Anchored clathrate waters bindantifreeze proteins to iceChristopher P. Garnham, Robert L. Campbell, and Peter L. Davies1

Department of Biochemistry, Queen’s University, Kingston, ON, Canada K7L 3N6

Edited by Michael Levitt, Stanford University School of Medicine, Stanford, CA, and approved March 8, 2011 (received for review January 10, 2011)

The mechanism by which antifreeze proteins (AFPs) irreversiblybind to ice has not yet been resolved. The ice-binding site ofan AFP is relatively hydrophobic, but also contains many potentialhydrogen bond donors/acceptors. The extent to which hydrogenbonding and the hydrophobic effect contribute to ice bindinghas been debated for over 30 years. Here we have elucidatedthe ice-binding mechanism through solving the first crystal struc-ture of an Antarctic bacterial AFP. This 34-kDa domain, the largestAFP structure determined to date, folds as a Ca2þ-bound parallelbeta-helix with an extensive array of ice-like surface waters thatare anchored via hydrogen bonds directly to the polypeptidebackbone and adjacent side chains. These bound waters makean excellent three-dimensional match to both the primary prismand basal planes of ice and in effect provide an extensive X-raycrystallographic picture of the AFP∶ice interaction. This unob-structed view, free from crystal-packing artefacts, shows the con-tributions of both the hydrophobic effect and hydrogen bondingduring AFP adsorption to ice. We term this mode of binding the“anchored clathrate” mechanism of AFP action.

Ca2+ binding protein ∣ repeats-in-toxin protein ∣ thermal hysteresis ∣Antarctic bacterium ∣ organized biohydration

Antifreeze proteins (AFPs) adsorb to the surface of ice crystalsand prevent their growth (1). This adsorption lowers the

freezing temperature of a solution below its melting point, en-abling the survival of many organisms that inhabit ice-ladenenvironments. Despite their common function, AFPs display re-markable diversity in their tertiary structures (2–7). This diversityresults partly from their independent evolutionary origins (8, 9)and partly from the surface heterogeneity of their natural ligand,ice (10). Hexagonal ice presents many different planes (expressedas Miller indices) of water molecules to which an AFP can devel-op affinity. Although specificity toward different ice planes is akey determinant of antifreeze activity (11), the mechanism bywhich an AFP binds to ice remains undefined.

Hydrogen bonds were originally proposed to be the main bind-ing force between an AFP and ice (12). Yet this hypothesis wasunable to explain how an AFP would preferentially bind ice whensolvated by 55 M water. Subsequent studies proposed that thehydrophobic effect was the main ice-binding force, where con-strained, clathrate-like water on the ice-binding site (IBS) isreleased into the solvent upon ice binding, resulting in a gain ofentropy (13, 14). However, several molecular dynamics (MD)simulations have indicated that the relatively hydrophobic IBSof an AFP is capable of ordering water molecules into an ice-likelattice (15–21) and, instead of shedding bound water moleculesupon ice binding, the ordered waters might facilitate the AFP'sinteraction with ice by matching certain ice planes (15). Althoughintriguing, these simulations fall short of describing at themolecular level how an AFP might order water molecules intoan ice-like lattice that has the specificity and binding strengthto irreversibly adsorb the AFP to ice.

The Antarctic bacterium Marinomonas primoryensis producesan exceptionally large (ca. 1.5 MDa) Ca2þ-dependent AFP(MpAFP) (22, 23). The protein contains two highly repetitivesegments that divide it into five distinct regions (Fig. 1A). Region

II of MpAFP accounts for 90% of the protein's mass and iscomprised of ca. 120 tandem copies of an identical 104-aa

Fig. 1. Structure of MpAFP_RIV. (A) MpAFP consists of five distinct regions.Highlighted in blue are the tandem 104-aa repeats that constitute region II.Region IV is colored orange. Numbers below each region indicate the numberof amino acids. (B) Cartoon representation of MpAFP_RIV. Ca2þ = greenspheres. (C) 2Fo-Fc electron density of the unit cell contoured at σ ¼ 1 show-ing four distinct beta-helical chains. (D) Cross-section view of MpAFP_RIVresidues 95–131. Carbon = white, nitrogen = blue, oxygen = red, Ca2þ =green. (E) Stereo view of Ca2þ-binding turn of MpAFP_RIV. Color scheme isas previously mentioned. Black hatched lines represent hydrogen bonds.Residues involved in Ca2þ-binding are indicated. Numbers indicate positionwithin the hexapeptide xGTGND motif. U = upper loop, L = lower loop.

Author contributions: C.P.G., R.L.C., and P.L.D. designed research; C.P.G. performedresearch; C.P.G. analyzed data; and C.P.G. wrote the paper with editorial input from P.L.D.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The atomic coordinates and structure factors have been deposited in theResearch Collaboratory for Structural Bioinformatics Protein Data Bank, www.rcsb.org(RCSB PDB ID code 3P4G).

See Commentary on page 7281.1To whom correspondence should be addressed. E-mail: peter.davies@queensu.ca.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1100429108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1100429108 PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7363–7367

BIOCH

EMISTR

YSE

ECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

July

25,

202

0

sequence. The antifreeze activity of MpAFP resides in region IV(MpAFP_RIV), a 322-aa segment of the protein that contains 13tandem 19-aa repeats. Recombinantly expressed MpAFP_RIVhas been shown to bind Ca2þ and depress the freezing pointof a solution in a hyperactive manner (23). Here, we have deter-mined the X-ray crystal structure ofMpAFP_RIV to a resolutionof 1.7 Å (Table S1). The structure explains MpAFP_RIV's Ca2þ-dependent hyperactivity and also provides direct experimentalevidence that anchored clathrate waters bind AFPs to ice.

Results and DiscussionMpAFP_RIV folds as a right-handed Ca2þ-bound parallel beta-helix roughly 70 Å long and 20 × 10 Å in cross-section (Fig. 1B).Four of these beta-helices are packed within the unit cell of thecrystal, each one oriented antiparallel to its two closest neighbors(Fig. 1C). Every coil of the beta-helix, typically 19 amino acidsin length, contains one 6-residue Ca2þ-binding turn and threeshort beta strands separated by Gly-rich turns (Fig. 1D). Distinctcapping structures reside at the N and C termini (Fig. S1 A and B)of the beta-helix. The consensus amino acid sequence of eachloop is xGTGNDxuxuGGxuxGxux, where x represents any resi-due (typically hydrophilic) and u residues are hydrophobic(Val, Leu, or Ile) (Fig. S1C). The u residues from each beta strandform a hydrophobic core that is screened from the solvent bymain-chain hydrogen bonds that run parallel to the long axisof the helix. Thirteen Ca2þ ions align down one side of the struc-ture and each is heptacoordinated between successive xGTGNDturns (Fig. 1E). In particular, the glycines at positions 2 and 4(xGTGND) of the upper loop contribute main-chain carbonylswhereas the aspartate at position 6 (xGTGND) contributesone side-chain carbonyl. In the lower loop, the x residue at posi-tion 1 and threonine at position 3 (xGTGND) contribute main-chain carbonyls, whereas the aspartate at position 6 (xGTGND)partially contributes both side-chain carbonyls. In this way, Aspresidues at position 6 of each loop lock Ca2þ ions into place andbridge each Ca2þ-binding site. This coordination scheme differsonly at the most C-terminal (13th) Ca2þ-binding site, where aglutamate residue extends from a recessed Ca2þ-binding loop andbinds Ca2þ along with two water molecules (Fig. S1D). This site isapparently the protein's most solvent-accessible Ca2þ-binding siteas the lanthanide Ho3þ, soaked into the crystal as a heavy atomreplacement for Ca2þ, substituted into this position at roughly50% occupancy in all four molecules of the unit cell. Ho3þ didnot substitute at any other Ca2þ-binding site.

MpAFP_RIV presents a long and flat IBS that runs the lengthof its Ca2þ-bound side (Fig. 2A). It consists of the Thr and Asx(usually Asn) residues that project outward from the xGTGNDCa2þ-binding turns. Previous site-directed mutagenesis studieshave confirmed the location of the IBS (23). Glycine typicallyseparates Thr and Asx in each turn (except in loops 6 and 8where it is substituted by aspartate), and this pattern maintainsthe IBS's flatness and regularity, two characteristics that definethe IBS of most AFPs, and in particular the hyperactive beta-helical insect AFPs from Tenebrio molitor (TmAFP) and sprucesubworm (sbwAFP) (Fig. 2 B and C) (2, 4). Side-chain O atomson the IBS's of all three beta-helices form rectangular arrays ca.7.4 −Åwide × 4.6 −Å long (Fig. 2D) and these distances closelymatch the spacing of O atoms found on the primary prism andbasal planes of ice. MpAFP_RIV uses a Thr-Gly-Asx motif lo-cated in a Ca2þ-binding turn as its IBS, whereas both TmAFPand sbwAFP use a Thr-x-Thr motif located in a flat beta sheetto bind ice. Asx residues are required on the IBS ofMpAFP_RIVto compensate for the change in beta-helical pitch that occursin the Ca2þ-binding turns of the protein, a phenomenon avoidedin the beta-stranded IBSs of both TmAFP and sbwAFP. Theextra length of each Asx residue allows it to align ca. 7.4 Å froma Thr of the ensuing Ca2þ-binding turn. If it is remarkable thatthe identical IBSs of the nonhomologous beta-helical insect AFPs

arose via convergent evolution, it is even more extraordinary thatthis precise IBS geometry has been replicated again using a dis-tinct ice-binding motif located within a unique beta-helical fold.

MpAFP_RIV is a homologue of the repeats-in-toxin (RTX)family of secreted virulence factors produced by numerous patho-genic Gram-negative gammaproteobacteria (24). RTX proteinsare defined by the presence of a tandemly repeated nonapeptidemotif (GGxGxDxux) that folds as a beta-helix with Ca2þ bounddown both sides of the structure (25, 26), and not just one as seenin region IV (Fig. 3A). The xGTGND repeats of MpAFP_RIVbind Ca2þ and fold identically to the GGxGxD repeats that definethe RTX proteins. However MpAFP_RIV binds Ca2þ only downone side of the helix because each 19-aa loop of the protein con-tains just one xGTGND Ca2þ-binding repeat, and not two as isseen in the RTX proteins. Positions 3 and 5 of the canonical RTXGGxGxD Ca2þ-binding turn are highly variable (Fig. 3B) andMpAFP_RIV's affinity toward ice therefore most likely devel-oped once sufficient Thr and Asx substitutions occurred withinthe Ca2þ-binding turns of the beta-helix.

The unique IBS of MpAFP_RIV arranges water moleculesinto a specific ice-like lattice. These organized surface watersalign down the entire length of one of the four helices in theunit cell (chain D) (Fig. 4A), and also down the bottom third ofchain A (Fig. 4B). These were the only two areas in the unit cellwhere the IBS was completely solvent-exposed, ensuring specificprotein∶solvent interactions free from crystal-packing artefacts.Waters in these two separate areas are coordinated in the samemanner, confirming the specificity of the interaction. The ice-likewater molecules enclose the γ-methyl of each Thr (xGTGND),and this cage is anchored to the IBS by hydrogen bonds to themain-chain nitrogen and side-chain hydroxyl of each Thr(Fig. 4C). Waters are also hydrogen bonded to the main-chainnitrogen of each Gly (xGTGND) and side-chain oxygen of eachAsx (xGTGND). The two Asp residues substituted for Gly (coils6 and 8) project outward and are located in the middle of theIBS. Their side-chain carbonyls mimic the location of surface

Fig. 2. IBS of MpAFP_RIV in comparison to those of TmAFP and sbwAFP.(A) The IBS of MpAFP_RIV. The Thr and Asx residues of each xGTGNDCa2þ-binding turn point outward and create a flat, repetitive array idealfor ice binding. (B) The hyperactive right-handed beta-helical AFP producedby the beetle T. molitor. Green = carbon, red = oxygen, and orange = sulfur.(C) The hyperactive left-handed beta-helical AFP produced by the spruce bud-worm moth (Choristoneura fumiferana). Yellow = carbon and red = oxygen.(D) Comparison of IBS geometry between MpAFP_RIV and the two hyperac-tive beta-helical insect AFPs. Distances between O atoms are indicated inangstroms.

7364 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1100429108 Garnham et al.

Dow

nloa

ded

by g

uest

on

July

25,

202

0

waters and also assist in their coordination (Fig. 4D). The ice-likewaters that encompass the IBS of MpAFP_RIV make an excel-lent 3-D match to both the basal and primary prism planes of ice(Fig. 4 E and F, Fig. S2 A and B). The solvent-exposed waterslocated on the lower third of chain A’s IBS have an rmsd of0.68 Å when aligned to 28 oxygen atoms on the primary prism

plane of ice and 0.73 Å when aligned to 24 oxygen atoms onthe basal plane of ice. This near-perfect alignment underscoresthe precision of the fit between the protein and ice providedby this glimpse of an AFP∶ice interaction.

Microscopic ice crystals formed in the presence of MpAF-P_RIV are shaped as hexagonal plates (Fig. 5A), confirming the

Fig. 3. Comparison between MpAFP_RIV and RTX-like beta-helices. (A) Cross-section of MpAFP_RIV residues 115–161 and alkaline protease (PBD ID code1KAP) residues 331–375. Oxygens are shown in red, nitrogens blue, Ca2þ are green. Carbons are white for MpAFP_RIV and yellow for alkaline protease.The side chains of solvent-exposed residues have been omitted for clarity. (B) Amino acid alignment of MpAFP_RIV to various known RTX-like proteins.The amino acid sequence of two Ca2þ-binding beta-helices of known tertiary structure found in alkaline protease (produced by Pseudomonas aeruginosa)and C5 epimerase (produced by Azotobacter vinelandii) are shown as are GGxGxD repeats of alpha-hemolysin (produced by Escherichia coli). Boxed residuesindicate GGxGxD Ca2þ-binding repeats and sequences in bold represent beta-strand. The highly variable positions 3 and 5 of the GGxGxD Ca2þ-binding repeatsare highlighted in red.

Fig. 4. Ordered surface waters on the IBS of MpAFP_RIV. (A) IBS of chain D freely exposed to solvent in the unit cell. (B) Section of chain A freely exposed tosolvent in the unit cell. The color scheme is the same as in Fig. 1. Ordered surface waters are represented as aqua spheres. (C) Ordered surface waters hydrogenbonded directly to the IBS or one molecule removed. (D) Asp residues on the IBS mimic surface waters and also assist in the coordination of two watermolecules. (E) and (F) The organized surface waters make an excellent three-dimensional match to both the basal (E) and primary prism (F) planes of ice.The direction of the c-axis is indicated in both figures.

Garnham et al. PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7365

BIOCH

EMISTR

YSE

ECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

July

25,

202

0

protein’s affinity toward the basal and primary prism planes.MpAFP_RIV tagged with GFP for fluorescence-based ice planeaffinity (FIPA) analysis (27) turned single ice-crystal hemispheresuniformly fluorescent (Fig. 5 B and C), consistent with the pro-tein’s affinity toward the basal and primary prism planes of ice, aswell as surfaces between these planes, a property also seen withthe hyperactive AFP from snow fleas (28). For a comparison,FIPA analysis was also performed on type I AFP. Type I AFPbinds only the f20–21g plane of ice (Fig. 5 E and F) (29) andshapes ice into a hexagonal bipyramid with a c∶a axial ratio of3.3∶1 (Fig. 5D), consistent with it binding only to this pyramidalplane and not the basal plane.MpAFP_RIV’s ability to bind to atleast the primary prism and basal planes of ice accounts for itshyperactivity and further strengthens the finding (11) that basalplane binding is a key determinant of AFP hyperactivity.

As previously mentioned, several MD studies have reportedice-like hydration and slowed water exchange around the IBSof an AFP (15–21). The structural determination ofMpAFP_RIVprovides a molecular explanation as to how this phenomenonmight occur. The relative hydrophobicity of an AFP's IBS orderswater molecules via the hydrophobic effect into an ice-like latticethat is anchored to the protein via hydrogen bonds. These an-chored waters then allow an AFP to bind ice by matching aspecific plane, or planes, of ice. This mechanism is applicable toany AFP regardless of activity level. Hyperactive AFPs will beable to order water molecules that allow them to adsorb to multi-ple planes of ice, one of which is the basal plane, while moder-ately active AFPs will be capable of ordering water molecules thatenable them to adsorb to multiple planes of ice, just not the basalplane. The energetics that govern AFP∶ice interactions havethus come full circle, with hydrogen bonds required to anchorwater molecules organized by the hydrophobic effect.

Fish type III AFP was recently shown by NMR to bind ice onlyonce the hydration shell surrounding the IBS of the protein froze,further supporting our observation (30). Though the structure ofMpAFP_RIV was solved at a temperature of 100 K, it wouldseem reasonable to assume that at temperatures above 0 °C,the stability of anchored clathrate waters on the solvent-exposedportions of the IBS of MpAFP_RIV would decrease. In solution,

waters near the IBS of an AFP are likely to be mobile andexchangeable with solvent waters. Another recent NMR studyrevealed that water molecules on the IBS of TmAFP displayedhigh mobility and exchanged with bulk water on a subnanosecondtimescale (31). The short lifetime of ordered water moleculeson the IBS of an AFP offers an explanation to both the concen-tration (11) and annealing time dependence (32) of antifreezeactivity. Only a small percentage of total AFP in solution at anyone time would have a quorum of water molecules in the correctorientation sufficient for ice-binding. The number of AFPs com-petent to bind ice would rise with increasing protein concentra-tion and/or annealing time, increasing the amount of AFP boundto the surface of the ice crystal and therefore raising thermalhysteresis levels. In this scenario, the energetics of an AFP–iceinteraction are driven primarily by enthalpic, and not entropic,gains. Shedding ordered surface waters for entropic gains wouldactually spoil the AFP∶ice interaction, leaving the protein lessable to recognize the specific plane, or planes, of ice toward whichit has evolved affinity. Our results indicate that an AFP partiallyforms its ligand before binding to it.

Materials and MethodsMpAFP_RIV Expression and Purification. The gene encoding MpAFP_RIV wasligated into the NdeI/XhoI sites of the pET28a expression vector placingan N-terminal 6X His-tag on the protein. The protein was expressed andpurified as previously described (23). The purified protein was dialyzedagainst 50 mM Tris-HCl (pH 8), 150 mMNaCl, 2 mM CaCl2, and then incubatedat room temperature overnight with one unit of thrombin per milligramof protein to remove the His-tag. The digestion solution was then adjustedto 5 mM imidazole and placed over an Ni-nitrilotriacetate (NTA) column(Qiagen) where the flow through was collected and dialyzed against20 mM Tris-HCl (pH 8), 2 mM CaCl2 (buffer A).

Buffer Screen to Identify Optimal Buffer for Crystallography. MpAFP_RIV wasconcentrated to 16 mg∕mL in buffer A and screened via hanging drop vapordiffusion in a manner similar to ref. 33 against buffers at a concentration of100 mM spanning a wide range of pH values (pH 3–10). Buffers with acidic toneutral pKa values caused the protein to precipitate (typically within minutesto days), while drops remained clear in wells containing alkaline buffers. Amore specific buffer screen was performed using solely buffers with alkalinepKa values, while testing increasing CaCl2 concentrations in the drop (up to10 mM). Drops containing 100 mM L-arginine (pH 9.5) remained clear atCa2þ concentrations up to 4 mM. MpAFP_RIV was then dialyzed against20 mM L-arginine (pH 9.5) and 4 mM CaCl2 and concentrated to 5.5 mg∕mLfor crystallographic screening.

MpAFP_RIV Crystallization. Initial crystal hits were obtained via hangingdrop vapor diffusion by mixing equal volumes of 5.5 mg∕mL MpAFP_RIVwith well solution containing 10% PEG 8000 and 200 mM MgAcetate. Crys-tals grew at room temperature and appeared roughly 4 wk following dropset up. The initial hit was multicrystalline in nature. Crystals suitable forstructure determination were obtained using both micro- and macroseedingtechniques. Briefly, the drop (4 μL) containing the initial hit was brought upto 50 μL with well solution and the crystals in the drop were crushed using aglass rod. The solution of crushed crystals was added at 1∕10th volume to adrop (4 μL) containing 2.75 mg∕mL MpAFP_RIV mixed with an equal volumeof well solution containing 10% PEG 8000, 400 mM MgAcetate. This condi-tion typically resulted in the growth of hundreds of small crystals (ca. 10–40-μm long). To obtain larger crystals, individual crystals were removed from themicroseeded drop and successively washed for 30 to 60 s each in solutions ofdecreasing PEG 8000 concentration (10%, 8%, 6%) containing 400 mMMgAcetate. Crystals were then placed in a drop (4 μL) contaning 2.75 mg∕mL MpAFP_RIV mixed with an equal volume of well solution containing10% PEG 8000, 400 mM MgAcetate. Plate-like crystals (dimensions ca.200 × 200 × 50 μm) grew within 1–2 wk. Crystals obtained following macro-seeding were still multicrystalline. Placing crystals in cryo solution resulted infracture along fault lines of individual crystal lattices, releasing single-crystalfragments (ca. 100 × 100 × 50 μm) suitable for structure determination. Thecryo solution for native crystals consisted of 12% PEG 8000, 400 mM MgA-cetate, and 20% glycerol. HoCl3 was used as a heavy atom for phasingand was included at a final concentration of 100 mM in an altered cryo solu-tion consisting of 5% PEG 8000, 600 mM MgAcetate, and 30% glycerol. Crys-tals were allowed to soak for 5 min prior to freezing and data collection.

Fig. 5. Ice-binding characteristics of MpAFP_RIV and type I AFP. (A) MpAF-P_RIV shapes ice into hexagonal plates at the microscopic level. The scale barrepresents a length of 20 μm. (B) Single ice-crystal hemisphere of GFP-taggedMpAFP_RIV. The crystal was mounted with a primary prism plane orientedperpendicular to the ice finger. The direction of the c-axis is indicated.(C) Same as in (B), here the basal plane of the crystal oriented perpendicularto the ice finger. (D) Type I AFP shapes ice into a hexagonal bipyramid.(E) Single ice-crystal hemisphere of tetramethylrhodamine-labeled type IAFP. The crystal was mounted with a primary prism plane oriented perpen-dicular to the ice finger. (F) Same as in (E), here the basal plane of the crystaloriented perpendicular to the ice finger. All hemispheres have a diameterof ca. 5 cm.

7366 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1100429108 Garnham et al.

Dow

nloa

ded

by g

uest

on

July

25,

202

0

Structure Determination. X-ray data were collected at 100 K at the X6A beam-line (National Synchrotron Light Source). Data were integrated and scaledusing HKL2000 (34). Ho3þ atoms were located using SHELX C/D/E (35) andca. 33% of the unit cell was built automatically using ARPwARP (36). Succes-sive rounds of model building using Coot (37) and refinement using RE-FMAC5 with the TLS protocol (38) were used to build the contents of theunit cell. A single protein chain from the heavy atom dataset was used asan initial search model of the native dataset using Phaser (39). The contentsof the unit cell from the native dataset were built and refined in the samemanner as the heavy atom dataset.

FIPA Analysis. FIPA analysis was performed as previously described (27). DNAcoding for MpAFP_RIV was ligated into the NdeI/XhoI cut sites of the pET24avector, placing a C-terminal 6X His-tag on the protein. GFP was ligatedupstream of MpAFP_RIV into the NdeI cut site of the vector. The correctclone was identified by restriction endonuclease digestion followed byDNA sequencing (Robart's Research Institute). The protein was purified viaNi-NTA chromatography and was eluted from the column in buffer

containing 50 mM Tris-HCl (pH 8), 500 mM NaCl, 400 mM imidazole, and2 mM CaCl2. The Ni-NTA eluate was dialyzed against 10 mM Tris-HCl (pH8.0), 2 mM CaCl2, and then used for FIPA analysis. Synthetic type I AFP waslabeled with tetramethylrhodamine using a previously described protocol(27). Following labeling, the protein was dialyzed against 10 mM Tris-HCl(pH 8) and then used for FIPA analysis. All hemispheres were grown at aconcentration of ca. 0.1 mg∕mL.

ACKNOWLEDGMENTS. We thank Jean Jakoncic and Vivian Stojanoff of theX6A beamline at Brookhaven National Laboratories for help with X-raydata acquisition and processing and Jack Gilbert for the initial discoveryof M. primoryensisAFP. We also thank Dr. John Allingham for access to aRigaku home X-ray source for initial diffraction experiments. This workwas funded by the Canadian Institutes of Health Research. C.P.G. is therecipient of an Natural Sciences and Engineering Research Council ofCanada three-year Postgraduate Scholarship (PGS D3). P.L.D. holds a CanadaResearch Chair in protein engineering.

1. Raymond JA, DeVries AL (1977) Adsorption inhibition as a mechanism of freezingresistance in polar fishes. Proc Natl Acad Sci USA 74:2589–2593.

2. Graether SP, et al. (2000) Beta-helix structure and ice-binding properties of a hyper-active antifreeze protein from an insect. Nature 406:325–328.

3. Ko T, et al. (2003) The refined crystal structure of an eel pout type III antifreezeprotein RD1 at 0.62-Å resolution reveals structural microheterogeneity of proteinand solvation. Biophys J 84:1228–1237.

4. Liou YC, Tocilj A, Davies PL, Jia Z (2000) Mimicry of ice structure by surface hydroxylsand water of a beta-helix antifreeze protein. Nature 406:322–324.

5. Nishimiya Y, et al. (2008) Crystal structure and mutational analysis of Ca2þ-indepen-dent type II antifreeze protein from longsnout poacher, Brachyopsis rostratus. J MolBiol 382:734–746.

6. Pentelute BL, et al. (2008) X-ray structure of snow flea antifreeze protein determinedby racemic crystallization of synthetic protein enantiomers. J Am Chem Soc130:9695–9701.

7. Sicheri F, Yang DS (1995) Ice-binding structure andmechanism of an antifreeze proteinfrom winter flounder. Nature 375:427–431.

8. Cheng CH (1998) Evolution of the diverse antifreeze proteins. Curr Opin Genet Dev8:715–720.

9. Fletcher GL, Hew CL, Davies PL (2001) Antifreeze proteins of teleost fishes. Annu RevPhysiol 63:359–390.

10. Davies PL, Baardsnes J, Kuiper MJ, Walker VK (2002) Structure and function ofantifreeze proteins. Philos Trans R Soc Lond B Biol Sci 357:927–935.

11. Scotter AJ, et al. (2006) The basis for hyperactivity of antifreeze proteins. Cryobiology53:229–239.

12. Devries AL, Lin Y (1977) Structure of a peptide antifreeze and mechanism of adsorp-tion to ice. Biochim Biophys Acta 495:388–392.

13. Chao H, et al. (1997) A diminished role for hydrogen bonds in antifreeze proteinbinding to ice. Biochemistry 36:14652–14660.

14. Baardsnes J, et al. (1999) New ice-binding face for type I antifreeze protein. FEBS Lett463:87–91.

15. Nutt DR, Smith JC (2008) Dual function of the hydration layer around an antifreezeprotein revealed by atomistic molecular dynamics simulations. J Am Chem Soc130:13066–13073.

16. Gallagher KR, Sharp KA (2003) Analysis of thermal hysteresis protein hydration usingthe random network model. Biophys Chem 105:195–209.

17. Jorov A, Zhorov BS, Yang DS (2004) Theoretical study of interaction of winter flounderantifreeze protein with ice. Protein Sci 13:1524–1537.

18. Smolin N, Daggett V (2008) Formation of ice-like water structure on the surface of anantifreeze protein. J Phys Chem B 112:6193–6202.

19. Wierzbicki A, et al. (2007) Antifreeze proteins at the ice/water interface: Threecalculated discriminating properties for orientation of type I proteins. Biophys J93:1442–1451.

20. Yang C, Sharp KA (2004) The mechanism of the type III antifreeze protein action: Acomputational study. Biophys Chem 109:137–148.

21. Yang C, Sharp KA (2005) Hydrophobic tendency of polar group hydration as a majorforce in type I antifreeze protein recognition. Proteins 59:266–274.

22. Gilbert JA, Davies PL, Laybourn-Parry J (2005) A hyperactive, Ca2þ-dependentantifreeze protein in an Antarctic bacterium. FEMS Microbiol Lett 245:67–72.

23. Garnham CP, et al. (2008) A Ca2þ-dependent bacterial antifreeze protein domain has anovel beta-helical ice-binding fold. Biochem J 411:171–180.

24. Coote JG (1992) Structural and functional relationships among the RTX toxindeterminants of gram-negative bacteria. FEMS Microbiol Rev 8:137–161.

25. Aachmann FL, et al. (2006) NMR structure of the R-module: A parallel beta-rollsubunit from an Azotobacter vinelandii mannuronan C-5 epimerase. J Biol Chem281:7350–7356.

26. Baumann U, Wu S, Flaherty KM, McKay DB (1993) Three-dimensional structure ofthe alkaline protease of Pseudomonas aeruginosa: A two-domain protein with acalcium binding parallel beta roll motif. EMBO J 12:3357–3364.

27. Garnham CP, et al. (2010) Compound ice-binding site of an antifreeze protein revealedby mutagenesis and fluorescent tagging. Biochemistry 49:9063–9071.

28. Mok YF, et al. (2010) Structural basis for the superior activity of the large isoform ofsnow flea antifreeze protein. Biochemistry 49:2593–2603.

29. Knight CA, Cheng CC, DeVries AL (1991) Adsorption of alpha-helical antifreezepeptides on specific ice crystal surface planes. Biophys J 59:409–418.

30. Siemer AB, Huang KY, McDermott AE (2010) Protein–ice interaction of an antifreezeprotein observed with solid-state NMR. Proc Natl Acad Sci USA 107:17580–17585.

31. Modig K, et al. (2010) High water mobility on the ice-binding surface of a hyperactiveantifreeze protein. Phys Chem Chem Phys 12:10189–10197.

32. Takamichi M, Nishimiya Y, Miura A, Tsuda S (2007) Effect of annealing time of an icecrystal on the activity of type III antifreeze protein. FEBS J 274:6469–6476.

33. Jancarik J, Pufan R, Hong C, Kim SH, Kim R (2004) Optimum solubility (OS) screening:An efficient method to optimize buffer conditions for homogeneity and crystalliza-tion of proteins. Acta Crystallogr Sect D Biol Crystallogr 60:1670–1673.

34. Otwinowski Z, Minor W (1997) Processing of X-ray diffraction data collected in oscilla-tion mode. (Academic, New York), Vol 276, pp 307–326 Methods in Enzymology(Macromolecular Crystallography Part A).

35. Sheldrick GM (2008) A short history of SHELX. Acta Crystallogr Sect A Found Crystal-logr 64:112–122.

36. Langer G, Cohen SX, Lamzin VS, Perrakis A (2008) Automated macromolecular modelbuilding for X-ray crystallography using ARP/wARP version 7. Nat Protoc 3:1171–1179.

37. Emsley P, Cowtan K (2004) Coot: Model-building tools for molecular graphics. ActaCrystallogr Sect D Biol Crystallogr 60:2126–2132.

38. Murshudov GN, Vagin AA, Dodson EJ (1997) Refinement of macromolecular structuresby the maximum-likelihood method. Acta Crystallogr Sect D Biol Crystallogr53:240–255.

39. McCoy AJ, et al. (2007) Phaser crystallographic software. J Appl Crystallogr 40:658–674.

Garnham et al. PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7367

BIOCH

EMISTR

YSE

ECO

MMEN

TARY

Dow

nloa

ded

by g

uest

on

July

25,

202

0