Antifreeze Proteins

"Even though we haven't directly observed it, the scenario seems pretty convincing as an example of Darwinian evolution by natural selection." — Michael Behe, The Edge of Evolution, 2007¹

Summary

Fish that live in ice-laden polar seas would freeze solid without help.  Their blood freezes at about −0.7 to −1 °C, and the sea sits at −1.9 °C.²  Antarctic notothenioids and several northern cods survive because their blood carries antifreeze glycoproteins (AFGPs):  chains of the three amino acids threonine, alanine and alanine, repeated, with a two-sugar unit on each threonine, which bind to small ice crystals and stop them growing.  The two groups of fish are not close relatives, and their proteins are almost the same.³  Their genes are not.

This article asks how strong the evidence is that these genes arose new, and what exactly arose.

1. The Antarctic gene was rebuilt from a digestive-enzyme gene. Its two ends and its first intron match a trypsinogen-like protease gene at 93% to 96% identity, about 560 letters in all. Between them sits a new coding region of about 2,100 letters, built from 170 copies of a nine-letter unit that codes threonine, alanine or proline, then alanine.² The founding nine letters sit in the trypsinogen gene where its first intron meets its second exon, partly in DNA that never coded for protein.² This site's own alignment of the deposited sequences reproduces the published identities.⁴
2. The northern cod gene was built from DNA that coded for nothing. Xuan Zhuang and colleagues compared seven cod species in 2019. Basal species carry a non-coding stretch that already resembles a signal sequence, next to a run of repeats. One letter change made the first threonine-alanine-alanine unit, copying expanded it, a one-letter deletion joined it to the signal sequence, and a promoter was gained by a rearrangement whose source nobody has found.⁵ A second team, using whole genomes, reached the same "non-coding" verdict by a different method.⁶
3. Both origins are reconstructed, not observed. Nobody watched either gene form. The steps, their order and their dates are inferred by comparing living fish. The experiments in these papers test the reconstruction, for instance by showing that the precursor DNA in relatives is not transcribed. They do not show the history happening. See Observed, or Only Inferred?, which grades both cases "reconstructed".
4. The reconstructions are strong but not seamless. The Antarctic gene's origin date has been revised from 5 to 14 million years to 42 to 22 million years.⁷ The two cod papers put the cod gene's origin at about 3 and at 13 to 18 million years ago, and disagree on whether the deepest cod relatives carry the precursor at all. A 2023 study of 24 notothenioid genomes still lists the derivation as "not yet been fully resolved", including where the spacers between antifreeze units came from.⁸
5. What arose is new genes with very little new information. Each origin took roughly four to seven mutation events of common kinds, followed by copying. The 2,124-letter new coding region of one Antarctic gene uses only eleven distinct versions of its nine-letter unit, and a standard compressor needs less description for it than for the 750-letter trypsinogen coding sequence beside it.⁴ The protein has no fixed fold and binds no partner protein.
6. Every other fish antifreeze follows the same pattern. Type III antifreeze came from a duplicated enzyme gene whose end already bound ice weakly.⁹ Type II is a borrowed sugar-binding domain. Type I, a single alanine-rich helix, arose at least four times, each time from a different gene, keeping the old gene's control regions and growing a new repetitive coding stretch.¹⁰ None of the known fish antifreezes is a new folded domain built from scratch.
7. The young-earth and design writers mostly concede the origin and contest its meaning. Michael Behe calls the Antarctic scenario "pretty convincing".¹ Creation Ministries International's Shaun Doyle grants that mutation and selection are "a likely mechanism" for antifreeze proteins and calls them "beneficial debris".¹¹ Others call the cod reconstruction "just storytelling".¹² The suggestion that the genes were present all along and merely switched on fails for the cod gene, whose relatives carry neither the coding repeats nor a promoter.
8. The evolution side has overstated it too. The cod paper's "We experimentally proved" refers to tests of a reconstruction.⁵ Its claim to have "fully resolved" the question leaves the promoter's source unexplained.⁵ The 1997 Antarctic date was called "reliable" and was off by a factor of about three.²

The verdict.  These are the best-documented cases of genes that did not exist before, and the evidence for them is strong.  The gene from non-coding DNA in cod is real by every test that has been applied to it, and no young-earth source found for this article explains why its precursor sits in relatives exactly as the reconstruction predicts.  What the cases show is also limited.  Selection under a life-or-death pressure built a protein whose job, sticking to ice, is one of the easiest jobs a protein can have:  a few letters of new sequence, copied many times.  That is a real gain of a new function.  It is not evidence that the same process builds folded enzymes or molecular machines, and nothing in these papers claims it is.

Part 1: The Antarctic gene, rebuilt from a trypsinogen gene

What the 1997 paper found

In 1997 Liangbiao Chen, Arthur DeVries and Christina Cheng, at the University of Illinois, sequenced an antifreeze gene and a trypsinogen gene from the Antarctic toothfish, Dissostichus mawsoni.  Trypsinogen is the inactive form of trypsin, a digestive enzyme made in the pancreas.  Their headline:  "the antifreeze glycoproteins (AFGPs) of the predominant Antarctic fish taxon, the notothenioids, evolved from a pancreatic trypsinogen."²

The evidence is a set of matches between the two genes.  The antifreeze gene's ends and its intron match the trypsinogen gene closely, and its middle does not.

The figures are Chen and colleagues'.²  This site aligned the two sequences they deposited in GenBank and got 94.0% for exon 1, 92% for intron 1 and 96.4% for the tail, 29 differing letters in 523 compared.⁴  Only the nine-letter stretch connects the new coding region to the old gene.  It "translates into Thr-Ala-Ala", and it straddles the junction of intron and exon, so its first letters lay in DNA that was cut out of the trypsinogen message and never coded for protein.²  John Logsdon and Ford Doolittle, commenting in the same issue, put it this way:  "the contemporary AFGP gene retains, as its birthmark, sequences at both ends which are nearly identical to trypsinogen."¹³

A second line of evidence is the spelling of the repeats.  Each amino acid can be spelled by several codons.  In this gene threonine is spelled aca 137 times out of 170, and the second alanine gca 162 times out of 171, which matches the trypsinogen stretch.  Copies of one founder would be expected to keep its spelling.²  The match is not perfect.  The middle codon of the trypsinogen stretch is gcg, while most repeats use gct, and the authors propose that the ancestral stretch read gct.²

How the gene is supposed to have formed

Chen and colleagues propose four events, and they say the order is open:  "The deletion, recruitment, and amplification events did not need to occur in the order given."²

1. Keep the front of a trypsinogen gene: exon 1, intron 1, and the first few letters of exon 2, including the nine-letter stretch.
2. Delete the middle: the rest of exon 2 through intron 5, joining the nine-letter stretch to exon 6. This joint shifted the reading frame, creating a stop codon at the start of the old exon 6, so the enzyme's last exon became the new gene's untranslated tail.²
3. Copy the nine letters many times, probably by slippage of the copying machinery at a run of gt repeats just upstream.
4. Gain a three-amino-acid spacer between antifreeze units, which is cut out after the protein is made. Its origin "is unclear at this point."²

Two further features of the gene are not explained.  The extra 1,700 letters in the intron may be "preexisting sequence or recombinatory events", and the original trypsinogen gene that was converted is gone, so it cannot be checked.²  Neither matters for the protein.

What the genomes have added since

Later work has filled in the neighbourhood and the copy numbers, and has also made the story less tidy.

• The gene sits among its relatives. In the toothfish the antifreeze genes lie in tandem with trypsinogen-like protease genes and with three chimeric genes that encode both an antifreeze chain and the protease.¹⁴ Cheng and Chen reported such a chimera in 1999 as "an important intermediate in this evolutionary process", "confirming the protease origin of AFGP".¹⁵ In the blackfin icefish, 11 antifreeze genes sit beside 10 trypsinogen genes and two trypsinogen-like protease genes.¹⁶
• The chimeras are not clearly intermediates. Iliana Bista and colleagues reported 24 new notothenioid genomes in 2023. They list as open questions "why copies of the chimeric afgp/tlp genes, proposed to be evolutionary intermediates, persist in extant genomes of the cryonotothenioids, the origin of the three-residue linker sequence in the AFGP polyprotein, and finally, the mechanism of expansion of afgp copies."⁸ The one chimera they found in the South Georgia icefish, Pseudochaenichthys georgianus, is broken: it lacks the signal-sequence exon and would encode "a single long run of 722 Thr-Ala-Ala tripeptide repeats".⁸
• The copy numbers are large and vary. Bista and colleagues count 14 antifreeze genes plus three chimeras in the toothfish, and 15 in P. georgianus, of which eight look intact.⁸ A 2025 preprint counts 17 genes in Notothenia coriiceps, which together "can generate a total of 612 AFGP molecules".¹⁷ Earlier counts of four to eight came from short-read assemblies that collapse repeats, and are minimums.⁸
• The family that never froze never had it. The Patagonian robalo, Eleginops maclovinus, a notothenioid that lives outside the Antarctic, "never had afgp".⁶

When it happened

The 1997 paper dated the origin by the differences between the antifreeze and trypsinogen genes:  "about 5–14 million years ago".²  It had no nuclear mutation rate for fish and used a rate from salmon mitochondrial DNA, which Logsdon and Doolittle called "probably an overestimate".¹³  The authors were confident anyway.  The agreement of their date with the freezing of the Antarctic Ocean and with the fishes' family tree, they wrote, "argue strongly that this time frame could not be due to mere coincidence, but is, in fact, reliable."²

Later work did not bear it out.  In 2012 Thomas Near and colleagues placed the origin on a fossil-calibrated family tree of 83 notothenioid species, from the presence or absence of the protein in 47 of them, and found that "the origin of AFGP occurred between 42 and 22 Ma".⁷  Its midpoint, 32 million years, is more than three times the midpoint of the 1997 range.  A 2018 whole-genome study put it nearer 13 to 20 million years.⁶  All three are inferences:  one from a clock on the gene, two from where the trait falls on a dated tree.

A clock on these genes has a deeper problem.  The trypsinogen gene compared today is a relative of the one that was converted, not the gene itself.  Their differences date the split between two gene copies, which may be older than the conversion, and gene copies lying side by side can exchange stretches of sequence, which makes them look more alike, and so younger, than their age.  At the mutation rate measured directly in families of nine-spined sticklebacks, 4.83 × 10⁻⁹ per letter per generation,¹⁸ the 5.5% difference counted here takes about 5.7 million generations to build up.  That fits Near's date if a notothenioid generation lasts four to seven years.⁴

Part 2: The northern cod gene, built from DNA that coded for nothing

Why the cod gene had to be a separate invention

The 1997 team also sequenced an antifreeze gene from the Arctic polar cod, Boreogadus saida.  Its protein is the same chain of threonine-alanine-alanine units, with an occasional arginine in place of a threonine.  Its gene is not related to the Antarctic one.  The cod gene "share[s] no sequence identity with the trypsinogen gene", has a different arrangement of introns, different spacers, and spells the same amino acids with different codons.³  The authors concluded that it "is unrelated to notothenioid AFGP gene or trypsinogen gene, and very likely arose from a different genomic locus."³  Cod and notothenioids belong to different superorders, and fossils of both from the same 40-million-year-old rocks in Antarctica show they had separated long before either needed antifreeze.³  For two decades nobody knew where the cod gene came from.

The 2019 reconstruction

Xuan Zhuang, Chun Yang, Katherine Murphy and Christina Cheng built clone libraries from seven cod species, chosen for where they sit on the family tree, and compared the antifreeze gene with whatever matched it in the species that lack antifreeze.⁵

The gene counts are from the full loci published in 2021;¹⁹ the 2019 paper gives the same numbers as working and broken genes.⁵  A working gene has three exons.  The first two and the first two letters of the third code the signal sequence, a short piece is cut off afterwards, and the rest codes the antifreeze chain, from about 22 to more than 550 repeat units long depending on the gene.⁵

The deeper the branch, the less the matching DNA looks like a gene.  Zhuang and colleagues read that series as the gene being assembled step by step, and propose this order:⁵

1. The starting material, in the common ancestor of the family: a stretch of about 240 letters of non-coding DNA that, if spliced, would encode a signal sequence, next to a 27-letter stretch rich in GCA. The authors call the signal-sequence part "fortuitous noncoding DNA sequence".⁵
2. The 27 letters duplicated into four copies. This kind of DNA copies itself readily, and the deep-branching cusk and burbot carry expanded runs of it, so it was there before the antifreeze.
3. One letter changed. Inside the repeats are several nine-letter stretches that read proline-alanine-alanine or alanine-alanine-alanine. "A chance 1-nt substitution (C → A or G → A) in the first position could give rise to the incipient three codons for the Thr-Ala-Ala tripeptide unit of AFGP".⁵ Copying then expanded the unit, still without a promoter. The Norway pout's threonine-proline-alanine repeat is taken as this stage.
4. A promoter arrived "through a stochastic translocation of the ancestral AFGP founder region to a new genomic site that happened to contain a TATA motif".⁵
5. One letter was deleted. Working genes lack a T that the repeat consensus, the Norway pout and the whiting all carry. Losing it shifted the reading frame so that the signal sequence, the short piece cut off later, and the repeats became one continuous reading frame.⁵
6. The repeats and then the whole gene multiplied under selection. The smallest working antifreeze, AFGP8, "comprises only four tripeptide repeats", reachable "through only two tandem duplications".⁵

The authors sum up the coding part as needing "only a single nucleotide change in the ancestral 9-nt GCA(Ala)-rich element".⁵  That sentence is about the coding stretch.  The whole gene also needed the deletion of step 5 and the promoter of step 4.

What was tested, and what "experimentally proved" means

The abstract says:  "We experimentally proved that all genic components of the extant gadid AFGP originated from entirely nongenic DNA."⁵  The experiments are two.  RNA blots of pancreas, where the antifreeze is made, found that "No transcripts of AFGP homologs were detectable in any of the three AFGP-lacking gadids", while the polar cod lane lit up strongly.⁵  And when the polar cod and tomcod clone libraries were probed with the signal-sequence and promoter region, the only clones that lit up were the antifreeze clones themselves, so no protein-coding gene elsewhere in the genome supplied those parts.⁵

Those are real tests, and the reconstruction could have failed them.  They test its predictions about the living fish.  They do not show the history.  The steps, their order and their timing remain inferences from comparing living species, which is why Observed, or Only Inferred? grades the case "reconstructed".  Neither paper made the proposed ancestral sequence or tested whether its product did anything.

The second team

Helle Baalsrud and colleagues in Oslo reached the same conclusion in 2018 by a different method:  whole-genome searches across 27 cod-family genomes and mapping of the gene's neighbours.⁶  BLAST searches with very loose settings found no gene or open reading frame in any cod genome resembling any part of the antifreeze gene.⁶  The antifreeze locus in Atlantic cod lies between the genes mak16/rab14 and dmtn/rsph14, whose order is conserved in stickleback, tilapia and pufferfish, and in those fishes the same place holds nothing like the antifreeze gene:  "We found no conserved non-coding elements or afgp-like sequences in the genomic region syntenic to afgp".⁶  Both papers read the whiting's broken gene as a loss, and Baalsrud's team adds the haddock.⁶

The hard step, and the seams

The step nobody can account for in detail is the promoter.  The three deepest species share an upstream sequence with each other that differs from the antifreeze species', and the cusk's matching region and the polar cod's antifreeze locus have "a complete lack of microsynteny":  none of the neighbouring genes are shared.⁵  The proposed translocation has not been found, and the authors say so:  "Although the specific mechanism of the translocation is currently unclear".⁵  One clue is in the second paper.  Sequences resembling the antifreeze gene's leader sit at 376 other places in the Atlantic cod genome,⁶ so copies of the leader and promoter region have spread through the genome before.  Neither paper connects that to the proposed translocation.

Four other seams are worth stating.

• The whiting. It nests inside the antifreeze group, its broken gene has the promoter of step 4, and it keeps the T that step 5 removed.⁵ If the deletion happened once in the common ancestor of the antifreeze species, the whiting copy should lack the T. It does not. So either the deletion happened more than once, or the ancestor carried both deleted and undeleted copies. Two broken polar cod genes do carry reading-frame errors upstream of the repeats,¹⁹ which may fit the second reading, but nobody has worked the question through.
• The date. Zhuang and colleagues give no clock date. They write that the gene "presumably evolved very recently, in response to the cyclic northern hemisphere glaciation that commenced in the late Pliocene about 3 Mya."⁵ Baalsrud's team put the origin at 13 to 18 million years ago by placing it on a dated family tree, and preferred that window partly because an origin "more than 10 Ma before the onset of freezing temperatures in the Northern Hemisphere is a less plausible scenario."⁶ In both papers the climate helps choose the date.
• The deepest relatives. Zhuang's team pulled a matching sequence out of cusk and burbot clone libraries. Baalsrud's team, searching short-read genome drafts, reported "no afgp-like sequences in Brosme brosme or the 20 codfish genomes investigated outside B. brosme in the phylogeny".⁶ A sequenced clone is direct evidence, while a failed search of a draft genome can be an assembly gap, so the clone result is the stronger. The two papers do not discuss each other's findings.
• The starting material. The non-coding ancestor already held a sequence that, once spliced, made a typical signal sequence. That is less of a coincidence than it sounds. In 1987 Chris Kaiser and colleagues replaced the signal sequence of a yeast enzyme with essentially random sequences and found that "about one-fifth of these sequences can function as an export signal".²⁰ Working splice sites in non-coding DNA are not priced here.

Part 3: The other fish antifreezes, and the pattern they share

The two glycoproteins are not the only fish antifreezes.  Three other families, called types I, II and III, do the same job with different proteins, in other fish.  Their origins show what kind of novelty this job calls for.

Type III is the clearest case of an old protein retrained.  A team led by Liangbiao Chen in Beijing and Christina Cheng in Illinois found that in an Antarctic eelpout an enzyme gene, "having both sialic acid synthase and rudimentary ice-binding activities, became duplicated. In one duplicate, the N-terminal SAS domain was deleted and replaced with a nascent signal peptide".⁹  They tested the living fish's enzyme:  its end domain, made on its own in the laboratory, bound ice about as well as the whole enzyme.⁹  Even the new signal sequence was not new:  "The capacity to code for a functional signal peptide (SP), in fact, existed in the precursor sequence".⁹  Their summary is that the antifreeze arose by "tapping into two inconspicuous functionalities in the same cytoplasmic ancestor".⁹  The type III family is put at about 18 million years old.²¹

Type II is an existing fold, a C-type lectin, the kind of domain that binds sugars.  Its odder story is how it spread.  The type II genes of herring and smelt "share up to 85% identity" in protein, have introns up to 97% identical, and occur in no other fish tested, although herring and smelt belong to distantly related groups.²²  Laurie Graham, Peter Davies and colleagues at Queen's University in Canada concluded that the gene jumped between fish lineages.  In smelt it sits alone in a stretch whose matching region in five other fish genomes holds no trace of it:  "Not even so much as a pseudogene was found".²³  An earlier study read the same scattered distribution as an old duplication of a lectin gene followed by losses in most lineages.²⁴  Either way, nothing new was folded.

Type I is a single alanine-rich helix, and it arose at least four times.  Graham's group showed in 2013 that the flounder, sculpin, cunner and snailfish proteins are not related at the DNA level.²⁵  Later work found the source in three of the four, and each is a different gene:

• Flounder: an antiviral gene, Gig2. The antifreeze "arose, in situ, at the original Gig2 locus via gene duplication and divergence". Its coding sequence "bears little similarity to the Gig2 progenitor", but a short stretch of about 35 letters, coding ten amino acids of helix, "was likely retained and amplified".²⁶ In one starry flounder the locus had two versions, one encoding 33 antifreeze proteins and the other only four.²⁶
• Sculpin: lunapark, a membrane protein of the endoplasmic reticulum. After duplication and "loss of all but three of the 15 exons", the last exon "was converted to an alanine-rich sequence by a combination of frameshifting and mutation".²⁷
• Cunner: a GIMAP gene, whose antifreeze coding region "stems from an alanine-rich region flanking the GTPase domain".²⁸
• Snailfish: no source gene found. The locus "does not share any similarity to functional genes" and is "replete with repetitive DNAs and transposons".²⁸

Nathan Rives, Vinita Lamba, Christina Cheng and Xuan Zhuang compared three of these lineages in 2024 and found the same arrangement each time:  the new gene keeps the old gene's untranslated regions, introns and flanking DNA, while "the essential ice-binding motif with the adaptive novel function — 11-aa Ala-rich repeats with evenly spaced Thr — is exclusively present in the new gene AFPI, not in any precursor genes."¹⁰

The pattern

Put the cases together and a pattern is plain.

1. No fish antifreeze whose origin is known is a new folded domain built from scratch. Where the protein has a real fold (types II and III), the fold is old, and in type III the ice binding was already there in rudimentary form.
2. Where the coding sequence is new (both glycoproteins and type I), it is a low-complexity repeat or a single helix: a short motif, copied.
3. The new genes reuse old frameworks. The Antarctic glycoprotein kept trypsinogen's ends. Type I kept its precursors' control regions and introns. Only the cod glycoprotein built its framework from non-coding DNA, and even there the signal sequence region and the repeats were present before the gene was.
4. The job is easy to reach. At least eight independent origins in fish, from at least seven different starting points, converged on a handful of products, twice on near-identical ones. Graham and Davies call the near-identity of flounder and sculpin type I proteins evidence of "strong convergent selection at the level of protein sequence".²⁷ That is selection finding a few simple solutions over and over.

Behe described the Antarctic protein in 2007 as "floppy and unfolded" and binding no partner proteins.¹  The Oslo team found the same for the cod protein:  its sequences score as highly disordered, with a mean predicted disorder of 0.68 against 0.36 for the average cod gene.⁶  Both sides describe the products the same way.

Part 4: What arose, counted and priced

How many changes

Counting the discrete mutation events each reconstruction needs, before the repeats and genes were multiplied:

Then come the copies.  Four repeat units, the smallest working antifreeze, take two duplications.⁵  The 170 units of the toothfish gene Dm1A take at least eight rounds if every round doubles the array, and more if they came one at a time.  A cod gene of more than 550 units takes at least ten.  The 16 genes of the polar cod take at least four rounds of whole-gene doubling.⁴

Every event on this list is of a kind that happens routinely:  single-letter substitutions and deletions, slippage in repeats, and gene duplication.  Behe makes the same point:  the Antarctic scenario is "convincing because each of the steps is tiny", and "each step is an improvement".¹  No step needs two specific changes at once.  The steps that are hard to account for are the ones whose mechanism is unknown, not the ones that look improbable:  the cod promoter and the Antarctic spacer.

The cod reconstruction adds one twist.  On its own account, the repeat duplications, the letter substitution and the first copying of the unit all happened before the DNA was transcribed, so selection could not have favoured them.  They had to arise and persist by chance, as neutral changes.  The Norway pout's untranscribed threonine-proline-alanine repeat shows that such a stretch can sit in a genome without being used.  And the gene could do nothing until both the promoter and the one-letter deletion were in place, so whichever came first also had to wait without help from selection.  Behe's "each step is an improvement" fits the Antarctic story better than the cod one.

How much new information

"Information" can be measured in several ways, and Measuring Biological Information discusses which are sound.  Every measure tried here gives the same answer for these genes:  very little is new.

• Distinct pieces. The 2,124-letter coding region of the toothfish gene Dm1A is 170 copies of the nine-letter unit and 39 copies of a nine-letter spacer. Among the 170 units only eleven different spellings occur, and three of them account for 129 copies. The second toothfish gene, Dm3L, uses seven spellings. The polar cod gene Bs3-1 is more varied, with 26.⁴
• Compressibility. A general-purpose compressor (deflate, the method inside zip files) shrinks the Dm1A coding region to 0.76 bits per letter, against 2.71 for the same letters shuffled. Ordinary coding DNA barely compresses: the trypsinogen coding sequence beside it comes out at 2.83 bits per letter against 2.91 shuffled. By this crude measure the 2,124 new letters carry less description than the 750 letters of the enzyme gene.⁴
• Tolerance. The working antifreezes range from 4 to at least 55 repeat units, tolerate proline for alanine and arginine for threonine, and are cut into pieces of many sizes.³ A protein that works across that range of sequences is a target that is easy to hit.

On the other side of the ledger, the gene as a whole is new:  a secreted product, made in quantity in the pancreas, cut into pieces, with a sugar unit on every threonine, doing a job the ancestor could not do.  The sugars take part in ice binding,³ and the gene codes only the chain;  the sugars are added by the cell.  None of the origin papers discusses how the new chain came to be decorated.

So both descriptions are true.  These are new genes, and the new sequence in them amounts to a few letters, copied.

Observed, or reconstructed

Nothing in either origin was watched.  The living fish, their genes, their proteins and the proteins' ability to stop ice growing are observed.  The precursor states, the order of steps and the dates are inferred from comparing living species.  This site's audit of the evidence for evolution grades both antifreeze origins "reconstructed" and concedes the cod gene as a real gain on that basis.  See Observed, or Only Inferred? for the grading and the case for treating strong reconstructions as evidence.

The difference from trypsinogen on a young-earth clock

No young-earth source found for this article proposes a date or a mechanism for either gene.  Answers in Genesis tells children that "God made some animals with special antifreeze proteins in their blood".²⁹  The young-earth model does say when polar seas froze.  The Creation Answers Book reports Michael Oard's estimate that "it would have taken only about 700 years to cool the polar oceans from a uniform temperature of 30°C at the end of the Flood to the temperatures observed today".³⁰  On that model the fish of the Southern Ocean needed no antifreeze until some centuries after the Flood, roughly 4,400 years ago on the usual young-earth chronology.

That leaves two options, and each has a cost.

1. The genes were created as they are. Then the Antarctic gene was made with a trypsinogen gene's front end, its intron, and its last exon read in a shifted frame as an untranslated tail, and the cod gene was placed in relatives' genomes beside a graded series of non-coding precursors. That is possible, but it is the pattern construction predicts, and no young-earth source has explained it.
2. The genes arose after the Flood. Then they arose in a few thousand years. The steps themselves are fast kinds of mutation. The difference between the Antarctic gene's inherited parts and today's trypsinogen gene is the problem: 29 substitutions in 523 compared letters. At the measured fish mutation rate, two copies diverging for 4,400 years would gain about 0.002 to 0.02 substitutions over those letters, with generations of one to ten years. The observed difference is 1,300 to 13,000 times larger.⁴ The way out is that the gene converted was not today's trypsinogen gene but a relative that already differed from it by about 5%. Copies that differ that much could have been created different. So this option survives, at the price of a created family of trypsinogen-like genes with differences that look like the products of time.

Mainstream dating has its own soft spot here, as Part 1 shows:  the same differences gave a date that later work tripled.

Part 5: The claims on both sides, tested

The origin conceded, the product called debris

The main design and young-earth writers who have looked at these genes accept that they arose by mutation and selection, and argue that the result is too simple to matter.

• Michael Behe, 2004: "Antibiotic and pesticide resistance, antifreeze proteins in fish and plants, and more may indeed be explained by a Darwinian mechanism. The critical claim of ID is not that natural selection doesn't explain anything, but that it doesn't explain everything."³¹
• Behe, 2007: the antifreeze is "not so much a molecular machine as it is a blood additive", and it "likely marks the far border of what we can expect of random mutation in vertebrates."¹
• Shaun Doyle, Journal of Creation, 2011, writing on the type III protein: "Random mutation and natural selection is a likely mechanism for how this [antifreeze protein] and many others were produced." But "Being an AFP is a very non-specific job that many different random proteins could perform", and antifreezes are "more akin to beneficial debris than a new complex and specified protein."¹¹
• David Berlinski and Tyler Hampton, 2013: "Antifreeze proteins are not specific; they contain no robust secondary structure; they are poor in information and highly repetitive, surviving on a simple Thr-Ala-Ala repeat; and they do not interact with other proteins."³²
• Casey Luskin, 2014: an antifreeze protein "acts more like 'debris' in the cytoplasm to prevent freezing rather than performing a specific and complex function. This is not likely a viable model for explaining the origin of typical proteins."³³

Tested.  The description is accurate, and this article's counts support it.  The new coding sequence is a repeat of a few spellings of nine letters, it compresses far more than ordinary coding DNA, and the products are disordered and bind no partners.  Every fish antifreeze whose coding sequence is new is a repeat or a single helix, and every one with a real fold reused an old one.  Two parts of the case are weaker than stated.  The job is not "non-specific" in the chemical sense:  the glycoproteins "adsorb to very specific faces (the prism planes)" of ice crystals.³  What the many origins show is that many different starting points can reach that specific job.  And "the far border" is a forecast, not a finding.  These cases show what selection built under an extreme, simple pressure;  they cannot show that it could build nothing harder.  That question is argued in How Rare Is a Working Protein? and Orphan Genes.  Behe's "ten million years" was the 1997 date.¹  On the 2012 date the protein has had two to four times longer and is still a simple repeat, which helps his argument.

Luskin's "debris in the cytoplasm" is wrong on one point:  the antifreezes are secreted into the blood, not left in the cell.

The reconstruction called storytelling

Writers who reject the origin attack the method.

• Ross Anderson, a biochemist writing on Creation-Evolution Headlines, 2019, on the cod paper: "one must bear in mind that all is hypothetical as there is no real way to experimentally test the conclusions of such hypothesizing. In short, it amounts to just storytelling."¹² He counts "six fortuitous steps".
• Cornelius Hunter, Evolution News, 2019, on the same paper's claim that the repeats came from duplicating a nine-letter element: "There is no scientific evidence for it. Zero."³⁴
• Luskin, 2011, on the type III protein: "The paper doesn't address any questions relating to the plausibility of its story". Taking a textbook rate of "about 0.01 duplication per gene per million years" for the eelpout's more than 30 gene copies, he concluded: "It should have taken some 3 billion years just to accomplish the last step of this little story".³⁵

Tested.  The reconstruction was tested, and could have failed.  It predicted that the precursor in relatives would not be transcribed, and the RNA blots found no transcript.  It predicted that no other gene in the genome supplies the signal sequence and promoter, and the clone screens found none.  It predicted a series of precursor states in relatives, and the series is there, ending in fish outside the cods whose matching chromosome region holds nothing.  A second team using whole genomes reached the same verdict.  Hunter's question, what evidence shows that repeats grow by duplication, has an observed answer:  repeat expansion is watched in human families, and Behe himself points out that such "stuttering is the kind of mutation that leads to Huntington's disease".¹  Anderson's "no real way to experimentally test" and Hunter's "Zero" are wrong.

The narrower complaint is right.  Nobody has made the proposed precursor and shown what it does, the promoter's arrival is unexplained, the order of steps is inferred, and the two cod papers disagree on the date by a factor of four to six.  Those are reasons to call the history reconstructed, which this article does.  They are not reasons to call it fiction.

Luskin's arithmetic applies a genome-wide average for single genes to tandem arrays, which gain and lose copies by unequal crossing-over, and it adds the copies one at a time from one gene, when every new copy can itself be copied.  The measured arrays vary within single species, which an average of one duplication per gene per hundred million years cannot allow:  the two versions of the starry flounder's antifreeze locus carry 33 and 4 genes,²⁶ and the two toothfish versions of the glycoprotein locus differ in length by about 100,000 letters because of different antifreeze copy numbers.¹⁴

The genes said to be present all along

Jerry Bergman, 2021:  "This new study indicates that what had been called “new genes,” supposedly created by random mutations and favored by natural selection evolution (such as the gene in fish that makes a novel antifreeze), were there all the time. The existing genes were activated when the conditions were appropriate!"³⁶

Tested.  The study Bergman cites is about fruit-fly genes and heterochromatin, not fish.  For the fish it fails on the data.  In the cod relatives that lack antifreeze, the matching DNA is not a silenced copy of the gene, except in the whiting, which carries a broken one.  The deepest relatives have no threonine-alanine-alanine coding stretch at all, the three basal species have no promoter, and in them the region sits at a different place in the genome.⁵  In fish outside the cods, the place where the cod gene sits holds nothing like it.⁶  Among the notothenioids, the Patagonian robalo "never had afgp".⁶

The fish without antifreeze said to have lost it

Some young-earth writers hold that mutation and selection "only lead to a loss of functional systems".³⁷  No published source found here applies that to the cod gene, but it is the obvious reading for a writer who holds it:  the gene was there first, and the fish without it lost it.

Tested.  Losses did happen.  Both cod teams read the whiting's broken gene as a loss, and Baalsrud's team adds the haddock.  But a loss story cannot produce the deeper relatives.  To turn a created gene into what the cusk and burbot carry, the coding repeats would have to be deleted, the promoter and leader lost, and the remaining stretch moved to a new place in the genome, while in the Norway pout lineage, and again in the whiting, a T would have to be inserted at exactly the place, and with exactly the letter, that restores the consensus of the surrounding repeat.⁵  The whiting's T is awkward for the gain story too, as Part 2 notes, but the pout's is not.  That is as many events as the gain story, run backwards, and they would have to land on exactly the intermediate states that the gain story predicts.

Genomes said to be designed to change

Robert Carter of Creation Ministries International writes:  "It looks like God engineered into life the ability to change DNA", and creationists are "also stating that some of these changes are controlled by genetic algorithms built into the genomes themselves."³⁸  No creationist source found here applies this to antifreeze genes, but it is the natural design reading:  stretches of repeat-prone DNA, built in so that fish could adapt to cold.

Tested.  The reading fits the data, because it can absorb any data.  It predicts nothing that the evolutionary reconstruction does not, and it does not explain the details:  why the Antarctic gene's tail is a trypsinogen exon read in a shifted frame, why its founder straddles an intron boundary, or why the cod precursor states line up along the family tree.

The evolution side's overstatements

• "From junk." The University of Illinois announced the 2019 paper as "Study of Arctic fishes reveals the birth of a gene – from ‘junk’", saying the fish built the gene "from tiny fragments of noncoding DNA, regions once considered 'junk DNA.'"³⁹ A 2006 release had already said the cod gene "has evolved from non-coding or 'junk' DNA".⁴⁰ Jerry Coyne wrote that the protein was "assembled out of scraps of genome that have no function at all."⁴¹ The paper says non-coding, not junk and not functionless. The precursor is untranscribed, and nobody has tested whether it does anything else. See Junk DNA and Genome Function for why "non-coding" and "junk" are different claims.
• "Experimentally proved." As Part 2 shows, the experiments tested the reconstruction's predictions about living fish.
• "Fully resolved." Zhuang and colleagues wrote: "With this paper, we fully resolved the lasting question of how two unrelated groups of fish at opposite poles … invented a near-identical AFGP."⁵ The same paper calls the promoter's arrival "putative" and its mechanism "currently unclear".
• "Reliable" and "confirming." The 1997 date was called "reliable" and was later revised to about three times older.² The chimeric genes were said in 1999 to be "confirming the protease origin"; the 2023 genome team lists the derivation as "not yet been fully resolved".⁸ The trypsinogen origin is well supported by the matching ends. The chimeras are not the proof they were said to be.
• "Whole cloth." The 1997 National Science Foundation release said the Antarctic gene evolved by "arising 'whole cloth' from trypsinogen".⁴² The paper says the opposite for the part that matters: the new protein came from "recruitment of segments of an existing protein gene plus de novo amplification of a short DNA sequence".²

Objections and responses

"A reconstruction from living fish is only an inference, so it proves nothing."  It is an inference, and this article grades it so.  But it is an inference that made risky predictions and passed them:  no transcript in relatives, no other source for the signal sequence and promoter, a graded series of precursor states along the family tree, and an empty site in fish outside the cods.  Calling it "storytelling" ignores those tests.  Calling it "observed" ignores that nobody saw it.

"The 'non-coding' precursor may have had a function, so this is not a gene from junk."  Correct, and the article does not say junk.  The papers say non-coding and untranscribed.  Whether the precursor DNA did anything else is untested.  That bears on the press headlines, not on whether the gene is new:  a protein-coding gene now exists where no protein-coding gene existed.

"Both fish have the same protein because a designer reused a good design."  The proteins are near-identical, and the genes that make them are built in different ways, from different raw material, with different codons and different cleavage sites.³  A designer reusing a protein design would not need to build two unrelated genes for it, though nothing forbids it.  Convergent Evolution and Design Reuse argues the general case;  for antifreeze, the reuse reading predicts nothing the separate-origins reading does not, and the separate-origins reading predicts the differing gene structures.

"Notothenioids have a trypsinogen-like gene because both genes were created with a shared module."  Then the shared module includes an intron, a stop codon created by a frame shift, and the coding sequence for an enzyme's last 50 amino acids used as an untranslated tail.  A designer could do that.  It is what descent from an enzyme gene predicts, and the design reading has to accept it as an unexplained choice.

"If antifreeze arose so easily, why do so few fish have it, and why so late?"  Because it is only useful in ice-laden water.  Fish that left the cold lost or degraded it:  the whiting and haddock among cods,⁶ Patagonotothen guntheri, which "has completely lost afgp",⁶ and a New Zealand notothenioid whose one intact-looking gene carries changes "that likely render the proteins inactive".¹⁷  The published dates are disputed, and each team ties its date to the cooling of the ocean concerned, which is partly how the date was chosen.

"Behe concedes too much."  Behe concedes the origin of a simple protein and argues that its simplicity marks a limit.  That position is consistent with the data here.  What the data do not settle is whether the limit is real or only where these particular cases stop.  A young-earth reader who wants to deny the origin itself has the harder task:  explaining the cod precursor series and the trypsinogen ends, which no young-earth source has attempted.

"Copying nine letters a hundred times is not new information."  By compression, distinct spellings, or tolerance of variation, the new coding sequence carries very little.  But the gene is new, it has a new function, and the function decides whether a fish lives in ice.  If "information" is defined so that this counts for nothing, the definition should say so in advance.  Measuring Biological Information takes up which definitions are sound.

"The dates keep moving, so the whole story is unreliable."  The dates moved because the first one used the wrong clock.  The sequence evidence for the origins did not move.  The date and the origin are separate claims, and only the date has been revised.

Issues and unanswered questions

1. The whiting's undeleted T. The whiting sits inside the antifreeze group and its gene has the promoter, yet it keeps the letter whose loss made the gene readable. Either the deletion happened more than once or the ancestor held both kinds of copy. The cod papers do not address it.
2. The two cod papers disagree on the date (about 3 against 13 to 18 million years) and on whether the deepest relatives carry the precursor (found in clone libraries, not found in draft genomes). Neither paper discusses the other's results.
3. The promoter's source is unknown. The proposed translocation has not been found. The 376 copies of leader-like sequence in the Atlantic cod genome suggest a family of mobile or copied sequences, but nobody has traced the antifreeze promoter to a parent copy.
4. The Antarctic spacer's origin is unexplained in 1997 and still listed as open in 2023.
5. The chimeric Antarctic genes were called intermediates in 1999 and are now treated as copies that persist or decay. Whether any was ever a step on the path is unsettled.
6. No ancestral sequence has been resurrected. The proposed cod precursor, or the proto-gene stage with its untranscribed repeat, has never been expressed and tested. That experiment is possible and would turn part of the reconstruction into a test.
7. Glycosylation. The ice-binding surface involves the sugars on each threonine. None of the origin papers discusses how the cell's sugar-adding enzymes came to decorate the new chain.
8. The young-earth account is missing. No young-earth source found here offers a mechanism or a date for either gene, or an explanation of the cod precursor series. Oard's 700-year cooling of the polar oceans implies that the fish needed antifreeze only after the Flood. If the genes arose then, the young-earth model needs the progenitor of the Antarctic gene to have already differed from today's trypsinogen gene by about 5%.
9. The type I sculpin origin is described as a frameshift within a duplicated gene in one 2024 paper and as "partial de novo" in another. Only the abstract of the first was read.

Verdict

The fish antifreeze genes are the best-documented cases of genes that did not exist before.  The Antarctic gene was rebuilt from the ends of a trypsinogen gene around a new repeated core grown from nine letters.  The northern cod gene was assembled from DNA that coded for nothing, and the stages of its assembly can still be seen in its relatives.  Both conclusions rest on comparison of living fish, not on observation.  They have passed every test applied to them, and no young-earth account explains the evidence they rest on.

What arose is also limited.  Each gene took a handful of ordinary mutations, followed by copying.  The new sequence amounts to a few letters, repeated, coding a floppy protein that sticks to ice.  Every other fish antifreeze tells the same story:  old folds retrained, or simple repeats grown.  These cases show that mutation and selection can make a new gene with a new, life-saving function when the function is simple and the pressure is extreme.  They do not show that the same process makes folded enzymes or machines, and their own authors do not claim it.  The design writers who concede the origin and contest its scope have read the evidence correctly.  The writers who call it storytelling, or a switched-on old gene, have not.

Sources