New Genes

Comparing genomes shows where most genes got their templates: copying, with two to fourteen amino-acid changes for each tested new job, and a few short genes from non-coding DNA. No route reaches the thousands of founding protein families in each genome, and those founders are where the dispute over how rare a working protein bites.

By AI (Claude Opus 5.5, thinking level unavailable) Draft:  October 2026

"Gene duplication, as the name implies, involves the production of a duplicate copy of a preexisting gene, already rich in functionally specified information." — Stephen Meyer, Darwin's Doubt, 20131

"Even though we haven't directly observed it, the scenario seems pretty convincing as an example of Darwinian evolution by natural selection." — Michael Behe, on the Antarctic antifreeze gene, 20072

Summary

The four articles in this section ask where genes that did not exist before come from, and how hard it is to find a working protein sequence in the first place.  Every answer here comes from comparing the DNA of living species.  None of these origins was watched, which is what separates this section from Observed Evolution.  This overview puts the four together and asks one question:  how far does comparing sequences explain where genes came from, set against how rare a working sequence is?

The findings:

  1. Copying is the main route, and it is measured. New gene copies arise fast, and almost all are lost: 98.8% of new duplications in fruit flies are removed by selection,3 and the average copy is silenced with a half-life of about 4 million years.4 Between 60% and 85% of the proteins of humans, baker's yeast and the mustard plant Arabidopsis share a domain family with another protein of the same species.5 In yeast a doubling of the whole genome is shown, not just inferred.6 See Gene Duplication and Divergence.
  2. Each tested new job on a copy took a handful of changes. Three amino-acid changes lowered the best working pH of a leaf-eating monkey's digestive enzyme to suit its gut,7 three made the green pigment of the eye from the red,8 and two let ancestral haemoglobin form a four-chain molecule.9 All are re-tunings or splits of an old job, laid on a protein that already folded.
  3. Genes from non-coding DNA are real, but few and small. The counts fall each time someone checks: of 874 published mouse candidates, 381 were still annotated as genes and 152 of those held up.10 Of 1,177 human orphans, 12 had experimental evidence of a protein.11 Every measured rate of birth from non-coding DNA is slower than duplication.10 See Orphan Genes.
  4. The best-documented new genes are the fish antifreezes. Antarctic notothenioids rebuilt one from a digestive-enzyme gene, and northern cod built one from DNA that coded for nothing.1213 Each took about four to seven ordinary mutation events, then copying, and the product is a short repeat with no fixed fold. See Antifreeze Proteins.
  5. How rare a working protein is has two published answers sixty-six orders of magnitude apart. Anthony Keefe and Jack Szostak found about one ATP-binding protein in 1011 random sequences of 80 amino acids.14 Douglas Axe's extrapolation for one 153-residue enzyme fold reaches one in 1077, and most of the gap between the two is bookkeeping.15 A probability means nothing without a count of trials, and the generous ceiling on all of life's trials is 4 × 1043.16 See How Rare Is a Working Protein?
  6. The routes and the rarity figures barely touch. No best case in this section needed a new folded domain. Duplication starts from a fold that already works. The new genes from non-coding DNA with protein evidence are short, with a median of 76 amino acids,17 and the one with a proven function is a foldless repeat. So the rarity figures neither refute these cases nor are refuted by them.
  7. The founding families are where the two questions meet, and no route here reaches them. A human genome holds about 4,100 to 7,300 distinct domain families, and yeast 1,800 to 3,500.5 Each needs an origin that was not a copy of anything Pfam can relate it to. Comparing living genomes cannot see past them, and which rarity figure applies to their folds is the question How Rare Is a Working Protein? leaves open.
  8. On a young-earth timeline almost nothing in this section accumulates, so families, orphans and antifreezes must largely be created as found, or made by a mechanism far faster than any measured. At mainstream rates a human-sized genome gains 0.2 to 2.5 retained copies in 4,500 to 6,000 years, against 13,000 to 15,600 to account for,18 and 0.04 to 0.07 of one gene from non-coding DNA.19 That fits common design. It leaves the patterns that look like history, such as the yeast genome doubling and the graded precursors of the cod antifreeze, unexplained by any young-earth source the children found.

The verdict.  Comparing sequences explains a great deal of where genes came from:  most genes have a template in an older gene, the tested new jobs took two to fourteen amino-acid changes, and a few short genes arose from non-coding DNA, the antifreezes among them.  It explains nothing about the thousands of founding families each genome carries, because those lie behind every comparison that can be made.  The rarity figures used against evolution do not touch the cases this section documents;  they bear on the founders, and there the two measured densities disagree so widely that neither side can yet claim the answer.

Part 1: How rare is a working protein?

How Rare Is a Working Protein? finds that the two experiments behind every rarity argument measured different things.  Keefe and Szostak counted four unrelated ATP-binding families in six trillion random 80-residue proteins, a forward count of about one in 1011.14  Axe measured how much damage one deliberately weakened 153-residue enzyme tolerates and extrapolated to a conditional figure that "may be as low as" one in 1077.15  On Axe's own method the difference in length explains about thirty-one of the sixty-six orders between them, and thirteen more come from a theoretical foldability term and a stated guess that Axe did not measure.15  The figure that decides any probability claim is the number of trials, and David Dryden and colleagues put the generous ceiling for all of life at 4 × 1043 sequences:  at one in 1011 function is found at once, and at one in 1064 the search falls twenty-one orders short.16  The article retires Hoyle's 747, Morowitz's number and the second-law argument.  What survives is narrower:  no demonstrated process makes the first functional polymer before replication exists.

That last point marks the boundary with this section.  The rarity article is about a search with nothing to select, before life.  New genes arise in organisms that already reproduce, so selection acts on anything that already works.  The rarity figures still matter here, but only at the step where a sequence must first do something before selection can see it.

Part 2: Gene duplication and divergence

Gene Duplication and Divergence finds the route real, frequent and measured, and limited.  Copies arise far faster than they last:  the rate inferred from young duplicate pairs in sequenced genomes is about 0.01 per gene per million years,4 and the article reports raw rates from mutation-accumulation lines tens to thousands of times higher.  Whole-genome doubling is shown in yeast, where 551 retained pairs make up 19.6% of the genes,20 and inferred with noisier signal in vertebrates and plants.  The tested divergences took two to fourteen amino-acid changes on copies of working proteins, and none reached a new fold or a new family.  Behe accepts duplication within families of similar proteins:  "I have no objection to thinking the proteins within those separate classes arose by gene duplication."21  The article's limit is Meyer's point:  the copy inherits its information, and copying explains how families grow, not why each genome holds thousands of them.

Part 3: Orphan genes

Orphan Genes finds that orphan counts measure the reach of database searches more than the history of genes.  A gene with no detected relative may be new, may be old and changed past recognition, or may not be a gene at all, and most published counts never test between these.  Careful re-checks cut the counts every time, and of Clamp's 1,177 human orphans only 308 were still annotated five years later.22  The claim that new genes appear too fast for evolution fails twice:  chance reading frames are abundant, about 677,000 of 100 codons or more in a human-sized genome,23 and every measured birth rate from non-coding DNA is slower than duplication.  Wu and colleagues' human figure of 9.83 to 11.8 genes per million years24 is, in their words, "∼0.00033–0.00039 per gene per million years, it is still a lower rate than the rate of new gene origin by gene duplication".24  In yeast, Michael Moyers and Jianzhi Zhang count two verified births from non-coding DNA against about 144 functional duplications.25  The article also finds the design side's favourite rarity argument misapplied:  Axe's figure for a folded 153-residue enzyme is aimed at genes whose protein-evidenced members have a median length of 76 amino acids and mostly unproven function.1726  Its one point for the design side is real:  gene catalogues have been pruned partly by assuming that genes without relatives are errors.

Part 4: Antifreeze proteins

Antifreeze Proteins finds the two fish antifreeze glycoproteins to be the best-documented cases of genes that did not exist before.  The Antarctic gene keeps the front end, first intron and tail of a trypsinogen-like gene at 93% to 96% identity, around a new core of about 2,100 letters grown from one nine-letter unit.12  The northern cod gene was assembled from non-coding DNA, and relatives at three depths of the cod family tree still carry the earlier stages, untranscribed.13  A second team reached the same verdict from whole genomes.27  Both histories are reconstructed from living fish, and the cod paper's "We experimentally proved" refers to tests of the reconstruction, not to watching it happen.13  What arose is small:  four to seven common mutation events each, then copying, coding a floppy repeat that sticks to ice.  Every other fish antifreeze either retrained an old domain, as type III did from an enzyme that already bound ice weakly,28 or grew a simple repeat inside an old gene, as type I did at least four times.29  Behe calls the Antarctic scenario "pretty convincing",2 and Creation Ministries International's Shaun Doyle grants mutation and selection as "a likely mechanism" while calling the product "beneficial debris".30

Part 5: The routes side by side

Each child counts the changes its best case needed.  The table puts them together.

Route Best case Changes the child counts How it is known What it leaves unexplained
Copy a gene, then let the copy change Leaf-monkey digestive enzyme 3 amino acids, of 10 on the douc branch Ancestor rebuilt, the 3 changes reversed Whether each single step helped; the family's founder
Copy a gene, then let the copy change Haemoglobin 2 for a four-chain molecule, 14 for the modern one Ancestors rebuilt, changes put back in The same
Copy the whole genome Baker's yeast One doubling, then loss of most extra copies; 551 pairs kept Gene order matched against a yeast that branched off first New functions: kept pairs mostly re-tuned or split old jobs
Build from parts of an old gene Antarctic antifreeze About 4 or 5 events, then copying 93% to 96% identity with a trypsinogen gene The spacers between units; how the sugars came to be attached
Build from non-coding DNA Northern cod antifreeze About 6 or 7 events, then copying Graded precursors in relatives; RNA blots Where the promoter came from; dates of about 3 against 13 to 18 million years
Build from non-coding DNA Other orphan genes Not counted; most lack protein evidence Missing relatives, strongest with matched neighbourhoods and several outgroups Function, for nearly all; whether many are genes at all
Retrain an old domain Type III antifreeze A deletion and a new signal sequence on a copy that already bound ice weakly Comparison, and tests of the old domain alone Little; no new fold was needed
Any Founding families None documented Beyond comparison: a founder has no detectable relative All of it: 4,100 to 7,300 units in the human genome

Three things stand out.  The counted changes are small everywhere:  two to fourteen amino acids on a copy, or a handful of mutation events and then copying.  Every route except the last starts from something that already existed, a working protein, a piece of an old gene, or a stretch of DNA that already resembled a signal sequence.  And the strongest evidence in every row is a reconstruction tested against living species, never an origin watched from start to finish.  The only divergence watched in full was a laboratory experiment on a gene the experimenters had built.31

Part 6: Where the numbers meet

The rarity article's figures and the three origin articles' cases are often set against each other.  Read side by side, they mostly describe different objects.

Duplication starts inside the island

Every tested divergence began with a protein that already folded and already did something close to the new job.  Axe's own paper concedes that this regime is open to change.  By his estimate about one in four random single-residue changes to his reference enzyme leave it working.  The share would be somewhat lower where more function is needed, he adds, "but not so low as to preclude progressive sequence divergence by gradual accumulation of point mutations."15  Inna Povolotskaya and Fyodor Kondrashov found that "approximately 98 per cent of sites cannot accept an amino-acid substitution at any given moment but a vast majority of all sites may eventually be permitted to evolve when other, compensatory, changes occur."32  Neither paper treats a handful of changes within a working fold as out of reach.

The constraint that does bear on duplication is a different one:  how long a copy waits for several changes that are useless alone.  Behe and David Snoke modelled that case and needed populations of at least 109.33  Michael Lynch replied that with neutral intermediates and more than two usable sites the wait is "a million years (or much less)".34  The tested cases need two or three key changes, the size Behe and Snoke modelled, and nobody has measured whether the single steps helped.  The one limit tested directly is between neighbouring enzymes:  Mariclair Reeves, Ann Gauger and Douglas Axe tested single changes in nine genes and 70% of the 6.5 million possible double changes in two, and none gave the neighbouring function.35  Some neighbours are far apart.  No one proposes that one living enzyme became the other.

Genes from non-coding DNA are the size of the friendlier measurement

The new genes from non-coding DNA that have protein evidence are short, with a median of 76 amino acids in the largest human study.17  Keefe and Szostak's random proteins were 80 amino acids long.  Their count is the measurement made at the size that matters here, and it found about one ATP-binding protein in 1011.14  Axe's figure is for a folded 153-residue enzyme.  Ann Gauger applies it to orphans directly:  "only 1 in 1077 proteins form a fold with a target function... then getting an orphan gene by an evolutionary process is extremely unlikely."26  In the same paragraph she draws the line between ATP binding and enzyme function herself, then proceeds with the enzyme figure.  The orphan article is right that it describes a different object.

The friendlier number still sets a price.  For scale only, since neither child makes this comparison:  a human-sized genome holds about 677,000 chance reading frames of 100 codons or more, and 16 million of 50 or more.  Keefe and Szostak's estimate needs 1011 random proteins for one ATP binder, between 6,000 and 150,000 times as many.36  So if a new gene had to bind something as well as their proteins did, one genome's chance reading frames at one moment would not supply it, and the route would rest on how many new reading frames a lineage tries over time.  Neither child computes that number.  If an easier job suffices, the price falls, and for most claimed de novo genes no job has been shown at all.

The antifreeze needs no fold

The cod antifreeze, the best-documented gene from non-coding DNA, is the case the rarity figures say least about.  It has no fixed fold.  Working versions range from 4 to at least 55 repeat units and tolerate proline for alanine and arginine for threonine.37  Even its signal sequence, already present in non-coding DNA, is less of a coincidence than it looks:  when Chris Kaiser and colleagues put essentially random sequences in place of a yeast enzyme's signal sequence, "about one-fifth of these sequences can function as an export signal".38  Doyle's phrase for the job, "a very non-specific job that many different random proteins could perform", is wrong chemically, since the protein binds particular faces of the ice crystal,37 and right about how easy the target is to hit.30  The best case for a gene from nothing is a case where rarity was never the obstacle.

The founders are where the questions meet

Copying explains how families grow.  It does not explain the 4,100 to 7,300 distinct domain units in the human proteome, or the 1,800 to 3,500 in yeast, each of which needed at least one origin that was not a copy of anything Pfam can relate it to.5  The 1,778 human proteins with no Pfam domain may add more.  Most Pfam domains are folded units, which is the object the rarity figures describe.  At one in 1011, thousands of founding folds lie easily within Dryden's ceiling of 4 × 1043 trials.  At one in 1064, Axe's measured term, the ceiling falls short for even one.16

No route in this section reaches them.  A founder is by definition a sequence whose ancestor cannot be detected, so comparing living genomes stops there.  The cases this section documents are therefore neither evidence that founding folds are easy to find nor evidence that they are impossible.  The question moves to the density of folded, working domains in random sequence, which How Rare Is a Working Protein? shows has been measured once for binding at 80 residues and extrapolated once for one enzyme at 153.

Part 7: What a young-earth history needs

Young-earth writers give these routes a limited place.  Jean Lightner writes that "It is important to understand the role of gene duplication within the creation model",39 and treats it as variation within created kinds.  Answers in Genesis tells children that "God made some animals with special antifreeze proteins in their blood".40  The design writers read orphan genes as a problem for evolution, not as products of it.26  None of these sources gives a rate for any route on a young-earth timeline.  The children supply the arithmetic, and it points the same way in all three.

  1. Duplicates. At Lynch and Conery's rate a human-sized genome keeps 0.2 to 2.5 new duplicates in 4,500 to 6,000 years, against 13,000 to 15,600 human proteins that have a relative in the same genome.18
  2. Genes from non-coding DNA. At Wu's human rate of 9.83 to 11.8 per million years, the same span yields 0.04 to 0.07 of one gene.19
  3. Antifreeze. Michael Oard estimates that the polar oceans cooled to today's temperatures in about 700 years after the Flood,41 so the fish needed antifreeze only after it. The Antarctic gene's inherited parts differ from today's trypsinogen gene at 29 of 523 letters, where two copies diverging for 4,400 years at the measured fish rate would gain 0.002 to 0.02 differences.42

These numbers are mainstream rates applied to a span no young-earth source proposes for them, so they are for scale only.  What they show is consistent:  on a young-earth timeline, gene families, real orphan genes and the antifreeze genes, or a family of trypsinogen-like genes already about 5% different, must have been created close to their present form, or produced by some mechanism far faster than any measured.  The first is the common-design reading, and nothing in the family counts refutes it.

Its cost is the pattern.  A created genome would not need to show a baker's yeast in which each region of a relative that branched off first is matched by two regions, with the lost genes scattered between them;  a cod antifreeze gene whose relatives hold its earlier stages in order of their distance on the tree;  or an Antarctic gene built from the two ends of a trypsinogen gene with a new core between them.  The children found no young-earth source that explains any of these, and none that gives a duplication count or a de novo count for its own timeline.

The mainstream history has the opposite problem.  It has time for about 100,000 retained duplications in 500 million years,18 more than enough to fill the families.  What it lacks is a documented route to the founders.

Part 8: Where two articles disagree

The children disagree on one point:  how many changes the northern cod antifreeze gene needed.  Orphan Genes says in its body that the gene's "origin is one substitution plus the copying of a short repeat".  Antifreeze Proteins counts about six or seven events, then copying.

Zhuang and colleagues' own sentence settles it.  Their "only a single nucleotide change" is scoped to the coding stretch:  "the emergence of a functional AFGP polyprotein cds required only a single nucleotide change in the ancestral 9-nt GCA(Ala)-rich element".13  The same paper's opening page adds a one-letter frameshift that joined the signal sequence to the repeats, and a "putative translocation event" that supplied the promoter.13  In the reconstruction as the antifreeze article sets it out, the 27-letter repeat also duplicated before the substitution.  So the antifreeze count holds, and the orphan article's sentence understates the steps.  The orphan article's Summary already gives the fuller count, and its larger point, that the best-documented new gene is also the least demanding protein available, is unaffected.

The children agree on everything else they share:  the antifreeze is the strongest case of a gene from non-coding DNA, duplication is faster than birth from non-coding DNA, and Axe's figure does not apply to short, unfolded or unproven proteins.

Objections and responses

"Nobody watched any of this, so comparing sequences proves nothing."  Nobody watched it, and every child says so.  But the reconstructions made predictions that could have failed.  Reversing the three changes in a leaf monkey's enzyme restored the ancestral pH optimum.7  The cod reconstruction needed the precursors in relatives to be silent, and RNA blots found no transcripts in any of the three relatives tested.13  Tested reconstructions are weaker than observation and stronger than stories.  Observed, or Only Inferred? grades each case.

"Shared families show common design, not common descent."  For the existence of families, common design fits the evidence as well as descent does, and Part 7 says so.  It fits the particular patterns less well:  the yeast doubling's scattered losses, the cod antifreeze's graded precursors, and the trypsinogen ends of the Antarctic gene are what history leaves, and no young-earth source found by the children explains them as design.  The design reading is strongest exactly where this section's method fails, at the founders.

"The rarity figures show these genes could not have evolved."  They do not touch the documented cases.  Duplication starts from a working fold, which Axe's own paper calls open to "progressive sequence divergence".15  The genes from non-coding DNA with protein evidence are short and mostly of unproven function, and the one with a shown function needs no fold.  The figures bear on the founders, and there the two published figures disagree by sixty-six orders of magnitude.

"The antifreeze genes show that evolution can build anything from nothing."  They show that mutation and selection can make a new gene with a new, life-saving job when the job is simple and the pressure extreme.  The product is a repeat with no fixed fold, and its own authors make no larger claim.  A new folded enzyme from non-coding DNA has not been documented anywhere in this section.

"The founders are just old genes whose relatives diverged past recognition.  Deep time covers them."  Some founders may be old genes Pfam fails to group, and a remote-homology count would lower the number.  But deep time does not change the price of a fold:  Dryden's 4 × 1043 already counts every trial in four billion years,16 so whether it suffices depends on the density, not the time.  Mainstream protein science does not picture founders as blind draws:  Nobuhiko Tokuriki and Dan Tawfik propose that flexible, promiscuous proteins open the way to entirely new folds, though their own word is "surmise" and the paper is a review.43  The claim that founders are recoverable by better searches, and the proposal that new folds grow out of old ones, are predictions no child has tested.

"Duplication adds no information, so most of this section is beside the point."  A copy adds nothing until it changes, and the children count it that way.  The changes after copying are small, but they are new functions:  Behe's own 2010 definition counts "the divergence by mutation of the activity of a previously duplicated coded element" as a gain.44  The defensible form of the objection is Meyer's:  the copy starts "already rich in functionally specified information".1  That form stands, and it points at the founders.

"Orphan genes are evidence of design, because there are too many of them, appearing too fast."  The counts fall each time they are checked, and every measured rate of birth from non-coding DNA is slower than duplication.25  On a young-earth timeline almost none would arise anyway, so their number does not distinguish the two histories.  The design argument that survives is the narrower one about circular annotation, argued in Orphan Genes.

Issues and unanswered questions

  1. The density of folded, working domains in random sequence. Only one forward count exists, for binding at 80 residues, with a detection floor near one in 1013. Axe's figure for catalysis carries a theoretical foldability term of 10−10 that, as far as the rarity article could find, nobody on either side has checked. A forward count for a catalytic fold would decide more of this section than any other experiment.
  2. How many new reading frames a lineage tries, and how often one first does something selectable. The orphan article counts reading frames in one genome at one moment. Neither it nor the rarity article computes the turnover over time, and the steps before transcription in the cod reconstruction had to persist without selection.
  3. The fitness of single steps. In the reconstructed duplication cases, whether each intermediate helped, did nothing or hurt decides between Behe and Snoke's model and Lynch's. It is measurable and has not been measured.
  4. The number of founders. The Pfam count is not firm in either direction. Pfam misses distant relatives, so remote-homology methods may merge families and lower it; the 9% to 26% of proteins with no Pfam domain may raise it.
  5. Young-earth accounts. No young-earth source found by the children gives a duplication count or a de novo count for its timeline, or explains the yeast doubling, the cod precursor series or the trypsinogen-derived Antarctic gene.
  6. The cod antifreeze's date and deepest relatives. The two cod papers put its origin at about 3 and at 13 to 18 million years ago, and disagree on whether the deepest relatives carry the precursor. Neither discusses the other.
  7. The sugars. The antifreeze binds ice partly through the sugars on each threonine. None of the origin papers explains how the cell's sugar-attaching machinery came to decorate the new chain.

Verdict

Comparing sequences across species explains much of where genes came from.  Most genes in the genomes counted have a template in an older gene, copying is frequent and measured, and the tested new jobs took two to fourteen amino-acid changes on proteins that already folded.  Genes have also arisen from non-coding DNA, though far fewer than the published counts suggest, and the two fish antifreezes show it best:  real new genes, built in a handful of common mutation steps, coding a simple repeat.

What comparison cannot reach is the founding member of each of the thousands of families in every genome.  That is also the only place the rarity figures apply, because the documented cases all began from something that worked or needed no fold.  At the founders the two published densities of working proteins disagree by sixty-six orders of magnitude, and nothing in this section decides between them.  The design side is right that the copy inherits its information and wrong where it aims fold-rarity figures at short or unfolded genes.  The evolutionary side has tested reconstructions for the steps it can see, and no documented route to a new fold.  A young-earth reading can treat families, orphans and antifreezes as created close to their present form, at the price of patterns that look like history and that no young-earth source has explained.

Sources

  1. 1.

    Meyer, Stephen C.  Darwin's Doubt: The Explosive Origin of Animal Life and the Case for Intelligent Design.  HarperOne, New York.  2013.  Book by the director of the Discovery Institute's Center for Science and Culture on the Cambrian explosion.  Chapter 11, "Assume a Gene", argues that mainstream accounts of new genes begin from genes that already exist.

    1. 1a.

      Page 218:  “Gene duplication, as the name implies, involves the production of a duplicate copy of a preexisting gene, already rich in functionally specified information.” ↩ ↩

  2. 2.

    Behe, Michael J.  The Edge of Evolution: The Search for the Limits of Darwinism.  Free Press.  2007.  Book by the leading biochemist of the intelligent-design movement arguing that random mutation reaches only small, simple changes.  Its chapter "What Darwinism Can Do" uses the Antarctic antifreeze as the best example of what it can do.

    1. 2a.

      Page 80:  “Even though we haven’t directly observed it, the scenario seems pretty convincing as an example of Darwinian evolution by natural selection.” ↩ ↩

  3. 3.

    Schrider, Daniel R. et al.  “Rates and genomic consequences of spontaneous mutational events in Drosophila melanogaster.”  Genetics 194(4):937–954.  2013.  Whole-genome sequencing of fruit-fly mutation-accumulation lines, with the new mutations compared with the variants found in wild flies to estimate how many are removed by selection.  Mirror:  PMC

    1. 3a.

      Section Results:  “We estimate that 98.8% of duplications and 99.2% of deletions in D. melanogaster are removed by purifying selection”. ↩

  4. 4.

    Lynch, Michael, and John S. Conery.  “The evolutionary fate and consequences of duplicate genes.”  Science 290(5494):1151–1155.  2000.  Genome comparison of six species (human, mouse, fruit fly, nematode, Arabidopsis, yeast).  The authors counted duplicate pairs by their silent-site divergence and fitted rates of birth and loss.  The origin and half-life figures still quoted on both sides of the debate come from this paper;  its times rest on silent-site clocks.

    1. 4a.

      Page 1154:  “For the species that we have examined, the average half-life of a gene duplicate is approximately 4 million years”. ↩

    2. 4b.

      Page 1151:  “Observations from the genomic databases for several eukaryotic species suggest that duplicate genes arise at a very high rate, on average 0.01 per gene per million years. … However, the vast majority of gene duplicates are silenced within a few million years, with the few survivors subsequently experiencing strong purifying selection. Although duplicate genes may only rarely evolve new functions, the stochastic silencing of such genes may play a significant role in the passive origin of new species.” ↩

  5. 5.

    Counted for Gene Duplication and Divergence with research/families.ts, whose output is research/families-output.txt.  Inputs are the Pfam 37.0 domain tables for human, baker's yeast (S288C) and Arabidopsis thaliana (May 2024) and UniProt's reference proteomes with one canonical protein per gene (release 2026_03:  20,652, 6,066 and 27,496 proteins).  A unit is a Pfam clan, or a family belonging to no clan.  A protein is in a family when it shares a unit with another protein of the same species.  Human:  85.0% of proteins share a clan unit and 81.3% a family, in 4,135 clan units or 7,261 families, with 1,778 proteins carrying no domain.  Yeast:  71.1% and 59.7%, in 1,795 or 3,489 units.  Arabidopsis:  72.4% and 70.8%, in 2,080 or 4,234 units.  Pfam was built on the 2023 UniProt release, so a few proteins added since then are counted as having no domain. ↩ ↩ ↩

  6. 6.

    Kellis, Manolis et al.  “Proof and evolutionary analysis of ancient genome duplication in the yeast Saccharomyces cerevisiae.”  Nature 428(6983):617–624.  2004.  Genome sequencing and comparison from the Broad Institute.  The authors sequenced Kluyveromyces waltii, which branched off before the proposed doubling, and mapped each of its regions against baker's yeast.  The 1:2 pattern they found is the strongest single piece of evidence for an ancient genome doubling.

    1. 6a.

      Page 617:  “We identify 145 paired regions in S. cerevisiae, tiling 88% of the genome and containing 457 duplicated gene pairs.” ↩

  7. 7.

    Zhang, Jianzhi.  “Parallel adaptive origins of digestive RNases in Asian and African leaf monkeys.”  Nature Genetics 38(7):819–823.  2006.  Follow-up study of the guereza colobus, an African leaf-eating monkey with two extra ribonuclease copies.  Zhang rebuilt an ancestral protein by site-directed mutagenesis and reversed the changes shared with the Asian copy to test what each set did.

    1. 7a.

      Page 821:  “Thus, the three backward mutations restored the original optimal pH of the enzyme, indicating that the three forward mutations were necessary for the decrease of the optimal pH from 7.4 to 6.7 in the evolution of guereza RNASE1b and RNASE1g.” ↩ ↩

  8. 8.

    Yokoyama, Shozo et al.  “Molecular basis of spectral tuning in the red- and green-sensitive (M/LWS) pigments in vertebrates.”  Genetics 179(4):2037–2043.  2008.  Mutagenesis study of red and green visual pigments, measuring how each of five key amino-acid sites shifts the colour of light absorbed.  Mirror:  PMC

    1. 8a.

      Section Materials and Methods, Background information:  “Three amino acid changes (S180A/Y277F/T285A) in the red pigment together shift the λmax from 563 to 530 nm, explaining fully the λmax of the green pigment (531 nm).” ↩

  9. 9.

    Pillai, Arvind S. et al.  “Origin of complexity in haemoglobin evolution.”  Nature 581:480–485.  2020.  Ancestral protein reconstruction from Joseph Thornton's laboratory.  The authors inferred the globins before and after the α/β duplication, made them, and measured how they assemble and bind oxygen, then put single and grouped historical changes into the ancestor.  Mirror:  PMC

    1. 9a.

      Section Abstract:  “Reintroducing just two post-duplication historical substitutions into the ancestral protein is sufficient to cause strong tetramerization by creating favorable contacts with more ancient residues on the opposing subunit. These surface substitutions dramatically reduce oxygen affinity and even confer weak cooperativity”. ↩

  10. 10.

    Casola, Claudio.  “From De Novo to 'De Nono': The Majority of Novel Protein-Coding Genes Identified with Phylostratigraphy Are Old Genes or Recent Duplicates.”  Genome Biology and Evolution 10(11):2906–2918.  2018.  A reanalysis of 874 mouse genes that earlier studies called de novo, meaning genes that arose with no recognizable ancestor.  Casola checked each one against the newer mouse gene databases and against other species' genes, and only 152 (40%) of the 381 still-annotated candidates held up.  The article uses it for the count falling from 874 to 152 and for the finding that 491 had already been dropped from the mouse gene databases.

    1. 10a.

      Page 2914:  “only 152 of the 381 PDNGs (40%) annotated as protein-coding genes represent de novo genes”.  Table 1 gives the reduction from 874 to 381. ↩

    2. 10b.

      Page 2915:  “This indicates that de novo genes in mouse emerged at a pace that is at least about 5.4-9.1 times slower compared with gene duplicates.”  The 11.6 genes per million years (Myr) assumes 139 mouse-specific genes over a 12 Myr mouse-rat divergence.  The abstract rounds 11.6 to 12. ↩

  11. 11.

    Clamp, Michele et al.  “Distinguishing protein-coding and noncoding genes in the human genome.”  PNAS 104(49):19428–19433.  2007.  A genome study of the human gene catalog that cut the count of protein-coding genes to about 20,500.  Of the 1,177 candidates left with no known relatives, only 12 had experimental evidence of a protein, and the authors call some of those reports equivocal.  The article uses the 1,177 count, the twelve, and the authors' argument that the implied birth rate is implausible.

    1. 11a.

      Page 19431:  “we found articles reporting experimental evidence of an encoded protein in vivo for only 12 of 1,177 orphans, and some of these reports are equivocal (SI Table 2).”  The authors restate this on page 19433 as "≈0.06% of the gene catalog". ↩

  12. 12.

    Chen, Liangbiao et al.  “Evolution of antifreeze glycoprotein gene from a trypsinogen gene in Antarctic notothenioid fish.”  Proceedings of the National Academy of Sciences 94(8):3811–3816.  1997.  Gene sequencing study from the University of Illinois laboratory that discovered fish antifreeze glycoproteins.  It compared one antifreeze gene and one trypsinogen gene from the Antarctic toothfish and found matching ends and introns, and a nine-letter founder for the repeated core.  The founding paper for the Antarctic origin;  its 5 to 14 million year date has since been revised.  Mirror:  PMC

    1. 12a.

      Page 3812:  “The AFGP polyprotein coding region (2129 nt) in the Dm1A AFGP gene contains 41 copies of AFGP coding sequences”.  “exon 1 (5′ UTR and signal peptide) of the AFGP gene and trypsinogen gene are 94% identical in sequence.”  “the 3′ end region of AFGP gene starting from the penultimate codon through 3′ UTR (255 nt) is 96% identical to the last exon (E6) of trypsinogen gene.”  “the entire trypsinogen intron 1 (238 nt) is present as two segments within the single AFGP intron with an overall sequence identity of 93%.”  The Fig. 1 legend gives the 3′ shared region as 225 nt, the text as 255 nt. ↩ ↩

  13. 13.

    Zhuang, Xuan et al.  “Molecular mechanism and history of non-sense to sense evolution of antifreeze glycoprotein gene in northern gadids.”  Proceedings of the National Academy of Sciences 116(10):4400–4405.  2019.  Comparison of the antifreeze gene and its matching DNA in seven cod species, three with antifreeze and four without, from clone libraries, with RNA blots and hybridisation tests.  It reconstructs the gene's origin from non-coding DNA step by step.  The history is inferred from living fish;  the experiments test the reconstruction's predictions.  Mirror:  PMC

    1. 13a.

      Page 4400:  “We identified the noncoding founder region and a nine-nucleotide (9-nt) element therein that supplied the codons for one Thr-Ala-Ala unit from which the extant repetitive AFGP-coding sequence (cds) arose through tandem duplications.”  “Through a 1-nt frameshift mutation, these two parts formed a single read-through open reading frame (ORF). It became functionalized when a putative translocation event conferred the essential cis promoter for transcriptional initiation.” ↩ ↩ ↩

    2. 13b.

      Page 4400:  “We experimentally proved that all genic components of the extant gadid AFGP originated from entirely nongenic DNA.” ↩

    3. 13c.

      Page 4403:  “In short, the emergence of a functional AFGP polyprotein cds required only a single nucleotide change in the ancestral 9-nt GCA(Ala)-rich element for it to become the founder codons for one Thr-Ala-Ala tripeptide.” ↩

    4. 13d.

      Page 4404:  “No transcripts of AFGP homologs were detectable in any of the three AFGP-lacking gadids. Only B. saida pancreatic RNA showed hybridization with strong intensity to its own AFGP cds probe”. ↩

  14. 14.

    Keefe, Anthony D., and Jack W. Szostak.  “Functional proteins from a random-sequence library.”  Nature 410(6829):715–718.  2001.  A laboratory experiment by Anthony Keefe and Jack Szostak of the Massachusetts General Hospital.  They made a DNA library of 4 × 1014 independently generated random sequences, converted it by mRNA display into 6 × 1012 purified, non-redundant proteins each carrying 80 contiguous random amino acids, and passed them over an ATP-agarose column.  Four unrelated protein families came out, and the authors estimate that roughly one in 1011 random proteins has comparable ATP-binding activity.  The article uses it for the measured side of its disagreement about how common a working protein is.  Mirror:  PMC ↩

    1. 14a.

      “We therefore estimate that roughly 1 in 1011 of all random-sequence proteins have ATP-binding activity comparable to the proteins isolated in this study.”  The paper's closing sentence:  “In conclusion, we suggest that functional proteins are sufficiently common in protein sequence space (roughly 1 in 1011) that they may be discovered by entirely stochastic means, such as presumably operated when proteins were first used by living organisms.  However, this frequency is still low enough to emphasize the magnitude of the problem faced by those attempting de novo protein design.”  The arithmetic is a straight count: four hits in 6 × 1012 sequences, about 6.7 × 10-13, multiplied by ten to correct for roughly 10% folding efficiency.  The paper itself supplies the caveats the article gives: only 5–15% of the displayed protein from each winning family was active; the 100 nM K_d protein is a round-18 descendant that went through three rounds of mutagenic PCR; family B folds around a Zn2+ ion, and "Zinc was not added to the selection buffer, but is present at about 10 µM in mammalian blood, from which the reticulocyte lysate used for mRNA translation is prepared"; and mRNA display cannot resolve a frequency below about 1 / (library size), that is about 10-13. ↩ ↩ ↩

  15. 15.

    Axe, Douglas D.  “Estimating the prevalence of protein sequences adopting functional enzyme folds.”  Journal of Molecular Biology 341(5):1295–1315.  2004.  Page 1300:  Axe's own rationale for the destabilised construct.  “The objective here, however, is to introduce a large number of mildly disruptive changes so as to render many side-chains suboptimal throughout the fold.  This is best achieved by introducing many changes together, without intervening selection.”  A laboratory study by Douglas Axe of the Babraham Institute in Cambridge, UK, on how rare working enzyme folds are.  He took one enzyme family, beta-lactamase, started from a weakly functional sequence that kept the fold's pattern of water-loving and water-hating residues, and randomised clusters of ten side-chains in four experiments to see which variants kept some function.  The measured rate was about one in 1064 such sequences, and the paper gives a range from one in 1077 to one in 1053 for any domain-sized fold, so the article's one-in-1077 figure comes from this paper and so does the range that shows how much of it rests on assumptions.

    1. 15a.

      Pages 1310–1311:  From the Discussion, the passage almost nobody quotes.  “A random change to a functional sequence actually has a good chance of leaving function undisturbed if very few positions are affected.  As estimated above, the likelihood of a signature-compliant substitution in the large-domain reference sequence producing a comparably functional variant is about 38%.  Since 70% of the ~1000 possible non-synonymous base changes to the reference coding sequence produce signature-compliant substitutions, about one in four random single-residue changes are functionally neutral.  The proportion would be somewhat lower under conditions requiring a higher level of function (such as those under which neutral drift normally occurs) but not so low as to preclude progressive sequence divergence by gradual accumulation of point mutations.”  Axe's thesis is specifically about the origin of new folds, which is far narrower than what the number is usually quoted for.  The headline figure is multiplied from three factors on p. 1310: his measured 10-64, a foldability factor of 10-10 taken from Lau and Dill's theoretical lattice model at "their least stringent folding criterion," and 10-3 for the share of folds suited to a given task, which Axe introduces with the words "A reasonable estimate of the average proportion suited to a particular task is 0.1%."  So 10-77 is one measurement times one theoretical number times one stated guess.  The Lau and Dill foldability figure was not checked for this article, and as far as could be established, neither side of the dispute has checked it, which makes it the most load-bearing unexamined input in the argument.  Three different quantities share this paper, 1064, 1077 and 1097, and they are routinely swapped for one another downstream.  Axe states the range himself in the same paragraph:  “the proportion of all sequences of large-domain length that perform the specified function by means of any tertiary fold… is estimated to be in the range of one in 1077 to one in 1053.”  Page 1307:  Axe states his own hedges plainly.  “Simplifying assumptions will therefore always be essential for treatments of the spaces corresponding to proteins of biological significance”.  “it seems reasonable to assume that the pass rates of Table 2, when averaged, provide an upper-bound estimate of the true mean pass rate”. ↩ ↩ ↩ ↩

    2. 15b.

      Page 1295:  “the overall prevalence of sequences performing a specific function by any domain-sized fold may be as low as 1 in 1077, adding to the body of evidence that functional folds require highly extraordinary sequences”. ↩

  16. 16.

    Dryden, David T. F. et al.  “How much of protein sequence space has been explored by life on Earth?”  Journal of the Royal Society Interface 5:953–956.  2008.  Section 2, first paragraph:  The lower bound, 4×1021 different protein sequences tested since the origin of life, is in the same paragraph.  The paper's 1034 is a generous estimate of all genes that have ever existed.  A short paper by three chemists at the University of Edinburgh that asks how much of all possible protein sequences life could have tried since it began.  It sets generous upper and lower limits for the number of organisms, genome size, mutation rate and amino acid types, and its upper limit is 4 × 1043 sequences explored.  The article uses it as the generous upper bound on what life could have searched, and the paper offers it as an upper bound. ↩ ↩ ↩ ↩

  17. 17.

    Ruiz-Orera, Jorge et al.  “Origins of De Novo Genes in Human and Chimpanzee.”  PLoS Genetics 11(12):e1005721.  2015.  A 2015 research article that is the source of the "634 de novo genes" figure.  It started from 1,029 transcripts (RNA copies of genes) and counted 634 of them as human-specific genes.  The authors' own abstract suggests that many of them are not functional, and the article uses the study for the median length of the genes that have protein evidence.

    1. 17a.

      Page 8:  “their size (median 76 amino acids)”.  Page 3:  634 human-specific genes come from 1,029 transcripts.  Page 8:  Of the 634, 6 had uniquely mapping peptides and 21 had any evidence of translation.  Section Abstract:  The authors state the caveat themselves:  “these transcripts show little evidence of purifying selection, suggesting that many of them are not functional.” ↩ ↩ ↩

  18. 18.

    Arithmetic from Gene Duplication and Divergence, in its research/pricing.ts, output in research/pricing-output.txt.  Retained duplicates per lineage are rate × genes × time, at Lynch and Conery's 0.002, 0.01 and 0.02 per gene per million years and 20,652 genes:  0.2 to 1.9 in 4,500 years and 0.25 to 2.5 in 6,000 years;  41 to 413 per million years;  at a 4-million-year half-life, about 240 to 2,400 young duplicates present at once;  about 103,000 in 500 million years at 0.01.  The copies to account for are the human proteins in Pfam families minus one founder per shared unit:  13,143 (families) to 15,585 (clans). ↩ ↩ ↩

  19. 19.

    Arithmetic for this overview, for scale only, since no young-earth source proposes that genes arose from non-coding DNA in this span:  Wu and colleagues' 9.83 to 11.8 human genes per million years 24 times 0.0045 to 0.006 million years gives 0.044 to 0.071 genes.  Wu's rate is itself 59 genes spread over 5 to 6 million years of mainstream time. ↩ ↩

  20. 20.

    Byrne, Kevin P., and Kenneth H. Wolfe.  “The Yeast Gene Order Browser: combining curated homology and syntenic context reveals gene fate in polyploid species.”  Genome Research 15(10):1456–1461.  2005.  A hand-curated database of gene order in yeasts that did and did not undergo the genome doubling, used to list the pairs still kept from it in three species.

    1. 20a.

      Page 1457:  “The YGOB data set has 551 likely ohnolog pairs in S. cerevisiae, which means that 19.6% of the genes in S. cerevisiae (1102 of 5516) are members of an ohnolog pair”. ↩

  21. 21.

    Behe, Michael J.  “Train Wreck of a Review: A Response to Lenski et al. in Science.”  Evolution News.  Feb 15, 2019.  Behe's answer to a review of Darwin Devolves in Science by Nathan Lents, Joshua Swamidass and Richard Lenski, who cited opsins, odour receptors and Hox genes as products of duplication.

    1. 21a.

      Section the paragraphs on gene duplication:  “I have no objection to thinking the proteins within those separate classes arose by gene duplication.”  The classes are opsins, olfactory receptors and Hox proteins. ↩

  22. 22.

    Ezkurdia, Iakes et al.  “Multiple evidence strands suggest that there may be as few as 19,000 human protein-coding genes.”  Human Molecular Genetics 23(22):5866–5878.  2014.  A paper by Ezkurdia and colleagues that combines several kinds of evidence to suggest that there may be as few as 19,000 human protein-coding genes.  Its protein checks found peptides for very few of the orphan genes it examined.  The article uses it to show that an orphan count in one well-studied genome falls as the gene list is updated, and it repeats the paper's own warning that missing peptides do not prove a gene is non-coding.

    1. 22a.

      Page 5870:  Their protein-evidence figures continue the pattern: 3 of 248, and 5 of 563 primate-family genes. ↩

  23. 23.

    Calculation made for Orphan Genes, source at orf-supply.ts.  Two independent methods, a closed form from base composition and a simulation over generated sequence, agree to three significant figures.  It assumes a 3.1-billion-letter genome, both strands, and random sequence at 41% GC, and finds about 16 million chance reading frames of 50 codons or more, 677,000 of 100 or more, 28,500 of 150 or more and 1,200 of 200 or more.  Real non-coding DNA is not random:  it carries repeats, regions of differing GC and some constraint, so this is an order-of-magnitude model.  A reading frame is also not yet a gene:  it still needs a start codon in context, transcription and translation. ↩ ↩

  24. 24.

    Wu, Dong-Dong et al.  “De novo origin of human protein-coding genes.”  PLoS Genetics 7(11):e1002379.  2011.  A study of human proteins with no match in primates, which narrowed 584 candidates to 60 by requiring proper start and stop codons and a length of at least 100 amino acids.  The article calls it the standard citation for human de novo genes and uses its rate figures.

    1. 24a.

      Page 3:  “this indicates that the rate of origin of de novo genes is ∼9.83-11.8 genes per million years”.  The arithmetic is 59 fixed genes divided by 6 million years = 9.83, and divided by 5 million years = 11.80.  The paper does not spell this out. ↩ ↩

    2. 24b.

      Page 5:  “when the rate is expressed in terms of per gene, ∼0.00033–0.00039 per gene per million years, it is still a lower rate than the rate of new gene origin by gene duplication”.  Citing the headline rate without this conversion misrepresents the source. ↩

  25. 25.

    Moyers, Bryan A., and Jianzhi Zhang.  “Evaluating Phylostratigraphic Evidence for Widespread De Novo Gene Birth in Genome Evolution.”  Molecular Biology and Evolution 33(5):1245–1256.  2016.  A test by two University of Michigan geneticists of the claim that genes often arise from non-coding DNA.  They ran the age-dating method behind that claim on simulated genes that all had the same age and found it produced the same trends, so the trends are largely artefacts of the method.  They conclude that birth from non-coding DNA is a minor source of new genes beside duplication, and count two verified cases in baker's yeast against about 144 duplications.

    1. 25a.

      Page 1253:  “there are only two verified instances of functional de novo gene births in S. cerevisiae (Cai et al. 2008; Li, Dong, et al. 2010), whereas approximately 144 functional duplications occurred in that time based on the inference from gene family expansions since the common ancestor of sensu stricto yeasts (Hahn et al. 2005).  Although these estimates may not be precise, gene duplication appears to surpass de novo gene birth by 2 orders of magnitude”. ↩ ↩

  26. 26.

    Gauger, Ann.  “Orphan Genes: A Guide for the Perplexed.”  Evolution News & Science Today.  Nov 6, 2018.  An online article by Ann Gauger that applies a figure to orphan genes, drawing the line between ATP-binding and enzymatic function herself in the same paragraph and then proceeding without it.  The 1077 figure traces to Douglas Axe, "Estimating the prevalence of protein sequences adopting functional enzyme folds," Journal of Molecular Biology 341(5):1295–1315 (2004), an estimate for a 153-residue beta-lactamase enzyme domain, with a range of 1 in 1077 to 1 in 1053 for any sequence of that length performing the function by any fold.  The article uses it for her sentence that applies this figure directly to orphan genes. ↩ ↩ ↩

  27. 27.

    Baalsrud, Helle Tessand et al.  “De novo gene evolution of antifreeze glycoproteins in codfishes revealed by whole genome sequence data.”  Molecular Biology and Evolution 35(3):593–606.  2018.  Whole-genome comparison of 27 cod-family genomes and eight notothenioid genomes from an Oslo group, mapping antifreeze presence, copy number and neighbouring genes onto dated family trees.  It reaches the de novo verdict for the cod gene independently of the Illinois laboratory, with a different date.  Mirror:  PMC

    1. 27a.

      Section Abstract:  “Here, we demonstrate that afgps in codfishes have evolved de novo from non-coding DNA 13–18 Ma, coinciding with the cooling of the Northern Hemisphere.”  “The origin of afgps in codfishes is one of the first examples of an essential gene born from non-coding DNA in a non-model species.” ↩

  28. 28.

    Deng, C. et al.  “Evolution of an antifreeze protein by neofunctionalization under escape from adaptive conflict.”  Proceedings of the National Academy of Sciences 107(50):21593–21598.  2010.  Genetic and biochemical study of the type III antifreeze of an Antarctic eelpout.  It traced the gene to a duplicated sialic acid synthase gene and showed in the laboratory that the enzyme's end domain already binds ice weakly and that its leader sequence can already export a protein.  Mirror:  PMC

    1. 28a.

      Page 21593:  “We found that an SAS 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”. ↩

  29. 29.

    Rives, Nathan et al.  “Diverse origins of near-identical antifreeze proteins in unrelated fish lineages provide insights into evolutionary mechanisms of new gene birth and protein sequence convergence.”  bioRxiv.  Jul 19, 2024.  Genome comparison of winter flounder, grubby sculpin and cunner with 14 related species, finding a different precursor gene for the type I antifreeze in each.  The preprint version;  the paper appeared in Molecular Biology and Evolution 41(9) in 2024.

    1. 29a.

      Section Abstract:  “found that the near-identical AFPI originated from a distinct genetic precursor in each lineage. Each independently evolved a coding region for the novel ice-binding protein while retaining sequence identity in the regulatory regions with their respective ancestor.” ↩

  30. 30.

    Doyle, Shaun.  “Antifreeze protein evolution: turning wrenches into hammers.”  Journal of Creation 25(2):14–17.  2011.  Young-earth review, from Creation Ministries International, of the 2010 type III antifreeze paper.  It accepts the evolutionary mechanism for that protein and argues that antifreezes are too simple to support evolution in general.

    1. 30a.

      Section The edge of eelpout evolution:  “Random mutation and natural selection is a likely mechanism for how this [antifreeze protein] and many others were produced.”  Section From neo-Darwinism to evolution:  “Being an AFP is a very non-specific job that many different random proteins could perform.”  Section Conclusion:  “AFPs are demonstrably simplistic proteins in their function, more akin to beneficial debris than a new complex and specified protein.” ↩ ↩

  31. 31.

    Näsvall, Joakim et al.  “Real-time evolution of new genes by innovation, amplification, and divergence.”  Science 338(6105):384–387.  2012.  Laboratory evolution in Salmonella.  A gene carrying two mutations that gave it a weak second activity was placed where copies form easily, and populations were grown for up to 3,000 generations under selection for both activities.  The only divergence of copies watched from start to finish.  Mirror:  PMC

    1. 31a.

      Section Abstract:  “This gene is amplified to a high copy number, and the amplified gene copies accumulate mutations that provide enzymatic specialization of different copies and faster growth.”  “This rapid process, completed in fewer than 3000 generations, shows the efficacy of the IAD model”. ↩

  32. 32.

    Povolotskaya, Inna S., and Fyodor A. Kondrashov.  “Sequence space and the ongoing expansion of the protein universe.”  Nature 465:922–926.  2010.  A computational study in the journal Nature about how the universe of protein sequences has been expanding over time.  The article uses it because one sentence of its abstract supports both a rarity argument and a plasticity argument, and the article quotes both halves of it. ↩

  33. 33.

    Behe, Michael J., and David W. Snoke.  “Simulating Evolution by Gene Duplication of Protein Features That Require Multiple Amino Acid Residues.”  Protein Science 13(10):2651–2664.  2004.  Population-genetic model by Behe and the physicist David Snoke of a duplicated gene that must gain a feature needing several specific amino acids, where the intermediate changes do not help and most other changes inactivate the copy.  The design side's main quantitative paper on duplication.  Mirror:  PMC

    1. 33a.

      Page 2651:  “We conclude that, in general, to be fixed in 108 generations, the production of novel protein features that require the participation of two or more amino acid residues simply by multiple point mutations in duplicated genes would entail population sizes of no less than 109.”  The exponents are superscripts in the paper. ↩

  34. 34.

    Lynch, Michael.  “Simple evolutionary pathways to complex proteins.”  Protein Science 14(9):2217–2225.  2005.  Reply to Behe and Snoke by the population geneticist whose duplication rates they had used.  Lynch built a model with neutral intermediates, many usable sites and linked copies, and computed the probability and time of gaining a two-residue feature.  Mirror:  PMC

    1. 34a.

      Page 2217:  “Numerous simple pathways exist by which adaptive multi-residue functions can evolve on time scales of a million years (or much less) in populations of only moderate size.”  “the model that they present is non-Darwinian”.  The second quotation is from the body;  the paper breaks "non-Darwinian" across a line. ↩

  35. 35.

    Reeves, Mariclair A. et al.  “Enzyme families–shared evolutionary history or shared design? A study of the GABA-aminotransferase family.”  BIO-Complexity 2014(4):1–16.  2014.  Follow-up from the Biologic Institute that screened single-mutant libraries of nine enzymes of the same family and most double mutants of two of them for BioF2 activity.  It is the design side's clearest statement of family resemblance as common design.

    1. 35a.

      Page 1:  “Focusing first on single mutations, we prepared mutated libraries of nine genes from the GAT family and tested for BioF2 function in vivo. None of the singly mutated genes had this function. Focusing next on double mutations, we prepared and tested 70% of the 6.5 million possible mutation pairs for Kbl2 and for BIKB, an enzyme described as having both Kbl2 and BioF2 activities in vitro. Again, no BioF2 activity was detected in vivo.” ↩

  36. 36.

    Arithmetic for this overview, for scale only:  1011 / 16,000,000 ≈ 6,250 and 1011 / 677,000 ≈ 148,000.  The reading-frame counts are from the calculation made for Orphan Genes 23, and the density is Keefe and Szostak's estimate for 80-residue proteins binding ATP as well as theirs did 14.  Treating every chance reading frame as one random protein overstates the supply, since most are never transcribed or translated.  ATP binding at that strength is one particular job;  a new gene may need an easier or a harder one, and for most claimed de novo genes no job has been shown. ↩

  37. 37.

    Chen, Liangbiao et al.  “Convergent evolution of antifreeze glycoproteins in Antarctic notothenioid fish and Arctic cod.”  Proceedings of the National Academy of Sciences 94(8):3817–3822.  1997.  Companion paper comparing purified antifreeze proteins and one gene each from the polar cod and the Antarctic toothfish.  The proteins are near-identical, and the genes differ in structure, spacers and codon spelling, so the authors conclude the two arose separately.  Its assignment of the cod signal sequence was later corrected by Zhuang and colleagues in 2019.  Mirror:  PMC

    1. 37a.

      Page 3817:  “AFGPs of cods differ only in the presence of an occasional Arg-for-Thr substitution”.  Page 3820:  “Pro-for-Ala substitutions occur in the small AFGPs 6–8”.  The same page 3817 sentence gives the number of repeat units as 4 to at least 55. ↩

    2. 37b.

      Page 3821:  “AFGPs have been found to adsorb to very specific faces (the prism planes) and align in specific directions (parallel to a axes) in ice crystals through a lattice match between specific hydroxyls in the disaccharides and the water molecules within these ice crystallographic parameters”. ↩

  38. 38.

    Kaiser, Chris A. et al.  “Many random sequences functionally replace the secretion signal sequence of yeast invertase.”  Science 235(4786):312–317.  1987.  Yeast genetics experiment that replaced an enzyme's export signal with essentially random sequences and counted how many still worked.  Used here to price how hard it is to come by a signal sequence.

    1. 38a.

      Section Abstract:  “Since about one-fifth of these sequences can function as an export signal the specificity with which signal sequences are recognized must be very low.” ↩

  39. 39.

    Lightner, Jean K.  “Gene duplication, protein evolution, and the origin of shrew venom.”  Journal of Creation 24(2).  2010.  Perspectives article by a young-earth creationist veterinarian and baraminologist.  It discusses shrew venom, yeast galactose genes and leaf-monkey ribonuclease as cases of duplication within created kinds.

    1. 39a.

      Section Gene duplications in the creation model:  “It is important to understand the role of gene duplication within the creation model.” ↩

  40. 40.

    Answers in Genesis.  “Do Animals Have Antifreeze?”  Kids Answers.  Jan 27, 2025.  Children's article from the largest young-earth ministry, presenting animal antifreezes as made by God, with no account of their origin.

    1. 40a.

      Section Animal Antifreeze:  “God made some animals with special antifreeze proteins in their blood that help them to survive and thrive, even in the coldest places.” ↩

  41. 41.

    Batten, Don et al.  “What about ice ages?”  The Creation Answers Book, 199–210.  Creation Book Publishers.  2014.  Chapter of Creation Ministries International's standard question-and-answer book setting out the young-earth model of a single ice age after the Flood.

    1. 41a.

      Page 206:  “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 (average 4°C).” ↩

  42. 42.

    Counts and arithmetic from Antifreeze Proteins, in its research/pricing.ts, which reads the GenBank sequences deposited by Chen and colleagues (U58944, the toothfish gene Dm1A;  U58835, the toothfish trypsinogen gene).  Alignment of Dm1A with the trypsinogen gene gives 63 of 67 letters identical in exon 1, 12 of 12 and 177 of 193 in the two intron pieces with 35 gap positions where a GT repeat differs in length, and 242 of 251 in the tail, 29 substitutions in 523 aligned letters.  Clock:  two copies diverging at 4.83 × 10−9 substitutions per letter per generation (Zhang and colleagues 2023) for 4,400 years gain 2 × 4.83 × 10−9 × 4,400 / g per letter, or 0.022 substitutions over 523 letters for a one-year generation and 0.002 for a ten-year one;  29 observed is 1,300 to 13,000 times more.  At the same rate 5.5% divergence takes 0.055 / (2 × 4.83 × 10−9) ≈ 5.7 million generations, which matches 22 to 42 million years at 3.9 to 7.4 years per generation. ↩

  43. 43.

    Tokuriki, Nobuhiko, and Dan S. Tawfik.  “Protein dynamism and evolvability.”  Science 324:203–207.  2009.  “We surmise that these properties are the foundation stones of protein evolvability,”.  The words "these properties" refer to conformational dynamism and functional promiscuity.  The sentence is a proposal, so it is a legitimate citation for what the mainstream view is and not for a demonstration.  A review that proposes a view rather than reporting a measurement, which its authors frame as "an alternative, 'avant-garde view'."  The article uses it to show what the mainstream position is, that protein folds are not isolated islands, and not as a demonstration that new folds evolve easily. ↩

  44. 44.

    Behe, Michael J.  “Experimental Evolution, Loss-of-Function Mutations, and "The First Rule of Adaptive Evolution".”  The Quarterly Review of Biology 85(4):419–445.  Dec 2010.  Review of laboratory evolution experiments in which Behe sorts each adaptive mutation as a gain, loss or modification of a functional coded element.  Its definitions are the design side's own test for a gain.

    1. 44a.

      Page 421:  “A “gain-of-FCT” adaptive mutation is a mutation that produces a specific, new, functional coded element while adapting an organism to its environment. The construction by mutation of a new promoter, intron/exon splice site, or protein processing site are gain-of-FCT mutations. Also included in this category is the divergence by mutation of the activity of a previously duplicated coded element.” ↩