitin and does not freeze together with the bulk water.

Antifreeze proteins (AFPs) can be found in a variety of cold-adapted organisms including bacteria, plants, insects, and fish (1). AFPs lower the freezing point of a given solution below its melting point, a process called thermal hysteresis. This thermal hysteresis is thought to result from the strong binding of the AFP to the surface of ice crystals making their further growth energetically unfavorable (2, 3). One of the most intensely studied classes of AFPs is the type III AFPs from the ocean pound. In particular the HPLC-12 isoform (in the following just termed AFP III), was studied with X-ray crystallography (4), NMR (5, 6), mutagenesis (4, 7, 8), and molecular dynamics (9). These studies showed that the ice-binding site of AFP III is a flat surface formed by predominantly nonpolar residues (5, 8), which is remarkable because the solvent accessible surface of other soluble proteins like ubiquitin are predominantly polar. There have been limited opportunities to study the water molecules at the ice-binding surface of AFPs, although recent computational and sub-angstrom X-ray structures indicate ordered water molecules at the ice-binding surface (9, 10).

Ice, with its very slow ¹H spin-lattice relaxation rate (R₁) and very fast spin-spin relaxation rate (R₂), is relatively difficult to study with NMR. Measurements of relaxation rates of ice showed that R₂ is on the order of 10⁶ s^-1 and R₁ on the order of 10^-2 s^-1 at a temperatures of about 260 K and a magnetic field of 7 T (11). Due to this slow R₁ and fast R₂, the measurement of direct ice–protein cross peaks is extremely difficult. Therefore, direct NMR measurements of ice are quite rare in the literature. On the other hand, there have been studies of the effect of ice on various properties of AFPs by NMR (6, 12).

Water has a much slower R₂ and faster R₁ than ice, which permits the investigation of protein–solvent interactions of protein crystals or fibrils using solid-state NMR: Detailed studies of protein–water interactions in crystals of Chr and the SH3 domain showed that chemical exchange is the dominant mechanism for polarization transfer in these systems and that it is possible to spectroscopically distinguish bulk from the crystal water (13–16). Recently, the protein–water interactions of the membrane channels KcsA and M2 were characterized using solid-state NMR, thereby identifying the parts of these proteins that are in direct contact with the solvent (17, 18). For the amyloid fibrils formed by the prion protein fragment HET-s(218–289) it was shown with a combination of solid- and liquid-state NMR that the dynamic loops framing the core of the amyloid fibril strongly interact with water (19).

The hydration shell of a protein is extremely important for its dynamics and function (20). Protein hydration water was shown to be only slightly less dynamic and 10–15% denser than bulk water due to defects in the H-bond network induced by the protein surface (21, 22). Interestingly, the hydration shell of soluble proteins does not freeze below the freezing temperature of the bulk solution (23, 24). For frozen ubiquitin solutions, Tompa and coworkers showed that the hydration layer starts to freeze at about -50 °C (25). These results about protein hydration shells raise the following questions regarding antifreeze proteins: If the hydration shell of proteins does not freeze, how can AFPs bind to ice? Do AFPs, in contrast to other non-ice-binding proteins, lose their hydration shell by directly binding to ice? If AFPs lose their hydration shell by binding to ice, does the entire hydration shell disappear due to the ice interaction?

To answer these questions, we developed previously undescribed NMR methods that measure protein ice binding and methods that characterize the interaction of a protein with its hydration shell. We applied these methods to study the hydration shell of AFP III and the non-ice-binding protein ubiquitin and to characterize the special interaction of AFP III with ice. Solid-state NMR provides atomic resolution structural information of frozen protein solutions. Therefore, solid-state NMR has been applied to study protein folding (26) and large protein complexes (27) in frozen solution, as well as AFPs bound to ice (28–30). Recently, we confirmed the location of the ice-binding surface of AFP III by comparing liquid-state NMR spectra of this protein in solution with solid-state NMR 2D spectra of AFP III in frozen solution (6). AFP III and ubiquitin have relatively narrow lines in frozen solution, making it possible to study these proteins with high resolution techniques and fully ¹³C and ¹⁵N labeled samples. We were able to detect significant chemical shift changes in AFP III upon freezing but only minor chemical shift changes for ubiquitin as a control. With the present study, we are continuing this work by giving direct spectroscopic evidence of protein–ice interaction, and we show that not the entire hydration shell of AFP III disappears when AFP III binds to ice.

Results

R₁ Measurements.

In our previous study, we noticed relatively long ¹H spin-lattice relaxation time T₁ for AFP III in frozen solution. We followed the hypothesis that a special molecular contact with ice could be related to these anomalous characteristics and pursued a quantitative analysis of the R₁ relaxation rate (in the following we use R₁ instead of T₁ where R₁ = 1/T₁) because R₁ measurements on multiple phase systems can be used to measure chemical exchange and spin-diffusion rates between the individual phases (31, 32). We measured the proton R₁ of AFP III and ubiquitin in frozen solution (-35 °C) using a ¹³C detected inversion-recovery pulse sequence. Fig. 1 shows the results of these measurements: The R₁ of AFP III in frozen H₂O (Fig. 1A) is surprisingly slow compared to the R₁ of ubiquitin in frozen H₂O solution (Fig. 1D). As can be seen in Fig. 1B, the addition of 5 mM Cu(II)-EDTA as paramagnetic relaxation agent increased the R₁ of AFP III in frozen H₂O solution (33). Furthermore, the ¹H R₁ of AFP III in frozen D₂O is much faster than in frozen H₂O solution (Fig. 1C). We fit all inversion-recovery curves to mono- and biexponential models to quantify these results. The relaxation curves of ubiquitin in H₂O and AFP III in D₂O can be nicely fit using monoexponential decays as illustrated with dashed lines in Fig. 1 C and D. However, both R₁ relaxation curves of AFP III in pure H₂O and in Cu(II)-EDTA doped H₂O could only be fit with a biexponential model (solid lines in Fig. 1 A and B)

[1]

where m_i(t) can be either the amplitude of the ¹H magnetization of ice (m_ice(t)) or protein (m_pr(t)). Saturation recovery measurements lead essentially to the same results (see Fig. S1). Eq. 1 was previously shown to be the solution of modified Bloch equations describing the relaxation and cross-relaxation of two magnetization baths (see SI Text) (31, 32).

Fig. 1.

¹³C detected ¹H R₁ inversion-recovery measurements of AFP III and ubiquitin in frozen solution (-35 °C sample temperature). The normalized intensities for the carbonyl lines are plotted. Biexponential fits are shown as solid lines, monoexponential fits as dashed lines. (A) AFP III in H₂O. (B) AFP III in H₂O + 5 mM Cu(II)-EDTA. (C) AFP III in D₂O. (D) Ubiquitin in H₂O. The curves of AFP III in D₂O and ubiquitin in H₂O can be fit to monoexponential decays giving relatively fast R₁ rates (3.44 ± 0.07 s^-1 for ubiquitin in H₂O and 8.3 ± 0.5 s^-1 for AFP III in D₂O). However, the curves of AFP III in H₂O and with Cu(II)-EDTA can only be fit biexponentially. The fits gave values of , , , and for AFP III in H₂O and , , , and for AFP III doped with Cu(II)-EDTA. Please note that all these values are concentration dependent. The biexponential curves indicate that AFP III is, in this case, in contact to a second proton bath.

Generally, the apparent relaxation rates and do not correspond to the spin-lattice relaxation rates of the ice () and protein () ¹H baths. Both m_ice(t) and m_pr(t) have to be measured separately to find a solution for the individual relaxation rates and as well as the cross-relaxation rates k_ice and k_pr. However, we were not able to measure m_ice(t) because the wide ice line decays already in the dead time of the detection (∼10 μs). The fact that AFP III in frozen H₂O has a biexponential relaxation curve indicates that the proton bath of AFP III is in intermediate exchange with the proton bath of ice. The biexponential behavior of AFP III in Cu(II)-EDTA doped H₂O supports this hypothesis because the copper changes the relaxation behavior of the individual proton baths but not the fact that they are in intermediate exchange. Following this picture, ubiquitin in H₂O and AFP III in D₂O both show monoexponential relaxation because ubiquitin does not bind to ice, and frozen D₂O does not provide a second proton bath, respectively.

Cross-Saturation Experiments.

To confirm the AFP III-ice coupling suggested by the relaxation measurements described above and to indirectly detect the NMR line due to protons in ice, we measured ¹³C detected ¹H cross-saturation spectra, also known as Z-spectra (34). In the original experiment, a weak off-resonance pulse on the broad ¹H line of the immobilized protein led to the decrease of the solvent ¹H line via cross saturation. We performed an inverted version of this experiment by applying a weak ¹H off-resonance pulse to the broad ¹H line of ice and measured its influence on the relatively narrower ¹H line of the predominantly perdeuterated protein. After the weak ¹H presaturation pulse with varying transmitter frequency, we transferred the remaining protein ¹H magnetization by ¹H-¹³C cross polarization (CP) and detected it on ¹³C. If the presaturation pulse saturates the magnetization of the protein ¹H spins, the intensity of the ¹³C spectrum is reduced. The protein protons are saturated when the ¹H transmitter frequency is on resonance with the protein ¹H line or when the spin baths of the ice and the protein are coupled and the ¹H transmitter frequency is on resonance with the ice line (i.e., via cross saturation). Fig. 2 shows the Z-spectra recorded on ²H-¹³C-¹⁵N labeled AFP III and ubiquitin in frozen H₂O solution.

Fig. 2.

¹³C detected ¹H cross-saturation spectra (Z-spectra) of ²H-¹³C-¹⁵N labeled AFP III and ubiquitin in frozen solution (-35 °C sample temperature). The normalized intensities of the carbonyl lines after 0.1 s, 0.5 s, and 2 s of ¹H presaturation are plotted with squares, circles, and diamonds, respectively (points are connected with dash-dotted, dashed, and solid lines to guide the eye). (A) ²H, ¹³C, and ¹⁵N labeled AFP III in H₂O. (B) ²H, ¹³C, and ¹⁵N labeled ubiquitin in H₂O. For short ¹H presaturation the Z-spectra of AFP III and ubiquitin are very similar. However, for long ¹H presaturation pulses the Z-spectra of AFP III are much broader than those of ubiquitin. This shows that AFP III establishes direct contact with the ice and can, therefore, be cross-saturated by pulsing on the much broader ice line.

Long presaturation pulses of 0.5 and 2 s lead to broad Z-spectra (∼100 kHz linewidth) for AFP III in frozen solution (Fig. 2A). For ubiquitin, the linewidth of the corresponding Z-spectra (Fig. 2B) is about 50 kHz. The linewidth of the Z-spectra of ubiquitin is relatively independent from the length of the presaturation pulse, only increasing slightly from approximately 40 kHz at a pulse length of 0.1 s to approximately 50 kHz at pulse lengths of 0.5 and 2 s. In contrast to this, the linewidth of the AFP III Z-spectra depends strongly on the length of the presaturation pulse: At a pulse length of 0.1 s the linewidth is the same as the corresponding linewidth of the Z-spectrum of ubiquitin (∼40 kHz). However, presaturation pulses of 0.5 and 2 s increase the linewidth to 80 and 100 kHz, respectively. The Z-spectra presented in Fig. 2 are essentially smooth, and we could not detect any spinning side bands by using smaller step-sizes or by rotor synchronizing the step size to the MAS frequency of 9 kHz.

The Z-spectra confirm our hypothesis that AFP III directly interacts with ice in frozen solution: At short ¹H presaturation times, the effect of cross saturation via ¹H-¹H spin diffusion is negligible and we only observe the relatively narrow line of the amide protons of AFP III. With longer ¹H pulse lengths, the Z-spectra of AFP III are dominated by the broad ¹H line of ice. The Z-spectra of ubiquitin show no broadening with increasing ¹H pulse length, which confirms that ubiquitin has no direct contact with ice.

¹H-¹³C HETCOR Spectra.

The experiments described above show that AFP III establishes direct contact to ice. But does the entire hydration shell of AFP III disappear due to the ice interaction? In order to answer this question, we measured a set of ¹H-¹³C CP heteronuclear correlation (HETCOR) spectra that detect H₂O-protein contacts. We recorded simple HETCOR spectra correlating carbons with their directly bound protons and a variant of the HETCOR experiment, which had an additional ¹H-¹H spin-diffusion period between t₁ on ¹H and the CP to ¹³C (13). This additional ¹H-¹H mixing permits the detection of correlations between protons and carbons that are further away or not even in the same phase.

These two variants of the HETCOR experiment gave very different spectra for ²H-¹³C-¹⁵N labeled AFP III (90% ²H labeling degree) in frozen H₂O solution as can be seen from Fig. 3. Because our proteins were purified in protonated buffers under denaturing conditions and redissolved in protonated buffers, we assume all exchangeable groups to be protonated. As expected, the simple HETCOR spectrum in Fig. 3A shows two large cross peaks between the amide protons and the C_α and C′ of the protein and a cross peak between methyl protons and methyl carbons. The HETCOR spectrum with additional ¹H-¹H mixing (Fig. 3B) does not show these cross peaks but shows strong correlations between the C_α and C′ and a ¹H line at about 5 ppm. We assigned this line to the protons in the hydration shell of AFP III. Furthermore, Fig. 3B shows weaker correlations between the ¹H line at 5 ppm and side chain carbons.

Fig. 3.

¹H-¹³C HETCOR spectra of ²H-¹³C-¹⁵N labeled AFP III in frozen H₂O solution. Spectra were recorded at 9.3 T, -35 °C, and a MAS frequency of 9 kHz. (A) CP HETCOR spectrum. The simple HETCOR spectrum shows correlations between the amid protons and the closest carbons, i.e., the C′ and C_α of the protein leading to the two broad peaks at about 170 and 50 ppm in the ¹³C dimension and about 8–9 ppm in the ¹H dimension as well as a cross peak between the methyl carbons and protons at about 20 ppm and 1 ppm, respectively. (B) HETCOR spectrum with additional ¹H spin-diffusion mixing of 25 ms between t₁ and the CP to ¹³C. This spectrum shows two strong cross peaks between C′ and C_α of the protein and the hydration shell at about 5 ppm in the ¹H dimension as well as less intense cross peaks between the hydration shell and the other aliphatic carbons at about 40–20 ppm. The resonance of the hydration shell is highlighted with a gray bar in both spectra. Interestingly, the cross peaks to the amid protons are missing in B because the ¹H mixing period leads to a decay of the amid–proton magnetization.

As a control, we recorded the same set of HETCOR spectra on fully ²H-¹³C-¹⁵N labeled ubiquitin in frozen H₂O solution. As can be seen from Fig. S2, the HETCOR spectra of ubiquitin and AFP III look essentially the same. Due to a higher ²H labeling degree (close to 100%), we could observe no cross peak between methyl protons and methyl carbons in ubiquitin in the simple HETCOR spectrum (Fig. S2A). The ¹H linewidth of the hydration shell protons is < 1 kHz for both proteins. The HETCOR spectra recorded on ubiquitin and AFP III give evidence of an, at least partially, intact hydration shell for both proteins.

²H Spectra.

As discussed above, we were not able to measure R₁ of the ice protons directly. Therefore we asked: Does AFP III influence the ²H R₁ relation rate of frozen D₂O, thus providing additional evidence for molecular contacts between AFP III and ice? To answer this question, we recorded ²H saturation recovery curves of frozen D₂O with AFP III and ubiquitin as well as of frozen D₂O only. The results of these measurements can be seen in Fig. 4A. The saturation recovery curves of frozen D₂O and frozen D₂O plus ubiquitin fit perfectly to single exponentials, and we found R₁ to be relatively slow for both samples. However, the R₁ of frozen D₂O with AFP III is much faster, and the saturation recovery curve is somewhat better explained with a double exponential fit (see solid line in Fig. 4A). This result shows that AFP III provides relaxation sinks by binding to ice and, therefore, influences the relaxation of ice. Neither AFP III nor ubiquitin have an influence on the quadrupolar ²H tensor of frozen D₂O, and all our ²H spectra of frozen D₂O look the same as the red spectrum shown in Fig. 4B. This broad ²H tensor has a quadrupolar splitting of about 185 kHz, which is slightly smaller than the values reported for immobilized D₂O quadrupolar coupling of 216 kHz for hexagonal ice or 192 kHz for polycrystalline ice (35, 36). The red spectrum in Fig. 4B was recorded on a D₂O plus ubiquitin sample with relatively long recovery delays of 10 min. The blue spectrum in Fig. 4B was recorded on the same sample with a much shorter recovery delay of 1 s. This spectrum shows an intense central peak (∼3.5 kHz) in addition to the less intense sideband manifold of the quadrupolar ice tensor. This sideband free central peak, which comes from the relatively dynamic, liquid ²H of the hydration shell, could not be detected in spectra of frozen D₂O only and spectra of frozen D₂O plus AFP III. In the case of D₂O with AFP III, we could not detect the liquid ²H line probably because the faster relaxation of the ice in this sample prevented successful use of relaxation filtering.

Fig. 4.

(A) Saturation recovery curves of frozen D₂O with AFP III (squares), ubiquitin (circles), and D₂O only (diamonds). The slow R₁ values of pure D₂O and the ubiquitin solution can be fit with a single exponential (dashed line). The relaxation rates are R₁ = 0.022 ± 0.002 s^-1 for D₂O plus ubiquitin and R₁ = 0.0136 ± 0.0001 s^-1 for pure D₂O. However, the significantly faster R₁ (0.24 ± 0.01 s^-1) of the AFP III solution is best fit with a biexponential fit (solid line). (B) ²H 1D direct excitation spectrum of frozen D₂O + ubiquitin solution, recorded at 9 kHz MAS, -35 °C, and 9.3 T. A spectrum with a short recovery delay of 1 s is depicted in blue and dominated by the sideband free central peak coming from the hydration water. The spectrum in red was recorded with a recovery delay of 600 s and is dominated by the broad quadrupolar tensor of the ice line. The quadrupolar splitting of 185 kHz is slightly smaller than theoretical value for immobilized D₂O quadrupolar coupling of 216 kHz for hexagonal ice or 192 kHz for polycrystalline ice.

Discussion

Because the slow R₁ and fast R₂ of ice impedes a lot of NMR spectroscopic techniques such as CP, we introduced indirect ¹H and ²H spin-diffusion based methods such as cross relaxation and cross saturation to characterize the ice contact of AFP III. The R₁ and cross-saturation experiments presented above suggest, by showing both cross-relaxation and cross-saturation effects, that AFP III is making direct contact with the bulk ice phase of the frozen solution. In contrast to this, ubiquitin in frozen solution did not show either of these effects and is essentially decoupled from ice.

The monoexponential R₁ rates we measured in our ¹H R₁ inversion-recovery experiments can be explained by fast exchange processes: AFP III in frozen D₂O has only one strongly coupled proton bath (i.e., the protein ¹H phase) and thus only a single R₁ rate. The ubiquitin in a frozen H₂O solution sample has three different ¹H phases (i.e., the proton baths of ice, hydration shell, and protein). The protein and the ice phase are not in exchange as shown by the Z-spectra and the relatively fast ¹H R₁ of this system. Our observation that ubiquitin does not significantly alter the R₁ of D₂O ice (Fig. 4A) further confirms this interpretation. The ¹H baths of the protein and the hydration shell, on the other hand, are in fast exchange, which explains the monoexponential relaxation and the strong hydration water–protein cross peaks in the ¹H-¹H spin-diffusion HETCOR (Fig. S2B). The three proton bath system of this sample is reminiscent of the three classes of protons used to describe the water relaxation of nonfrozen polymer solutions by Halle (37). However, the exchange between ice and hydration shell protons is expected to be much slower than the hydration shell water exchange in Halle’s model.

The biexponential nature of the inversion-recovery curves measured for AFP III in frozen H₂O and frozen Cu(II)-EDTA doped H₂O solutions can only be explained with the presence of slow to intermediate exchange processes; namely the ¹H-¹H spin diffusion between ice and AFP III’s ice-binding surface. This spin diffusion occurs independently of the R₁ values of the individual spin baths, leading to very different but still biexponential inversion-recovery curves for AFP III in Cu(II)-EDTA doped and nondoped, frozen H₂O solutions. Besides that, the proton bath of AFP III is also in fast exchange with the remaining hydration shell of its non-ice-binding surface as shown by the spin-diffusion HETCOR spectrum (Fig. 3B).

We could not calculate the exact cross-relaxation rates k_ice and k_pr from and because we were not able to measure m_ice(t). The R₁ measurements of frozen D₂O with AFP III qualitatively confirm the direct interaction of AFP III with ice but cannot replace the measurement of m_ice(t). However, a semiquantitative estimation of the cross-relaxation rates tells us a lot about the strength of the AFP III–ice interaction: In order to get an approximation for the cross-relaxation rates, we did a least square fit of the inversion-recovery curve shown in Fig. 1A (i.e., AFP III in H₂O) using a reduced biexponential model and fixed values for and (see SI Text). With and , i.e., the reported by Nunes and coworkers (11) and the we measured of ubiquitin in H₂O, the reduced model still gave a good fit. We used an F-test (38) to show that the full biexponential model does not fit the data better than the reduced model (see SI Text). Using the fit with fixed R₁s, we obtained cross-relaxation rates of k_ice = 0.3 ± 0.1 s^-1 and k_pr = 6 ± 1 s^-1. This leads to an ice–proton to protein–proton ratio of 1∶20, which is smaller than the ice–proton to AFP III plus hydration shell–proton ratio of about 1∶50 we calculated (see SI Text). This difference in the ice–proton to protein–proton ratios suggests that AFP III is not in equivalently strong contact with all ice in the sample, which is likely due to irregularities in the ice lattice and the fact that not all the ice is right at the ice-binding surface.

What do these cross-relaxation rates mean regarding the ice-binding strength of AFP III? To answer this question, we estimated the cross-relaxation rate using equations describing spin diffusion in a spherical two-component system (see SI Text and ref. 39). The estimated rate of k_est ≈ 3000 s^-1 is many orders of magnitude faster than the cross-correlation rates we obtained from our constrained fit of the inversion-recovery data. Therefore, the approximation suggests that AFP III binds only weakly to ice by either dedicating a small fraction of its surface to ice binding or by having a weak ice-binding constant. Our estimation for the cross-relaxation rate k_est is expected to be a better description of the protein below the freezing temperature of the hydration shell where ice will completely surround the protein. In this case, the whole protein surface will be in contact with ice and both proton phases will be in fast exchange (i.e., ) and will share a common, slow R₁ relaxation rate.

The cross-saturation experiments (Z-spectra) on ubiquitin support the results of our ¹H and ²H R₁ measurements: At -35° ice and ubiquitin have no direct contact, the corresponding proton baths are practically decoupled, and the Z-spectra recorded on ubiquitin in frozen solution showed essentially no ice–protein cross saturation. AFP III binds directly to ice via its ice-binding surface and, therefore, shows cross saturation in our Z-spectra. However, the cross-saturation rate is not fast, and about 0.5–2 s of presaturation are needed to reach full cross saturation. This presaturation time is on the same order as the rate constants, further confirming that we see intermediate magnetization exchange between the ice and AFP III. The linewidth of the Z-spectra is a function of R₁, R₂, k, and the rf-field strength (34) and can, therefore, not be compared to the ¹H linewidth of the HETCOR spectra without spin diffusion of about 1 kHz.

We recorded HETCOR spectra and HETCOR spectra with additional spin diffusion on ubiquitin and AFP III to confirm the non-ice-binding of ubiquitin and to probe whether AFP III bound ice with its entire surface or not. The ¹H linewidth of less then 1 kHz that we measured for the hydration shell of ubiquitin and AFP III corresponds well to previously reported hydration water linewidths of up to 0.85 kHz (23, 25). The chemical shift we measured for the hydration water is 5ppm, which, given the broad linewidth, is not distinct from the value for bulk liquid water. We were also able to detect the hydration shell in our frozen D₂O spectra in the presence of ubiquitin via a liquid central ²H peak. Similar ²H spectra of the hydration shell of crystalline crambin were reported by Usha and Wittebort (24). Together with the absence of any cross relaxation and cross saturation between ice and ubiquitin (discussed above), this shows that ubiquitin is completely immersed into its hydration shell at -35°. The corresponding spin-diffusion HETCOR spectrum of AFP III in frozen H₂O solution (Fig. 3B) shows that at least parts of its hydration shell do not freeze together with the bulk water. Therefore, AFP III is not binding ice with its entire surface but only with its ice-binding surface. This explains why only the ice-binding surface of AFP III experienced significant ¹³C chemical shift changes upon freezing in our previous study (6). The nonfrozen hydration shell presumably leaves ubiquitin and the non-ice-binding surface of AFP III in a similar environment as in solution, explaining the negligible ¹³C chemical shift differences we observed for these proteins between solution and frozen solution.

Fig. 5 presents a model for the different solvent interactions of AFP III and ubiquitin in frozen solution. Most soluble proteins are likely to behave like ubiquitin (Fig. 5 Right) at moderate freezing temperatures: Their hydration shell does not freeze until a certain temperature (around -50 °C in the case of ubiquitin) (25). However, as soon as the bulk solution freezes, they stop tumbling isotropically, become arrested inside the ice, and can be observed using dipolar-based solid-state NMR methods. When the sample temperature is lowered further, the protein hydration shell will eventually freeze, and the protein will then be in direct contact with the ice lattice. This picture would explain the relatively slow R₁ rates and broad lines reported by Tycko and coworkers for solid-state NMR spectra on frozen protein solutions well below -100 °C (i.e., likely below the freezing point of the protein hydration shell) (26, 40, 41).

Fig. 5.

Model illustrating the different solvent interactions of AFP III and ubiquitin in frozen solution at moderate freezing temperatures of -35 °C: The hydration water at the ice-binding interface of AFP III can be displaced, permitting direct ice–protein contacts. The rest of AFP III’s hydration shell stays intact under these conditions. In contrast to this, the non-ice-interacting protein ubiquitin keeps its entire hydration shell, which prevents ubiquitin from interacting with the ice lattice.

What distinguishes AFPs from other, non-ice-binding proteins? In the present study we showed that part of the surface of AFP III binds directly to ice crystals at the moderate freezing temperature of -35 °C. Our previous chemical shift perturbation study on AFP III indicated that the protein binds ice already at -15 °C, and functional studies done by Davies and coworkers showed that AFP III binds to ice crystals right at the freezing point of the bulk solution (3, 6). Molecular dynamics simulations suggested that AFP III preforms a more ordered, ice-like water structure at its ice-binding interface at the freezing point (9). This explains why AFP III does not hold onto its hydration shell once the bulk solution freezes but displaces its hydration shell at the ice-binding surface and establishes direct contact to ice. We showed that this is only true for the ice-binding interface of AFP III, because we found other parts of its hydration shell to be nonfrozen at -35 °C.

Conclusions

In this study we used ¹H and ²H R₁ measurements as well as Z-spectra to probe directly the interaction of the ice-binding protein AFP III with ice. This interaction appears to be specific to ice-binding proteins in that ubiquitin did not exhibit any evidence for ice binding. Besides probing the protein–ice interface of AFP III, our results show that spin-diffusion HETCOR spectra and ²H NMR on frozen protein solutions at moderate freezing temperatures is a very efficient way to directly study the hydration shell of proteins. The hydration shell can be distinguished from the bulk ice because it has a very different R₁ rate. Using these techniques, we were able to prove the presence of a nonfrozen hydration shell for ubiquitin and AFP III at moderate freezing temperatures of -35 °C. Antifreeze proteins are special in that they are able to displace part of their hydration shell (i.e., at its ice-binding surface) at this temperature and establish direct contact to the ice lattice.

Materials and Methods

Protein Expression and Purification.

Uniformly ¹³C and ¹⁵N labeled AFP III was expressed and purified as described previously (6). The expression of uniformly ²H, ¹³C, and ¹⁵N labeled AFP III and ubiquitin was done following a protocol similarly to the one described by Tugarinov and coworkers (42) except that Escherichia coli cells were slowly adapted to deuterated M9 minimal medium by going from a starter culture in 50 ml LB, 0% D₂O into a sequence of M9 minimal media containing varying amounts of D₂O: 350 ml 0% D₂O, 2 l 70% D₂O, and 1 l 95% or 100% D₂O for AFP III or ubiquitin, respectively. Uniformly ²H, ¹³C, and ¹⁵N labeled AFP III was purified exactly as nonperdeuterated AFP III. Uniformly ²H, ¹³C, and ¹⁵N labeled ubiquitin was purified as follows: The cell pellet was redissolved in 20 ml glacial acetic acid. The cell lysate was centrifuged at 4,000 g for 1 h. Afterward, the supernatant was neutralized to pH 5 with addition of 5 M KOH. The supernatant was then dialyzed twice against deionized water and a third time against 50 mM pH 4.5 ammonium acetate solution. The cell lysate was loaded onto a SP sepharose column equilibrated with 50 mM ammonium acetate, pH 4.5 and eluted with 50 mM ammonium acetate, pH 5.5. Purity and labeling degree (close to 100% for ubiquitin, about 90% for AFP III) for both proteins was confirmed by mass spectrometry.

NMR Spectroscopy.

All solid-state NMR spectra were recorded on a Varian 400 MHz Infinityplus spectrometer using a 4 mm T3 probe. Samples were spun at a MAS frequency of 9 kHz. The sample temperature of -35 °C was determined externally using Pb(NO₃)₂ (43). ¹³C chemical shifts were referenced externally with adamantane and ¹H shifts were referenced indirectly using the ratio of Ξ = 25.144953% (44). Generally, the CP rf-field strengths were 50 and 59 kHz for ¹³C and ¹H, respectively, and the contact time of the CP was 1 ms. ¹³C and ¹H hard-pulses were done with rf-field strengths of 50 and 100 kHz, respectively. All ¹³C dimensions were decoupled with 80–100 kHz of ¹H TPPM decoupling (45). The ¹³C detected ¹H inversion-recovery curves were recorded with 16 acquisitions per τ step and a recycle delay of 180 s. Z-spectra (34) were recorded with a presaturation rf-field strength of 1 kHz, a recycle delay of 3 s, and 128 acquisitions per presaturation rf-field offset that was varied in steps of 10 kHz around the on-resonance condition at about 3 ppm. ¹H-¹³C HETCOR with and without additional ¹H-¹H spin-diffusion period of 25 ms were recorded with a spectral width of 50 kHz in the direct and 25 kHz in the indirect dimension. For each of the 64 t₁ increments 256 and 512 acquisitions were recorded for ubiquitin and AFP III, respectively. ²H spectra were recorded with an rf-field strength of 100 kHz. The ²H saturation recovery curves were recorded with 32 acquisitions for each τ recovery step.