Shared Retroviruses and Pseudogenes

"We found 41 sites that hosted two independent integration events at exactly the same base pair in the human genome." — Wang, Ciuffi, Leipzig, Berry and Bushman, on 40,569 HIV integrations, 2007¹

"The primate and muroid syncytins are clearly not orthologous genes — they are not syntenic in the two species clades — indicating that they are the result of independent gene captures" — Lavialle and colleagues, 2013²

"The findings of comparative genomics around the syncytin-1 loci is especially challenging to creationists." — Yingguang Liu and Charles Soper, Answers Research Journal, 2009³

Summary

About 8% of the human genome is made of endogenous retroviruses, or ERVs:  retrovirus copies that entered a sperm or egg cell and were passed down.  Nearly all sit at the same positions in the chimpanzee genome.  Humans and chimpanzees also share broken genes, called pseudogenes, with the same disabling mutations at the same places.  The mainstream reading is that both species inherited these from one ancestor.  The creationist reading is that they were put there separately or arose independently.  This article tests both against the measurements, and against the arithmetic that neither side has done.

The general theory of shared genetic changes as markers, including hot spots, incomplete lineage sorting, and the Alu literature, is covered in Shared Genetic Errors.  This article owns the human-chimpanzee cases.

What the numbers say.

1. The sharing is lopsided by three orders of magnitude. The 2005 chimpanzee genome paper counted ERV insertions found in one species and not the other. In the aligned parts of the two genomes there are 361: 234 and 45 chimpanzee-specific, 5 and 77 human-specific.⁴ The background is about 98,000 human ERV elements and fragments.⁵ The round figure of "200,000 ERVs" that circulates has no source in either paper.
2. Insertion is not random, and the popular "1 in 3 billion" argument is dead. Bushman's lab mapped 40,569 HIV integration sites and found 41 pairs that landed on the identical base pair.¹ That is one collision per 990 integrations, far more than chance over a 3.1 billion base genome. Any mainstream writer who says two viruses could never hit the same base is wrong.
3. Measured preference explains about 1% of the sharing, and forcing it further needs a pool no measurement supports. Computed for this article from Wang's collision rate: the effective pool of integration sites is 1 to 2 × 10^7. Two independent sets of 98,000 insertions would share 480 to 960 positions. The observed sharing is of order 10^5. To close that gap by preference alone, the pool must shrink to about 98,000 sites, 100 to 200 times smaller than measured, and for HERV-K alone to about 1,570 sites. A pool that small would still leave a few hundred unshared insertions in each genome, so the 361 lineage-specific ERVs do not refute it. The measurements do.
4. The syncytins are the control experiment. Six retroviral envelope genes were captured independently in primates, mice, rabbits and carnivores to build placentas. They are at six different, non-syntenic loci.² When capture really is independent, position says so. The human and chimpanzee ERVs do not carry that signature.
5. Function is proved by deletion, and it does not answer the ancestry question. Delete an ERV enhancer and a human immune gene stops responding to interferon.⁶ Delete the HBBP1 pseudogene region and fetal haemoglobin switches back on.⁷ Knock out a mouse syncytin and the embryo dies.² The creationist claim that these are not junk is correct. A co-opted insertion is still an insertion.
6. Independent loss does converge on the same site far more often than chance. Humans and guinea pigs lost the vitamin C gene GULO separately. Nishikimi's lab found the identical substitution at 47 of the 129 positions where the human copy differs from rat, a 36% overlap with a chance probability of 1.84 × 10^-12.[inai-2003:hotspots] In Gilad's 60-gene olfactory receptor sample, 12 of 27 shared disruptions are at the identical nucleotide across species, and three species carry three different deletions inside one 14-base window.⁸ Creationist writers cite Inai's paper and never use this result.
7. Of the three most-cited creationist pseudogene papers, two fail on their own chosen ground and one was right about function. Tomkins's GULO similarity figures were computed by a BLASTN method he himself documented as broken in 2015 and never recomputed,⁹ and his independent-loss claim rests on a difference in flanking elements that one old deletion would also produce.¹⁰ His vitellogenin "genomic address messenger gene" is a Genscan prediction whose functional reference is a United States patent.¹¹ His HBBP1 paper is the exception. He said the pseudogene is transcribed into a working regulatory RNA. A 2017 start-site deletion left globin output unchanged,⁷ but a larger 2021 study found the RNA essential for human red-cell development, by deletion, knockdown and rescue.¹² On ancestry his paper argues nothing, and the same study found the RNA absent from chimpanzee blood.
8. No published creationist quantitative account of shared ERV positions exists. The fullest attempt, Liu and Soper 2009, proposes that orthologous ERVs "were created to occupy similar genomic loci in separate species by a single designer."³ Liu's 2018 conference paper repeats the position without a number: the homology is "more likely due to functional necessity rather than common descent."¹³ Neither gives an expected count, a prediction of which loci, or a reason the shared loci should sort into a nested series by species, which they do in creationist data as much as in anyone's.

Verdict in one paragraph.  Shared position is strong evidence of shared inheritance for the ERVs, and the strength is now a number:  measured preference falls short by a factor of about 100 to 200, and the fix that closes the gap needs a target pool that no measured retrovirus comes near.  Shared pseudogene lesions are weaker evidence than mainstream writers say, because independent loss hits the same site often.  Function is real, conceded in full, and irrelevant to origin.  A young-earth account needs a model that predicts the count of shared positions and their nesting by species.  None has.

Part 1: What is shared, counted

An endogenous retrovirus is a retrovirus that integrated its DNA into a germ cell.  Its copy has two long terminal repeats, or LTRs, at each end, with gag, pol and env between them.  Most copies later lose the middle by recombination between the LTRs, leaving a solo LTR.  Solo LTRs are "typically 10–100 times more numerous than their more intact, undeleted counterparts."⁵

How many there are.  Belshaw and colleagues put the human total at "98,000 elements and fragments", about 5 to 8% of the genome.⁵  That counts elements and fragments, not independent insertion events, and any argument using it as a number of insertions overstates.  The cleaner census is for HERV-K(HML-2), the youngest family and the only one recently active.  Subramanian and colleagues in John Coffin's lab found "91 HML-2-related full-length and near full-length proviruses in the human genome"¹⁴ plus 944 solo LTRs, 1,035 elements in all.  Writers on both sides slip between these numbers.

How many are not shared.  The 2005 chimpanzee genome paper compared the aligned parts of the two genomes and counted transposable elements present in one and absent in the other.  Table 2 gives, for ERV class 1, 234 chimpanzee-specific elements and 5 human-specific;  for ERV class 2, 45 chimpanzee-specific and 77 human-specific.⁴  The sum is 361.  The 234 chimpanzee class-1 elements are PtERV1 and PtERV2;  the class-2 elements are mostly HERV-K.  The paper reports HERV-K "active in both lineages, with at least 73 human-specific insertions (7 full length and 66 solo long terminal repeats (LTRs)) and at least 45 chimpanzee-specific insertions (1 full length and 44 solo LTRs)."⁴  Two cautions.  Table 2 covers the aligned regions only, so it is not a whole-genome census.  And the figure "391" that appears in some creationist notes is an arithmetic slip;  the four cells sum to 361.

So the shape of the data is this.  Of order 10^5 ERV-derived loci sit at the same position in both genomes.  A few hundred do not.  That ratio, roughly 300 to 1, is what any explanation must reproduce.

The shared loci are not a flat list.  Subramanian's Tables 1 and 2 give each HML-2 provirus an "oldest common ancestor", the most distant primate that carries it at the same position.¹⁴  Counted for this article from all 91 rows:  30 are human-specific, 14 date to the human-chimpanzee ancestor, 6 to the gorilla node, 25 to the orangutan node, 3 to the gibbon node, 10 to the rhesus node, and 3 have no match in any non-human primate genome and are recorded as unknown.¹⁴  So 58 of the 91 are shared with chimpanzee or something older.  Two things follow.  About a third of the youngest family's proviruses are human-specific, so HERV-K is not a wall of shared positions, and creationist writers are right to say so.  The other two-thirds sort into a nested series by node, which is what a tree predicts and what a design account has to explain.

The same nesting appears in creationist data.  Tomkins measured HBBP1 against four primate genomes and found it "97.8%, 97.8%, 92.2%, and 92.1% identical" to chimpanzee, gorilla, orangutan and macaque.¹⁵  He does not comment on the shape.

The family is still moving.  HERV-K "contains >12 elements that have integrated since the divergence of humans and chimpanzees, as well as at least two that are polymorphic among humans."⁵  Insertion is not a frozen ancient event, which matters for the creationist side:  a compressed timescale needs a family that stayed infectious, and this one did.

The awkward case for the mainstream:  PtERV1.  Yohn, Eichler and colleagues characterised a retrovirus family with about 200 copies in chimpanzee and none in human.  They mapped 287 integration sites across chimpanzee, gorilla, baboon and macaque and found "approximately 95.8% of the insertions occur at non-orthologous regions between closely related species."¹⁶  The chimpanzee genome paper says PtERV1-like elements "are present in the rhesus monkey, olive baboon and African great apes but not in human, orang-utan or gibbon, suggesting separate germline invasions in these species."⁴  A virus in African apes and Old World monkeys but absent from humans and orangutans does not fit the species tree, and Yohn says so:  "the retroviral phylogeny is inconsistent with the generally accepted primate species tree based on molecular data."¹⁶  His explanation is separate infections of each lineage, with human and orangutan ancestors unexposed or resistant.

One sentence from this paper circulates on creationist sites:  "Such a series of independent deletion events at the same precise locations in the genome is unlikely."¹⁶  Read in place it is Yohn's argument that the apparently shared PtERV1 sites are not really shared, and he then tests it:  "In two cases, we were able to refine the map location to single basepair resolution.  Based on this analysis, we determined that the sites were not orthologous between chimpanzee and macaque."¹⁶  Quoted alone, the sentence reverses the paper's meaning.

What PtERV1 genuinely gives the creationist side is that apparent sharing at map resolution can dissolve at base-pair resolution.  Applied to the human-chimpanzee HERV-K loci, that test predicts the sharing should dissolve under refinement.  It does not.

The viral gene that stayed conserved.  A retrovirus needs its env gene to get out of one cell and into another.  Belshaw's group built the family tree of the HERV-K copies and measured selection separately on its internal branches, the lineage that went on producing new copies, and on its terminal branches, each copy's history after it settled in place.  env was held tightly on the internal branches and much more loosely on the terminal ones, with ratios of protein-changing to silent change (dN/dS) of 0.26 against 0.79, and seven of eight other HERV families showed the same internal-branch constraint.⁵  Their reading is that "all genes, including the env gene, which is necessary only for movement between cells, have been under continuous purifying selection," so that "an infectious pool of endogenous retroviruses has persisted within the primate lineage throughout the past 30 million years."⁵  The family multiplied by moving between cells, and each copy decayed once it settled.

This is weaker evidence against the creationist account than it first looks.  A compressed-timescale reader can drop the 30 million years.  And Liu and Soper do not say env has no job in a created genome.  They list host roles for it:  cell fusion in the placenta, shielding the fetus from the mother's immune system, and blocking related outside viruses by occupying their receptor.³  They also propose that ERVs were designed to move between cells for gene transfer and have degenerated since the Fall.³  Constraint while moving and decay after settling fits that proposal as well as it fits an ancient infection.  What the tree does establish is that HERV-K kept making new copies by infection.  Whether the copies humans and chimpanzees share were among those insertions, or were placed separately, is the position question of Part 2.

Part 2: Could insertion preference put them in the same places?

The strongest creationist objection to the shared-ERV argument is that retroviral integration is not random.  This is correct, and the premise-level case is made in Shared Genetic Errors.  What no source on either side has done is the arithmetic.

How strong the preference is.  Mitchell and colleagues in Bushman's lab mapped integration sites for three retroviruses in human cells against matched random controls.  "For HIV the frequency of integration in transcription units ranged from 75% to 80%, while the frequency for MLV was 61% and for ASLV was 57%.  For comparison, about 45% of the human genome is composed of transcription units."¹⁷  MLV, and only MLV, favours transcription start sites strongly (p = 1.4 × 10^-14);  ASLV shows no such bias.¹⁷  A 75% against 45% enrichment is a bias of 1.7-fold.  Even MLV's start-site preference concentrates integration into a few thousand bases around each of about 20,000 promoters, which is of order 10^7 to 10^8 bases, not 10^5.

Subramanian's HML-2 census points the same way for a real human ERV family:  about 60% of elements are in or within 30 kb of genes, about 20% are inside introns, and about 80% of those antisense.¹⁴  Chromosome 17 "is relatively gene-rich, but is devoid of HML-2 proviruses and solo LTRs, implying a selective pressure against HML-2 integration or maintenance on this chromosome."¹⁴  Preference and selection are both real.

The one experiment that measured collisions directly.  In 2007 the same lab mapped 40,569 HIV integration sites in a T-cell line and counted how often two independent integrations hit the same base pair.  "We found 41 sites that hosted two independent integration events at exactly the same base pair in the human genome.  Sites were only included in the analysis if the proviruses integrated at a single site were in opposite orientations, indicating independent events."¹  They attribute the collisions to "the favorable local DNA sequence and a globally favorable chromosomal environment."¹

Forty-one identical-base-pair collisions in 40,569 integrations is one per 990.  Random placement over 3.1 × 10^9 bases would give a collision probability per pair of about 3 × 10^-10;  the observed rate is about 5 × 10^-8, 150 times higher.  So the argument that two viruses could never land on the same base pair is refuted by measurement.  The question is whether the preference is strong enough to do what the design account needs.

Computed here

Everything below was computed for this article with a Deno script kept beside the research files, research/insertion-collision.ts.  No published source on either side has attempted it.  The method reads the preference off the experiment rather than modelling it.  Wang's collision count fixes an effective number of integration sites, N_eff, that already contains every preference the virus has:  chromatin state, transcription units, target sequence.  The counted pairs are opposite-orientation only, which detects about half of true collisions, so the script runs both the counted rate and a doubled rate.  The expected number of shared positions between two independent sets of n insertions is then n² divided by N_eff.

Output, run 6 September 2026 and extended 27 October 2026:

--- as counted (opposite-orientation pairs only)
  collision probability per pair : 4.982e-8
  effective integration sites    : 2.007e+7
  two independent sets of  98000 -> expected shared: 478.5  (0.488% of the set)
  two independent sets of   1035 -> expected shared: 0.1  (0.005% of the set)
  two independent sets of     91 -> expected shared: 0.0  (0.000% of the set)

--- orientation-corrected (x2)
  collision probability per pair : 9.965e-8
  effective integration sites    : 1.004e+7
  two independent sets of  98000 -> expected shared: 957.0  (0.977% of the set)
  two independent sets of   1035 -> expected shared: 0.1  (0.010% of the set)
  two independent sets of     91 -> expected shared: 0.0  (0.001% of the set)

--- what N_eff would preference alone need?
  to share 97639 loci: N_eff = 9.836e+4, i.e. each genome fills 99.6% of every viable site
  to share 49000 loci: N_eff = 1.960e+5, i.e. each genome fills 50.0% of every viable site
  to share 9800 loci: N_eff = 9.800e+5, i.e. each genome fills 10.0% of every viable site

--- free sites left in one genome
  N_eff = 1.0e+5: fraction of viable sites still free =  2.0%
  N_eff = 1.0e+6: fraction of viable sites still free = 90.2%
  N_eff = 1.0e+7: fraction of viable sites still free = 99.0%

--- unshared insertions the saturated pool itself predicts
  N_eff = 9.836e+4: about 361 per genome, 722 in the two together
  (observed, aligned regions only: 361)

--- HERV-K(HML-2): pool size preference alone would need
  shared fraction 66% -> pool ~ 1570 sites;
    Wang's N_eff 1.00e+7 is 6.39e+3 times larger
  shared fraction 66% -> pool ~ 1570 sites;
    Wang's N_eff 2.01e+7 is 1.28e+4 times larger

Two results.  First, measured preference gets the design account about 1% of the way.  With an effective pool of 1 to 2 × 10^7 sites, two independent sets of 98,000 insertions would share 480 to 960 positions.  Any claim that shared position is impossible by chance is therefore overstated.  But the observed sharing is of order 10^5, so preference is short by a factor of about 100 to 200.  On the HERV-K subset the shortfall is worse.  Two independent sets of 1,035 elements would share a tenth of one locus;  Subramanian's tables show 58 of 91 proviruses shared with chimpanzee or older.

Second, turning the crank the other way shows what the rescue costs.  To make preference alone produce the observed sharing, N_eff must fall to about 9.8 × 10^4 sites, so that each genome occupies 99.6% of every site a retrovirus could ever use.  That pool does not forbid the lineage-specific insertions, as it can seem to.  Two genomes that each fill 98,000 of about 98,400 sites still differ at about 360 sites apiece, which is the right order for the 361 the chimpanzee genome paper counted in the aligned regions.  So the count of unshared insertions does not refute the small pool.  The measurements do.  The pool has to be 100 to 200 times smaller than the one Wang measured.  For HERV-K alone it is worse:  about two-thirds of Subramanian's dated proviruses are shared with chimpanzee or an older primate, and preference alone would need the family to be confined to about 1,570 sites in 3.1 billion bases, roughly ten thousand times fewer than Wang's pool.¹⁸  A reader who proposes that "1 in 100 sites" match the virus's preferences has taken N_eff from 10^7 to 10^5 and needs a measurement to back it.

What is wrong with the computation, stated plainly.  Wang measured HIV in an immortalised T-cell line, and no collision measurement exists for a germline-transmitted retrovirus.  That is the largest weakness here.  The nearest germline data are Yohn's PtERV1 map, where 95.8% of copies are at non-orthologous positions, which points the same way.  The 98,000 figure counts fragments, not insertion events, though using a smaller n strengthens the conclusion.  And the model assumes both genomes sample the same preference distribution.  A design account can reply that the same designer used the same sites in both.  That reply is available, but it is no longer the insertion-preference argument;  it is the design argument, and it needs its own evidence.

Part 3: Function, and the syncytins as the control

The creationist literature's best material is on function, and it is right.  Three results, each proved by deletion.

Syncytins.  Lavialle, Heidmann and colleagues reviewed the placental genes their lab spent a decade characterising.  Syncytins "are 'new' genes encoding proteins derived from the envelope protein of endogenous retroviral elements that have been captured and domesticated on multiple occasions and independently in diverse mammalian species", and "Knockout of syncytin genes in mice provided evidence for their absolute requirement for placenta development and embryo survival".²  Six are known:  syncytin-1 and syncytin-2 in primates, syncytin-A and syncytin-B in mice, syncytin-Ory1 in rabbits, and syncytin-Car1 in carnivores, found in all 26 carnivore species tested.²

Interferon enhancers.  Chuong, Elde and Feschotte showed that a retrovirus family called MER41 seeded interferon-inducible binding sites across the anthropoid genome.  They deleted the MER41 element upstream of the immune gene AIM2:  "IFNG-induced AIM2 protein levels were undetectable in ΔMER41.AIM2 cells, thus demonstrating that MER41.AIM2 is necessary for endogenous IFNG-inducible regulation of AIM2."⁶

HBBP1.  Huang, Blobel and colleagues deleted the HBBP1 pseudogene region in adult human erythroid cells.  Contact maps rewired toward the fetal pattern, "accompanied by strong increases in γ-globin transcription."⁷

Anyone who says every ERV is genomic garbage is wrong, and creationist writers said so before these experiments.

Why it does not answer the question.  Each of these papers establishes function by reading the insertion history first.  Chuong's frame is that "lineage-specific ERVs have dispersed numerous IFN-inducible enhancers independently in diverse mammalian genomes",⁶ and independence is established because they are different elements in different places:  MER41 in anthropoids, a separate family called RLTR30B in mice.⁶  A DNA segment can be a retroviral insertion and a working enhancer at once.  The second fact says nothing about whether two species inherited the first.

The syncytins are the control experiment.  How does Heidmann's group know the primate and mouse syncytins were captured separately?  "Although they share similar features, the primate and muroid syncytins are clearly not orthologous genes — they are not syntenic in the two species clades — indicating that they are the result of independent gene captures that occurred separately in the genome from ancestors of each lineage."²  They apply exactly the test the shared-ERV argument applies.  When two lineages independently domesticate a retroviral env gene for the same job, the result is six loci in six different places, and no biologist mistakes them for shared inheritance.  The human and chimpanzee ERVs, including the ERV-W provirus that carries syncytin-1, sit at the same loci.

Peer Terborg put the creationist objection in the Journal of Creation in 2013.  Assembling the primate syncytin-1 locus by the standard route, he argues, means "the most parsimonious evolutionary scenario relies on a great number of highly unlikely genetic events of which the selective value of individual arrangements remains highly doubtful",¹⁹ and the improbability is multiplied because "there must have been at least six independent gene-capture events by mammalian genomes."¹⁹  That is an argument about mechanism, and it deserves an answer on its own terms.  It does not touch the position evidence, and Terborg reports that evidence himself in the same paragraph:  "The syncytin genes identified in different orders of mammals do no [sic] show a high level of sequence homology and they map to different genomic locations."¹⁹  That is the whole point.  Six independent captures at six different loci is not a coincidence needing explanation;  it is the signature by which independence is recognised, and the creationist account of the locus records the signature without drawing the inference.

The creationist side's own best paper names the problem.  Liu and Soper's 2009 Answers Research Journal paper is the fullest creationist treatment of ERVs.  It proposes that ERVs were created as part of the host genome and that exogenous retroviruses derive from them, "exogenization, as opposed to endogenization."³  Its answer to shared position is direct:  "The best alternative explanation is that orthologous ERVs were created to occupy similar genomic loci in separate species by a single designer to carry out similar physiological functions."³  Then it says what no other creationist paper says.  "The findings of comparative genomics around the syncytin-1 loci is especially challenging to creationists."³  And of the broken ERV-W copy in Old World monkeys:  "the presence of an inactivated ERV-W in OWMs simultaneously argues for the idea that ERVs are optional add-ins to the primate genome and the idea that humans and monkeys have a common Catarrhine ancestor!"³  That is to the authors' credit.  It is also the state of the creationist account:  a restatement of the observation, with no expected count and no prediction of which loci should match.

Liu also offers "directed integration", drawing on yeast Ty elements that target Pol III promoters tightly, and says it "may also partially explain the positional 'conservation' of orthologous ERVs".³  The yeast targeting is real.  The primate measurements are Mitchell's 1.3- to 1.7-fold biases and Wang's 1-in-990 collision rate, and Part 2 shows what those numbers can and cannot do.  Nine years later, in the 2018 paper, the account had not acquired a number.

Part 4: Three broken genes

A pseudogene is a gene with disabling mutations.  The mainstream argument is that when two species carry the same disabling mutation at the same nucleotide, they inherited it.  The creationist reply has two parts:  the gene is not broken, or the same break happened twice.  Three cases carry the debate.

GULO, the vitamin C gene

Most mammals make vitamin C with the enzyme GULO.  Humans, other anthropoid primates, guinea pigs, some bats and some birds cannot.  Inai, Ohta and Nishikimi described the human GULO region in 2003.  "Only five exons, as compared to 12 exons constituting the functional rat GULO gene, remain in the human genome."²⁰  The survivors carry two single-base deletions, a three-base deletion, an insertion, two stop codons and two broken splice sites.  The vertebrate-wide picture, including reactivation in bats and passerines, is in Shared Genetic Errors.

The result the creationist side should be using.  Inai lined up the human and guinea pig remnants against the working rat gene.  Everyone agrees, on every model, that these two losses were independent.  "The same substitutions from rats to both humans and guinea pigs occurred at 47 nucleotide positions among the 129 positions where substitutions occurred in the human sequences.  A high percentage of the same substitutions in the total substitutions (36%) indicates that there were many hot spots for nucleotide substitution throughout the sequences examined."²⁰  Under equal-chance substitution the probability of that overlap is 1.84 × 10^-12.[inai-2003:probability]  Two lineages that lost the gene separately converged on the identical change at more than a third of the positions where either changed.  That is the mechanism the creationist argument needs, measured by a lab with no stake in the debate.

Its limits go in the same paragraph.  These are substitutions, with four states per site;  retroviral insertions have of order 10^9 possible positions, so a 36% convergence rate in point mutations says nothing directly about a virus landing on the same base pair.  Transition bias and CpG hypermutability inflate the rate by shrinking the effective state space below four.  And it is 47 of 129 substituted positions in 647 bases of five exons, not 36% of all its mutations.

Tomkins 2014.  Jeffrey Tomkins's Answers Research Journal paper is the strongest published creationist treatment of GULO, and it cites Inai without using this result.  Its case is discontinuity:  "The 28,800 base human GULO region is only 84% and 87% identical compared to chimpanzee and gorilla, respectively.  The 13,000 bases preceding the human GULO gene, which corresponds to the putative area of loss for at least two major exons, is only 68% and 73% identical to chimpanzee and gorilla, respectively."¹⁰  Since gorilla comes out closer to human than chimpanzee, he argues the standard tree is negated.  He built similarity trees from each retained exon and found "the only set of sequences that produced the standard inferred phylogeny was for exon 5".¹⁰  And he argues from the transposable elements around the lost exons that the losses "occurred independently in each taxon".¹⁰

The first three claims do not survive.  The 84% and 68% figures were produced "using the previously established technique of optimized sequence slices and the BLASTN algorithm (Tomkins 2013b)".¹⁰  The next year Tomkins himself reported that "recent versions of the BLASTN algorithm employed as a local installation may be omitting significant levels of query sequences in large-scale genomic searches.  This claim potentially affects the results of past research done by this author".⁹  Only one version returned hits for nearly all queries;  "Every later version of the algorithm omitted significant levels of query sequence with the problem generally increasing with progressive algorithm releases."⁹  He revised his whole-genome figure from 70% to 88%.⁹  He deserves credit for finding and publishing the error.  The GULO figures were computed by the uncorrected method and have never been recomputed.  They should not be cited until they are.

The exon-by-exon trees are a method error that needs no new data.  Exon 3 is 107 bases, so one difference is 0.93%.  His Table 2 row for it, 98.2% to chimpanzee, 85.0% to gorilla, 98.2% to orangutan and 74.8% to baboon and rhesus, is fourteen or so differences spread over five taxa.  That is too little signal to recover a tree, no confidence measure is given, and incongruence at that sample size is expected under any model.  He attributes it to incomplete lineage sorting, a common-descent mechanism.

The fourth claim rests on a disagreement Tomkins states openly.  Lachapelle and Drouin found that the transposable-element analysis "did not reveal conclusive evidence that illegitimate recombination between repeats has contributed to the loss of exons", and dated the primate inactivation to a single event, "61 and 14 MYA for the primate and guinea pig genes, respectively."²¹  Tomkins reports that conclusion accurately and rejects it:  "their evolutionary presuppositions caused the rejection of otherwise strong genomic data that implicated TE related unequal recombination at the GULO locus."¹⁰  That is an argument, not a misquotation, and it should be answered as one.

The answer is that a different element landscape today does not distinguish one old deletion from several.  Transposable elements insert and are lost continuously, so a deletion shared by two lineages accumulates different flanking elements in each afterwards.  Tomkins's evidence fits either history, and the datum that separates them is one he prints himself:  "While the exon deletion events in both humans and apes appear to involve the same exons, the transposable element signatures surrounding the events are clearly non-similar and taxonomically restricted."¹⁰  The same exons are gone.  He raises the problem and answers it by saying "we really don't know what the original genes looked like in their pristine created state."¹⁰  An appeal to unknowable initial conditions is not a prediction, and it is the weakest point in the paper.

What Tomkins never addresses:  his own Table 2 shows every retained exon 97.6 to 99.4% identical between human and chimpanzee, and he offers no model that predicts that number for two separately created genomes.  He concedes GULO is a broken gene "tolerated by the organism".¹⁰  His argument is about discontinuity, not function.

HBBP1, the beta-globin pseudogene

HBBP1 sits in the beta-globin cluster on chromosome 11, between the fetal and adult globin genes.  Tomkins's 2013 paper argued that it "appears to be a highly functional and cleverly integrated feature of the human genome that is intolerant of mutation."¹⁵  Three claims:  it is unusually invariant, its variants associate with thalassaemia severity, and it is transcribed into regulatory RNAs, "at least 16 different exon variant transcripts and 42 different intron variant transcripts" in "at least 251 different human cell and/or tissue types".  Those last figures are Tomkins's own reading of browser tracks, with no independent published source.

On function he is vindicated.  Moleirinho and colleagues sequenced the cluster in 1,092 people from the 1000 Genomes Project plus ten chimpanzees and found that "HBD and its neighbor pseudogene HBBP1 have mainly evolved under purifying selection, suggesting that their roles are essential and nonredundant."²²  Then Huang's deletion proved it.  A broken relic does not, when removed, reconfigure a chromatin domain and switch on fetal haemoglobin.

On mechanism he is mostly right, and the direct test came from a secular lab.  Tomkins's headline evidence is transcription.  Two results seemed to sink it.  Moleirinho, whom Tomkins cites repeatedly, says the constraint on HBD and HBBP1 is "not related to protein function, as both are either marginally or not transcribed at all."²²  And Huang deleted only the transcription start site in a cultured adult red-cell line, HUDEP-2, and saw no change in globin output.⁷

Then a Beijing group led by Yanni Ma, Yong Zhang and Jia Yu tested the transcript directly, in five human systems.¹²  HBBP1 RNA is abundant in developing red cells.  On day 8 of differentiation it is more highly expressed than 98.9% of protein-coding genes, and it is found "exclusively in bone marrow".¹²  In human embryonic stem cells, deleting either the whole locus or only its 260-base start site cut early red-cell colonies from about 58 to about 2.¹²  Lowering the RNA with short hairpin RNAs, which leaves the DNA untouched, did the same in cord-blood and adult bone-marrow cells and in mice carrying human cells.  The authors raised Huang's objection themselves, that removing DNA might remove a chromatin element rather than an RNA, and tested it.  Splice-site mutations that destroyed the mature RNA but left the locus in place also impaired red-cell development, and HBBP1 RNA supplied from elsewhere in the genome partly rescued knockout cells.¹²  The RNA works by binding the protein HNRNPA1 and so stabilising the messenger RNA of TAL1, a master regulator of red-cell development.

So Tomkins's central claim, that HBBP1 is transcribed into a working regulatory RNA, now has direct experimental support, and Moleirinho's "marginally or not transcribed" is wrong for HBBP1 in red-cell precursors.  It is one laboratory's result, and a 2025 correction replaced one of its gel images, which the authors say leaves the conclusions intact.  Two parts of his case still fail.  His transcripts in "at least 251 different human cell and/or tissue types" sit badly with an RNA that Yanni Ma's group found only in bone marrow.  And the two labs disagree on one point:  Huang's start-site deletion left globin output at control levels in HUDEP-2 cells, while Yanni Ma's splice mutants in the same cell line lowered it.  They used different mutations and measured different things, γ-globin silencing against red-cell maturation, so both results may stand.  A separate 2021 paper by Shu-Ping Ma and colleagues reports that extra HBBP1 raises γ-globin through the transcription factor ELK1, but it is a gain-of-function result in cultured cells, and its abstract names its own confound:  "Surprisingly, ELK1 also directly bound to and activated the γ-globin proximal promoter."²³

The function is human-specific, and that cuts both ways.  Yanni Ma's group found HBBP1 "not or poorly expressed" in chimpanzee blood and confirmed its absence in five more chimpanzees.¹²  The authors read this as a dead gene recently put back to work in the human line.  A creationist can read it as a design difference between two species.  Neither reading touches the ancestry evidence below:  chimpanzee carries the same disabling lesions whether or not it uses the RNA.

One more correction.  Moleirinho's Table 3 gives human-chimpanzee divergence in HBBP1 exons as 1.45%, against 0.26% for HBB and 0.80% for HBD.²²  HBBP1 is not more conserved between species than the working genes beside it.  What is unusually low is within-species polymorphism.  Both sides have run the two measurements together.

On ancestry the paper does not argue.  Its one sentence is:  "Creationists would postulate that the sequence similarities were a matter of code re-use for a common functional purpose, a theme inherent to designed and engineered systems."¹⁵  No prediction, no number, and no account of why re-used code would carry the same disabling lesions and a nested 97.8 / 97.8 / 92.2 / 92.1 series.  Tomkins also quotes Chang and Slightom's 1984 finding that the three hominoid beta-globin pseudogenes differ by "only about one-third of that predicted for DNA sites evolving at the neutral rate."²⁴  He quoted it accurately and located it correctly.  What he does not quote is the sentence beside it.  Chang and Slightom found that human, gorilla and chimpanzee "each share the same substitutions in the initiator codon (ATG----GTA), a substitution in codon 15 which generates a termination signal TGG----TGA, nucleotide deletion in codon 20 and the resulting frame shift".²⁴  Three identical disabling lesions at three identical codons in three species.  A low divergence rate is the argument for function;  the shared lesions are the argument for inheritance, and they sit in the same abstract.

Vitellogenin, the egg-yolk gene

Brawand, Wahli and Kaessmann screened mammalian genomes for remnants of the three chicken egg-yolk genes.  Their claim has three legs, and the creationist reply addresses none.  The remnants lie "in regions syntenic to those containing these VIT genes in chicken", with synteny set by aligning the flanking genes.  The VIT1 matches score above the genomic background of the same alignment chain by a Mann-Whitney U test.²⁵  And "Exon 3 of VIT1 reveals shared indels between human and dog, which indicates that VIT1 was inactivated before the separation of the human and dog lineages".²⁵  Armadillo VIT1 exon 3 "shares two indels with the human/dog sequences."²⁵  The alignment behind those sentences was checked for this article.  It is 152 columns wide and carries exactly three gaps.  Human and dog share all three;  armadillo shares two and keeps the chicken's base at the third.²⁵  The authors also print their own cautions:  the VIT2 and VIT3 matches "may be too short to provide statistical significance or partially spurious",²⁵ and VIT genes were independently inactivated more than once in mammals.

Tomkins 2015 answers a different paper.  He treats the human remnant as one 150-base fragment, 62% identical to chicken over that stretch and 39% over a 36,450-base window,¹¹ and argues it "is not a pseudogene remnant at all, but a functional enhancer element in the fifth intron of a 'genomic address messenger' (GAM) gene."¹¹  He never mentions indels, dog, armadillo, or the Mann-Whitney test.

Both pillars of his own claim are weaker than they look.  The GAM gene is "predictively annotated" by "the gene prediction software Genscan",¹¹ a program that predicts protein-coding genes and has a high false-positive rate, here used as evidence for a long non-coding RNA gene.  And the statement that GAM genes "have been demonstrated to selectively inhibit the translation of known target genes"¹¹ cites Bentwich 2007, which his own reference list identifies as United States Patent 7,250,496, held by Rosetta Genomics.¹¹  Nothing in that reference is peer-reviewed or an experiment on this locus.  His synteny check finds one shared flanking gene in 360,000 bases, "LTD1",¹¹ which is ELTD1, the gene immediately beside the locus;  finding the nearest neighbour conserved is a point for Brawand.  A pseudonymous rebuttal at EvoGrad claims the mammalian remnant is about 3,100 bases across five fragments, not 150.²⁶  That is checkable at the UCSC browser and was not re-checked here;  it is attributed, not endorsed.

Part 5: The one dataset that shows both patterns

Olfactory receptor genes settle the shared-lesion question at the right resolution, because one lab sequenced enough of them in enough species to show both patterns in one table.  Gilad, Pääbo and Lancet sequenced olfactory receptor loci in human, chimpanzee, gorilla, orangutan and rhesus and recorded, for every species, the position and nature of each coding disruption.⁸  Their headline is that humans have lost these genes fastest:  "humans have a rate of OR gene disruption 3.3 times higher then [sic] the mean rate throughout the phylogeny."⁸

Counted here from Table 1, by a script kept beside the research files, research/gilad-table1.ts.⁸  The table has 60 loci.  Ten carry one cell marked ND, where that species failed to amplify;  the undetermined cell is dropped and the locus kept, because discarding a locus over one distant species throws away real shared lesions.  No ND falls in the human or chimpanzee column, so that comparison uses all 60.  Of the 60 loci, 27 carry a disruption in two or more of the five primates.

• Twelve carry the identical disruption at the identical nucleotide in every species that has one. Four of them are identical in all five species: a deletion at position 178 in 4G3p, at 723 in 52B1p, at 605 in 8F1p, and a stop at 236 in 5AK4p.
• Fifteen carry disruptions at different positions in different species, which is independent loss. 13C6p is a deletion at 235 in human and at 744 in chimpanzee. 5E1p is a stop at 335 in human and a deletion at 384 in chimpanzee. 52H2p is a deletion at 270 in human, chimpanzee, gorilla and orangutan but a stop at 540 in rhesus.
• Human against chimpanzee alone: of the 19 genes where both carry a disruption, 16 share the identical position and 3 do not.
• The hotspot warning is in the same table. OR2Q1p is a deletion at 505 in human, at 507 in orangutan and at 518 in rhesus. Three separate deletions inside a 14-base window.

Nobody on either side of the creation debate appears to have used this table.  It says three things at once.  Independent loss is common and lands in obviously different places, fifteen times in a sixty-gene sample, so a creationist appeal to independent breakage is not idle.  Independent lesions sometimes cluster within a few bases, so one shared lesion proves little alone.  And the human-chimpanzee ratio is better than five to one for identical position.

Shared pseudogene lesions are therefore real evidence, weaker than mainstream writers say and stronger than creationist writers allow.  Shared ERV positions are a different case, because the state space is a billion positions rather than four bases.

Objections and responses

"The odds of two viruses hitting the same base pair are one in three billion, so shared ERVs prove common ancestry."  Retire this argument.  Wang measured 41 identical-base-pair collisions in 40,569 HIV integrations, about 150 times the random rate.¹  The correct expectation is 500 to 1,000 coincidental shared positions between two independent sets of 98,000 insertions, not zero.  What survives is a comparison of magnitudes, not an appeal to impossibility.

"Integration site preference explains the shared positions."  It explains about 1% of them.  The measured pool of effective sites is 1 to 2 × 10^7;  closing the gap needs about 10^5, and for HERV-K alone about 1,570.  A pool that small would also reproduce the few hundred lineage-specific insertions, so the objection is not self-refuting;  it is unsupported.  Anyone who wants it to work has to supply an N_eff and defend it against Wang's measurement.

"ERVs are functional, so they are not viral insertions."  Function and origin are separate questions, and the papers that prove function prove it by reading the insertion history.  Syncytin-1 sits inside an ERV-W provirus with recognisable LTRs, gag and pol remnants and a target-site duplication.

"Six syncytins captured independently is convergent evolution six times over, which is absurd."  Whatever its force against the mechanism, it misses the position argument.  The six captures are how independence is recognised:  six different, non-syntenic loci.²  Terborg, making this objection, records the same fact.¹⁹  If human and chimpanzee ERVs had been installed separately, they should look like the syncytins.  They do not.

"Mutational hot spots explain the shared pseudogene lesions."  Partly, and Inai's guinea-pig result is the best evidence for it:  47 of 129 substitutions identical in two independent losses, p = 1.84 × 10^-12.[inai-2003:hotspots]  The limit is the state space.  Hot spots work on four-state substitutions.  They do not scale to insertion positions, and the Gilad table that shows clustering also shows fifteen cases of independent lesions landing apart.

"PtERV1 shows ERVs do not follow the species tree."  It shows that a virus infecting several lineages after they separated leaves a pattern that does not match the tree, which is Yohn's conclusion.¹⁶  His data put 95.8% of 287 insertions at non-orthologous positions,¹⁶ and the apparently shared sites dissolved under base-pair refinement.  Independent infection leaves a recognisable signature.  Human-chimpanzee HERV-K does not carry it.

"Missing ERVs disprove the tree, since the pattern should be perfect."  It should not.  Belshaw's group measured recombinational deletion of HERV loci and found recent HERV-K loci deleted about 200 times faster than older ones.²⁷  ERVs are removed cleanly, so absence where a tree predicts presence is weak evidence.  This cuts against both sides and is argued in Shared Genetic Errors.

"Tomkins showed the human and chimpanzee GULO regions are only 84% identical."  That figure came from a BLASTN method Tomkins himself documented as dropping large fractions of query sequence, and he revised the whole-genome number it produced from 70% to 88%.⁹  The GULO figures have not been recomputed.  That is a small job and should be done before they are used again.

"Shared position settles the ancestry question by itself."  This is the objection to our own verdict from the other side, and it does not settle it alone.  It rules out one alternative, independent insertion at coincidentally matching sites, by a factor of about 100 to 200.  A designer who chose to place identical sequences at identical loci in two genomes is not ruled out by any measurement of position, because the proposal says nothing about how many loci should match or which.  It is not empty, though.  Liu and Soper wrote predictions into their 2009 paper, and the first was that "creationists expect to find more examples of complex, interdependent functions between ERVs and the host genome".³  That has held at particular loci since they wrote:  the AIM2 enhancer deletion of Part 3 is from 2016.  It does not separate the two accounts, because co-option of ERVs was already a mainstream research program in 2009.  What the proposal does not predict is the count of shared positions, which loci match, or their nesting by species.  On those, the position data behave exactly as inheritance predicts and need an unconstrained assumption to behave otherwise.

Issues and unanswered questions

Two papers were read only in abstract.  Lachapelle and Drouin 2011 is paywalled at Springer;  the exon breakpoints and element positions have not been read, so nothing here rests on them.  Shu-Ping Ma and colleagues' 2021 ELK1 paper is closed at Elsevier;  the abstract shows the ELK1 route is confounded, and nothing here rests on it.

The two HBBP1 laboratories have not been reconciled.  Huang's start-site deletion in HUDEP-2 cells left globin transcription at control levels;  Yanni Ma's splice-site mutants in the same cell line lowered it and impaired red-cell maturation.  The mutations and end points differ.  A test of both mutations side by side, in one cell line, measuring both γ-globin silencing and maturation, would settle whether the RNA and the DNA element do separate jobs.

Terborg's improbability argument is not answered here.  This article answers only the position half of it, because that is what the shared-ERV question turns on.  Whether a retroviral env gene can be domesticated at all by known mechanisms belongs to the origin-of-new-genes question, and nothing here settles it either way.

Three computations would settle what is still open, and none has been run.

1. A collision rate for a germline retrovirus. Wang's number comes from HIV in a T-cell line, the largest weakness in Part 2. The substitute is to take Subramanian's 944 solo LTR and 91 provirus coordinates, measure what fraction of the genome their positions occupy, and compute the collision expectation from that distribution instead of from HIV.
2. Divergence between orthologous HERV-K copies. Belshaw showed HERV-K stayed an actively reinfecting pool, so two independent insertions at one position would come from viruses that had already diverged. Under shared inheritance the human and chimpanzee copies should differ only by post-split drift. No source obtained here reports this measurement, and it would be the sharpest single test in the area.
3. Recompute the GULO identities. Tomkins's 84% and 68% figures, redone with BLASTN 2.2.25+, nucmer and LASTZ, as he did for the whole genome in 2015.

Where this article's own side is weakest.  There is no published creationist quantitative account of orthologous ERV positions.  Liu and Soper 2009 is the fullest attempt and contains no number;  Liu's 2018 paper adds none.  Todd Wood, a young-earth creationist biologist, said as much of shared pseudogenes in his 2006 review of the chimpanzee genome:  "While common design could be a reasonable first step to explain similarity of functional genes, it is difficult to explain why pseudogenes with the exact same substitutions or deletions would be shared between species that did not share a common ancestor."²⁸  Such a model would have to predict an expected fraction of shared positions, predict which loci are shared, and predict the nested sorting by node that appears in Subramanian's ancestor column and in Tomkins's own HBBP1 series.  It would also have to work out what Liu and Soper only propose, that ERVs were designed to move between cells, against Belshaw's pattern of constraint on the copying lineage and decay after each insertion.  None of that has been attempted, and saying otherwise would be dishonest.

Two original counts here are reproducible but not independently confirmed.  Both were re-derived by scripts kept beside the research files, research/subramanian-nodes.ts and research/gilad-table1.ts, and both read the published tables rather than the underlying sequences.  The Subramanian tally accounts for all 91 rows and matches the paper's own arithmetic.  The Gilad tally has not been re-derived from the GenBank accessions AY192803 to AY192982, and its treatment of the ten undetermined cells is a judgement call:  keeping those loci gives 27 disrupted in two or more species and 12 identical, discarding them gives 23 and 8.  The human-chimpanzee figures do not depend on the choice.

Verdict

Shared ERV positions are strong evidence of shared inheritance, and the strength is now a number.  Measured integration preference predicts 480 to 960 coincidentally shared positions between two independent sets of 98,000 insertions.  The observed sharing is of order 10^5.  The shortfall is a factor of about 100 to 200, and on the HERV-K subset it is four to five orders of magnitude.  The fix that closes the gap needs a pool of integration sites 100 to 200 times smaller than the measured one, and for HERV-K about ten thousand times smaller.  It does not conflict with the 361 lineage-specific insertions the chimpanzee genome paper counted, since a pool that small leaves a few hundred unmatched sites in each genome;  it conflicts with the measurements of where retroviruses integrate.

The creationist function claim is right and answers a different question.  Deleting an ERV enhancer kills the interferon response of a human immune gene.  Deleting the HBBP1 region switches fetal haemoglobin back on, and removing its RNA stalls human red-cell development.  Knocking out a syncytin kills a mouse embryo.  All of that is conceded without hedging.  None of it bears on whether human and chimpanzee inherited the sequence or received it separately, and the strongest creationist papers slide between the two questions inside one paragraph.

The syncytins are the control, and they run against the creationist reading.  Six independent captures of a retroviral envelope gene for the same job produced six different, non-syntenic loci.  Independent capture leaves a signature.  Human and chimpanzee ERVs do not carry it.

The creationist side's best card is real and under-used by that side.  Human and guinea pig, which lost GULO separately, share 47 of 129 substitutions at identical positions, chance probability 1.84 × 10^-12, and Nishikimi's own conclusion is "many hot spots".  Gilad's table shows independent lesions clustering within 14 bases.  Wang's 1-in-990 collision rate kills the "one in three billion" argument outright.

What an honest young-earth position looks like here.  Say that many ERVs and pseudogenes are functional, and cite the deletion experiments, including the 2021 HBBP1 study that bore out Tomkins on the RNA.  Say that integration is measurably non-random and give Wang's number.  Say that independent loss converges on the same site far more often than chance, and give Inai's.  Then say plainly that the count of shared ERV positions is two orders of magnitude beyond what measured preference can produce, that the shared loci sort into a nested series in creationist data as much as anyone's, that no creationist model predicts either fact, and that two of the three most-cited creationist pseudogene papers fail on the method they chose while the third, on HBBP1, was right about function and silent on ancestry.  Anything stronger is not supported by the data on the table.

Companion pages

• The literature — every source consulted, with what it measured, whether it was read in full, and which way it cuts.
• Quarantine — the claims held back from this article and why.
• Shared Genetic Errors — the general theory of rare genomic changes as markers: hot spots, homoplasy, hemiplasy and incomplete lineage sorting.
• Human and Chimpanzee DNA Similarity — where the percent-identity numbers, including Tomkins's, are argued.
• Chromosome 2 Fusion — the other structural argument for common ancestry, and what it can and cannot show.

Sources