Introduction

Antifreeze proteins (AFPs) are found in a variety of cold-adapted organisms including fish, insects, plants, bacteria and fungi and promote survival at subzero temperatures by binding to ice crystals and decreasing the freezing temperature of body fluids (Davies et al. 2002; Ewart et al. 1999; Fletcher et al. 2001; Jia and Davies 2002; Yeh and Feeney 1996). Fish AFPs comprise several structurally diverse classes of proteins and are grouped into four types on the basis of their structures: type I AFPs are alanine-rich α-helical proteins (Harding et al. 1999); type II AFPs are cysteine-rich globular proteins containing five disulfide bonds (Liu et al. 2007); type III AFPs are ~7-kDa globular proteins (Jia et al. 1996); and type IV AFPs are glutamate- and glutamine-rich α-helical proteins (Gauthier et al. 2008).

One of most widely studied classes of AFPs is the type III from arctic fish such as the ocean pout (Zoarces americanus) and Japanese notched-fin eel pout (Zoarces elongates Kner). The type III AFPs have been categorized into two subgroups, quaternary-amino-ethyl (QAE) and sulfopropyl (SP) Sephadex-binding, based on differences in their isoelectric points (Fig. 1a) (Hew et al. 1988). The QAE proteins can be further divided into two subgroups, QAE1 and QAE2 (Fig. 1a) (Nishimiya et al. 2005). Among them, only the QAE1 isoforms exhibit full thermal hysteresis (TH) activities, whereas the others have extremely low or no such activities (Nishimiya et al. 2005). The HPLC12 isoform from the ocean pout, Z. americanus, which belongs to the QAE1 subgroup, has been extensively studied by NMR (Chao et al. 1994; Sönnichsen et al. 1996), X-ray crystallography (Jia et al. 1996), mutagenesis (Chao et al. 1994; DeLuca et al. 1998; Graether et al. 1999; Jia et al. 1996), and molecular dynamics (Antson et al. 2001; Chen and Jia 1999). The X-ray structure of the HPLC12 isoform contains one α-helix (residues 37–40), three 310-helices (residues 19–21, 33–36, and 57–59), and two β-strands (residues 4–7 and 22–25) (Fig. 1a) (Jia et al. 1996). The ice-binding surface (IBS) is comprising residues Q9, P12, N14, T15, A16, T18, V20, M21 and Q44 (Jia et al. 1996) and the flatness of IBS is thought to be important for antifreeze activity (Yang et al. 1998). The molecular dynamics study reported that five additional residues, I13, L19, V41, S42, and K61 are also potentially able to interact with ice (Chen and Jia 1999). Among them, five hydrophilic residues (Q9, N14, T15, T18, and Q44) have the potential to form hydrogen bonds (H-bonds) with water molecule in the ice crystal (Fig. 1b) (Sönnichsen et al. 1996; Jia et al. 1996). The residues, L10, I13, L19, V20, and V41 contributed to ice-binding via hydrophobic interaction with ice and thus replacement by alanine caused significant loss of antifreeze activity (Baardsnes and Davies 2002). Recently, it was reported that the IBS consists of two adjacent surfaces that are flat and relatively hydrophobic (see Fig. 1b) (Garnham et al. 2012). The QAE1 isoform binds both pyramidal and primary prism planes of ice crystal and is able to halt the growth of ice, whereas the SP and QAE2 isoforms exhibit low or no TH activities by binding to only pyramidal ice plane (Garnham et al. 2012). The inactive QAE2 isoform could be converted into a fully-active QAE1-like isoform that is able to bind primary prism ice plane by changing four residues (V9Q/V19L/G20V/I41V) (Garnham et al. 2012).

Fig. 1
figure 1

a Multiple sequence alignment of the type III AFP family and AFL domains. NfeAFP2, nfeAFP8, and nfeAFP11 are the isoforms 2, 8, and 11 of the Notched-fin eelpout type III AFP, respectively. hAFL, mAFL, and nmAFL are the AFL domains of human, mouse, and Neisseria meningitides SAS, respectively. Numbering and secondary structure elements for HPLC12 are shown above the sequence. Red and green bars indicate residues composing the pyramidal and primary prism ice plane segments of the IBS, respectively. Five residues (Q9, N14, T15, T18, and Q44) that can form H-bonds with ice crystals are highlighted in yellow letters. The asterisk and dot symbols represent invariant or nearly invariant residues in the type III AFP family, respectively. Blue bars indicate the hydrophobic residues that are invariant or nearly invariant residues in the type III AFP family. Substrate-binding R314 in Neisseria meningitides SAS is highlighted in orange. b Two IBS of HPLC12 and c the corresponding surfaces of hAFL. Coordinates were derived from the HPLC12 X-ray crystal structure (PDB id = 1MSI) (Jia et al. 1996) and hAFL NMR structure (PDB id = 1WVO) (Hamada et al. 2006). The pyramidal and primary prism ice plane segments of the IBS in the HPLC12 isoform and the corresponding segments in hAFL are shown in red and green, respectively

Full size image

Sialic acids (N-acetylneuraminic acids) play a vital role in a variety of biological functions, such as transmembrane signaling, cell growth, and cell adhesion (Schauer 2004). Sialic acids are synthesized by a conserved enzymatic pathway that includes sialic acid synthase (SAS). Bacterial SAS converts phosphoenolpyruvate (PEP) and N-acetylmannosamine (ManNAc) into N-acetylneuraminic acid (NeuNAc, or sialic acid) (Masson and Holbein 1983). Mammalian SAS catalyzes the condensation of PEP and ManNAc 6-phosphate (ManNAc-6P) into NeuNAc 9-phosphate (NeuNAc-9P) which is dephosphorylated into sialic acid (Lawrence et al. 2000). Both enzymes consist of two distinct domains, an N-terminal catalytic NeuB domain (~250 amino acid residues) and a C-terminal antifreeze-like (AFL) domain (~75 amino acid residues). The crystal structure of an SAS from Neisseria meningitides, in complex with Mn2+, PEP, and ManNAc, revealed a unique domain-swapped homodimer consisting of an N-terminal (α/β)8 barrel-type fold and a C-terminal AFL domain with high structural similarity to the type III AFPs (Gunawan et al. 2005). The AFL domain of the N. meningitides SAS (nmAFL) contributed to the substrate binding of the NeuB domain of the opposite monomer via hydrogen bonding (H-bonding) interaction between the side chain of R314 (see Fig. 1a) and the acetyl oxygen of ManNAc (Gunawan et al. 2005). It was proposed that the deletion of the AFL domain in human SAS (hAFL) would abolish the SAS activity, indicating that the hAFL domain might play an important role in substrate binding by the SAS (Hamada et al. 2006). An NMR study of the hAFL domain found that the structure was similar to those of the type III AFPs (Hamada et al. 2006).

Despite having very similar structures, the type III AFP and AFL have very different temperature dependencies of stability and activity. The type III AFP is considered as psychrophilic protein because of the lower working temperature (around 0 °C) while AFL is thought to be mesophilic protein with the working temperature in the range of 20–37 °C. Comparative molecular dynamics studies have been used to elucidate temperature dependence of the dynamics of psychrophilic and mesophilic proteins (Brandsdal et al. 1999; Olufsen et al. 2007; Papaleo et al. 2006, 2007, 2008). A recent molecular dynamics simulation study suggested that the type III AFP showed increased overall flexibility at higher temperature compared to the AFL protein (Kundu and Roy 2009). However, the molecular basis of their distinctive behaviors is not well understood.

Heteronuclear NMR spectroscopy is a suitable experimental tool to compare the thermostability and structural flexibility of homologous psychrophilic and mesophilic proteins. In this study, we have investigated backbone dynamics of the type III AFP from ocean pout, Z. americanus (HPLC12 isoform) and the AFL domain of human SAS (hAFL) at various temperatures. We also characterized the structural/dynamic properties of the IBS of the HPLC12 and hAFL proteins by analyzing the temperature gradient of the amide proton chemical shift and its correlation with chemical shift deviation (CSD) from random coil. Our results show that the two proteins have entirely different backbone dynamics, especially at low temperature (5 °C). HPLC12 exhibits mostly rigid at low temperature and this rigidity of IBS might be crucial for efficient hydrophobic and H-bonding interaction with ice crystal. Interestingly, most residues of hAFL are relatively rigid at higher temperature (25 °C) while the substrate-binding residues experience distinctive flexible backbone dynamics. Based on our study, we suggest that the dynamic properties of each protein give insight into the underlying causes for the different biological working temperatures of HPLC12 and hAFL. It is very likely that each protein maintains a relatively rigid backbone structure overall, but that the functionally important residues exhibit distinctive dynamics for substrate binding (rigid for ice-binding or flexible for sialic acid binding).

Materials and methods

Sample preparation

The DNA coding sequences for the HPLC12 (M1-A66) and hAFL (L291–S359) were purchased from BIONEER Inc. (Korea) and cloned into E. coli expression vectors pET20b (Novagen, USA) with a C-terminal histidine-affinity tag (His-tag) and pET28a (Novagen, USA) with an N-terminal His-tag, respectively. To produce uniformly 15N-labeled or 13C, 15N-labeled HPLC12, C41(DE3) cells were grown in M9 minimal medium that contained 1 g/L 15NH4Cl and/or 2 g/L 13C-glucose as the sole nitrogen and carbon sources. Isotopically-labeled hAFL domain was expressed in a similar manner but in BL21(DE3) cells. The expressed proteins were purified by Ni–NTA affinity and followed by Sephacryl S-100 gel filtration chromatography (GE Healthcare, USA) on a GE AKTA Prime Plus. The concentrations of both proteins were measured using a Pierce BCA Protein Assay Kit (Thermo Scientific, USA). For NMR experiments, the purified proteins were concentrated to 1 mM in a 90 % H2O/10 % D2O buffer containing 10 mM sodium phosphate (pH 8.0) and 100 mM NaCl.

NMR experiments

All of the 1H, 13C, and 15N NMR experiments were performed on an Agilent DD2 700-MHz spectrometer (GNU) equipped with a cold probe. All three-dimensional (3D) NMR experiments were carried out with 1 mM 13C, 15N-labeled HPLC12 and hAFL proteins at 25 °C. All 2D 1H/15N-HSQC spectra for HPLC12 and hAFL were obtained using 1 mM 15N-labeled protein samples at various temperature. 1D NMR data were processed with either VNMR J (Agilent, USA) or FELIX2004 (FELIX NMR, USA) software, while the 2D and 3D data were processed with NMRPIPE (Delaglio et al. 1995) and analyzed with Sparky (Goddard and Kneller 2003). 2-2-Dimethyl-2-silapentane-5-sulfonate was used as an external reference.

The backbone resonance assignments for HPLC12 and hAFL were obtained from the following 3D experiments: CACB(CO)NH, HNCACB, HNCA, HNCO, and NOESY-1H/15N-HSQC. The 1H/15N-HSQC spectra of the HPLC12 and hAFL were acquired at temperatures of 5–25 °C in 5 °C increments. The Δδ/ΔT values of the amide protons were determined from linear variation of the NH chemical shifts with temperature. The amide proton CSDs (δobs–δrc) at 25 °C were calculated from the observed chemical shifts (δobs) and the corresponding random coil (δrc) values taken from the literature (Andersen et al. 1997).

Backbone dynamics parameters, longitudinal R 1 relaxation rates, transverse R 2 relaxation rates, and {1H}-15N heteronuclear NOEs were measured using 15N-labeled HPLC12 and hAFL protein samples at 5, 15, and 25 °C. 2D 1H/15N-HSQC spectra for backbone dynamics measurements were acquired with 2,048 (HN) × 64 (N) data points and 8 scans. Spectral widths of 7,716 Hz (11 ppm) and 2,414 Hz (34 ppm) were used in the HN and N dimensions, respectively. R 1 values were measured in a series of spectra with relaxation delays of 50, 100, 150, 200, 250, 300, 350, 400, 500, and 600 ms. R 2 measurements were taken with relaxation delays of 10, 30, 50, 70, 90, 110, 130, 150, 190, and 230 ms. For evaluation of {1H}-15N heteronuclear NOE values, four different data sets with and without an initial proton saturation (2.5 s period) were measured. Reduced spectral density analysis was performed using the following relationships as described by Bracken et al. (1999):

$$ \sigma_{NH} = R_{1} (NOE - 1)\gamma_{N} /\gamma_{H} $$
(1)
$$ J(0.87\omega_{H} ) = 4\sigma_{NH} /(5d^{2} ) $$
(2)
$$ J(\omega_{N} ) = (4R_{1} - 5\sigma_{NH}^{{}} )/(3d^{2} + 4c^{2} ) $$
(3)
$$ J_{e} (0) = (6R_{2} - 3R_{1} - 2.72\sigma_{NH} )/(3d^{2} + 4c^{2} ) $$
(4)

where d = (μo hγNγH/8π2)(r −3NH ) and c = ωNΔσ/√3, μo is the permeability of free space, h is Planck’s constant, γH and γN are the gyromagnetic ratios of the 1H and 15N nuclei, respectively, ωH and ωN are the Lamor frequencies, r NH is the average 1H–15N bond length (1.02 Å), and Δσ is the 15N chemical shift anisotropy (−160 ppm). The effective J(0) [J e(0)] contains contributions from the chemical exchange term, if present.

For well-ordered sites in a macromolecule, the correlation time of overall tumbling motion (τm) is obtained from the spectral density functions as (Bracken et al. 1999):

$$ \tau_{m} = \frac{{\sqrt {\{ J(0) - J(\omega_{N} )\} /J(\omega_{N} )} }}{{\omega_{N} }}. $$
(5)

Results and discussion

NMR resonance assignments

The 1H/15N HSQC spectra of the HPLC12 isoform and hAFL domain at 25 °C are shown in Supplementary Fig. S1. Amide backbone resonance assignments for the HPLC12 were made by heteronuclear three-dimensional (3D) NMR experiments at pH 8.0. In hAFL, some amide cross-peaks were missing at pH 8.0, so backbone resonance assignments were conducted at pH 6.0 (see Supplementary Fig. S2). Resonance assignments of hAFL at pH 8.0 were made by comparing of 1H/15N-HSQC spectra acquired at pH 6.0, 6.5, 7.0, and 8.0 and were confirmed by 3D NOESY-1H/15N-HSQC and TOCSY-1H/15N-HSQC spectra at pH 8.0.

Temperature dependence of 1H/15N-HSQC spectra of HPLC12 and hAFL

The NMR chemical shift is sensitive to the environment of the observed nuclei as well as to temperature-related structural or dynamic changes (Hong et al. 2010; Ohnishi and Urry 1969). As the temperature was lowered, significant cross peak movements were observed in the 1H/15N HSQC spectra of the HPLC12 and hAFL proteins (Supplementary Fig. S3). Figure 2 shows the temperature gradients of amide proton chemical shifts (Δδ/ΔT) as a function of residue number (in hAFL, the residue number is starting with residue L291 of SAS as residue L1). In both proteins, most residues have Δδ/ΔT values >−4.6 ppb/K, indicating that these amides are H-bonded (Cierpicki and Otlewski 2001; Cierpicki et al. 2002). Residues with Δδ/ΔT values <−4.6 ppb/K (Fig. 2b, c) are in good agreement with the X-ray crystal structure of HPLC12 (Jia et al. 1996) or the NMR structure of hAFL (Hamada et al. 2006), where these amide protons were not protected from exchange by H-bonding.

Fig. 2
figure 2

a Temperature dependence of the chemical shifts (Δδ/ΔT) of HPLC12 (closed blue circles) and hAFL (open red circles) as a function of residue number. Secondary structure models for previously determined b X-ray crystal structure of HPLC12 (PDB id = 1MSI) (Jia et al. 1996) and c NMR structure of hAFL (PDB id = 1WVO) (Hamada et al. 2006). Residues with Δδ/ΔT values <−4.6 ppb/K are colored blue and residues with Δδ/ΔT values >0 ppb/K are in magenta. The amide proton Δδ/ΔT–CSD correlation plots of c HPLC12 and d hAFL. Solid lines indicate the statistical cutoff, Δδ/ΔT = –2.41 × CSD−2.11. The residues composing the pyramidal and primary prism ice plane segments of the IBS in the HPLC12 isoform and the corresponding segments in hAFL are shown in red and green, respectively. The hydrophobic residues that are nearly invariant in the type III AFP family are highlighted in blue

Full size image

Residues N14, T15, I37, and V45 in HPLC12 have Δδ/ΔT values >0 ppb/K. Similarly, in hAFL, the corresponding residues G14, I37, and V45 exhibit positive Δδ/ΔT values. Residues D20, G27, G42, and E50 in hAFL also have Δδ/ΔT values >0 ppb/K. The most surprising feature is that some residues (V5, D20, N39, and N62) in hAFL have significantly different Δδ/ΔT values compared to the corresponding residues in HPLC12 (Fig. 2a). For example, the Δδ/ΔT value of D20 in hAFL is 0.8 ppb/K, but the corresponding residue V20 in HPLC12 has a Δδ/ΔT value of −4.7 ppb/K. Instead, the neighboring M19 amide proton exhibits a large negative Δδ/ΔT (−6.1 ppb/K).

To characterize in detail the structural and dynamic properties of the HPLC12 and hAFL proteins, a combined analysis of Δδ/ΔT values with CSD (Andersen et al. 1997) was performed. Correlation plots of Δδ/ΔT versus CSD are shown in Figs. 2c, d. The hydrophobic residues, except I32 and V60, that nearly invariant in the type III AFP family were all below the statistical correlation line between Δδ/ΔT and CSD (Δδ/ΔT = −2.41 × CSD−2.11) (blue in Fig. 2c), suggesting that these residues are H-bonded and in slow-exchange (Andersen et al. 1997). Similar results were also observed in hAFL with the exception of V5, I55, and V60 (Fig. 2d).

In HPLC12, residues I11 and V49 have Δδ/ΔT values significantly lower than the statistical cutoff (Fig. 2c), reflecting strong H-bonding interactions between the s1 and s5 strands [the strands s1–s6 were random coil in HPLC12 but were assigned as β strands in the type III AFP variant, RD3, from Rhigophila dearborni (Miura et al. 2001)] (Supplementary Fig. S4). Similarly, residues N14 and V45 contributed to strong H-bonding interactions between the s2 and s4 strands via N14-NH↔V45-CO and V45-NH↔T15-CO. In addition, four residues in the β1 strand (S4, V5, V6, and A7) and three residues in the second 310 helix (A34, E35, and I37) had Δδ/ΔT that deviated significantly from random coil values (Fig. 2c, Supplementary Fig. S4). In hAFL, significant deviations of Δδ/ΔT values from the statistical cutoff were observed for I11, V49, G14, and V45 (Fig. 2d). However, residues S4 and V5 in the β1 strand and three residues in the α-helix (residues 38, 39 and 41) have significantly higher Δδ/ΔT values compared to the statistical cutoff (Fig. 2d, Supplementary Fig. S4), suggesting that these residues are not strongly H-bonded but are in fast-exchange.

15N NMR relaxation dynamics of HPLC12 at various temperatures

The longitudinal R 1 relaxation rates, transverse R 2 relaxation rates, and {1H}-15N heteronuclear NOEs for uniformly 15N-labeled HPLC12 were measured at three different temperatures (5, 15 and 25 °C) to gain insight into the temperature-dependent dynamic motions of these two proteins (see Figs. 3, 4, 5). R 1 values of HPLC12 are fairly uniform throughout the proteins and increase with temperature (Fig. 3a; Table 1). The R 2 values of HPLC12 are also mostly uniform, with a few exceptions. For example, R 2 of T53 (25.39 Hz) and M59 (19.66 Hz) showed large deviation from the average value at 5 °C (Fig. 3c). The R 2/R 1 ratio was significantly larger than the average value, which implies conformational exchange on the μs–ms timescale (Hong et al. 2010; Tjandra et al. 1995). Interestingly, residues I32, Q44, T53, and M59 exhibited R 2/R 1 ratios above the upper cutoff line [upper/lower cutoff = average ± 1.5 × standard deviation (SD)] at 5 °C, whereas at 25 °C the R 2/R 1 ratios of these residues were below the upper cutoff line (Fig. 4a).

Fig. 3
figure 3

15N relaxation parameters of HPLC12 and hAFL as a function of residue number. R 1 relaxation rates of a HPLC12 and b hAFL are determined at 5 °C (lower, blue), 15 °C (middle, green), and 25 °C (upper, red). R 2 relaxation rates of c HPLC12 and d hAFL are determined at 5 °C (lower, blue), 15 °C (middle, green), and 25 °C (upper, red). The error bars represent curve fitting errors during the determination of R 1 and R 2 values from experiment data. The solid lines indicate the average values. In hAFL, the R 1 and R 2 relaxation rates of A7 at 5 and 15 °C could not be determined because of overlapping with other resonance. The black square symbols indicate residues whose relaxation data could not be determined because of severe line-broadening at 5 °C

Full size image
Fig. 4
figure 4

R 2/R 1 ratios of a HPLC12 and b hAFL as a function of residue number are determined at 5 (lower, blue), 15 (middle, green), and 25 °C (upper, red). The error bars represent the standard deviations of R 2/R 1 ratios (σ R2/R1) determined by using equation, \( \sigma_{R2/R1} = R_{2} /R_{1} \sqrt {(\sigma_{R2} /R_{2} )^{2} + (\sigma_{R2} /R_{2} )^{2} } \), where σ R1 and σ R2 are curve fitting errors during the determination of R 1 and R 2 values, respectively. The thick and two thin solid lines indicate the average and upper/lower cutoff values (average ± 1.5 × SD), respectively. The black square symbols indicate residues whose relaxation data could not be determined because of severe line-broadening at 5 °C. Mapping of the R 2/R 1 ratios onto the secondary structure models of c HPLC12 and d hAFL. Residues which have deviated R 2/R 1 ratios than the upper/lower cutoff lines are shown in blue. Residues composing the pyramidal and primary prism ice plane segments of HPLC12 and corresponding residues of hAFL are shown in pink and light green, respectively. Residue G27 of hAFL, which is corresponding the substrate-binding R27 in Neisseria meningitides SAS, is in orange

Full size image
Fig. 5
figure 5

Heteronuclear {1H}-15N NOEs of a HPLC12 and b hAFL as a function of residue number are determined at 5 °C (lower, blue), 15 °C (middle, green), and 25 °C (upper, red). The error bars of the heteronuclear NOE data represent standard deviations of four different data sets. In hAFL, the heteronuclear NOE of A7 at 5 and 15 °C could not be determined because of overlapping with other resonance. The black square symbols indicate residues whose relaxation data could not be determined because of severe line-broadening at 5 °C. Mapping of the {1H}-15N NOEs onto the secondary structure models of c HPLC12 and d hAFL. Residues which have smaller heteronuclear NOE <0.8 or <0.75 are shown in light blue and dark blue, respectively. Residues composing the pyramidal and primary prism ice plane segments of HPLC12 and corresponding residues of hAFL are shown in pink and light green, respectively. Residue G27 of hAFL, which is corresponding the substrate-binding R27 in Neisseria meningitides SAS, is in orange

Full size image
Table 1 The average 15N relaxation parameters of HPLC12 and hAFL determined at 5, 15, and 25 °C
Full size table

With the exception of the termini, the {1H}-15N heteronuclear NOEs of HPLC12 were mostly >0.8 at all three temperatures, except for V27 at 5 °C (Fig. 5a). This indicates that there are no ps–ns timescale motion in the temperature range studied here.

15N NMR relaxation dynamics of hAFL at various temperatures

The R 1 values of hAFL showed temperature dependence similar to those of HPLC12 (Fig. 3b). Residues G27–G31 and E35 showed slightly elevated R 1 values compared to other residues except both termini at all three temperatures (Fig. 3b). The R 2 values increased and showed larger deviation from the average values as the temperature decreased (Fig. 3d; Table 1). Residues K30, G31, and I37 exhibited significantly larger R 2/R 1 ratios than the average at 25 °C (Fig. 4b). Line broadening prevented the R 2/R 1 ratios for E28, K30, and G31 at 5 °C from being determined (Supplementary Fig. S3). At 5 °C, three more residues (K25, E58, and V60) show larger R 2/R 1 ratios than the upper cutoff value (Fig. 4b). Residues K3, G27, and D51 have significantly smaller R 2/R 1 ratios than the average value at 5 °C (Fig. 4b).

The {1H}-15N heteronuclear NOEs of hAFL exhibit very different patterns from those of HPLC12 (Figs. 4, 5). Several residues had NOE values ≤0.8 and this number increased when the temperature dropped (Fig. 5d). At 25 °C, some residues (K3, K10, V24, V26–G31, K44, E50, I55–E57 and N62) had NOE values ≤0.8. The NOE values for residues K3, K30, and G31 could not be determined due to line broadening at 5 °C. Residues S4, E13, I16, D20, V26–G31, L40–G42, K44, V45, T48–E50, D53, T54, and E57–N62 had NOEs ≤0.8 or showed very weak resonances at 5 °C (Fig. 5d). It is noteworthy that the structural flexibility of hAFL on the ps–ns timescale increased as the temperature decreased, in contrast to HPLC12. Interestingly, residues K10 and K44 that showed NOEs <0.8 are located on the corresponding surfaces of ice-binding site even at 25 °C (Fig. 5d). Residues G7, V24, V26, G27, E28, K30 and G31 that showed NOEs <0.8 at 25 °C (Fig. 5d) are clustered on the substrate-binding surface for bacterial SAS (Hamada et al. 2006). Residue K44 of hAFL showed NOE values ≤0.8 at all three temperatures, while the corresponding Q44 of HPLC12 had NOE values >0.8 through all temperatures. As mentioned above, Q44 of HPLC12 may experience conformational exchange on the μs–ms timescale at low temperature. Taken together, our data indicate that the IBS of HPLC12 and the corresponding surface of hAFL have different dynamic properties at low temperature.

Analysis of 15N NMR relaxation parameters in terms of reduced spectral density

The 15N-relaxation data at three different temperatures were used to calculate reduced spectral density function J-values. Figure 6 shows J e(0), J(ωN), and J(0.87ωH) of HPLC12 and hAFL determined at three temperatures. The temperature dependence of J e(0) was similar to R 2 , and that of J(ωN) was similar to the R 1 pattern. HPLC12 had fairly uniform J e(0) values with the average values of 4.24 ns/rad at 5 °C, 3.22 ns/rad at 15 °C, and 2.63 ns/rad at 25 °C (Table 1). Residues I32 (4.94 ns/rad), T53 (6.64 ns/rad), M59 (5.12 ns/rad), and Y63 (4.94 ns/rad) had significantly larger J e(0) values than the average at 5 °C (Fig. 6a). This indicates that these residues are involved in μs–ms timescale motions (Lefevre et al. 1996). J(ωN) and J(0.87ωH) of HPLC12 were also mostly uniform (Fig. 6a). The small J(0.87ωH) values imply that HPLC12 is well structured with limited internal mobility at all three temperatures, which is consistent with the uniformly high {1H}-15N heteronuclear NOE values.

Fig. 6
figure 6

Reduced spectral density functions of HPLC12 and hAFL. J e(0) (left); J(ωN) (middle); and J(0.87ωH) (right) are calculated for a HPLC12 and b hAFL as a function of residue number determined at 5 °C (lower, blue), 15 °C (middle, green), and 25 °C (upper, red)

Full size image

In contrast, hAFL showed higher and more varied J e(0) because of more deviation in R 2. The average J e(0) values were 5.48 ns/rad at 5 °C, 4.16 ns/rad at 15 °C, and 3.33 ns/rad at 25 °C (Table 1). The elevated average J e(0) values of hAFL imply that there are more residues that experience slow motions. Residues K25, E58, and V60 which showed larger R 2/R 1 ratios as well as significantly higher J e(0) values than the average at 5 °C (Figs. 4b, 6b). As mentioned previously, abnormally large values of J e(0) indicate that these residues are experiencing μs–ms timescale motions. At all three temperatures, the J(0.87ωH) values of hAFL are much larger and show more deviation than those HPLC12, implying that the hAFL domain exhibits greater ps–ns internal motion compared to HPLC12.

The correlation times of overall tumbling motion (τm) of HPLC12 and hAFL are obtained from the spectral density functions as using Eq. 4 (Supplementary Fig. S5). The residues that have R 2/R 1 ratio deviated from upper/lower cutoff values or heteronuclear NOE <0.7 are excluded from calculation of the average τm values. We determined that the average τm values of HPLC12 were 9.74 ns at 5 °C, 7.30 ns at 15 °C, and 5.78 ns at 25 °C (Table 1). Similarly, hAFL have the average τm values of 12.16 ns at 5 °C, 9.02 ns at 15 °C, and 7.15 ns at 25 °C (Table 1). If HPLC12 and hAFL are monomeric, we would expect the average τm values of approximately 8.1 ns for HPLC12 (8.1 kDa with C-terminal His-tag) and 9.9 ns for hAFL (9.9 kDa with N-terminal His-tag) at 5 °C by using relation, τm~Mw, at 5 °C (Graether et al. 2003). Our experimental τm values for both proteins are slightly larger than what would be predicted from their molecular weights.

Functional implications of HPLC12 and hAFL

In this study, 15N NMR relaxation experiments in the temperature range of 5–25 °C were used to characterize the unique backbone dynamics of the psychrophilic protein, HPLC12 and compare them to those of a homologous mesophilic protein hAFL. HPLC12 had fairly uniform R 1 and R 2 values at all three temperatures, whereas the R 2 values of hAFL showed much larger deviations from the average values as the temperature decreased (Fig. 3). This means that HPLC12 has a more rigid backbone structure and lacks conformational exchange on the μs–ms timescale for most residues even at low temperature. This result is consistent with previous reports that the insect β-helical AFPs from Tenebrio molitor and spruce budworm are well structured and become more rigid as the temperature is lowered (Daley et al. 2002; Graether et al. 2003). This study also found that the mesophilic protein, hAFL, shows significant flexibility and exchange in backbone dynamics at 5 °C compared to the psychrophilic protein, HPLC12. This is evident from: i) large deviations in R 2 values (μs–ms timescale motion) and ii) {1H}-15N heteronuclear NOEs ≤0.8 (ps–ns timescale motion) (Figs. 3, 5). These results indicate that the backbone of hAFL displays significant conformational exchange at low temperature, even though the protein is ordered at 25 °C.

With the exception of the termini, most residues of HPLC12 had {1H}-15N heteronuclear NOEs >0.8, whereas in hAFL, the number of residues displaying {1H}-15N heteronuclear NOEs ≤0.8 increased as the temperature decreased (Fig. 5). These data indicate that HPLC12 has very little backbone flexibility on the ps–ns timescale over the temperature range of 5–25 °C, which is consistent with its small J(0.87ωH) values. These results imply that HPLC12 is well structured and has limited internal mobility at all three temperatures, hAFL, on the other hand, show greater backbone flexibility on the ps–ns timescale with remarkably higher values of J(0.87ωH) for several residues at 25 °C, and this structural flexibility increased as the temperature decreased. In addition, hAFL displayed significant slow motions (μs–ms timescale) at the low temperature. Taken together, our data suggest that psychrophilic proteins like HPLC12 exhibit a rigid backbone structure with internal dynamics that remain unaffected by temperature, whereas mesophilic proteins like hAFL have less rigid backbones and specific residues whose conformational mobilities increase significantly at low temperature.

Since the biologically relevant temperature of HPLC12 is subzero, the NMR dynamics data at low temperature may provide insight into the molecular mechanism of its anti freezing activity. We found several unique features of the backbone structure and dynamics of HPLC12 at low temperature from our temperature-dependent NMR dynamics study. First, most of hydrophobic residues which are invariant or nearly invariant in the type III AFP family were below the statistical cutoff line in the correlation plot of Δδ/ΔT versus CSD (Fig. 2c). These results suggest that the amide protons of these residues are protected from exchange with solvent via H-bonding interactions and they can play important roles in protein folding of the type III AFPs. As shown in Supplementary Fig. S4A, the H-bonding interactions established between the s1 and s5 strands and between the s2 and s4 strands as well as in the β1 sheet and the second 310 helix may contribute energetically to the rigid backbone structure of HPLC12 for efficient ice-binding at low temperature. However, in hAFL, the β1 strand (S4 and V5) and α-helix (F38, N39, and V41) are not H-bonded and show a fast-exchange pattern, which could partially explain why hAFL, especially in the s3 region, displays conformational flexibility at low temperature.

The ice-binding residues of HPLC12 exhibit distinct dynamics features compared to the corresponding residues of hAFL. The structural study of insect β-helical AFPs proposed an “anchored clathrate mechanism” where both the hydrophobic effect and H-bonding contributed to adsorption of AFP to ice (Garnham et al. 2011). Previous NMR relaxation studies of the insect β-helical AFPs revealed that the TXT side-chains on the IBS are highly rigid (Daley et al. 2002; Graether et al. 2003; Graether and Sykes 2004). It was reported that the Thr side-chains on the IBS of the Tenebrio moliter AFP adopt a preferred rotameric conformation at low temperature (Daley et al. 2004). The rigidity of the IBS suggests that AFP maintains surface-to-surface complementarity between the AFP and ice crystal, wherein the rigid array of Thr side-chains is capable of forming efficient hydrophobic and H-bonding interactions (Daley et al. 2004). Our backbone dynamics study also showed that all residues on the IBS of HPLC12 are rigid with high heteronuclear NOE values (Fig. 5) as well as little deviations of the R 2/R 1 ratio from the average value (Fig. 4). In the active QAE1-isoform of the type III AFP, the surface pocket created by residues Q9, T18, V20, and M21 traps the ice-like water that anchor the AFP-ice interaction (Kumeta et al. 2013). The hydrogen-deuterium exchange experiments revealed that these four residues in the active QAE1-like isoform are rigid and capable of forming the surface pocket, while the corresponding residues of the inactive QAE2-isoform cannot form the pocket (Kumeta et al. 2013). Taken together, our NMR study support an “anchored clathrate mechanism” of the type III AFP where the IBS exhibits rigid structure for efficient hydrophobic and H-bonding interaction with ice crystal and is capable of trapping ice-like water into surface pocket that anchor the AFP-ice interaction.

Although HPLC12 has fairly uniform R 1 and R 2 values at all three temperatures, some residues (I32, Q44, T53, and M59) have significantly larger R 2/R 1 ratios than the average value at 5 °C (Fig. 4a). As expected from R 2/R 1 ratios, these residues have significantly larger J e(0) values than the average at 5 °C (Fig. 6a), which strongly support the slow exchange of these residues on the μs–ms timescale. The internal motion of residue Q44 might be required for efficient H-bonding of its side-chain with the ice crystal. However, residues I32, T53, and M59 are far from the IBS of HPLC12, indicating that the conformational exchange on the μs–ms timescale of these residues at low temperature might not be crucial for interaction with ice crystal.

At 25 °C, the R 1 and R 2 values of the mesophilic protein, hAFL, are uniform throughout and the {1H}-15N heteronuclear NOEs >0.8 for most residues, indicating that hAFL displays a rigid backbone structure at ambient temperature. Interestingly, residues located at the loop region (V26–Y32) between the β2 sheet and the second 310 helix show significant flexibility and exchange in backbone dynamics even at 25 °C evident from: i) large deviations in R 1 and R 2 values (μs–ms timescale motion) and ii) {1H}-15N heteronuclear NOEs ≤0.8 (ps–ns timescale motion) (Figs. 3, 5). In contrast, the corresponding residues of HPLC12 show no deviated R 2/R 1 ratios (Fig. 4a) as well as heteronuclear NOE >0.8 (Fig. 5a) at 25 °C, indicating this loop region of HPLC12 has rigid backbone structure. In addition, residues I55–E58 also experience flexible internal motion (see Figs. 4b, 5b). Among them, residues V26, E28, and G31 are class-specific residues of the vertebrate AFL domain (Hamada et al. 2006). The crystal structure of an N. meningitides SAS found that the highly conserved R314 (highlighted in orange as R27 in Fig. 1a) protrudes to form direct of water-mediated H-bond with ManNAc (Gunawan et al. 2005). The importance of R314 in the catalytic activity of SAS is supported by inhibition study of Streptococcus agalactiae SAS through the addition of an arginine directed modifying reagent (Suryanti et al. 2003). In the vertebrate AFL domains, this Arg residue is missing from the ligand-binding site, but some class-specific residues (K8 and K25 in Fig. 1a) with positively charged side-chains lie near the substrate (Hamada et al. 2006). Interestingly, we found flexible regions clustered around the corresponding residue in hAFL, G27 (Figs. 4d, 5d). Thus, it is reasonable to think that the flexibility of these regions may contribute to the biological function of the AFL domain in SAS. In contrast, the IBS of the insect β-helical AFP exhibits rigid backbone structure and the ice-binding Thr residues participate in intrastrand H-bonds that stabilize the flat surface required for optimal ice-binding (Daley et al. 2004). Instead, the Thr side-chains on the IBS are flexible to adopt a preferred ice-binding conformation without an entropic penalty (Daley and Sykes 2003).

Conclusions

In this study, we have compared the dynamic properties of the type III AFP (HPLC12 isoform) and the AFL domain of human SAS (hAFL) by analyzing the temperature gradient of the amide proton chemical shift, its correlation with CSD from random coil, backbone dynamics, and reduced spectral density. Our data show that the two homologous proteins have distinctive backbone dynamics. Most residues of HPLC12 are rigid while a few residues on the IBS experience conformational exchange on the μs–ms timescale at 5 °C. In contrast, hAFL is remarkably flexible at 5 °C, and its flexibility decreases at higher temperatures (25 °C). At the same time, the residues that are involved in substrate binding show fast internal motion. Based on our study, we suggest that the dynamic properties of each protein give insight into the basis for the different physiological working temperatures of HPLC12 and hAFL. It is very likely that each protein maintains a relatively rigid backbone structure overall, but that the functionally important residues exhibit distinctive dynamics for substrate binding (rigid for ice-binding or flexible for sialic acid binding).