Peptidic Antifreeze Materials: Prospects and Challenges

Necessitated by the subzero temperatures and seasonal exposure to ice, various organisms have developed a remarkably effective means to survive the harsh climate of their natural habitats. Their ice-binding (glyco)proteins keep the nucleation and growth of ice crystals in check by recognizing and binding to specific ice crystal faces, which arrests further ice growth and inhibits ice recrystallization (IRI). Inspired by the success of this adaptive strategy, various approaches have been proposed over the past decades to engineer materials that harness these cryoprotective features. In this review we discuss the prospects and challenges associated with these advances focusing in particular on peptidic antifreeze materials both identical and akin to natural ice-binding proteins (IBPs). We address the latest advances in their design, synthesis, characterization and application in preservation of biologics and foods. Particular attention is devoted to insights in structure-activity relations culminating in the synthesis of de novo peptide analogues. These are sequences that resemble but are not identical to naturally occurring IBPs. We also draw attention to impactful developments in solid-phase peptide synthesis and ‘greener’ synthesis routes, which may aid to overcome one of the major bottlenecks in the translation of this technology: unavailability of large quantities of low-cost antifreeze materials with excellent IRI activity at (sub)micromolar concentrations.

IBPs expressed in microbes, plants, fish, algae and insects are both structurally and functionally diverse. To date, five biological functions have been characterized: freezing point depression [ 11 ], inhibition of ice recrystallization [ 12 , 13 ], modulation of ice nucleation [ 14 ], retention of liquid pockets [ 15 ], and ice adhesion [ 16 ]. In the following we briefly summarize the structural characteristics of selected AFPs which are glycosylated (AFGPs) and AFPs which are α-helical, rich in β-strands (β-clips, β-solenoids), lectin-like, and display polyproline II helices [ 1 ]. These inspire the design of (peptidic) IBP mimics, which often contain (a portion of) naturally occurring ice-binding sites and strive to emulate the three-dimensional folds found in native IBPs. 2.1. α-Helical AFPs Several species of fish including the winter flounder, yellowtail flounder and shorthorn sculpin express α-helical AFPs. These fish type I AFPs are typically present as a mixture of isoforms differing in antifreeze activity [ 17 , 18 ]. Widely investigated is the HPLC-6 isoform of the winter flounder type I AFP, in short wf AFP ( Figure 1 a). It has an α-helical structure consisting of 37 amino acids arranged in a sequence of three 11-amino acid repeats of threonine residues separated mainly by alanine residues [ 17 , 19 ]. The high alanine content (62%) contributes to the propensity of the protein to adopt an α-helical conformation [ 20 ]. This helical periodicity places all hydrophilic residues such as Thr on one face of the helix with a spacing of around 16.6 Å. Molecular modeling and docking studies showed that the opposite, Ala-rich face binds ice, suggesting that hydrophobic interactions are involved [ 17 , 18 ]. These results were supported by 13C solid-state NMR spectroscopy experiments, which revealed four times lower relaxation times for the methyl groups of alanine residues in the IBS compared to non-IBS alanine residues [ 21 ]. This decrease in relaxation time was attributed to the interaction between protons from water and methyl nuclei in alanine, suggesting interaction with the ice surface in the ice/water interface. Computational studies showed that the hydroxyl groups of the four Thr residues in the wf AFP (HPLC 6 isoform) are aligned within the helical axis and the oxygen-oxygen spacing between hydroxyl groups is 16.1 Å, 16.0 Å and 16.2 Å, respectively [ 23 ]. This matches the oxygen-oxygen spacing on the pyramidal plane of ice, thereby favoring binding of the AFP to the pyramidal plane ( Figure 1 b).

AFGPs are a group of eight structurally related glycoproteins present in the blood serum of Antarctic notothenioids and northern cod. These proteins consist of 4-55 tripeptide units of AAT that are O -glycosylated at the threonine side chains with the disaccharide β-D-galactosyl(1–3)-α- N -acetyl-D-galactosamine ( Figure 3 b). AFGPs exist as a mixture of compounds rather than as a single compound, where eight different isoforms have been isolated and categorized by electrophoretic techniques. AFGP 1 is the largest glycoprotein and has a molecular weight of roughly 34 kDa. AFGP 8 is the smallest of them with a molecular weight of 2.6 kDa ( Figure 3 b). Detailed conformational studies by circular dichroism (CD) and NMR spectroscopy did not reveal a well-defined three-dimensional structure for large AFGPs in solution. Presumably, AFGPs are rather flexible and dynamic and—akin to synthetic polymers—many configurations co-exist in solution. Simulations [ 29 ], CD spectroscopy and solid-state NMR on AFGP 8 , the smallest AFGP isoform, suggest that adopts PPII structures due to the discrete distribution of prolines in the amino acid sequence [ 30 ].

Ice-binding is a prerequisite for all AF(G)Ps to function. Yet, the physico-chemical principles of the interaction between AF(G)Ps and ice remain elusive, as both experimental and theoretical studies of protein binding to ice are challenging. In recent years, computational studies [ 29 , 33 , 34 , 35 , 36 , 37 ] and microscopy experiments [ 38 , 39 , 40 ] have been particularly illuminating in this area. The simulations of Midya, and Bandyopadhyay [ 41 ] beautifully illustrate how AF(G)Ps can arrest growth via the Kelvin effect as originally proposed by Knight and DeVries [ 42 , 43 ]. Antifreeze proteins adsorb onto growing ice surfaces via their IBS. Growth ceases at the location of the bound AFPs, but continues in between the adsorbed proteins until the addition of water molecules to the convex surface is no longer energetically favorable. Ice growth is now completely halted and the freezing point depressed. Growth remains prohibited as long as the temperature remains above the non-equilibrium freezing point, which depends on the magnitude of the local curvature set by the spacing between bound proteins and the undercooling ( Figure 4 ). In the following sections, we will briefly describe various assays tailored to probe the thermal hysteresis activity, ice recrystallization inhibition activity, and ice-plane affinity of ice-binding proteins. 3.1. Ice Plane Recognition Ice-binding proteins binds ice, but not in the same manner. Preferential adsorption of IBPs onto specific sets of ice crystal planes ( Figure 5 a) in specific orientations can be monitored in ice-etching experiments [ 26 , 44 , 45 , 46 ] assisted by fluorescence detection in a so-called fluorescence-based ice plane affinity (FIPA) assay [ 28 , 44 ]. These yield characteristic ice-etching patterns upon adsorption of IBPs onto ice hemispheres, which are related to the specificities of IBPs for certain ice planes and binding orientations [ 44 ] ( Figure 5 b). AFGPs from the Antarctic notothenioids and northern cod bind to the primary prism plane (1010) [ 47 ]. Type III AFP from the ocean pout also binds to the primary prism (1010) plane, but in addition also recognize the pyramidal plane (2021) [ 48 ]. This pyramidal plane is also recognized by type I AFP from the winter flounder, whereas the shorthorn sculpin AFP adsorbs onto the (2110) secondary prism plane [ 49 ] ( Figure 5 c). Hyperactive proteins such as the insect protein D AFP-1 and the Mp AFP from the Antarctic bacterium Marinomonas primoryensis bind up to three different crystallographic planes. These proteins bind with high affinity to the basal plane (0001) and can also recognize both the primary and secondary prism (1120) [ 26 , 50 ].

Ice-binding proteins in freeze-tolerant species such as overwintering plants inhibit recrystallization in e.g., plant apoplasts to prevent cell and tissue damage [ 3 ]. Ice recrystallization is a thermodynamically driven process wherein large crystals grow at the expense of smaller ones, thereby lowering the free energy of the system. Ice recrystallization inhibition (IRI) is assayed in various ways all of which rely on detection of ice crystal growth by microscopy. In a classical ′splat assay′ ( Figure 7 a), the time-evolution of the mean (largest) grain size (MGS or MLGS) is monitored in a thin polycrystalline wafer of ice, which is kept for several hours at a fixed annealing temperature. Gibson and coworkers proposed that the relative reduction in MLGS by an inhibitor compared to a negative control is a useful metric to distinguish between potent, moderate, and poor inhibitors with high, modest and low/no IRI activity [ 60 ]. Davies et al. proposed an endpoint analysis where the lowest AF(G)P concentration shows IRI activity [ 59 ], wherein up to 12 samples of a series of dilution are analyzed at the same time. Samples are placed side-by-side into wells etched from a sapphire crystal slide ( Figure 6 a). After snap-freezing in liquid nitrogen, the multi-crystalline wafer is observed by optical microscopy to monitor the grain size over time. Quantitative IRI assays use low molecular weight solutes such as salts or sucrose to accelerate recrystallization processes and ensure a sufficiently large non-frozen fraction to avoid false positives. In ′sandwich assays′ of IRI activity microliter samples ′sandwiched′ between two coverslips are observed of AF(G)Ps in 20–45 wt% sucrose solutions with ice crystal volume fractions < 0.3 thus focusing on migratory recrystallization ( Figure 7 b). Quantification is done using automated image analysis to extract the time-evolution of the cubic number average mean radius < R n > 3 from the micrographs [ 61 , 62 ], which is analyzed in the framework of the Lifshitz, Slyozov and Wagner (LSW) theory for Oswald ripening ( Figure 6 c) [ 61 ].

De novo antifreeze analogues offer unprecedented means to advance our understanding of the working mechanism of ice-binding proteins and may be prepared in a more straightforward and less time-consuming manner than native AFPs. Different orthogonal chemical ligations not present in natural AF(G)Ps have been investigated to improve synthetic accessibility and probe whether it impacts antifreeze activity. Constructs varying considerably in composition and architecture from native AFGPs have been investigated to identify the minimal requirements for activity and the impact of various structural elements, such as secondary folds, tethered sugars, and free N- and C-termini. Inspired by type I AFPs, Wierzbicki and coworkers prepared a series of de novo constructs by SPPS comprising 43 amino acid long polypeptides with repetitive sequences of alanine and lysine residues [ 92 ]. Like wf AFP, these constructs had a high content of Ala facilitating the formation of an α-helical fold. Unlike wf AFP, these constructs were devoid of the threonine residues supposedly essential for ice binding. Instead, the constructs contained lysine residues to improve solubility in water. Surprisingly, despite the absence of Thr, the peptidic constructs exhibited thermal hysteresis activity, albeit low with a modest TH of 0.3 °C at 50 mM in 0.1 M ammonium bicarbonate buffer [ 93 ]. Glycine rich snow flea AFP is not available in large quantities, which hampers detailed mechanistic investigations. Kent and coworkers synthesized D- sf AFP, the right-handed amino acid sequence by employing chemical ligations. The sequential ligation was done with N-terminal cysteines incorporated as 1,3-thiazolidine-4-R-carboxylic acid (Thz) and a thioester at the C-terminal [ 32 ]. Aiming to prepare peptidic antifreezes that are not susceptible to acidic, basic or enzymatic hydrolysis, Ben′s group synthesized various C-linked glycoconjugates. In these AFGP-inspired analogues, the natural AAT repeat was replaced by GGK to tether by C-linked glycosylation a variety of sugar moieties. Proline is unique amongst the canonical amino acids in that it has no amide N–H. Consequentially, polyproline cannot form intramolecular hydrogen bonds. In combination with the pyrrole functionality, this makes the polypeptide water-soluble and quite hydrophobic at the same time. Graham and coworkers hypothesized that polyproline helices can be considered the minimal AF(G)P mimic due to their implicit amphiphilicity. To demonstrate that, they performed a series of experiments with polyproline constructs of different length and stereochemistry [ 3 ]. These polyproline constructs showed mild IRI activity since they reduced the mean largest grain size by 50% at concentrations around 15 mM. In another study, a set of different PPII helices were synthesized using SPPS. In this case, peptide sequences resembled AFGP sequences wherein each threonine was replaced by hydroxyproline (Hyp) ( Figure 9 ). The investigators functionalized the hydroxyl group from the Hyp residues with the carbohydrate present in natural AFGPs. Surprisingly, none of the peptides with the carbohydrates showed either TH activity, IRI activity or ice shaping. Peptides without carbohydrates had IRI activity since they were able to reduce the mean larger grain size of ice crystals by more than 80% compared to the PBS control. Moreover, the N-acetylated peptide induced some ice-shaping [ 3 ]. CD spectroscopy studies confirmed the PPII structure of both glycosylated and non-glycosylated compounds. Drori and coworkers recently investigated the activity of self-assembling organic dyes that display specific functional groups in a periodic manner resembling the periodic display of functional groups in IBPs like AFGPs. The organic dye Safranine O was able to show TH activity of −0.32 °C at a concentration of 28 mM, a value comparable to AFGP8 (the smallest isoform) [ 6 ]. Moreover, IRI activity tests through ′sucrose sandwich′ assays demonstrated that Safranine O is able to inhibit ice recrystallization at a concentration of 4.2 mM. Based on this idea, Robert Ben′s group used perylene bisimides (PBI), a well-studied polyaromatic dye, to assemble one-dimensional aggregates displaying ice-binding motifs through supramolecular interactions and π-π stacking between subsequent PBIs ( Figure 10 ). A regular PBI and a tetrachloro derivative were coupled to an AFGP analogue with modest IRI activity (80% MGS in splat assay) aiming to study the possible enhancement of IRI activity upon stacking of multiple short AFGP analogues [ 74 ]. PBI conjugates assembled into micron sized fiber-like structures at μM concentrations and inhibited ice recrystallization at 20 mM. Interestingly, neither of the constructs exhibit TH presumably due to a rather low affinity of the constructs for ice.

Virtual screening techniques, such as the basic local alignment search tool (BLAST) and quantitative structure-activity relationship (QSAR) models, have become an important tool in drug discovery [ 102 , 103 ] aiding the screening of large libraries of small molecules with the objective to find new compounds with therapeutic potential. These tools may speed up the drug discovery process and save costs through reducing the number of tested candidates. Several recent studies demonstrated that screening technologies also have the potential to discriminate between AFP and non-AFP sequences [ 104 ], to predict the activity of AFPs [ 22 ], and to inform efforts to design and synthesize de novo antifreezes [ 67 , 102 ]. Yang and coworkers compiled an effective predictor of new AF(G)Ps based on protein similarity using BLAST. A training dataset and an independent dataset consisting of 481 AFPs and 9,193 non-AFPs were used to create the basis of the predictor. These datasets contained all the diverse primary sequences and structures of known AF(G)Ps. The predictor used features of the AFP such as functional domain, the evolutionary conservation of these domains across proteins of the same family and statistical cross-validation to classify the best protein candidates that may show TH [ 105 ]. Ben′s group used a set of 124 previously synthesized compounds with known IRI activities to calibrate a 3D-QSAR model [ 66 ]. Input to train the 3D-QSAR model comprised of measured IRI activities and modelled three-dimensional structures, which were optimized by Monte-Carlo to identify the lowest energy conformations. On top of that, the team used an independent alignment of the 3D-descrpitors to correlate molecular surface curvature and electrostatical potential. The QSAR model was then used to prescreen a small library of novel putative IRI active compounds, and predicted that 11 small molecules in the library would be IRI active. Nine out of the eleven (82%) predicted compounds did reduce the main grain size in splat assays at concentrations between 11 and 22 mM. Kozuch and collaborators combined molecular dynamics of water molecules near an AFP surface and geometric structure of the IBS with neural networks aiming to predict the magnitude of the thermal hysteresis gap of AFPs [ 106 ]. To build the predictive model, the team used 17 AFP structures desposited in the Protein Data Bank and their corresponding TH activity as a function of AFP concentration, c AFP . In addition, 5 proteins that show no TH but have exposed planar surfaces and/or secondary structures similar to AFPs were used as a negative control set. The generated model successfully predicted the antifreeze activity of known AFPs not included in the training set based on the assumption of a linear relation between TH activity and √ c AFP [ 22 ]. The deviation between predicted and experimental activity values was ≤0.19 °C.

Materials and tools have become available to elucidate the structure-function relations of the broad spectrum of structurally and functionally diverse ice-binding proteins that have been discovered. Long-standing open questions related to the working mechanism of ice-binding proteins may now be addressed both computationally and experimentally. Computationally guided design approaches and further optimization of synthetic strategies will facilitate further development of novel IBP analogues. Various ice-binding motifs may be incorporated into a single analogue to maximize efficacy. Assays of e.g. ice recrystallization activity could be further standardized to facilitate a fair comparison between the activiy of diverse classes of IBPs and their analogues. We also anticipate more in-depth studies on the application of IBPs and analogues in non-native environments geared towards an improved understanding of the factors that impact performance, so as to inform the selection of an appropriate IBP and/or the modular design of a suitable analogue. For example, we have recently investigated the impact of salts which are commonly present in cryopreservation media on the ice recrystallization activity of native AF(G)Ps. Interestingly, the study revealed that sodium borate has little impact on IRI-activity, while it dramatically reduced TH activity [ 12 ]. These findings suggest it is worthwhile investigating whether the salt composition of AF(G)P-containing cryopreservation media can be tuned to maximize the efficacy of AF(G)P on e.g., cell survival. Synthetic approaches to produce peptidic antifreezes by solid-phase peptide synthesis can be further optimized to maximize yield, purity, and time-efficiency without generating large amounts of toxic waste, which constitute an environmental burden and health risk as chronic diseases may arise upon continued exposure. This motivated various successful attempts to adapt current protocols to utilize more environmentally benign solvents and coupling agents. Aiming to find a greener and safer alternative for DMF, McMillan and collaborators studied the effect of various solvents on amide bond-forming reactions [ 117 ]. Their results suggest that dimethylcarbonate (DMC) and 2-methyl tetrahydrofuran (2-MeTHF) can replace DMF and dichloromethane (CH 2 Cl 2 ). Albericio´s team substituted DMF by THF or acetonitrile (MeCN) [ 118 ]. Synthesis in THF or MeCN diminished the degree of racemization of the tripeptide GCF, while the amide bond formation in THF or MeCN remained the same as amide bond formation in DMF. Strikingly, SPPS of hindered pentapeptides with the fungal amino acid 2-aminoisobutiric acid (Aib) resulted in higher synthesis yields (ranging from 2 to 4-fold increase) when THF or MeCN were used as solvents instead of DMF. Aiming to employ less hazardous coupling reagents, many researchers used COMU, OxymaPure and K Oxyma to suppress racemization instead of HCTU and HATU with excellent results [ 119 , 120 ]. Third-generation coupling reagents such as COMU are the most appealing as they are proven safer, give high yields, and disfavor racemization [ 120 ]. In sum, revisited peptide synthesis protocols must retain the simplicity and reaction yields from previous protocols, while implementing improved solvents and coupling reagents. Furthermore, they should be inexpensive for academic and industrial purposes and applicable to various chemical modifications, including cyclization and N-methylation, to be commercially interesting. The approaches discussed in the above have yet to be examined for amenability to long sequences and aggregation-prone peptides.