Variations on a theme: non-canonical DUF3494 ice-binding proteins Authors: James A Raymond Published: 2024 ABSTRACT AbstractAmong the many ice-binding proteins (IBPs) found in microorganisms (bacteria, archaea, fungi and algae), the canonical DUF3494 beta-barrel type is the most common. Until now, little variation has been found in this structure: an initial coil leads into an alpha helix that directs the following coils into a reverse stack, with the final coil ending up next to the initial coil. Here, I show that there exist many bacterial proteins whose AlphaFold-predicted structures deviate from the DUF3494 structure so that they are not recognized as belonging to an existing DUF or Pfam family. In these non-canonical DUF3494 (ncDUF3494) proteins, the number of coils in the alpha helix is highly variable, often being as high as 14. The putative ice-binding sides of each of 13 proteins modeled have a well-aligned row of hydrophilic residues, with spacings that are close to the repeat distance on the ice a-axis. A recombinant protein made for one of the proteins showed that it had ice-binding activity, even in the µg/ml range. The ncDUF3494 proteins appear to be found only in bacteria, the great majority of which live in icy habitats. C-terminal PEP-Cterm motifs, which are rare in DUF3494s, are present in most of the ncDUF3494s, possibly indicating a secretory function. The relatively narrow distribution of ncDUF3494 proteins suggests that they are a later development in DUF3494 evolution.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00792-024-01374-y.Keywords: DUF3494, Ice-binding proteins, Antarctica, Arctic, Bacteria Introduction Many ice-associated microorganisms have genes for ice-binding proteins (IBPs) that aid in survival in icy habitats in various ways. These include protection from freeze–thaw damage, anchoring to an ice substrate and changing the structure of sea ice to provide liquid habitats (Raymond et al. 2009; Davies 2014; Bar Dolev et al. 2016). The most common type are proteins with a triangular, β-solenoid structure called DUF3494 (Vance et al. 2019), whose structure was determined from a crystallized IBP from a snow mold (Kondo et al. 2012). Subsequently, the structures of eight other DUF3494 were determined and they all followed the same pattern, although differences were found in the capping regions at ends of the β-solenoids (Arai et al. 2019; Vance et al. 2019). In most cases, one of the three sides of the structure is populated with amino acids that promote binding either directly to ice or to water molecules that glue the organism to an ice surface. They are found in dozens of ice-associated bacteria, archaea, fungi and unicellular algae. They are also found in hundreds of other bacteria, in which their functions are not known but they may serve to bind to surfaces other than ice. Here, I describe a novel type of DUF3494 proteins that differ from canonical DUF3494s in having a larger and variable number of coils in their beta roll structure. Two proteins of this type were recently found in Arctic sea ice, although their hosting organisms and coil structures were not reported (Winder et al. 2023). The first of the proteins in the present study was found in a patch of the chlorophyte alga Prasiola crispa in a rocky penguin colony in Antarctica, and dozens of homologs were found in the databases. Unlike DUF3494 proteins which are found in a variety of eukaryotic microorganisms and often in warmer climates, the non-canonical DUF3494 (ncDUF3494) proteins appear to be found only in bacteria, almost all of which live in cold, icy habitats. Materials and methods Prasiola An environmental sample of Prasiola crispa from an Adelie penguin colony at Cape Crozier, Antarctica (77.508 S 169.358 E) was kindly collected and provided to this study by David Ainley in January 2000. The sample was stored at −25 °C. Ice-binding activity Ice-binding activity was assayed as described previously (Raymond and Fritsen 2000) in which an ice single crystal was immersed in the supernatant at a temperature just below its freezing point, and the growth of the crystal was observed with a dissecting microscope. Distortions of the growing ice surface were taken as indications of ice-binding activity. Sequencing Genomic DNA was obtained from the sample in May 2015 at the University of Nevada Las Vegas after storage at −25 °C for 15 years. DNA was extracted from a liquid nitrogen-ground sample with a Nucleospin kit (Macherey–Nagel) using lysis buffer PL2 and sent to the Nevada Genomics Center (NGC) at the University of Nevada Reno for sequencing. A library was prepared using the Ion Xpress Plus fragment library kit (Life Technologies) and sequenced on an Ion Torrent Proton platform using Hi-Q chemistry and a P1 chip (Life Technologies), yielding about 86.1 million reads with an average length of 181 bp. The reads were assembled into 877,292 contigs (average length, 702 bp) with CLC Genomics Workbench, with a total length of about 616 million bp. The contigs were searched with blastp for known IBPs (Bredow and Walker 2017). Only the DUF3494 queries (59 bacterial and eukaryotic DUF3494 domains) produced hits. A contig of interest (#82,040) was expanded by finding individual reads that reliably connected it with other contigs. DNA sequences were assembled and imaged with Geneious Prime. Protein structure prediction The structures of selected ncDUF3494 proteins were predicted with the ColabFold notebook (Mirdita et al. 2022) which is based on AlphaFold21 (Jumper et al. 2021) and implemented by UCSF ChimeraX v. 1.6rc (Pettersen et al. 2021) using the default parameters. AlphaFold depicted all the structures in dark blue, indicating high levels of confidence. The PDB files generated were viewed with Yasara View (https://www.yasara.org/viewdl.htm) and free energy minimized with Yasara energy minimization server (https://www.yasara.org/minimizationserver.htm). Three-dimensional images were generated with Yasara View by rotating the images 5° around the y-axis. Recombinant protein One of the ncDUF3494s was selected for making a recombinant protein to test its ice-binding activity. The beta solenoid portion of the Antarctic Cand. Aquiluna sp. (WP_091849830) was sent to GenScript (Piscataway, NJ, USA) for generation of the recombinant. This protein has a predicted N-terminal beta sandwich structure, which was omitted from the recombinant. The sequence of the beta solenoid, based on its AlphaFold predicted structure, together with an N-terminal His-tag was: MHHHHHHDYSLLAGAAITTGASNSFTNSLGAGAAITTGAGNVIPGSLNAGAAITTGADNVIGVDLNAGAAITIGASNTISGSQNFGVVTPISGYSVAMAAFDVAMADSLSRPSTLLAAELGETTLGAGVYSAPAFFTLTGALTLDAKNDPTAVFIIRSPGYISTAAGASIVLANGAQASNVFWVSGSYFSAGAGAVLSGNILATSYVTLGADAGLTGRIFSQSGYITLGANVTVKN* The sequence was codon optimized for Escherischia coli and inserted into the pET30 vector, which was used to transform E. coli BL21 Star™ (DE3) competent cells. A single colony was inoculated into LB medium containing kanamycin; cultures were incubated for 16 h at 15 °C with shaking at 200 rpm. Once cell density reached OD600 = 0.6–0.8, 0.5 mM IPTG was added to induce protein expression. The expressed protein was obtained from the cell lysate using a Ni column. A soluble protein was confirmed to have the expected size (23.159 kDa) by SDS-PAGE and Western blot using an anti-His monoclonal antibody (Fig. S1). The protein was delivered at a concentration of 0.31 mg/ml in 50 mM Tris–HCl, 500 mM NaCl, 0.25 M L-Arginine, 1 mM Tris(2-carboxyethyl)phosphine (TCEP), 10% glycerol, pH 8. On arrival, the protein was dialyzed to remove glycerol: 1 ml was dialyzed for 1 h against 150 ml ice-cold stirred deionized water using 6–8 k MW cut-off dialysis tubing (SpectraPor). Water absorbed during dialysis diluted the sample to about 0.2 mg/ml. Final osmolality (~ 200 mOsm/kg) was measured with a Wescor model 5520 vapor pressure osmometer. Ice-binding activities were determined as described above. Other characteristics Ice-binding domains were detected with NCBI’s Conserved domain database (CDD) (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi). Signal peptides were detected with SignalP 6.0 (https://services.healthtech.dtu.dk/services/SignalP-6.0). Molecular weights and pIs were predicted with Expasy’s pI/MW tool (https://web.expasy.org/compute_pi/). A weblogo of 87 motifs from seven proteins was obtained from an online tool (https://weblogo.berkeley.edu/logo.cgi). Results Prasiola crispa grows well at an Adelie penguin colony at Cape Crozier (Fig. 1a). An environmental sample of the alga was found to have strong ice-structuring activity, as shown by distortions of a growing ice crystal (Raymond and Fritsen 2000) and, in a more recent assay, by pitting of the ice basal plane (Fig. 1b). The activity was initially assumed to be due to P. crispa, which accounted for essentially all of the biomass in the sample, but years later (in 2015), a metagenome of the Prasiola sample showed that other sources of activity could be present. A search of the metagenome for known IBP genes (Bredow and Walker 2017) found only DUF3494-type genes. Prasiola crispa. a Photo of a patch of the alga at Cape Crozier, Antarctica taken by D. Ainley in January 1983. Width of field is estimated at about 50 cm. b A densely pitted ice crystal grown in the presence of a supernatant of a homogenate of an environmental sample of the alga similar to that shown in (a). The pits are sharply faceted hexagons on the ice basal plane. They are evidence that the homogenate contains ice-binding proteins, but the source of the activity was not identified. Scale bar, 1 mm About 60 contigs with these genes were found but no chlorophyte genes could be found in them, suggesting they were all bacterial IBP genes. In one of the longer DNA contigs (#82,040), NCBI’s CDD detected a ~ 100-a.a. “ice-binding” region that differed from the full ~ 200-a.a. DUF3494 domain. The full protein (413 a.a., GenBank PP848465) had a predicted acidic pI and closely matched (Table S1) a protein from an Antarctic alpha proteobacterium (PZN94751) from a microbial mat on the opposite side of the Antarctic continent. The contig was manually expanded to about 18 kbp and annotated to identify its source (Fig. S2). The contig contained 12 genes, all most closely matching genes from alpha proteobacteria, especially Sandarakinorhabdus and Polymorphobacter. The protein was modeled with AlphaFold, which predicted a right-handed, triangular beta roll structure with a parallel alpha helix (Fig. 2a), similar to the structures of cDUF3494 proteins. The prediction was made with high confidence (Fig. S3a). However, its length was more than twice the normal length with 14 aligned coils instead of the normal six. Such a structure has not previously been observed, which explains why the CDD only partially recognized it. (Many IBPs have two and even three DUF3494 domains (Raymond et al.2008; Lee et al. 2012; Winder et al. 2023) but in these cases the domains are separate and not aligned). Stereo images of two bacterial ncDUF3494 IBPs as modeled by AlphaFold. The views are on to the putative ice-binding faces with aligned, outward-pointing hydrophilic Ser and Thr residues. a The 14-coil protein in contig 82,040 of the Prasiola metagenome. b A 9-coil protein from an Antarctic Cand. Aquiluna sp. (WP_091849830) with an N-terminal beta sandwich structure. Portions that could not be reliably modeled by AlphaFold are not shown To determine whether the ncDUF3494 proteins have ice-binding activity, a recombinant protein was made for the beta solenoid portion of the protein shown in Fig. 2b. At a concentration of approximately 0.2 mg/ml, the protein caused ice to grow in a highly irregular pattern (Fig. 3a) and treating a similar sample to 50 °C for 10 min destroyed this activity (Fig. 3b), strongly suggesting that a protein is responsible for the activity. Even when the recombinant was diluted 100x (to a concentration of about 2 µg/ml) it retained activity (Fig. 3c) that was also destroyed by heat treatment (Fig. 3d). Ice-binding activity of a recombinant non-canonical DUF3494 protein, the nine-coil region of Cand. Aquiluna sp. shown in Fig. 2b. Each panel shows an ice seed crystal submerged in a solution of the recombinant. The ice c-axis is normal to the page, as well as normal to the major surface of the ice called the basal plane. a, b Each panel shows a part of the seed crystal on the left and growth of new ice in the presence of ~ 0.2 mg/ml recombinant on the right. a Untreated recombinant in which new growth is shown propagating in the ice a-axis direction in a thin, highly disordered layer. b Exposure to 50 °C for 10 min destroyed the activity shown in (a), causing ice to grow in smooth dendrites. c, d Seed crystals in the presence of ~ 2 µg/ml recombinant. c Recombinant inhibited growth, resulting in the development of facets on the edges and distortions on the basal plane. d Denatured recombinant did not inhibit growth or distort the basal plane. Scale bar, 1 mm. All panels have the same scale Blast searches yielded over 100 homologs of this protein (excluding canonical DUF3494s) with evalues < 1e−15. Some genera (e.g., Colwellia) had highly similar (virtually duplicate) sequences. After removing the duplicates, 64 proteins remained (Table S2). Unlike canonical DUF3494s, which are found in many phyla and many climates, all of the ncDUF3494 proteins were bacterial and almost all of the bacteria were from icy environments. Prasiola contig82040 appears to have the only ncDUF3494 protein in the metagenome as using the remaining 63 proteins as queries yielded no strong additional hits. AlphaFold predictions of 17 of the hits revealed structures with between 8 and 16 coils (see column b of Table S2). For example, a 9-coil ncDUF3494 from a bacterium from the Shetland Islands, Antarctica is shown in Fig. 2b). Note that in the two structures shown in Fig. 2, the putative ice-binding faces are predicted to be flat. Similar flat surfaces were predicted for each of the other ncDUF3494s modeled, a feature that may result in improved ice-binding (see Discussion). All of the ncDUF3494s modeled retain a peculiar structural feature of canonical DUF3494s: a discontinuous coil structure in which the order of coils in the protein sequence differs from the order in the molecule. This results in the first and last coils in the sequence being adjacent to each other and the top coil in the stack being linked to the bottom coil through an alpha helix. Some examples are shown in Fig. 4, in which 9-, 11- and 14-coil proteins are compared with a canonical 6-coil DUF3494. Another way of viewing the discontinuous structure is to color one of the ncDUF3494s in rainbow format (Fig. S1b). Comparison of the structures of canonical and non-canonical DUF3494 proteins. The beta strands (red) each represent a 3-strand triangular coil and together form a flat, five-residue surface that is the putative ice-binding side of the protein. An alpha helix (blue) runs parallel to the stack of coils and separates them into two sections, leaving the first and last coils adjacent to each other. The N- and C-termini of the barrel are denoted by N and C. The order of the strands in the diagram is the order in which they appear in the sequence, while their order in the stack of coils is shown at the right. Note that most of the variation is in the number of coils on the N-terminal side of the alpha helix. a A canonical 6-coil DUF3494 from a glacier alga (Ancylonema nordenskioldii, WHL30856). b–d Antarctic bacterial ncDUF3494 IBPs with varying numbers of coils. b A 14-coil protein (PP848465) from the P. crispa metagenome. The same structure (not shown) is predicted for an alpha proteobacterium from an Antarctic microbial mat (PZN94751). c An 11-coil protein found in two bacteria from Prydz Bay (Colwellia sp., WP_146880230) and Taynaya Bay (Glaciecola sp., WP_006014494), respectively. d A 9-coil protein from Cand. Aquiluna sp. from the Shetland Islands (WP_091849830) The three sides of the triangular DUF3494 proteins are conventionally referred to as the a, b and c faces, in which the alpha helix is on the a face and the b face presents a flat face that is usually considered to be the ice-binding face. In some cases, the c face has also been implicated in ice binding (Mangiagalli et al. 2018). In each of the ncDUF3494s, the b face (the side shown in Fig. 4) appears to be well structured for ice-binding, as their AlphaFold structures show a central row with a high density of outward-pointing hydrophilic Thr and Ser residues. This is illustrated in the Prasiola-associated protein and two other ncDUF3494s in Fig. 5. In the former (Figs. 2a and 5a), the spacing of these residues in the (energy-minimized) AlphaFold structure is 4.62 Å, which is close to the ice a-axis repeat distance of 4.52 Å. The predicted structures of the other modeled ncDUF3494s (not shown but listed in Table S1) are similar, with high incidences of hydrophilic residues on the ice-binding side (Fig. 5d). DUF3494 IBPs have similar patterns on their ice-binding sides although their Thr/Ser contents appear to be not as high as those observed in the ncDUF3494s. Examples of alignment of threonine and serine residues on the putative ice-binding sides (red) of three Antarctic bacterial ncDUF3494 proteins. The central, outward-pointing residue is highlighted in yellow. a Prasiola metagenome contig 82,040. b Colwellia sp. (WP_146880230). c Glaciecola sp. WP_006014494. The coils are shown in the order in which they appear in the sequence, while their orders in the proteins are shown at the left. Alpha helices are shown in blue. d Weblogo diagram of the 5 residues, in which hydrophilic residues (green) dominate ncDUF3494 proteins resemble DUF3494s in some respects but differ in others. A shared feature is a high incidence of N-terminal signal peptides (67% for the ncDUF3494s in Table S2). A second feature that is found only in bacteria is a C-terminal motif called PEP-Cterm (a proline-glutamic acid-proline motif followed by two or three alkaline residues). Although they are not common in DUF3494 IBPs, interestingly, most (58%) of the ncDUF3494 proteins have them (Table S2). Additionally, three of the Antarctic ncDUF3494s from Cand. Aquiluna sp. (WP_091849830), Colwellia sp. (WP_182241509) and Shewanella sp. (MBO1896578), respectively, have an unusual structure recently observed in an algal DUF3494 IBP from a Swiss glacier (Prochazkova et al. 2024). The ice-binding domains of these IBPs have a flanking beta sandwich structure, either preceding the ice-binding domain (Colwellia) (Fig. 2b) or following it (Cand. Aquiluna and Shewanella). One feature that was present in 10 of the ncDUF3494s, notably in the Colwellia accessions (while unreported in DUF3494s), is a ~ 500-a.a. C-terminal domain called DUF6701 (Table S2). Possible functions of these features are discussed below. Discussion DUF3494 proteins have been identified in hundreds of bacteria (and some archaea) from both freezing and non-freezing environments. The latter include stalactites in caves, lake sediments, groundwater and even hot springs, in which they may bind to mineral substrates other than ice. They have also been found in dozens of eukaryotes (mostly algae and fungi), so far all of which live in icy habitats. Their structures are highly conserved across these phyla. However, many bacterial genomes also have proteins auto-annotated as “ice-binding”, apparently based on partial similarity to the DUF3494 consensus sequence. Until now, these ice-binding domains have not been modeled. The present results show that there exist many variations of the DUF3494 structure in bacteria, all retaining the triangular beta roll structure with (usually) one face well suited for ice-binding and two sets of coils separated by an alpha helix that stack in opposite directions. The highly ordered arrays of hydrophilic residues on the putative ice-binding sides of the molecules strongly suggests that the models are accurate. Furthermore, the finding of ice-binding activity in a recombinant ncDUF3494, even in the µg/ml range (Fig. 3) suggests that canonical and non-canonical DUF3494s have similar functions. The main difference between the two appears to be a variation in the number of coils on the N-terminal side of the alpha helix, being one in cDUF3494s and up to ten in ncDUF3494s. As mentioned above, the putative ice-binding faces (the b faces) of ncDUF3494s are extended compared to those of canonical DUF3494s, but the surfaces are predicted to remain flat. This would be expected to increase the ice-binding surface, and thus might result in stronger ice binding. In fact, a strong correlation was found between the number of coils in the antifreeze protein of the beetle Tenebrio molitor and antifreeze activity on a mg/ml basis (see Fig. 5b in (Marshall et al. 2004)). Thus, the longer lengths of ncDUF3494s might give them increased activities per mg compared to those of cDUF3494s. Among the shared features of canonical and non-canonical DUF3494 proteins are N-terminal signal peptides. These could either be cleaved during secretion of the protein or they could act as transmembrane regions that anchor the protein to the bacterial outer membrane. Whether they are secreted or membrane-bound, IBPs can mitigate the damaging effects of ice-recrystallization during warming periods (Raymond and Janech 2003; Raymond and Knight 2003) as well as provide liquid habitats in freezing ice (Raymond 2011). A second shared feature is the presence of beta sandwich structures in DUF3494 proteins. Many bacterial DUF3494 proteins have one or more beta sandwich domains (with two beta strands per sheet), called bacterial immunoglobulin (Big) domains, that are suspected of contributing to ice-adhesion (Vance et al. 2018, 2019). A larger beta sandwich structure (with four beta strands per sheet) was recently found in the A. nordenskioldii glacier alga IBP (Prochazkova et al. 2024). The finding of similar structures, each with between three and five beta strands per sheet, in three of the 17 modeled ncDUF3494s provides further support that these structures have a role related to ice-binding. On the other hand, PEP-Cterm motifs are much more common in ncDUF3494s (Table S2). They were also common among the IBPs found in Arctic sea ice (Winder et al. 2023) These motifs are present in many bacterial proteins and are typically found in species that secrete extracellular polymeric substances (EPSs) (Haft et al. 2006) and have been shown to have roles in adherence to surfaces, biofilm formation and desiccation tolerance (More et al. 2014), suggesting that the ncDUF3494s function outside the cell. This idea is supported by the presence of C-terminal DUF6701 domains in many of the ncDUF3494s. These domains are found in a wide range of bacterial cell surface proteins, which suggests that these ncDUF3494s reside on the cell surface. It is likely that DUF3494 proteins first evolved in bacteria because bacterial DUF3494s greatly outnumber eukaryotic DUF3494s in the databases and the latter were almost certainly acquired from bacteria (Raymond and Kim 2012). However, because bacterial DUF3494s are found in both icy and ice-free environments, their original substrate remains unclear. A separate question is the origin of the ncDUF3494s. It seems likely that they are a more recent development that evolved from DUF3494 IBPs because they appear less diverse (they are about an order of magnitude less abundant than bacterial DUF3494s in the databases), they have limited geographic distributions and there is no evidence yet that any of them have been horizontally transferred to eukaryotes. Further studies of the relative diversities and endemisms of ice- and non-ice-binding DUF3494-family proteins should shed some light on the order in which they evolved. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 661 KB) Supplementary file2 (XLSX 16 KB) Acknowledgements I thank Dr. David Ainley (H.T. Harvey Associates) for collecting the Prasiola sample for this study and for a photo of the site, the School of Life Sciences, University of Nevada Las Vegas for providing laboratory facilities to conduct this research, and Drs. Michael Miller and Elmar Krieger for technical help with Geneious and Yasara, respectively. This study was partly funded by National Science Foundation grants OPP-9814294 and OPP-0088000 to J.R. and OPP-9814882 to D.A. Abbreviations Domain of unknown function Ice-binding protein Non-canonical DUF3494 IBP Proline-glutamic acid-proline Author contributions J.R. planned experiments, performed experiments, analyzed data, wrote the paper. Data availability PDB files are available for the 17 ncDUF3494 proteins shown in Table S2 on request. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary file1 (DOCX 661 KB) Supplementary file2 (XLSX 16 KB) Data Availability Statement PDB files are available for the 17 ncDUF3494 proteins shown in Table S2 on request.