Shared Genetic Mistakes and Common Ancestry

"the probability that homoplasy will obscure phylogenetic relationships [based on SINEs and LINEs] is, for all practical purposes, zero … there is no need for statistical analysis and definitive conclusions [about phylogeny] can almost always be drawn" — Nikaido, Rooney and Okada, as quoted by David Hillis¹

"Similar claims have been made for other kinds of data in the past, and in every case examples have been found to refute the claim." — David Hillis, replying to them in the same issue¹

"Our results also suggest that TEs should not be viewed as perfect characters exempt from homoplasy.  Instead, TE insertions present many of the same challenges for phylogenetic analyses as other types of data, such as nucleotide sequences." — Han and sixteen co-authors, after testing the claim on the avian tree²

Summary

Suppose two students hand in the same essay with the same misspelling in the same sentence. You do not conclude that they made the same mistake twice. You conclude that one copied the other, or that both copied a third source. The mistake carries no information by itself. What carries the information is that a pointless detail is shared.

Biologists make the same argument about genomes. Human DNA contains a gene for making vitamin C. It does not work. Chimpanzee DNA contains the same broken gene, broken in overlapping ways. Human DNA contains roughly a million copies of a 300-base parasite sequence called Alu, scattered at particular positions. Chimpanzees have most of those copies at the same positions. So do gorillas, for a slightly smaller set. The pattern nests.

This is the strongest genetic argument for common ancestry, and it is stronger than the argument from overall similarity. Similarity can be explained by a designer reusing a working part. A broken gene shared at the same nucleotide is harder to explain that way, because it is not a working part.

This article asks how well the argument holds up. The short answer is that it holds up better than most critics admit, and that its advocates have overstated it in a specific way that the technical literature has already corrected.

The conclusions, in order of confidence

1. The "perfect character" claim is dead, and it was killed by evolutionary biologists, not by creationists. Nikaido and colleagues wrote in 1999 that homoplasy in these markers is "for all practical purposes, zero" and that "there is no need for statistical analysis."¹ David Hillis replied in the same issue of PNAS that this claim had been made before for DNA–DNA hybridization, for genome rearrangements, and for indels, "and in every case examples have been found to refute the claim."¹ In 2011 a seventeen-author team tested it directly on the bird tree and concluded that transposable elements "should not be viewed as perfect characters exempt from homoplasy."² The same group reached the same verdict for microinversions: they "are not perfect characters and should be interpreted with caution, just as with any other character type."³

2. But the measured failure rates are small — and this cuts against the creationist use of the point. Across more than 11,000 primate Alu loci, precise parallel insertion at the same site occurs at "about 0.0005 events/insertion."⁴ That is one in two thousand. The retention index of transposable-element insertions on the avian tree is 0.97, against 0.52 to 0.58 for the DNA sequence data measured on the same tree.² In the study that settled the human–chimpanzee–gorilla branching order with Alu elements, exactly one locus out of 133 conflicted.⁵ The direct experiment gives the same picture from the other end.  Bushman's laboratory mapped 40,569 HIV integrations in human cells and found 41 pairs that had landed on exactly the same base pair.⁶  That is far more often than a uniform genome would allow:  it means HIV effectively chooses among only 10 to 20 million sites, not billions.⁷  Insertion preference is strong.  It is still about a hundred times too weak to explain shared endogenous retrovirus positions between humans and chimpanzees, as the companion article Shared Retroviruses and Pseudogenes works out.  Documented homoplasy at these rates cannot produce a nested hierarchy of thousands of markers. Anyone who cites the hot-spot literature as if it could is misusing it.

3. The markers can be precisely deleted, and this specifically refutes a stated falsification criterion. Douglas Theobald's widely cited 29+ Evidences for Macroevolution allows for deletion but says "deletions are never clean and usually part of the transposon sequence remains."⁸ That is false. Van de Lagemaat and colleagues showed retroelements are removed by recombination between their own flanking target-site duplications, and that "the deleted loci are indistinguishable from pre-integration sites, effectively reversing the insertion."⁹ They estimate 0.5% to 1% of apparent retroelement insertions distinguishing humans from chimpanzees are actually deletions.⁹ Full-length human endogenous retroviruses collapse to solo long terminal repeats routinely; the deletion rate among recently inserted HERV-K loci is roughly 200 times that among loci older than the human–chimpanzee split.¹⁰ Absence of a marker is much weaker evidence than presence of one.

4. Insertion is not random, and hot spots are documented at the single-nucleotide level. Cantrell and colleagues found two hot spots inside one mouse element and, at each, two independent SINE insertions at identical sites. Their conclusion is worth quoting in full because it is the clearest statement of the problem by people who use the method: the results "indicat[e] the need for caution when one concludes that the existence of a SINE at a specific locus in multiple individuals is indicative of common ancestry. Although independent insertions at the same locus may be rare, SINE insertions are not homoplasy-free phylogenetic markers."¹¹ Han and colleagues found six hot spots among only 66 avian insertions.² Retroviruses have measurable, family-specific integration-site preferences.¹² Theobald's stated premise that transposons "insert fairly randomly"⁸ is an approximation, not a fact.

5. Where the markers fail completely is short internodes — and there they fail spectacularly. This is the largest effect in the literature and it is not homoplasy. Churakov and colleagues screened about two million loci for the root of the placental mammal tree and found five retroposon insertions supporting one answer, nine supporting a second, and eight supporting a third, concluding there is "significant support for a 'soft' polytomy."¹³ Doronina and colleagues repeated the exercise one node down and got 14, 8, 9 and 12 markers for four mutually exclusive orders of divergence: "All possible evolutionary scenarios received considerable retrotransposon support, leaving us with a network of affiliations."¹⁴ Suh, Smeds and Ellegren scored 2,118 retrotransposon markers across the bird radiation and found 745 incongruent, concluding that such radiations "may neither resemble a fully bifurcating tree nor are they resolvable as such."¹⁵

6. That failure is mostly hemiplasy, not homoplasy, and the distinction is the single most important thing to get right here. Homoplasy means the marker arose twice, or was cleanly removed — the marker is not what it looks like. Hemiplasy means the marker is genuinely unique and genuinely inherited, but it tracks a gene tree that differs from the species tree, because the ancestral population carried it as a polymorphism through more than one speciation.¹⁶ Guerrero and Hahn showed hemiplasy is "as likely—or more likely—than homoplasy for a wide range of conditions, even when levels of discordance are low."¹⁷ Lineage sorting is also tested, not just invoked.  It predicts that the ordinary DNA around a conflicting insertion will show the same conflicting tree, because that whole stretch of chromosome was inherited as one piece.  In the great apes, 157 of 187 conflicting insertions passed that test.  The other 30, about one in six, did not, and they are the honest residue in the best ape dataset.¹⁸ Hemiplasy destroys the claim that a shared marker fixes a branching order. It does not touch the claim that the marker was inherited from a common ancestor. Any argument that quotes Suh's 35% figure as evidence against common descent is quoting a paper that assumes common descent throughout and says explicitly that its markers "contain negligible homoplasy."¹⁵

7. Pseudogenes as a class are the least reliable marker of the set, and in one documented case they made the analysis impossible. Nuclear copies of mitochondrial DNA — numts — are pseudogenes that get amplified alongside the real thing. Thalmann and colleagues found "gorillas are notable for having such a variety of numt sequences bearing high similarity to authentic mtDNA that any analysis of mtDNA using standard approaches is rendered impossible."¹⁹ Song and colleagues showed numts inflate species counts in DNA barcoding, and that careful filtering still misses some.²⁰ In plants, ribosomal ITS supplied 66% of published phylogenies over five years and 34% of them used nothing else — and Álvarez and Wendel showed rDNA pseudogenes and paralogy make "homoplasy … higher in ITS than in other DNA sequence data sets."²¹

8. "Mistake" is an assumption, and it is failing for a growing minority of these sequences. The plagiarism analogy requires the shared feature to be useless. Pseudogenes are transcribed, some regulate their parent genes, and some act as microRNA decoys.²² Chuong, Elde and Feschotte deleted a set of endogenous retroviral elements from the human genome with CRISPR and found they were required for normal interferon-induced gene expression, including activation of the AIM2 inflammasome.²³ This does not rescue the design position by itself — a shared function at a shared position is still shared, and needs explaining — but it does remove the rhetorical force of "why would a designer put a broken gene there?"

9. Mutation is far less random than the plagiarism analogy needs. Achondroplasia is the cleanest case: James Crow reported that 153 of 154 analysed cases came from the same codon, and 150 of them from the same guanine-to-adenine transition at a single nucleotide with the highest known mutation rate.²⁴ Nobody argues that all people with achondroplasia share a recent common ancestor. The same thing happens in broken genes.  Humans and guinea pigs lost the vitamin C gene separately, yet where the human remnant differs from the rat gene, the guinea pig carries the identical change at 47 of 129 positions, a match its discoverers put at a chance probability of 1.84 × 10^-12 and explained as "many hot spots."²⁵²⁵ About 13% of the human genome folds into non-B structures, and substitution frequencies are systematically elevated there.²⁶ Deletions between short direct repeats recur: Cunningham and colleagues propagated six bacteriophage lineages and got nine independent deletions, seven of them with identical breakpoints, and warned "against giving undue phylogenetic weight to insertion-deletion events."²⁷ This matters most for single shared point mutations and small indels, and least for insertion positions of long elements, where the state space is genuinely enormous.

10. The best-known case study, vitamin C, is more complicated than either side usually says. Anthropoid primates and guinea pigs both lost the GLO gene, and both lost whole exons — but different exons, so the losses are independent.²⁸ That part fits common descent within primates and refutes any simple "same damage everywhere" story. Then the same review reports that in bats, GLO has been reactivated: "Given the currently accepted phylogeny of bats, these results therefore conclusively demonstrate that inactive genes can be reactivated during evolution."²⁸ And in songbirds, on the accepted phylogeny, vitamin C synthesis has been reacquired three or four times.²⁸ Loss and regain of this pathway is not a clean nested hierarchy across vertebrates. Within primates it is.

11. Gene resurrection is documented in humans. Zhang and colleagues identified 76 unitary pseudogenes in the human lineage, 11 of them still polymorphic, and found that two "are in fact pseudogenes in non-human primates, suggesting that they represent cases of a gene being resurrected in the human lineage."²⁹ Broken genes do not always stay broken.

12. The honest score. The shared-marker argument survives, and it is the strongest thing on the table. Every documented failure mode operating at its measured rate would corrupt a few markers in a hundred; the pattern to be explained involves thousands of congruent markers. What the failures do overturn is the rhetoric — the claim that these are perfect characters, that no statistics are needed, that absence is as informative as presence, and that a shared marker settles a branching order. The creationist answer, stated by its authors, is not mainly about hot spots.  Yingguang Liu and Charles Soper, the fullest creationist treatment, propose that "orthologous ERVs were created to occupy similar genomic loci in separate species by a single designer to carry out similar physiological functions," and offer directed integration only as something that "may also partially explain" the shared positions.³⁰³⁰  The partial part has been measured for HIV, and at that strength it covers about 1% of the shared ERV positions.  The created-in-place part escapes the rate arithmetic altogether, because it needs no insertion event, but it has not yet been turned into a prediction:  no count of shared loci, no list of which loci should be shared, and no reason the shared set should nest species inside species the way inheritance does.  Todd Wood, a creationist biologist, put the problem plainly in 2006:  "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."³¹

13. Where the argument is genuinely weakest is exactly where the dispute is loudest. These markers work well at moderate depths — SINE insertions placed hippos as the closest living relatives of whales with no homoplasy across 20 events.¹ They fail at rapid radiations, which include the origin of placental orders, the origin of Laurasiatheria, and the origin of modern bird orders. Those are precisely the deep branchings a creation model most disputes, and they are precisely where the marker method returns a network instead of a tree.

The verdict. Shared genetic mistakes are real evidence, they are not fabricated, and the popular creationist responses to them are weaker than the popular creationist literature suggests. The technical literature has already corrected the overreach on the other side — these are excellent characters, not perfect ones, and they cannot resolve short branches. Anyone arguing this topic should quote Han 2011 and Cantrell 2001 and Ray 2006 and Wang 2007, because the last two give the rates, and the rates decide the question.  The creationist side's best data point is not about insertions at all:  independent losses of the same gene do converge on identical point changes, as Inai 2003 showed for vitamin C, so a single shared point lesion proves little.  Shared insertion positions are untouched by that result, and the created-in-place proposal that would answer them has not yet produced a testable count.

What the argument actually claims

The argument is old, and its clearest online statement is Douglas Theobald's, so it is worth quoting rather than paraphrasing. Theobald sets it out in three linked predictions covering transposons, redundant pseudogenes, and endogenous retroviruses.

On transposons:

"As explained above, finding the same transposon in the same chromosomal location in two different organisms is strong direct evidence of common ancestry, since they insert fairly randomly and generally cannot be transmitted except by inheritance. In addition, once a common ancestor has been postulated that contains a certain transposition, all the descendants of this common ancestor should also contain the same transposition."⁸

On endogenous retroviruses:

"Again, this process is rare and fairly random, so finding retrogenes in identical chromosomal positions of two different species indicates common ancestry."⁸

And the stated falsification:

"It would make no sense, macroevolutionarily, if certain other mammals (e.g. dogs, cows, platypi, etc.), had these same retrogenes in the exact same chromosomal locations."⁸

That is a fair and testable statement of the case. It rests on four premises, and each is worth checking separately:

1. Insertion position is effectively random, so the same position twice is vanishingly unlikely.
2. Insertions are irreversible, so presence is inherited and absence means the ancestor lacked it.
3. The elements are non-functional, so shared position cannot be explained by shared requirement.
4. The pattern nests, so it reflects a branching history rather than a scatter.

Premises 1, 2 and 3 are all approximations that fail at measurable rates. Premise 4 is the one that carries the argument, and it is the one that holds.

What creationists actually propose

The creationist replies attack different premises, so each is worth stating in its authors' own words before it is tested.

1. Created in place. Yingguang Liu and Charles Soper wrote the fullest creationist treatment of endogenous retroviruses, in the Answers Research Journal in 2009. Their answer to shared position denies premise 3 and sidesteps premise 1: "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."³⁰ On this view there was no insertion event to explain.
2. The virus came from the genome, not the genome from the virus. The same paper reverses the usual direction of travel: "we speculate exogenous retroviruses were derived from endogenous retroviruses (exogenization, as opposed to endogenization)."³⁰ If that is right, a sequence that looks retroviral is not by itself the scar of an infection. Peer Terborg's "VIGE-first hypothesis" in the Journal of Creation argues the same direction: it "holds that RNA viruses originate in the genomes from genetic elements commonly known as endogenous retroviruses."³²
3. Directed integration. Liu and Soper add that targeted insertion "may also partially explain the positional 'conservation' of orthologous ERVs in human and primate genomes."³⁰ This is the hot-spot reply, and its authors offer it as partial.
4. Independent loss. For broken genes, Jeffrey Tomkins argues that the vitamin C gene's exon losses "in human, chimpanzee, and gorilla occurred independently in each taxon."³³ That reply denies that the shared damage is one event. It is tested in detail in Shared Retroviruses and Pseudogenes.

Not every creationist thinks these replies are enough.  Todd Wood, a creationist biologist then at the Center for Origins Research at Bryan College, reviewed the chimpanzee genome in 2006 and wrote:  "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."³¹  Liu and Soper concede a specific case:  the comparative genomics around the syncytin-1 locus is "especially challenging to creationists."³⁰

The markers, and how good each one is

Not all of these are equally strong. Ranking them honestly matters, because critics attack the weakest and advocates cite the strongest.

SINE and LINE insertions (including Alu). The strongest class. Alu elements are about 300 bases long, primate-specific, and make up roughly 10% of the human genome.³⁴ An insertion can land at a very large number of positions in two orientations, and different Alu subfamilies are distinguishable by sequence — so when two species share an insertion, it is usually possible to check whether it is the same insertion or two different ones. That check is what makes the class strong, and it is why Han and colleagues emphasise "the importance of sequencing TE insertions" rather than merely scoring presence and absence.²

Endogenous retroviruses. Similar logic, and rhetorically the most striking, because the shared feature is the scar of an ancient infection. Weaker than SINEs in one specific respect: full-length ERVs collapse into solo LTRs by recombination, so ERV absence data are unreliable in a way SINE absence data are less so.¹⁰ Weaker in a second respect too: retroviral integration is measurably non-random, with different families showing different target-site preferences.¹²

Unitary pseudogenes — a gene broken with no working copy left, like GLO for vitamin C. Rhetorically the most intuitive, because the loss is visible in the organism. Analytically weaker than it looks, because the same gene can be lost independently under similar selection, and because loss can be reversed.²⁹²⁸

Processed and duplicated pseudogenes. Weaker again, because they are subject to gene conversion from the parent gene, which erases the very differences a phylogeny needs.²¹

Numts. Weakest of all, and actively hazardous. These are mitochondrial sequences that moved into the nucleus. They look like mtDNA, amplify like mtDNA, and are not mtDNA.¹⁹²⁰

Microinversions, indels, gene order, miRNA families. Various. Microinversions have documented hot spots.³ miRNA presence/absence data were promoted hard as an ideal marker and then found to be badly behaved: "highly heterogeneous rates of microRNA gain and loss, pervasive secondary loss, and sampling error collectively render microRNA-based inference of phylogeny difficult."³⁵

The umbrella term for all of these is rare genomic changes, named by Rokas and Holland in 2000. Their appeal is that they are supposed to accumulate slowly and have a huge state space, so independent identical events should be almost impossible.²

Failure mode one: the same marker, twice

Hot spots

The cleanest demonstration is Cantrell and colleagues' 2001 study of an ancient retrovirus-like element called mys-9 in deer mice. They looked at SINE insertions inside it and found something they did not expect:

"We have identified two hot spots for SINE insertion within mys-9 and at each hot spot have found that two independent SINE insertions have occurred at identical sites."¹¹

They could tell the insertions apart because the second one used a nick site a few bases upstream and belonged to a distinguishable subfamily. Without sequencing, both would have been scored as one shared insertion. Their conclusion is the standard citation for this whole problem:

"These results have major repercussions for phylogenetic analyses based on SINE insertions, indicating the need for caution when one concludes that the existence of a SINE at a specific locus in multiple individuals is indicative of common ancestry. Although independent insertions at the same locus may be rare, SINE insertions are not homoplasy-free phylogenetic markers."¹¹

Han and colleagues found the same phenomenon in birds. Out of only 66 transposable-element insertions, they identified six potential hot spots, four of them cases where two independent insertions had landed at the same site in the same taxon.² They also found two insertions that were homoplastic in the strict sense — present in a pattern that cannot be explained by a single insertion on the accepted tree — and they explicitly ruled out lineage sorting as the explanation, because the branches involved are long and well supported and because each conflicting insertion conflicts with other insertions.²

The same team ran the exercise for microinversions and found two hot-spot loci, again with hemiplasy ruled out.³

How rare is rare?

This is where the argument is usually won or lost.  Liu and Soper offer directed integration as a partial explanation of shared positions, but give no rate,³⁰ and no creationist paper found for this article puts a number on it.  The mainstream literature does.

Ray, Xing, Salem and Batzer set out in 2006 to answer exactly this question. They catalogued four confounders — lineage sorting, near-parallel insertion, precise-parallel insertion, and precise excision — across the published primate Alu datasets, more than 11,000 loci in total. They found "very low levels of each potential confounding factor."⁴ Their headline number:

"The results indicate that of the 11,000+ loci examined, precise parallel insertions are exceedingly rare—occurring at only about 0.0005 events/insertion of the examined loci."⁴

They also re-audited a large mammalian study that had claimed 23 homoplastic Alu insertions among about 2,500 loci, and found the true number was 13.⁴ Their title — "SINEs of a Nearly Perfect Character" — is a deliberate downgrade of Hillis's 1999 "SINEs of the perfect character." The word that changed is "nearly," and it is the honest word.

Han and colleagues report the same thing in a different currency. The retention index measures how much of a character set is free of homoplasy on a given tree; 1.0 is perfect.

"the retention index (RI) of TE insertions on the Hackett et al. (2008) tree is 0.97, much greater than that of Hackett et al. (2008) sequence data on the same tree (RI_intron = 0.52, RI_{coding exons} = 0.54, RI_UTR = 0.58)."²

Read that number carefully. It is the strongest thing in this article for the other side. Insertions are roughly twice as clean as DNA sequence data on the same tree. A creationist who cites Han 2011 for "even transposons show homoplasy" and stops there has cited a paper whose main quantitative result is that transposons are the best characters available.

Non-random insertion

The premise that insertion is effectively random is an approximation. Mitchell and colleagues compared where three retroviruses integrate in human cells and found sharply different, family-specific preferences — murine leukaemia virus near transcription start sites, HIV inside active transcription units, avian sarcoma-leukosis virus more nearly uniform.¹² Different families, different targeting rules.

How strong the targeting is has been measured directly.  Wang, Bushman and colleagues infected a human T-cell line with HIV in fifty independent infections and mapped 40,569 integration sites.  They counted only pairs of proviruses lying in opposite orientations, so that each pair was certainly two separate events:

"We found 41 sites that hosted two independent integration events at exactly the same base pair in the human genome."⁶

If HIV chose uniformly among the 3.1 billion bases of the genome, a sample that size would be expected to hold about one-eighth of one such pair, which is to say probably none.  Forty-one opposite-orientation pairs, and so about 82 in all, mean the virus behaves as if it were choosing among only 10 to 20 million sites.⁷  That is the strongest single measurement on the creationist side of this question, and it kills the popular "one chance in three billion" version of the argument.

It does not rescue the shared positions.  Humans carry roughly 98,000 endogenous retrovirus elements and fragments.³⁶  The 2005 chimpanzee genome paper found only 361 ERV insertions, in the aligned parts of the two genomes, present in one species and not the other.³⁷  Two genomes each filled independently with 98,000 insertions from a pool of 10 to 20 million sites would share about 480 to 960 positions by chance, roughly 1% of what is observed.  Shrinking the pool until it explains the rest leaves no room for the hundreds of insertions found in one species and not the other.  The full calculation is in Shared Retroviruses and Pseudogenes.  This test uses HIV, one virus in one cell line;  germline insertions by other families may be more or less targeted, and no comparable collision count exists for Alu or L1.

At the sequence level, roughly 13% of the human genome folds into non-canonical DNA structures, and substitution frequencies are systematically elevated there, with patterns specific to each structure type.²⁶ Mononucleotide runs are deletion hot spots. James Shapiro's summary of the general point is that "It is difficult (if not impossible) to find a genome change operator that is truly random in its action within the DNA of the cell where it works."³⁸

None of this gets a critic to the finish line. Preference is not determinism. A hot spot that raises the probability of a repeat insertion from one in a billion to one in two thousand is still one in two thousand — which is exactly the number Ray and colleagues measured. What it does do is retire the claim that homoplasy is "for all practical purposes, zero" and that statistics can be dispensed with.

Failure mode two: the marker vanishes

Theobald's version of the argument allows for deletion but says the deletions are messy and leave a trace:

"A possible exception is if this transposition were removed due to a rare deletion event; however, deletions are never clean and usually part of the transposon sequence remains."⁸

This is not right, and the paper that shows it is not right was published in 2005. When a retroelement inserts, it duplicates a short stretch of target-site DNA on both sides of itself. Those two identical repeats can recombine with each other and excise the element between them:

"we show that retroelements can, in rare cases, be precisely deleted from primate genomes, most likely via recombination between 10- to 20-bp target site duplications (TSDs) flanking the retroelement. The deleted loci are indistinguishable from pre-integration sites, effectively reversing the insertion."⁹

"Indistinguishable from pre-integration sites" is the phrase that matters. There is no trace. And the rate is not negligible:

"Through human–chimpanzee–Rhesus monkey genomic comparisons, we estimate that 0.5%–1% of apparent retroelement 'insertions' distinguishing humans and chimpanzees actually represent deletions."⁹

Endogenous retroviruses have a much worse version of the same problem, because they carry long terminal repeats at both ends and those recombine readily. Belshaw and colleagues measured it: "The fate of most human endogenous retroviruses (HERVs) has been to undergo recombinational deletion," leaving a solo LTR, and the rate among recent insertions is almost 200 times the rate among insertions older than the human–chimpanzee split.¹⁰

The practical consequence is asymmetry. Presence of a shared marker at a shared position is strong evidence. Absence of a marker is weak evidence, because it can mean the marker was never there or that it was cleanly removed. Ray and colleagues concede the point and argue it is manageable, noting that precise excision requires the flanking direct repeats to still be intact, which becomes less likely as they accumulate mutations.⁴ That is a good argument, and it means precise excision is mainly a problem for recent, closely related comparisons — which is where most of the popular examples live.

Failure mode three: the marker is real and still gives the wrong tree

This is the largest effect in the literature by a wide margin, and it is the one most often misquoted.

Hemiplasy

Populations are not individuals. An ancestral population can carry an insertion as a polymorphism — some individuals have it, some do not — and that polymorphism can survive through more than one speciation event before it fixes. When it finally fixes, it can fix in a pattern that does not match the species tree at all, without anything having happened twice.

Avise and Robinson named this hemiplasy in 2008 precisely to stop it being confused with homoplasy.¹⁶ Guerrero and Hahn quantified the comparison ten years later:

"We find that hemiplasy is as likely—or more likely—than homoplasy for a wide range of conditions, even when levels of discordance are low."¹⁷

Hillis had already flagged the underlying issue in 1999, and his numbers are worth having:

"Even if the speciation events are separated by N generations, where N is the effective population size, the probability of incongruence between a single locus and the true phylogeny may still be as high as 0.3 … Thus, even in the absence of any homoplasy, it is highly inadvisable to dispense with statistical analysis altogether."¹

What this looks like in practice

The root of the placental mammal tree. Churakov and colleagues extracted about two million retroposon-containing loci from representatives of the major placental lineages, looking for the first split. They found:

"The Exafroplacentalia hypothesis (Afrotheria as the sister group to all remaining placentals) is significantly supported by five retroposon insertions, the Epitheria hypothesis (Xenarthra as the sister group to all remaining placentals) by nine insertion patterns, and the Atlantogenata hypothesis … by eight insertion patterns. These findings provide significant support for a 'soft' polytomy of the major mammalian clades."¹³

Three mutually exclusive answers, each with marker support, from the character class advertised as unambiguous.

Laurasiatheria. Doronina and colleagues screened 162,000 candidate loci down to 102 "virtually homoplasy-free" markers to settle the order of divergence of bats, horses, cattle and carnivores. The screen "revealed 14 markers supporting Chiroptera as the most basal order, eight for Perissodactyla, nine for Cetartiodactyla, and 12 supporting Carnivora as the earliest diverged order."¹⁴ Their summary:

"All possible evolutionary scenarios received considerable retrotransposon support, leaving us with a network of affiliations."¹⁴

Modern birds. Suh, Smeds and Ellegren scored 2,118 retrotransposon presence/absence markers across the neoavian radiation. 745 were incongruent — 35% of loci on the autosomes and 34% on the Z chromosome.¹⁵ Their conclusion:

"given the extreme level of up to 100% ILS per branch in super-radiations, particularly rapid speciation events may neither resemble a fully bifurcating tree nor are they resolvable as such. As a consequence, their complex demographic history is more accurately represented as local networks within a species tree."¹⁵

A note on how not to use these numbers. All three papers assume common descent throughout, and all three attribute the conflict to incomplete lineage sorting rather than to false markers. Suh and colleagues state directly that their "2,118 RE markers contain negligible homoplasy, and conflicts are instead due to ILS-derived hemiplasy."¹⁵ Quoting the 35% figure as though it showed the markers are unreliable indicators of ancestry misrepresents the source. What it shows is that they are unreliable indicators of branching order at short internodes — a real and important limitation, and a different one.

The same effect in primates

The two best-known conflicting Alu loci in hominids are worth naming, because they are the hard cases inside the strongest dataset.

Salem and colleagues sequenced 133 Alu loci to settle the human–chimpanzee–gorilla trichotomy and got a clean result. One locus did not fit: "We identified one Alu element (Ye5AH137) that was present in the genomes of human and gorilla but was absent from bonobo and common chimpanzee genomes."⁵ Their overall assessment: "Only a single insertion from the entire data set showed any potential evidence of insertion homoplasy; however, the distribution of this element appears to be the result of lineage sorting of an ancestral polymorphism."⁵

Hedges and colleagues found the mirror image: "one element, designated CS12, was determined to be exclusive to gorilla and chimpanzee genomes and not present in human, implying a relationship contrary to the orthodox phylogeny."³⁴ They too attributed it to lineage sorting.

Both attributions were reasonable, and in 2003 and 2004 neither was tested. The honest statement is that in the best hominid Alu datasets, conflicting loci appear at roughly one in a hundred, and the standard explanation for them is lineage sorting rather than error. A critic is entitled to ask how that explanation would be distinguished from independent insertion.  In Salem's case the answer was molecular-clock dating of the element, which is itself model-dependent.

A better test came later.  If a conflicting insertion was inherited as an old polymorphism, the ordinary DNA around it was inherited with it, so its single-letter differences should show the same conflicting tree.  If instead the insertion happened twice, the surrounding DNA should follow the normal species tree.  Hormozdiari and colleagues ran this test on whole genomes of 72 great apes:

"Our analysis of these 72 samples identified 187 MEI events (43 Alu and 144 L1 events) that were inconsistent with the great ape phylogeny (e.g., shared between human and gorilla but not chimpanzee). A total of 84% (157/187) of these loci were also flanked by SNPs of a similar phylogeny, confirming that most arose as a result of ILS"¹⁸

Doronina and colleagues found the same in Laurasiatheria:  for nine of ten groupings, the DNA flanking the insertions gave "the same order relationships as given by the presence/absence information of the retrotransposons."¹⁴  So lineage sorting is a tested explanation, not a label applied after the fact.  The 30 great ape loci that failed the test, about 16% of the conflicts, are a real residue.  The same paper assumes a false-genotyping rate of 10% and warns that mapping short reads across species makes errors worse,¹⁸ so some of the 30 are probably miscalls, and the rest are the candidates for true homoplasy, parallel insertion or clean deletion, in the largest great ape dataset.

Failure mode four: pseudogenes that are not what they seem

Numts

A numt is a chunk of mitochondrial DNA that has been copied into the nuclear genome, where it stops evolving at mitochondrial speed and starts evolving at nuclear speed. It is a pseudogene, and it is a very good imitation of the real thing.

Thalmann and colleagues tested how bad the problem is in humans and apes by generating guaranteed-authentic mtDNA with long-range PCR and comparing it against sequences obtained the ordinary way:

"Results revealed that gorillas are notable for having such a variety of numt sequences bearing high similarity to authentic mtDNA that any analysis of mtDNA using standard approaches is rendered impossible."¹⁹

Their broader warning is that "some taxa may not be amenable to analysis of mtDNA variation at all."

Song, Buhay, Whiting and Crandall quantified the downstream damage in DNA barcoding, in grasshoppers and crayfish. Numts drove the estimated number of species above the truth, and "even such rigorous quality control measures fail to identify certain numts."²⁰ They also note that "the distribution of numts is lineage-specific and the presence of numts cannot be known a priori."

Ribosomal pseudogenes in plants

The single most-used marker in plant systematics is the internal transcribed spacer of the ribosomal DNA array. Álvarez and Wendel surveyed 244 papers over five years: 66% used ITS data, and 34% of published phylogenetic hypotheses were based on ITS alone.²¹ Their assessment of that dependence:

"Among the most prevalent complications for phylogenetic inference is the existence in many plant genomes of extensive sequence variation, arising from ancient or recent array duplication events, genomic harboring of pseudogenes in various states of decay, and/or incomplete intra- or inter-array homogenization."²¹

"Homoplasy is shown to be higher in ITS than in other DNA sequence data sets … Despite the near-universal usage of ITS sequence data in plant phylogenetic studies, its complex and unpredictable evolutionary behavior reduce its utility for phylogenetic analysis."²¹

This is a pseudogene problem in a major branch of systematics, stated by mainstream botanists, and it is much less well known among apologists on either side than it should be.

Failure mode five: the "mistake" may not be a mistake

The plagiarism analogy has a hidden premise. A shared misspelling is evidence of copying because misspellings are useless. If the shared feature turns out to be doing a job, the argument weakens — not to zero, but noticeably, because now shared position could reflect shared requirement.

The evidence here has moved a long way since 1980.

Pseudogenes. Pink and colleagues summarise: "Far from being silent relics, many pseudogenes are transcribed into RNA, some exhibiting a tissue-specific pattern of activation. Pseudogene transcripts can be processed into short interfering RNAs that regulate coding genes through the RNAi pathway. In another remarkable discovery, it has been shown that pseudogenes are capable of regulating tumor suppressors and oncogenes by acting as microRNA decoys."²²

Endogenous retroviruses. The strongest single result is Chuong, Elde and Feschotte's. They found ERV-derived enhancers scattered through the interferon-response network in several mammals, then deleted a subset with CRISPR: the deletions "impaired expression of adjacent IFN-induced genes and revealed their involvement in the regulation of essential immune functions, including activation of the AIM2 inflammasome."²³

How much does this actually help the design side? Less than it is usually made to. Three problems:

1. A functional element at an identical position in two species is still shared, and still needs an explanation. If anything, function makes the shared position more surprising under independent origin, not less, because it means selection is now maintaining it — but selection cannot explain how it got to the same nucleotide twice.
2. Chuong and colleagues describe these as "lineage-specific ERVs" dispersed "independently in diverse mammalian genomes." The functional story they tell is one of repeated capture of viral sequence by different lineages. That is an evolutionary story, not a design one. The creationist reply is that the arrow runs the other way: Liu and Soper propose that the genome's elements came first and the free viruses were "derived from endogenous retroviruses."³⁰ That reply has a test case. Where both sides agree a retroviral gene was taken up separately in different mammals, as with the syncytins that build the placenta, the copies sit at unrelated places in each group's genome: the primate and mouse syncytins "are not syntenic in the two species clades—indicating that they are the result of independent gene captures."³⁹ Terborg, arguing the creationist side, reports the same fact, that the syncytin genes of different mammal orders "map to different genomic locations."³² Independent origin leaves its mark in position. The human and chimpanzee ERVs carry no such mark; they sit at the same places.
3. Function is documented for a minority of these sequences. Nobody has shown a job for the great majority of the million Alu copies. The honest position is that the blanket term "junk" was wrong and the blanket term "functional" is also wrong, and the fraction is still being argued over — see the ENCODE dispute, where Graur and colleagues' reply to ENCODE's 80%-functional claim is as sharp a piece of scientific polemic as exists.⁴⁰⁴¹⁴²

Failure mode six: mutation is not a fair coin

The plagiarism analogy also assumes that a given "mistake" is unlikely to be made twice. For the position of a long insertion that is roughly true. For a single base change it can be badly false.

Achondroplasia. James Crow's summary is the standard reference:

"The most striking is achondroplasia, in which 153 of 154 analysed cases are due to a glycine to arginine substitution at codon 1,138 … Of the 153 mutations, 150 were guanine to adenine transitions and three were guanine to cytosine transversions of the same nucleotide. This means that all the cases of achondroplasia are due to changes in one nucleotide — a nucleotide with the highest known mutation rate (about 10⁻⁵ per generation)."²⁴

If you found two unrelated people sharing that exact substitution and reasoned as the shared-error argument reasons, you would conclude they shared a recent ancestor, and you would be wrong.

Deletions between direct repeats. Cunningham and colleagues serially propagated six lineages of bacteriophage T7. Every lineage evolved a deletion fusing the same two genes. Nine independent deletions occurred, and seven had identical breakpoints, because of a 13-base-pair repeat at the endpoints. Their conclusion:

"This study provides a compelling reason to avoid the assumption that parallel evolution of deletions is rare until the mechanisms underlying insertions and deletions are better understood."²⁷

Their warning is explicit: "caution against giving undue phylogenetic weight to insertion-deletion events."²⁷

Genome-wide. About 13% of the human genome adopts non-B structures — G-quadruplexes, Z-DNA, mirror and inverted repeats — and substitution frequencies are elevated at those loci in patterns specific to each structure type.²⁶

Where this bites and where it does not. It bites hard on arguments built from a single shared point mutation or a single small indel in a mononucleotide run. It bites much less on arguments built from the position of a 300-base insertion.  Even HIV, a strongly targeted virus, chooses among some 10 to 20 million effective sites,⁷ while a point substitution has only three possible outcomes, so targeting shrinks the insertion state space without collapsing it. The strongest form of the shared-error argument uses insertion positions, and this objection barely touches it. The weakest form uses shared point mutations in a pseudogene, and this objection is decisive against treating any one of them as proof.

Case studies

Vitamin C

The most-cited example. Most mammals make their own vitamin C using an enzyme encoded by GULO (also written GLO). Anthropoid primates, guinea pigs, some bats and some songbirds cannot.

What supports common descent. The primate pseudogene is broken in a shared way, and the pattern of accumulated mutations across humans, chimpanzees, orangutans and macaques is nested and matches the primate tree.⁴³

What complicates it. Drouin, Godin and Pagé's review lays out the details. Anthropoid primates have lost exons 1, 5 and part of 6 — but guinea pigs have lost a different set.²⁸ So the two famous losses are independent, which is what common descent predicts and which no simple "same damage" story explains.

The two independent losses also show how often separate damage lands in the same place.  Inai, Ohta and Nishikimi compared the surviving human and guinea pig exons against the working rat gene.  Of the 129 positions where the human copy has changed, the guinea pig carries the identical change at 47:

"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."²⁵

Under uniform mutation the chance of that much overlap is 1.84 × 10^-12.²⁵  This is the strongest published evidence that shared point changes in a broken gene can arise independently, and it comes from the laboratory that first characterised the human GULO pseudogene.  It does not explain the shared exon deletions within primates, which are larger events, and its two species are a rodent and a primate, so the overlap cannot be put down to shared inheritance.

Then the same review reports two things that do not fit a clean hierarchy. In bats, GLO cDNAs from Rousettus leschenaultia and Hipposideros armiger produce functional protein:

"Given the currently accepted phylogeny of bats, these results therefore conclusively demonstrate that inactive genes can be reactivated during evolution."²⁸

And in songbirds:

"If one assumes that the inability to synthesize vitamin C is ancestral in the Passeriformes, then the ability of synthesizing vitamin C has been reacquired four times."²⁸

Honest assessment. Within primates, the vitamin C evidence is good and the nesting is real. Across vertebrates, loss and regain of this pathway is messy enough that it cannot be presented as a clean demonstration of hierarchy. Both statements are true at once and both are usually left out by whichever side is arguing.

A separate caution belongs here. The deletion at position 97 of exon 10 of the primate pseudogene is sometimes presented as decisive on its own. Small deletions can recur at the same place in independent lineages,²⁷ and Inai's data show how often independent changes coincide in this very gene.  Whether this particular deletion sits in an especially deletion-prone sequence has not been checked for this article. It should not be argued as a stand-alone proof in either direction; the strength of the primate GULO case is the whole pattern, not any one lesion.

Urate oxidase

Oda and colleagues examined the broken urate oxidase gene across the apes and found a stop codon at codon 107 present in gorilla, orangutan and gibbon — but not in human or chimpanzee. On any accepted ape phylogeny, that requires the same stop codon to have arisen more than once.⁴⁴ The paper also notes that four of the five CGA codons in the gene converted to the TGA stop,⁴⁴ which is exactly what would be expected if methylated CpG hypermutability, not shared inheritance, is driving those particular changes.

This and Inai's GULO overlap are the best documented cases of shared damage that is not shared inheritance, and both belong in any honest treatment. Each is a point change in a mutational context known to be hot.  Oda's figure legend adds that in the orangutan, codon 107 is still polymorphic, CGA in some animals and TGA in others,⁴⁴ which fits a recent mutation in that lineage rather than one inherited from an ancestor shared with the gorilla.

Egg yolk genes

Brawand, Wahli and Kaessmann traced the three vitellogenin genes — egg yolk proteins — across mammals and found them pseudogenised as milk and placentation took over. Their dating analysis "revealed multiple independent inactivation events (e.g., the independent inactivation of VIT1 in therians and monotremes)."⁴⁵ Independent inactivation of the same gene in separate lineages is exactly what a selection-driven loss predicts, and it is a warning against reading every shared pseudogene as a shared event.

Tooth enamel

Meredith and colleagues showed that the enamelin gene ENAM is pseudogenised in placental mammals that lack enamel, and that the pattern of decay tracks the fossil record of enamel loss.⁴⁶ This is one of the strongest correlations in the field and deserves to be stated as such. It is also, for the same reason, an example of a loss that is driven by loss of function — so the same gene going the same way in unrelated toothless lineages is expected, and shared brokenness cannot by itself distinguish shared ancestry from shared relaxation of selection. The nesting of which particular lesions are shared is what does the work.

Human and chimpanzee mobile elements

This is the hard case for a young-earth position and it should not be soft-pedalled. Alu elements make up about 10% of the human genome, they are present in the great majority of the same positions in chimpanzees, and the differences between the two species are themselves informative — Hedges and colleagues found roughly a twofold excess of recent Alu insertions in the human lineage over the chimpanzee lineage, with different subfamilies active in each.³⁴

There is no published creationist model that accounts for the shared positions quantitatively.  The fullest proposal, Liu and Soper's, says orthologous ERVs were "created to occupy similar genomic loci in separate species,"³⁰ and Liu's later experimental paper with Tam Nguyen restates it without a number:  shared position is "more likely due to functional necessity rather than common descent."⁴⁷ "Hot spots" is not a model until somebody produces the hot-spot map and shows it predicts the shared positions;  the one direct measurement, for HIV, accounts for about 1% of them. "Common design" is not a model until somebody says what function the shared positions serve and which species should share which ones. Both are available as research programmes; neither has been carried out.

Objections and responses

"Insertion is not random, so shared positions are expected." Partly true, and the effect is measured. Insertion preference is real and hot spots exist.¹¹¹² But the measured rate of independent insertion at the identical site in primates is about 1 in 2,000,⁴ and that will not generate a nested hierarchy of thousands of shared positions. The objection needs a rate near 1, and the data give 0.0005.  At the strength of targeting measured directly for HIV, preference would explain about 1% of the shared ERV positions between humans and chimpanzees.⁶⁷  Liu and Soper, who raise directed integration on the creationist side, offer it only as a partial explanation.³⁰

"These sequences turn out to be functional, so they are shared design, not shared error." The function evidence is real and growing.²²²³ But function does not explain position by itself.  The creationist version, from Liu and Soper, is that orthologous ERVs were created at matching loci "to carry out similar physiological functions," and that the free viruses came from them rather than the reverse.³⁰³⁰  If so, shared function at a shared place is what design predicts, and the retroviral look of the sequence proves nothing about its origin.  The weakness is that the proposal makes no count:  it does not say which loci should be shared between which species, or why the shared set nests the way inheritance does.  And where both sides agree retroviral genes were captured more than once, as with the placental syncytins,³² the copies sit at different loci in each group.³⁹  Liu and Soper themselves call the syncytin-1 locus "especially challenging to creationists."³⁰

"Markers give contradictory trees, so they are worthless." This misreads the sources. The contradictions are concentrated at rapid radiations, are attributed by the authors to lineage sorting rather than false markers, and appear in papers that state their markers "contain negligible homoplasy."¹⁵ The correct conclusion is that markers do not resolve short internodes, not that markers are unreliable evidence of inheritance.

"Pseudogenes have been shown to give wrong trees." True for specific classes — numts and ribosomal ITS in particular.²¹¹⁹ Not true as a general statement, and the failures are recognised and controlled for by the people who use the method.

"The argument assumes what it sets out to prove, because 'pseudogene' already assumes an evolutionary origin." There is something to this as a matter of naming, but not as a matter of evidence. Whether or not a sequence is called a pseudogene, the fact requiring explanation is that two species have the same sequence at the same chromosomal position with the same internal lesions. That fact is observational.

"Shared errors could come from a shared mutational mechanism operating in both lineages after creation." This is the strongest available reply, and it is a research programme rather than an answer. It would predict that shared positions cluster in identifiable structural contexts — non-B DNA, direct repeats, particular chromatin states — at rates far above background. Guiblet and colleagues' non-B DNA maps make this testable.²⁶ Nobody has run the test. Until someone does, it is a promissory note.

"Reverse the objection: if these are perfect markers, why do evolutionary biologists disagree about the placental root?" Fair, and this is the strongest point on the sceptical side. The answer given in the literature is lineage sorting, which is a real population-genetic process with a worked-out mathematical treatment.⁴⁸ It is not merely invoked where the data conflict:  it predicts that the DNA flanking each conflicting insertion will share its conflicting tree, and that prediction held for 157 of 187 conflicting great ape insertions and for nine of ten groupings in Laurasiatheria.¹⁸¹⁴  Doronina's group also found signals of ancient hybridization between lineages, not only sorting,¹⁴ and Hallström and Janke's paper on the same problem is titled "Mammalian Evolution May not Be Strictly Bifurcating."⁴⁹ The tree, at these depths, is being replaced by a network in the primary literature.

What would change the verdict

Three things, in order of how decisive they would be.

1. A demonstrated hot-spot mechanism with high specificity. If someone showed that a defined class of genomic sites accepts insertions at rates thousands of times background, and that shared human–chimpanzee Alu positions fall preferentially in that class, the central inference would be in serious trouble. The one direct measurement, for HIV, puts targeting at a few hundred times background, enough to explain about 1% of shared ERV positions.⁶⁷ No comparable collision count exists for Alu or L1, and the data to make one exist.¹²²⁶
2. Systematic function for shared positions. If shared insertion positions turned out to be predictable from function — that is, if the position of an Alu were derivable from what the surrounding region does — the design reading would become live. The CRISPR work on ERV enhancers is the closest anyone has come.²³
3. A quantitative model. Winston Ewert's dependency-graph work is the only design model fitted to real genomic data, and it is fitted to gene presence/absence, not to marker positions. Nobody has attempted the equivalent for insertion positions.

None of these has been done. Until one is, the honest description of this evidence is that it is strong, that its advocates have overstated its precision, and that its critics have understated its weight.

Related pages

• Phylogenetic Inference and Congruence — how trees are built, what support values mean, and whether designed systems produce nested hierarchies.
• literature-review — every source worked up for this article, with what each contributes and whether it is used.
• quotes — the verification ledger for every quotation on this page.

Footnotes