Results and Discussion

We first determined how AFPIII evolved from SAS. We constructed a bacterial artificial chromosome (BAC) library for the AFPIII-bearing Antarctic eelpout Lycodichthys dearborni and isolated and sequenced the clones comprising the genomic SAS and AFPIII loci. The SAS and AFPIII loci are spatially distinct (Fig. 1). The SAS locus (Fig. 1A) contains two SAS genes, LdSAS-A and LdSAS-B, both with six exons. The two SAS genes are separated by a chicken repeat (CR)-type retrotransposon (LdCR1-3) and are flanked by the CLTA and NCKX2 genes, respectively. This CLTA-SAS-NCKX2 microsynteny is conserved in annotated fish genomes, as illustrated for Gasterosteus aculeatus (Fig. 1A). L. dearborni and G. aculeatus (both percomorphs) have two SAS genes, whereas other teleost species that we examined have one copy. The AFPIII locus spans ∼400 kbp and contains an estimated 30 or more AFPIII genes (Fig. 1B) that share 97–99% nucleotide sequence identities. AFPIII genes have a two-exon structure encoding a signal peptide (exon1) and the mature ice-binding AFPIII (exon2). Except for the 5′ fringe copy ΨAFPIII, they are arrayed in ∼8-kbp tandem repeats with one gene per repeat (ref. 18 and this study) (Fig. 1B), indicative of gene family expansion through in situ tandem duplications. The 5′ copy ΨAFPIII (Fig. 1B), despite having a 214-nt frame-shift insertion, bears greater sequence similarity to the SAS homologs (SAS exon6) (Fig. S1) and is immediately preceded by a partial CR1-3 retrotransposon (LdCR1-3′) similar in sequence to LdCR1-3 in the SAS locus; thus, we reasoned it to be the most deeply diverging member of the AFPIII gene family. The AFPIII locus is flanked at the two ends by the genes Glud1b and Synuclein and LIM domain binding 3b and Melanopsin, respectively. This four-gene microsynteny (without CR1-3′ and AFPIII locus in its middle) is conserved in other teleosts, as illustrated for G. aculeatus (Fig. 1B). Chromosomal fluorescence in situ hybridization (FISH) localized L. dearborni SAS and AFPIII loci to distinct metaphase chromosome pairs (Fig. 1 C and D). Collectively, these results strongly suggest that a genomic region (estimated at ∼12 kbp) containing the LdSAS-B gene and its immediate neighbor sequences (including LdCR1-3) was duplicated and translocated to a site between Synuclein and LIM domain binding 3b genes; from this, the primordial AFPIII gene evolved, and the large AFPIII locus arose from in situ gene family expansion under selection pressure from polar sea-level glaciation.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Genomic organization and chromosomal localization of SAS and AFPIII loci in the Antarctic eelpout L. dearborni. (A) SAS genomic locus organization is conserved between L. dearborni and other teleosts (G. aculeatus shown), except for the insertion of a chicken repeat-type retrotransposon (LdCR1-3) between the two SAS genes in the L. dearborni locus. (B) AFPIII locus of L. dearborni, ∼400 kbp in size, occurs in the middle of the shared four-gene (gene names as indicated) microsynteny with non-AFPIII teleost (G. aculeatus shown). The locus comprises a 5′ pseudogene ψAFPIII followed by >30 AFPIII genes predominantly arrayed in 8-kbp tandem repeats. The regions spanned by a brown line (including genes) in the L. dearborni SAS locus (LdCR1-3 and LdSAS-B) and the AFPIII locus (5′-truncated LdCR1-3 and AFPIII tandem repeats) share strong sequence homology. The relevant BAC sequences were deposited in GenBank under accessions numbers GQ368892, GQ368893, and GQ368894. (C and D) FISH localized L. dearborni SAS (C) and AFPIII (D) loci in separate chromosome pairs, indicative of an interchromosomal translocation of the AFPIII progenitor gene.

Through comparative analyses of LdSAS-A, LdSAS-B, and ΨAFPIII sequences, we deduced that LdSAS-B is the closest relative of the AFPIII progenitor and the molecular events in the LdSAS-B to AFPIII transformation (Fig. 2). The six-exon LdSAS-A and LdSAS-B genes share high nucleotide identities (∼92% between exons and ∼67% between introns) but have differentiated in their 5′ flanking region (5′FR; ∼594 nt) with no sequence homology (Fig. 2, blue line and Fig. S2A). ΨAFPIII shares greater nucleotide identities with LdSAS-B (Fig. 2) than with LdSAS-A (Fig. S2 B and C), including 74% identity to the LdSAS-B 5′FR, which is nonhomologous between LdSAS-B and LdSAS-A. Thus, AFPIII 5′FR, intron1 (I1), exon2 (E2; ice-binding mature AFPIII), and 3′FR were derived from the 5′FR, I5, E6 (SAS C-terminal domain), and 3′FR, respectively, of the ancestral LdSAS-B (Fig. 2 and Fig. S2B). The emerging AFPIII would require a signal peptide for extracellular export of the mature protein. We discovered a precursor signal peptide coding sequence appropriately located in the extant LdSAS-B, starting from 54 nt upstream of the translation start site through the first six codons of LdSAS-B E1 (Fig. 2 and Fig. S2B). An intragenic deletion from the seventh codon of E1 through E5 of LdSAS-B and linkage of the new E1, the old I5, and E6 would complete the formation of the nascent two-exon AFPIII gene encoding the secretory antifreeze protein.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Molecular process of the evolution of AFPIII from SAS-B. One daughter SAS-B duplicate (SAS-B’) underwent N-terminal domain deletion (seventh codon of E1 through E5) and neofunctionalization into AFPIII. Regions in SAS-B’ corresponding to the regions in the two-exon AFPIII gene are indicated with the same colors for the two genes, with nucleotide sequence identities given. The partly nonprotein coding signal peptide (SP) precursor sequence in SAS-B’ that was modified to become a coding sequence for the AFPIII signal peptide is shown at the bottom. LdSAS-A lacks the 5′ flanking sequence homology (blue bar) with LdSAS-B and AFPIII; thus, it is not the evolutionary progenitor to AFPIII.

We then examined for sequence and functional properties that might have compelled the ancestral LdSAS-B duplication and neofunctionalization of one duplicate into AFPIII, and we found strong evidence that would fulfill the predictions of the EAC model. EAC predicts that (i) the ancestor is bifunctional and subject to selection before gene duplication, (ii) adaptive conflict between the ancestral and new function constrains improvement of the selected function(s) before duplication, and (iii) adaptive changes and functional improvement occur in the daughter genes after duplication. To assess LdSAS-B bifunctionality (prediction i), we cloned the SAS cDNAs from the liver and expressed the recombinant proteins in bacteria. We found that the recombinant SAS-B indeed has incipient antifreeze activity (Fig. 3). A single seed ice crystal in pure water (without antifreeze or other ice-active compounds) grows as a discoid and continues to expand at 0 °C, the equilibrium freezing point (fp) (Fig. 3 A and B). In contrast, LdSAS-B at 2 mg/mL causes strong hexagonal faceting of the test ice crystal at the equilibrium fp, indicative of ice binding and growth interference by the protein at or near the prism faces (Fig. 3C). LdSAS-B could also inhibit bulk ice growth at small cooling, producing a nonequilibrium fp depression ranging from 0.004 °C (Fig. 3D) to 0.015 °C (Fig. 3E). When temperature was lowered below nonequilibrium fp, ice grew rapidly as a hexagonal disk (Fig. 3F). Purified AFPIII protein from L. dearborni at the same concentration also caused strong faceting of the seed ice crystal (as a hexagonal bipyramid) at the equilibrium fp (approximately −0.0005 °C) (Fig. 3G), but in marked contrast to LdSAS-B, bulk growth of the ice crystal was effectively arrested at large cooling (0.39 °C to 0.67 °C below the equilibrium fp) (Fig. 3 H and I). Only when the temperature reached the nonequilibrium fp did the unrestricted burst spicular growth characteristic of AFP solutions occur (Fig. 3J). The fp depression activity was measured to be ∼0.58 °C ± 0.08 °C for 2 mg/mL LdAFPIII solution and 1.46 °C ± 0.15 °C for 20 mg/mL solution. This large noncolligative fp depression or thermal hysteresis is the unique property of a bona fide antifreeze protein. To determine which structural domain in the LdSAS-B protein contributed to the ice-binding activity, we chemically synthesized its C-terminal domain (69-residue E6 peptide), which is homologous to mature AFPIII, and tested for its ice-binding ability. We found similar ice-binding activity as the full-length LdSAS-B protein, indicating that the ice-binding activity is an intrinsic property of the SAS C-terminal domain (Fig. S3). To verify that the LdSAS-B–encoded protein possesses SAS enzymatic activity, we measured the sialic acid synthetic function using the recombinant SAS proteins. Indeed, recombinant SAS-B catalyzed in vitro sialic acid synthesis at twofold greater activity (2.7 ± 0.2 mU/mg) than recombinant LdSAS-A (1.4 ± 0.2 mU/mg) (Fig. S4B). Thus, LdSAS-B clearly possesses two activities—the original SAS function and a minor ice-binding activity in its C-terminal domain.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Ice-binding activity of the recombinant LdSAS-B compared with native AFPIII. Single ice-crystal growth behavior in (A and B) water, (C–F) LdSAS-B (2 mg/mL), and (G and H) LdAFPIII (2 mg/mL). The crystal morphology of the test ice at the equilibrium fp of water (A) is discoid but strongly faceted in LdSAS-B (C; hexagonal) and the eelpout antifreeze (G; hexagonal bipyramid), indicating growth inhibition at the prism faces. The ice crystal in water expands unrestricted at the equilibrium fp of water (B). In contrast, ice expansion is inhibited by LdSASB at its equilibrium fp (C and D) and at small coolings (∼0.004 °C to 0.015 °C) below the equilibrium fp (E); at the nonequilibrium fp, (F) ice expands quickly as a hexagonal disk. In comparison, the bona fide antifreeze protein LdAFPIII is much more effective in arresting ice-crystal growth to temperatures significantly (H, 0.39 °C; I, 0.67 °C; data from one of three replicates) below the equilibrium fp until fast spicular growth occurs when the nonequilibrium fp is reached (J).

To assess EAC prediction iii that daughter duplicates released from adaptive conflict acquire adaptive changes that would improve function, we used the branch model and modified branch-site model A (19) to test for positive selection on the LdSAS-B and LdAFPIII lineages and found its occurrence in both (Table 1 and Fig. S5). The input SAS tree includes various teleost SAS coding sequences from GenBank and SAS-A and SAS-B orthologs from the sympatric Antarctic eelpout Pachycara brachycephalum that we obtained in this study (Fig. S5A). The foreground branch (leading to the Antarctic eelpout SAS-B clade) has a large nonsynonymous (dN)/synonymous (dS) substitution rate ratio (branch-site dN/dS of ω2 = 9.44) that is highly significant [likelihood ratio tests (LRT) P < 0.01] (Table 1), whereas all SAS lineages, including eelpout SAS-A genes, have ω < 1, (Fig. S5A), suggesting positive Darwinian selection occurring specifically in the Antarctic eelpout SAS-B genes. Ten residues in SAS-B, mostly (seven) in the N-terminal domain, were identified to be under positive selection (Table 1), suggesting adaptive evolution of the SAS-B enzymatic function after the ancestral SAS-B gene duplication. Likewise, for the tree of homologous AFPIII and SAS E6 sequences (Fig. S5B), we detected a large ω (branch site ω2 = infinite; LRT P < 0.01) in the zoarcoid AFPIII branch, with a much greater percentage (P₂ = 63%) of residues under positive selection than in the eelpout SAS-B branch (P₂ = 16%) (Table 1). This suggests that, after gene duplication and deletion of the N-terminal domain coding sequence of the translocated SAS-B daughter duplicate, adaptive amino acid changes in the nascent AFPIII gene accelerated, rapidly improving the antifreeze function to the full-fledged thermal hysteresis illustrated in Fig. 3. Nozawa et al. (20) recently questioned the statistical basis of the branch-site model and claimed that it generates too many false positives. However, the false-positive rate found by those authors (20) was lower than the significance level (5%) (21). The branch-site test has been successfully applied in many studies, generating interesting biological hypotheses that have been validated by experimentation (22–24). In this study, 7 of 10 amino acids in SAS-B and 8 of 12 in AFPIII detected under positive selection (Table 1) showed hydrophobicity or charge switches from the corresponding sites of their paralogs or the precursor, implying biochemical property or activity changes. Indeed, mutational analyses of AFPIII have shown these amino acids, such as I13, N14, A16, Q44, and 61K, to be functionally important for antifreeze activity (25, 26).

The foregoing results added to existing zoarcoid AFPIII structural information, in turn, shed light on the selection and adaptive conflict that might have transpired in the ancestral SAS-B before gene duplication (EAC predictions i and ii). Of the six conserved residues (Q9, T18, A16, T15, Q44, and N14) in Atlantic ocean pout AFPIII identified through structural studies to constitute a putative flat ice-binding surface (25, 27), two (T15 and T18) are ancestral, preexisting in the C-terminal domain of both SAS-A and SAS-B (T305 and T308, respectively) of L. dearborni (Fig. S6) and other teleost SAS. These might have constituted an accidental structural basis for rudimentary ice affinity in SAS. Indeed, L. dearborni SAS-A and its E6 peptide also has incipient ice-binding activity similar to LdSAS-B (Fig. S3). However, only in SAS-B are adaptive residue changes detected (Table 1). One of the three adaptive changes detected in the LdSAS-B C-terminal domain (Table 1, double dagger), K351 (as opposed to D351 in LdSAS-A and other teleost SAS) corresponds to K61 in AFPIII. K61 in AFPIII forms intramolecular hydrogen bonds with two putative ice-binding residues (Q44 and N14) that are important in stabilizing the flat ice-binding surface (25) or the global protein fold (27), and both are essential for antifreeze activity. Thus, the D351/K351 change in SAS-B might have resulted from positive selection on the ancestral SAS-B before gene duplication for improving the accidental ice affinity of its small C-terminal domain. To test whether further improvement of ice-binding activity in the ancestral SAS-B would likely create conflict for its original SAS function, we substituted four residues (V299, G304, V306, and T334) in the LdSAS-B C-terminal domain with their homologs (Q9, N14, A16, and Q44) in the AFPIII ice-binding surface to mimic adaptive changes to form the AFPIII active site in the C terminal. This mutant (LdSAS-Bm4) showed no detectable SAS activity (Fig. S4B), supporting our hypothesis that residue change to ice-binding capability in the C-terminal domain of the bifunctional ancestor would create pleiotropic conflict. The conflict can be explained by the structure of the SAS holoenzyme. SAS monomer structure resembles an asymmetric dumbbell (16), and the active enzyme consists of a dimer or tetramer through juxtaposition of the N-terminal domain of one monomer with the C-terminal domain of another, forming the site for substrate (sugar) binding (28) between the interacting surfaces of the swapped domains of each monomer (Fig. S7). It is clear from these studies and the inactive LdSAS-Bm4 in this study that structural coordination between the C- and N-terminal domains of the SAS neighbors is essential for SAS enzyme activity. In addition, for the emerging AFPIII function, the bulky nonice-active N-terminal domain of the ancestral SAS would very likely hinder the small E6 domain in contacting ice crystals because of the propensity of the SAS monomers to form dimers. Furthermore, to test the hypothesis that improvement of the SAS enzyme function in LdSAS-B might conflict with the ice-binding activity of its C-terminal domain, we measured the fp depression capability of recombinantly expressed LdSAS-A (Fig. S3), the outgroup of the LdSAS-B and AFPIII genes (Fig. S5) and the putative ancestral SAS state, to infer the original level of ice-binding activity in the ancestral SAS-B before gene duplication. We found that the noncolligative fp depression by LdSAS-A ranged from 0.004 °C to 0.07 °C and that the noncolligative fp depression by LdSAS-B has a lower maximum, between 0.004 °C and 0.015 °C (both at 2 mg/mL concentration, the maximum solubility of the recombinant SAS proteins in aqueous buffer). The lower fp depression capability observed of LdSAS-B compared with LdSAS-A is consistent with improvement of SAS enzyme function in LdSAS-B (2.7 ± 0.2 mU/mg vs. 1.4 ± 0.2 mU/mg) (Fig. S4B) after the gene duplication having an adverse effect on the ice-binding function, supporting the presence of adaptive conflict within the ancestral SAS-B. These conflicts were resolved through a duplication of the ancestral SAS-B, whereby each daughter duplicate could freely improve one of the functions. Additionally, conflict resolution was likely quickened and made permanent by deleting the SAS N-terminal domain coding region in the presumptive AFPIII duplicate, eliminating potential dimer formation between the two paralogs. Accelerated adaptive changes subsequently occurred in the nascent AFPIII gene, indicated by nearly two-thirds of the residues in AFPIII experiencing positive Darwinian selection (Table 1); this resulted in rapid optimization to a full-activity AFP capable of preventing freezing of the fish body fluids.

The genesis of the extracellular secretory signal in the primordial LdAFPIII gene by incorporating a piece of 5′ UTR into a functional protein coding sequence with some modifications (Fig. 2) is an added evolutionary innovation. The capacity to code for a functional signal peptide (SP), in fact, existed in the precursor sequence in LdSAS-B, because we found comparable levels of AFPIII-exporting activity by the SP precursor-mature AFPIII construct and the native pre-AFPIII cDNA (Fig. S8). Thus, the evolution of Antarctic eelpout AFPIII has entailed tapping into two inconspicuous functionalities in the same cytoplasmic ancestor and remarkably, transforming them into a quintessential lifesaving antifreeze function. The partly nonprotein coding origin of AFPIII signal peptide represents an example of recruiting a hidden function by incorporating a translation start site, and sheds light on how products of duplicated genes could be targeted to different cellular localizations.

A remaining question is how the weakly ice-active ancestral SAS protein could be initially beneficial for natural selection to act on. Observations from our laboratory testing of SAS ice inhibition offers an explanation. Ice growth in SAS solution could be inhibited at temperatures slightly below its equilibrium fp when the seed ice crystals were very small (∼10 μm) and the cooling rate was very slow (∼0.0009 °C/min). These two conditions quite certainly apply in the wild over evolutionary time. Sea surface ice formed before water temperatures of the Antarctic Ocean reached freezing in its entirety, and thus, ice crystals that could appear in the water column were likely thermally unstable and therefore, small. The cooling rate of the polar waters was very slow, an overall cooling of ∼12 °C in the past 50 Myr in the Antarctic deep water (29), compared with the laboratory rate of ∼0.0009 °C/min (473 °C/y). Thus, the incipient ice activity of the ancestral SAS molecule or its detached C-terminal peptide would be quite sufficient in inhibiting ice-crystal expansion in the fish body fluids. By the time water temperatures chilled to the equilibrium fp of the fish and fp depression became essential for survival, one can assume that the ongoing process of natural selection would have led to the refinement of the incipient ice-binding activity to a full-fledged antifreeze protein.

To summarize, through tracing the processes in the SAS-B to AFPIII evolution in the Antarctic eelpout, we provided strong and comprehensive molecular and functional evidence for a clear example of EAC-compelled duplication of a bifunctional ancestral gene and additionally, acceleration of conflict resolution through intragenic domain deletion in one duplicate and its neofunctionalization into a protein of distinctive function. It also reveals that an ancestral molecule (SAS) can be subject to a separate modality of natural selection (advent of freezing marine conditions) for an accidental functional property (ice binding) concomitant with preexisting cellular selection on improving the ancestral function (enzymatic). Thus, this study provides a fresh evolutionary perspective that gene duplicates, although bearing identical sequence at birth, could be exposed to divergent mutational and selection pressures that propel them on different evolutionary trajectories (2). The evolutionary process of zoarcid AFPIII, in total or part, likely typifies the evolution of other antifreeze proteins from functionally unrelated ancestors. For example, the antifreeze glycoprotein (AFGP) gene in Antarctic notothenioid fishes evolved from a duplicated trypsinogen-like protease (TLP) gene and entailed intragenic deletion of most of the coding region for the protease in the chimeric AFGP-TLP evolutionary intermediate (30, 31). Type II AFPs in sea raven, herring, and smelt are homologous with the carbohydrate-binding domain of C-type (Ca²⁺-dependent) lectins (13, 32), and thus, EAC-driven gene duplication may apply to their evolution. In principal, any protein gene with more than one function that may be at odds with each other because of incompatible structural requirements or regulation of expression is candidate for EAC-based gene duplication and subsequent optimization. We envisage further interesting empirical investigations and evidence for EAC as an important and common underlying mechanism in generating genetic and functional novelty in time to come.